Untitled

On Shell Structure
This volume collects core papers by Richard K. Larson developing what has
since come to be known as the “VP Shell” or “Split VP” analysis of senten-
tial structure. The volume includes five previously published papers together
with two major unpublished works from the same period: “Light Predicate
Raising” (1989), which explores the interesting consequences of a leftward
raising analysis of “NP Shift” phenomena, and “The Projection of DP (and
DegP)” (1991), which extends the shell approach to the projection of nominal
and adjectival structure, showing how projection can be handled in a uniform
way. In addition to published, unpublished and limited-distribution work, the
volume includes extensive new introductory material. The general introduction
traces the conceptual roots of the VP Shell analysis and its problems in the face
of subsequent developments in theory, and offers an updated form compatible
with modern Minimalist syntactic analysis. The section introductions to the
material on datives, complex predicates and nominals show how the updated
form of shell theory applies in the empirical domains where it was originally
developed.
Richard K. Larson is Professor of Linguistics at Stony Brook University, US.

Routledge Leading Linguists
Edited by Carlos P. Otero, University of California, Los Angeles, USA
1 Partitions and Atoms of Clause 10 Theoretical Comparative Syntax

Structure Studies in Macroparameters
Subjects, Agreement, Case and Naoki Fukui
Clitics
Dominique Sportiche 11 A Unification of Morphology
and Syntax
2 The Syntax of Specifiers and Investigations into Romance and
Heads Albanian Dialects
Collected Essays of Hilda M. Rita Manzini and Leonardo
J. Koopman M. Savoia
Hilda J. Koopman
12 Aspects of the Syntax of
3 Configurations of Sentential Agreement
Complementation Cedric Boeckx
Perspectives from Romance
Languages 13 Structures and Strategies
Johan Rooryck Adriana Belletti
4 Essays in Syntactic Theory 14 Between Syntax and Semantics

Samuel David Epstein C.- T. James Huang
15 Regimes of Derivation in Syntax

5 Comparative Syntax and
and Morphology
Language Acquisition
Edwin Williams
Luigi Rizzi
16 Typological Studies
6 Minimalist Investigations in Word Order and Relative Clauses
Linguistic Theory Guglielmo Cinque
Howard Lasnik
17 Case, Argument Structure,
7 Derivations and Word Order
Exploring the Dynamics of Syntax Shigeru Miyagawa
Juan Uriagereka
18 The Equilibrium of Human
8 Towards an Elegant Syntax Syntax
Michael Brody Symmetries in the Brain
Andrea Moro
9 Generative Grammar
Theory and its History 19 On Shell Structure
Robert Freidin Richard K. Larson
On Shell Structure
Richard K. Larson
First published 2014
by Routledge
711 Third Avenue, New York, NY 10017
Simultaneously published in the UK
by Routledge
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN
Routledge is an imprint of the Taylor & Francis Group,
an informa business
© 2014 Richard K. Larson
The right of Richard Larson to be identified as author of this work
has been asserted by him in accordance with sections 77 and 78 of the
Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or
utilized in any form or by any electronic, mechanical, or other means, now
known or hereafter invented, including photocopying and recording, or in
any information storage or retrieval system, without permission in writing
from the publishers.
Trademark Notice: Product or corporate names may be trademarks or
registered trademarks, and are used only for identification and explanation
without intent to infringe.
Library of Congress Cataloging-in-Publication Data
Larson, Richard K.
On shell structure / By Richard K. Larson.
pages cm. — (Routledge leading linguists ; #19)
1. Grammar, Comparative and general—Syntax. 2. Structural linguistics.
3. Generative grammar. 4. Minimalist theory (Linguistics) I. Title.
P291.L333 2013
415—dc23
2013001621
ISBN: 978-0-415-16773-4 (hbk)
ISBN: 978-0-203-42920-4 (ebk)
Typeset in Sabon
by Apex CoVantage, LLC
Contents
Acknowledgmentsvii
General Introduction 1
PART I
Datives: Background 35
1 On the Double Object Construction 137
2 Double Objects Revisited: Reply to Jackendoff 191
3 Promise and the Theory of Control 230
PART II
Complex Predicates: Background 265
4 Light Predicate Raising 303
5 Some Issues in Verb Serialization 367
6 Sentence-Final Adverbs and “Scope” 387
PART III
Nominal Structure: Background 407
7 The Projection of DP (and DegP) 427
Index 481
This page intentionally left blank
Acknowledgments
When series editor Carlos Otero approached me years ago about collecting my
papers on the VP Shell structure, I had not thought about the issues discussed
in them for many years. My initial plan was to write a brief general introduc-
tion drawing out the main themes of the works and their contribution to sub-
sequent theory, leaving matters at that. What resulted instead was a complete
rethinking of the whole VP Shell project, including the basic nature of syntactic
projection. Carlos Otero encouraged me in embarking on the larger project
even though it dramatically slowed completion of this collection as I worked
through not only the ideas now found in the introductions but many others
not presented here, which form part of a larger manuscript in preparation. I
acknowledge Professor Otero’s encouragement and patience with enormous
gratitude. I thank a succession of editors at Routledge for their support and
patience as well.
Special thanks Noam Chomsky for inviting me to teach 24.958 Linguistic
Structure in 1988 at the Massachusetts Institute of Technology, where many of
these ideas got their first public airing, as well as the students and visitors who
attended the class. I thank Hiroko Yamakido for editorial work on several of
the papers in this volume. I thank Norbert Hornstein and Beth Levin for com-
ments on the dative materials, as well as audiences at the University of Mary-
land, CRISP (Brussels. Belgium), ICFL 5 (Ghangzhou, PRC), and Fuji Womens
University (Sapporo, Japan), where various parts of this work were presented.
General Introduction
The papers collected in this volume represent an approach to syntactic structure initially
developed during the late 1980s while the author was at the Department of Linguistics
and Philosophy at MIT. Five of the papers have previously appeared in print: “On the
Double Object Construction” (1988), “Double Objects Revisited: Reply to Jackendoff”
(1990), “Promise and the Theory of Control” (1991b), “Some Issues in Verb Serializa-
tion” (1991c), and “On Sentence-Final Adverbs and ‘Scope’” (2004). The remaining
two papers, “Light Predicate Raising” (1989) and “The Projection of DP (and DegP)”
(1991a), have not been previously published. The material assembled here was first pre-
sented as a connected set of topics in 24.958 Linguistic Structure taught by the author
at MIT in the fall of 1988.
The essays in this book are divided into three thematic groups: (i) papers on the dative
construction, (ii) papers that explore the general complex predicate view resulting from
shell structure, and (iii) a paper showing how the basic notions of shell structure might
be extended to nominals (DPs, or determiner phrases) and degree phrases (DegPs). All
of the topics discussed here have received considerable development in subsequent lit-
erature, and indeed the idea of VP shells itself has since been integrated into mainstream
generative grammar following Chomsky (1993). Accordingly, I provide both a general
introduction to the collection and introductions to the separate parts, in an attempt
to explain the historical development of my own proposals, the modern theoretical
context, and the ways in which the former might be accommodated within the latter.1
I make no attempt at exhaustivity in the discussion of modern developments, however.
A complete, up-to-date review of research on dative and double object constructions,
for example, would require a volume of its own, and the same holds for the other core
topics treated here. I also attempt no systematic comparison between shell structures
and other theories adopting a broadly right-descending view of phrase structure, for
example, the influential Antisymmetry Theory of Kayne (1994), the PredP Analysis of
Bowers (1993), or the Cascade Theory of Pesetsky (1995). Again, such a goal would
be too ambitious. Rather, I simply hope to provide a broader picture of shell structure,
sketching its motivations and some of its empirical applications and properties, and
noting its divergences from superficially similar ideas that have developed and some
respects in which it might still remain an attractive alternative to them.
1. BACKGROUND
Many of the structures proposed in the essays of this book have counterparts in the earli-
est work in transformational generative grammar, particularly Chomsky (1955/1975).
However, the direct debt of shell structure is to the work of David Dowty and Emmon Bach
2 On Shell Structure
within classical, Categorial Montague Grammar.2 Indeed, shell structures represented
my own attempt, following in the footsteps of Pauline Jacobson (1983, 1984, 1987),
to import certain ideas from Montague Grammar into a more traditional approach to
phrase structure.
1.1. Montague Grammar

A core feature of classical Montague Grammar is that syntactic and semantic composi-
tion proceed in parallel. Each basic expression is paired with an interpretation, and each
syntactic rule putting together expressions to yield a larger expression is paired with a
corresponding semantic rule putting together the interpretations of the first to yield the
interpretation of the second (the so-called rule-to-rule hypothesis).
For example, a rule-to-rule analysis of the sentence John walks might begin by
assigning John and walks the individual interpretations John’ and walk’, where the
former denotes an individual and the latter a function from individuals to truth-values
(the denotation of a sentence). A syntactic rule (A) concatenating John and walks to
form a sentence might then be matched to a semantic rule (X) applying walk’ to John’
(1a). Similarly, a rule-to-rule analysis of John kisses Mary might interpret kiss by a
function from pairs of individuals to truth-values. As shown in (1b), a syntactic rule (B)
performing the complex concatenation might then be matched by a semantic rule (Y)
that applies the interpretation of the verb to a pair consisting of the interpretations of
the nominals (in the appropriate order).
(1) Syntax Semantics
a. John walks walk’(John’)

/ \ A / \ X
John walk John’ walk’
b. John kisses Mary kiss’(<John’, Mary’>)

/ \ B / \ Y
John kiss Mary John’ kiss’ Mary’
Taken on its own, rule-to-rule correspondence imposes few constraints on either

syntax or semantics. However, it is not hard to appreciate that when they are imposed,
constraints in one domain will immediately be reflected in the other. For example, in
Montague Grammar, the functions that interpret verbs and other predicational expres-
sions are constrained to combine with a single argument at a time. This means that
transitive verbs cannot actually work as shown in (1b), where they combine directly
with their two arguments to yield a sentence. Rather, the derivation must proceed more
indirectly. The function corresponding to the verb is analyzed as combining first with its
direct object argument to yield another function, which then combines with the subject
argument to yield a truth-value (2b):
(2) a. Syntax b. Semantics
John kisses Mary (kiss’(Mary’))(John’)

/ \ / \
John kiss Mary John’ kiss’(Mary’)
/ \ / \
kiss Mary kiss’ Mary’
Now, because of the single-argument requirement, the syntactic rule combining heads
and complements must be one in which two expressions combine: one representing the
function and the other representing its unique argument. This entails that syntax must
be binary branching in its most general form: syntactic composition mirrors semantic
composition (2a).
By the mid- to late 1970s, the Montague Grammar program had yielded a number
of interesting results in regard to basic argument structure. Dowty (1978, 1979) had
argued for a semantic approach to grammatical relations in which status as sub-
ject, direct object, and so on was determined by the semantic combining order of an
argument. More precisely, subjects were identified as those phrases that combined
with intransitive verb phrases (IVPs): expressions denoting functions that mapped
their arguments to a truth-value. Direct objects (DObjs) were what combined with
transitive verb phrases (TVPs): expressions denoting functions that mapped their
arguments to the kind of function that took a subject (IVPs). Indirect objects (IObjs)
were what combined with ditransitive verb phrases (DVPs): expressions denoting
functions that mapped their arguments to the kind of function that took a direct
object (TVPs). And so on.
(3) Syntax Semantics
John give Fido to Mary ((give’(Mary’))(Fido’))(John’)

/ \ / \
Subj John give Fido to Mary IVP John’ (give’(Mary’))(Fido’)
/ \ / \
DObj Fido give to Mary TVP Fido’ give’(Mary’)
/ \ / \
DVP give to Mary IObj give’ Mary’
This kind of approach yields novel, and rather unexpected, results for syntactic
composition in many cases.3 For example, note that if Mary is the direct object in John
persuade Mary to leave, then under binary composition the remainder of the verb
phrase must form a constituent TVP (4a). The same conclusion will follow for complex
predicates like resultatives (4b) and control adjuncts like (4c).
(4) a. Control Complements b. Complex Predicates (Resultatives)
John persuade Mary to leave John hammer the metal flat

/ \ / \
John persuade Mary to leave John hammer the metal flat
/ \ / \
Mary persuade to leave TVP the metal hammer flat TVP
/ \ / \
persuade to leave hammer flat
c. Control Adjuncts d. Discontinuous Idioms
John buy the book to read John put Mary on the spot
/ \ / \
John buy the book to read John put Mary on the spot
/ \ / \
the book buy to read TVP Mary put-on-the-spot Idiom
/ \ TVP
buy to read
The constituencies expressed by the derivation trees in (4) may appear odd at first,
but they had attractions that were actively explored by Montague Grammarians.
Consider, for example, control and predication relations. In (4a) Mary constitutes the
first argument that combines with the predicate after the control complement has been
merged in; in other words, it is derivationally the closest argument. Mary is also under-
stood to control the complement; leave applies to Mary. The same is true of the relation
between the direct object the metal and the inner predicate flat in (4b). The former is
the argument closest to the latter, and the latter is understood as predicating of the for-
mer. Accordingly, we can obtain something like Rosenbaum’s (1970) Minimal Distance
Principle in this framework. (4c) is even more interesting: here the object the book is
understood simultaneously as the object of buy and the object of read, and the subject
John is understood simultaneously as the subject of buy and the subject of read. In
other words, buy-to-read can be understood semantically as forming a kind of complex
transitive verb.
Derivation trees like those in (4) also promised insight into selectional relations
and idioms. In (4b), the verb (hammer) combines directly with resultative predicate
(flat), presumably exercising selection on it. This at least suggested an approach to the
familiar observation that whereas certain adjectives could appear as resultatives, other
apparently similar ones could not (cf. *John hammer the metal curved).4 Consider
also the point, noted by Bach (1979), that composition of the kind in (4) allows us
to understand in a simple way certain kinds of surface-discontinuous VP idioms first
observed by Emonds (1976). Expressions like put-on-the-spot, meaning ‘confront,’
will in fact form an underlying syntactic constituent with which the direct object com-
poses (4d). Hence what is understood as a semantic unit will also occur as a syntactic,
derivational unit.
1.2. “Right Wrap” and Merge

In examining the Montague Grammar composition trees in (3)–(4), one is immediately
struck by the operation merging the direct object. Unlike what occurs with the subject
or the lowest complement of V, the form of Merge present here is not simple Right or
Left Concatenation. Instead, the direct object phrase is inserted into the predicate. Alter-
natively, the predicate is “wrapped around” the direct object so that the latter ends up
just to the right of the predicate’s head (5):
(5) a. persuade Mary to leave b. hammer the metal flat

/ \ / \
Mary persuade to leave the metal hammer flat
c. buy the book to read d. put Mary on the spot

/ \ / \
the book buy to read Mary put-on-the-spot
This syntactic operation was in fact recognized to be special and was given its
own name: “Right Wrap.” Right Wrap is interesting insofar as it has no clear coun-
terpart in traditional phrase structure. Whereas the derivations in (1) and (2), which
involve only concatenation, can be matched with equivalent phrase markers, no
such counterpart is evident with (3)–(4). Suppose, then, that one wished to retain
the general compositional structure of Montague Grammar–style sentence deriva-
tions, importing them into more standard phrase structure representations. How
would one do it?
Dowty (1982) suggested a general answer when he observed that the order of
arguments in VSO languages might be achieved by Right Wrap, as indicated in (6a).
In transformational grammar, VSO ordering had been widely analyzed as due to an
operation of verb raising (6b), following ideas by Joseph Emonds (1980).
(6) a. kisses John Mary b. CP

/ \
John kiss Mary C S
/ \
kiss Mary kisses NP VP
John V NP
kisses Mary
A natural idea, then, was to try to extend the equivalence Right Wrap ≈ ∼ V-Raising into
VP, analyzing that too as the product of V-Raising.
Pauline Jacobson (1983, 1984, 1987) was the first to propose such an extension.
Compare the Montague Grammar derivation in (7a) (= (3)) with the counterpart phrase
marker in (7b) suggested by Jacobson (1987). In both diagrams the verb give is first
composed with the indirect object argument (to) Mary to form a small predicate phrase
give to Mary. In Jacobson’s tree, the latter is analyzed as a VP. A larger predicate phrase
is then formed by combining the direct object in such a way that the verb ends up to the
left of the object. In the former case this is done with the familiar Right Wrap operation;
in the latter case it is done by raising the verb from the smaller VP to a larger one of
which it becomes the head (7b):
(7) a. John give Fido to Mary b. S

/ \
John give Fido to Mary NP VP
/ \
Fido give to Mary John V NP VP
/ \
give to Mary give Fido V PP
give to Mary
Jacobson thus offered the first translation from Montague Grammar–style composition
inside the VP to standard phrase structure, achieving this by appeal to V-Raising.
1.3. Questions and Puzzles

My own work entered the scene at this point. I arrived as an assistant professor at
MIT in 1985, having been trained in classical Montague Grammar as a PhD student
at the University of Wisconsin–Madison. I had heard two public talks by Jacobson
(1983, 1984) and was intrigued by the general idea of bringing the results of Montague
Grammar into the Extended Standard Theory. At the same time I had doubts and ques-
tions, both about the specific structures that Jacobson had proposed and about certain
broader assumptions within Montague Grammar generally.
One difference immediately apparent in comparing (7a,b) is that whereas the former
is binary branching, the latter is not: Jacobson’s upper VP is ternary. This constitutes a
significant departure from Montague Grammar derivation trees and connects to a wider
question about the source of binary branching in the theory. As noted above, Mon-
tague Grammar derives binary branching from its semantics. Predicates are analyzed
as expressing unary functions—functions taking a single argument—and thus semantic
composition must be binary. Imposing structure on the semantic objects induces the
same structure in the corresponding expressions in the syntax. Hence syntactic composi-
tion must be binary as well.
But what sense does it make to locate binary branching structure in the semantics?
In what sense is this structure semantic structure? On reflection, the answer was (and
is) far from clear to me. If by “semantic structure” we mean structure indispensable to
deriving the correct truth conditions for a sentence, then the structure in unary func-
tions is plainly not essentially semantic.5 This is clear from the fact that first-order logic
and other logical languages typically make no appeal to this device. It can also be seen
in alternative formal analyses of natural language (e.g., Larson and Segal 1995), where
appropriate truth conditions are compositionally derived but where predicates are ana-
lyzed as “flat” relations in the usual way.6
A more basic question, however, was simply the source of the VP structures in (7b)
in the first place. Jacobson was working within an early Generalized Phrase Structure
Grammar framework in which rules of the form VP  V NP VP could be stipulated. But
research in the Extended Standard Theory had moved steadily away from such devices,
attempting to reduce the contribution of phrase structure rules to a general templatic
format, with particular configurations the product of specific, local properties of the
individual elements being combined (selectional features, Case, agreement, etc.).7 What,
then, was the status of the V element/position that was the target of raising in (7b)?
What was its relation to the phrases (NP and VP) within the larger VP that it headed?
How was the core configuration for the V-Raising structure “licensed,” in the popular
jargon of the day?
2. VP SHELLS
In a paper on the double object construction (Larson 1988), I attempted to for-

mulate a theory of syntactic projection that would allow one to capture Montague
Grammar–style composition using conventional phrase markers with verb raising
and would address the core questions noted above. This account had three basic
parts.
First, the assumption of binary branching was extracted from the semantics (where
it didn’t seem to belong) and made a part of syntax. At the time, the general tem-
platic structure of expressions was assumed to be fixed by an independent X-bar theory.
Hence I proposed a restricted X-bar format requiring heads and predicates to combine
with only a single phrase at a time. In point of fact, predicates were already assumed to
take a unique subject (SpecX′) under X-bar theory (8a); hence all that was needed was
to assume that heads could take a unique complement as well (ZP). This was done by
means of the X-bar rule in (8b):
(8) a. XP SpecXʹ Xʹ
b. Xʹ X ZP (Single Complement H

ypothesis)
(= (76) in Larson 1988, p. 381)

Second, there was a principle of locality (P1) determining the domain in which thematic
roles could be assigned by a head:
(9) P1: If is a predicate and is an argument of , then must be realized within a
projection of . (Larson 1988, p. 382)
Finally, as mentioned in note 4, all semantic theories must assume that relations have
a “direction” that determines the role that a given argument plays in the relation. It is
often assumed that the arguments of a relation can be distinguished according to their
thematic roles. In Larson (1988) the following principle was assumed to map the the-
matic roles of a predicate to the arguments that received them in structure.
(10) a. Thematic Hierarchy
AGENT > THEME > GOAL > OBLIQUE (manner, location, time, …)
b. P2: If a verb determines -roles 1, 2, n, then the lowest role on the Thematic
Hierarchy is assigned to the lowest argument in constituent structure, the next lowest
role to the next lowest argument, and so on. (Larson 1988, p. 382)
To illustrate these principles, consider first the transitive verb kiss, which assigns an
agent and a theme role. (8)–(10) determine a VP headed by kiss, as in (11). This struc-
ture conforms to the restricted X-bar theory (8); all arguments of V are contained within
a projection of V (9); finally, the argument bearing the agent role is projected into a
position c-commanding the argument bearing the theme role, in compliance with (10b)
and the fact that θAGENT > θTHEME
(11) VP
DP V′
John V DP
kiss Mary
Ditransitive put, which assigns an agent, theme, and location, presents a more compli-
cated and more interesting case. Assuming θAGENT > θTHEME > θLOC, we project a minimal
VP as in (12), containing arguments corresponding to θTHEME and θLOC, with the former
higher than the latter:
(12) VP
DP V′
the key V PP
put on the table
This structure leaves θAGENT unassigned, with no position for its bearer. Larson (1988)
proposes that such circumstances license the projection of a “VP shell,” as in (13a),
which contains a higher specifier for the agent and brings along an empty verbal head
position purely as a consequence of X-bar theory. (13a) was suggested to constitute
something like the “minimal structural elaboration of (12),” allowing the principles in
(8)–(10) to be met. The surface word order then is derived by raising the verb to [v e]
(13b), basically following the idea of Jacobson:
(13) a. VP b. VP
DP V′ DP V′
John V VP John V VP
DP V′ put DP V′
the key V PP the key V PP
put on the table put on the table
The resulting structure now achieves full binary branching (no ternary structures) and a
composition that directly matches Montague Grammar–style derivations using Right Wrap.
This general account of shell licensing appeared extensible to the full range of
configurations within the framework of classical, Categorial Montague Grammar
noted earlier, so long as the lowest phrases (CP, AP, PP) were analyzed as oblique
verbal complements, projected below agents and themes. Compare (4a–d) to (14a–d),
respectively:
(14) a. Control Complements b. Complex Predicates (Resultatives)

VP VP
DP V′ DP V′
John V VP John V VP
persuade DP V′ hammer DP V′
Mary V CP the metal V AP
persuade PRO to leave hammer flat
c. Control Adjuncts d. Discontinuous Idioms

VP VP
DP V′ DP V′
John V VP John V VP
buy DP V′ put DP V′
the book V CP Mary V PP
buy OPi PRO to read ti put on the spot

2.1. Properties of Shell Structures

The specific attractions of structures like (13a) and (14a–d) were pursued and developed
in the papers in this book, and by others in their own work as well. Without rehearsing
particular points, it’s worth noting certain general features of the analysis, especially in
light of related proposals that came afterwards.
One feature immediately apparent in (13a) and (14a–d) is the generalized structural asym-
metry imposed on the arguments of the predicate. At the time, such asymmetry was widely
recognized to hold between subjects and complements but was not hypothesized among
complements themselves. Thus, double PP sentences like Max talked to John about Mary, or
double object sentences like Mary gave John a present, were routinely assigned VP configura-
tions like those in (15a,b), respectively, in which the complements were basically symmetric:
(15) a. VP b. VP
V PP PP V DP DP
talk to John about Mary give John a present
And even where two complement elements were known to behave asymmetrically with
respect to well-known domain relations, for example, the two boldfaced DPs in (16),
this was widely assumed to result from the structure of the complements themselves—
here the additional PP structure imposed by the about-PP:8
(16) VP
V DP PP
told John P DP
about Mary
By contrast, with shell structures, subject-object asymmetry is generalized across all argu-
ments. Thus double PPs receive a structure like (17a), in which the first PP is structurally
superior to the second. DP–PP constructions get a structure like (17b), in which an asym-
metry between John and Mary is imposed that is fully independent of the presence of PP:
(17) a. VP b. VP
DP V′ V′
Max V VP Max V VP
talk V′ V′
to John V PP John V PP
talk about Mary tell about Mary
A second feature of shell structures is their recursive potential. As we saw earlier, Right
Wrap in classical Montague Grammar is a general structure-building operation, a form
of complex merger between strings. This operation was hypothesized as taking place
in the combination of complex TVPs with their objects, and in the combination of VPs
with their subjects under Dowty’s approach to VSO languages. But these by no means
exhausted the operation’s possibilities. Consider the use of Right Wrap in analyzing the
buy-construction in (18), where the verb is assumed to select four arguments: a subject,
an object, a from-phrase, and a for-phrase:
(18) John buy Fido from Mary for $200
/ \ (Left) Concatenation
John buy Fido from Mary for $200
/ \ Right Wrap
Fido buy from Mary for $200
/ \ Right Wrap
from Mary buy for $200
/ \ (Right) Concatenation
buy for $200
The lowest complement (for $200) merges by (Left) Concatenation, followed by two
applications of Right Wrap: one to form the complex DVP buy from Mary for $200,
and another to form the complex TVP buy Fido from Mary for $200. The derivation is
coherent, and indeed quite natural given the example. In classical Montague Grammar,
nothing blocks Right Wrap from applying recursively in the same derivation.9
This feature is inherited by Jacobson’s (1987) approach to shells, given the recursive
nature of its core rule: VP  V NP VP. It is also inherited by the approach in Larson
(1988). Assuming oblique thematic roles for the for- and from-phrases, where the first
ranks lower than the second, and both rank lower than that of the object, we build the
initial VP tree shown in (19a). This leaves buy’s θTHEME role unassigned and provides no
position for its bearer. Hence we project (19b), the minimal structural elaboration of
(19a) allowing the principles in (8)–(10) to be met; the verb raises.
(19) a. VP b. VP
PP V′ DP V′
from Mary V PP Fido V VP
buy for $200 e PP V′
from Mary V PP
buy for $200
(19b) still leaves buy’s θAGENT role unassigned, with no position for its bearer. Hence
we project (20), the minimal structural elaboration of (19b) allowing the principles in
(8)–(10) to be met; again the verb raises:
(20) VP
DP V′
John V VP
buy DP V′
Fido V VP
buy PP V′
from Mary V PP
buy for $200
As with (18), the derivation appears coherent, and in fact natural given the example.
The recursive potential of Right Wrap is thus carried over into the shell account as well.
Finally, we note the category neutrality of shell structures. As a general structure-
building operation, Right Wrap in classical Montague Grammar was available to any
domain of syntactic composition, and hence (barring stipulation) was expected across
the full range of syntactic categories. And, indeed, applications outside VP were natural.
Consider candidate derivations like (21a,b) involving APs and DPs:
(21) a. dependent on Mary for help b. the man that John saw
/ \ / \
on Mary dependent for help man the that John saw
/ \ / \
dependent for help the that John saw
(21a) instantiates Jackendoff’s (1977) proposal that the adjective dependent, like the cor-
responding verb depend, selects two complements, here two PPs.10 (21b) is a variant of the
analysis of relative clauses first proposed in Smith (1964), wherein articles/determiners (the)
combine with relative clauses before combining with the noun (man)—the classical “Article-
S” analysis of relatives. In both cases, derivations involving Right Wrap are natural.
Once again, with some extensions, shell structures inherit this property of Right Wrap.
Assume the adjective dependent—like the verb depend—to be triadic in argument structure.
For concreteness, take the roles it assigns to be θTHEME, θLOC, and θSOURCE, ranked in that order
on the thematic hierarchy. Then the shell structure for (21a) can be projected as in (22),
where be is analyzed as a subject-raising verb, following proposals by Stowell (1978, 1981)
and Couquaux (1981), and where the projection of AP shells fully parallels that of VP shells:
(22) VP
V AP
be DP A′
Max A AP
dependent PP A′
on Mary A PP
dependent for help
Similarly, as discussed in detail in Larson (1991a,c), example (21b) can be analyzed as in

(23) under interesting assumptions about the thematic structure of determiners deriving
from Generalized Quantifier Theory (Barwise and Cooper 1981, and especially Keenan
and Stavi 1983):
(23) DP
Pro D′
D DP
the NP D′
man D CP
the that John saw
Shell structures thus also show category neutrality: projection counterpart to Right Wrap
is expected to be available in all categories.
2.2. Problems
The account of shell structures in Larson (1988) not only inherited key properties of
Right Wrap analyses from classical Montague Grammar but also improved on them in
some respects. Consider, for example, the fact that in all the derivations in (4), and in the
recursive derivation in (18), the first/lowest argument combines by Right Concatena-
tion, the last/highest argument combines by Left Concatenation, and all intermediate
arguments combine by Right Wrap. This pattern is unexplained in classical Montague
Grammar but actually follows in the shell account: the first/lowest argument is a com-
plement of the head X, the last/highest argument is a specifier of the maximal XP, and
all arguments in-between invoke an XP shell involving X-Raising—Right Wrap.
These virtues notwithstanding, there were also serious, unresolved aspects of the
analysis that could not help but provoke worry. For one thing, the exact motivation for
head raising between shells was obscure. The initial suggestion made by Larson (1988)
in connection with examples like (13b) was that V-Raising occurred in order to permit
Case assignment rightward to the direct object. But this proposal was discarded later in
the same paper since V-Raising was also required for structures in which Case assignment
was not an issue. Thus, in double PP complement structures like (17a), the higher comple-
ment is a PP (to John), not a DP, and hence presumably does not require Case. The same
holds for structures like (19b). Alternative general proposals, for example, the idea that
V-Raising occurred to “cover up” an empty category ([v e ]) that would otherwise have
raised problems for Proper Government, became likewise dubious as notions like Proper
Government themselves became suspect within the developing Minimalist Program.
Most worrisome, however, was simply the mechanism for projecting higher shells in
the first place. The basic idea seemed simple enough: invert the logic of the Extended
Projection Principle (EPP), in which the need to introduce a head (T) brought along with
it a subject specifier position, purely as a consequence of X-bar theory. In Larson (1988),
the need to project a specifier for XP was taken to introduce an empty X head for essen-
tially the same reason. But in fact the two cases are not truly symmetric. The former can
be accommodated within standard views of structure building: the T head merges with
VP under the relation ℜ of selection; X-bar theory (perhaps in combination with some
particular morphological property of T, as suggested in Chomsky 1986) then licenses the
additional TP structure (24a). In contrast, under the shell derivation, we were to merge
the specifier XP and VP, and then project the empty-headed X-bar structure between
them (24b), where the relation ℜ’ between XP and VP was plainly not selection:
(24) a. TP EPP Projection of Empty Spec
T VP XP T′ = selection
e T VP
b. VP EPP Projection of Empty Head?
XP ′ VP XP V′ ′ selection
V VP
e
Indeed, the merging of the specifier in (24b) required the additional, admittedly vague,
notion of a “minimal structural elaboration compatible with X-bar theory.” Despite its
broadly “last-resort” flavor, suggesting a connection to notions of economy of projection
and derivation that were then under development (Chomsky 1991), no real clarification
of this notion was forthcoming. And whatever prospects the idea might have had effec-
tively ended when Chomsky (1994) argued convincingly for the abandonment of X-bar
theory as a distinct module of grammar, removing the chief tool for the projection of VP
shells with empty heads. With X-bar theory gone, the concept of a “minimal structural
elaboration compatible with X-bar theory” became unformulable within grammatical
theory. Hence shell theory, in its original form, simply could not be maintained.
That the original account of VP shell projection became untenable does not, of
course, entail that the structures it yielded were wrong, nor that other means for pro-
jecting them could not be found. Indeed, in the years following, a number of alternative
proposals were advanced by researchers who found the general approach attractive.
Predictably, the key variation in these proposals centered around the problematic V
position that is the target for raising, and its relation to the VP with which it combined:
how was this element to be conceptualized and integrated into an explicit, coherent
account of structure building?11
3. EMPTY V AS NULL CAUSATIVE?
One response was to look backward toward older ideas within the field. Trees like (13a)
superficially resemble classic, Generative Semantics–style decompositional structures
like (25a), where put is factored into the abstract predicates cause and be/go and where
the latter are joined by a rule of predicate raising and subsequently “spelled out” pho-
nologically as put (25b):12
(25) a. VP b. VP
V′ V′
John V VP John V VP
CAUSE V′ V V DP V′
the key V PP CAUSE BE/GO the key V PP
BE/GO on the table BE/GO on the table
One simple model of shell structure projection would therefore be to reanalyze the empty
V introduced above as a phonologically null causative (Chomsky 1993)—a lexical head
with independent semantic content that thematically selects both its complement VP
(essentially a propositional structure) and its DP specifier. This effectively eliminates
the puzzle of how to motivate structure building with an empty V by eliminating the
latter altogether: V is not empty after all, and the account of structure building returns
to largely conventional lines.
Of course, accounts like this bear their own substantial burden of explanation. Full
decomposition incurs a commitment to exhaustive semantic analysis that is notoriously
hard to carry beyond initial intuitions. What precisely is the semantics of the capitalized
predicates in (25), and how do they differ from their pronounced counterparts? How
does the theory accommodate differences within the class of transitive locative verbs to
which put belongs (put, place, position, stand, etc.); apparently these cannot all be equal
spell-outs of cause-be/go, and so on. Furthermore, how can we extend the analysis
appropriately, if at all, to other cases analyzed through VP shells, for example, resulta-
tives or the double PP construction? What abstract predicate (??) together with cause
will yield wipe and talk in (26a,b)?13
(26) a. VP b. VP
DP V′ DP V′
Max V VP Max V VP
CAUSE DP V′ CAUSE PP V′
the table V AP to John V PP
?? clean ?? about Mary
CAUSE + ?? = wipe CAUSE + ?? = talk
An alternative, and more popular, version of this view is what might be called the
“partially decompositional” model, which, roughly put, tries to factor only the caus-
ative semantics and/or the role of the external argument from the associated predicate
(Chomsky 1993; Kratzer 1996; Harley and Folli 2007). As above, the empty verb, often
notated with a small v, is reanalyzed as an empty causative, or a head responsible only
for the subject’s θ-role. But the remainder of the predicate is left unanalyzed (27a,b):
(27) a. vP b. vP
DP v′ DP v′
John v VP John V VP
CAUSE DP V′ v V DP V′
the key V PP CAUSE put the key V PP
put on the table “put” put on the table
Although less daring, this view also carries significant commitments. If v is analyzed as
the uniform source of the agentive thematic role, then it must be present whenever the
latter is assigned. Thus, v must project with active transitives like (28a) and with unerga-
tives like (28b), in contrast to the simplest form of shell theory:
(28) a. vP b. vP
v′ v′
John v VP John v VP
V DP performed
kissed Mary
Matters become more complex still if v is further identified as the source of accusa-
tive Case in transitives, as is in fact widely assumed (Chomsky 1995, Adger 2003).
Many transitive verbal constructions involve accusative objects without an agentive
subject (John sees Mary, John loves Mary, John hurt his foot). On the partially decom-
positional model, this entails a proliferation of light verbs, some responsible for the
agentive role and some for other roles (e.g., Experiencer) (see Harley and Folli 2007).
A serious issue of appropriate coordination thus arises between the upper little v and
lower “big V.”
But whatever the strengths and weaknesses of the fully and partially decompositional
approaches, we may simply note that these are very different theories from the shell
theory account given in Larson (1988) insofar as neither preserves the core properties
of the Right Wrap analysis in their full generality. For example, although (25a) and
(27a) encode a structural asymmetry among subject, object, and prepositional object
arguments based on VP structure, this result is not a general one. Without additional
assumptions, there is nothing that would permit general rightward descent of the com-
plements in (29a) as in (29b):
(29) a. John bought [Fido] [from Mary] [for $200].

b. vP
DP v′
John v vP
v′
buy v Fido v VP
buy v V′
from Mary V PP
buy for $200
This follows directly from another difference between decompositional theories and
that in Larson (1988), namely, recursivity. Since “little v” is tied to a specific semantic
or thematic content (causation/agency), there is no general expectation of recursion
in the theory. A tree like (29b), for instance, could not deploy such a little v in both
upper and lower positions since this would entail the causative component entering
the derivation twice, or result in John and Fido both receiving the agent role. Of
course, it might be possible to ascribe different content to the little v’s in the two posi-
tions, in effect forcing further decomposition on ourselves. For example, in the partial
decompositional theory, the lower v would presumably extract the theme role from the
predicate. But along with posing for us the question of why the theme role isn’t also
factored into the put derivation in (27) above, as well as the more general question of
how the grammar “knows” which little v to insert where, this move would again do
nothing to ensure recursion below the theme level. In decompositional theories, each
vP shell, and each v heading it, must be treated as a special case, with its own separable
content or role.
Finally, and again unlike the original shell theory, decompositional theories of
projection appear essentially tied to the verbal domain—they aren’t category neu-
tral. The notion of “little v” relevant in this family of accounts is associated with
specifically verbal semantics (causation) or verbal thematic content (agency) in con-
trast to the original shell theory, wherein empty v is simply a structural placeholder.
Decompositional accounts therefore yield no expectation that structure building in
other complement-selecting categories should parallel that of VP, at least not among
those involving a nonverbal semantics and nonverbal θ-features. But on the analysis
of DP structure developed in Larson (1991a) and discussed later in this volume, such
parallels do appear to hold. DPs appear to project from their D heads according to
principles parallel to those involved with VPs, albeit through a very different set of
θ-features.14
4. RECONSTRUCTING SHELL THEORY
Given these points, let us now reconsider whether a variant of the original shell analysis
might be found that addresses its basic problems but without losing its essential proper-
ties, as the decompositional theories appear to do.15 To set the stage, I begin by revisiting
thematic roles and their relation to semantics.
4.1. θ-Theory and Semantic Composition

The original shell theory of Larson (1988) assumed a Government and Binding–style
θ-theory wherein predicates bear a set of θ-roles that are assigned to arguments in
the course of derivation, with the relation among these elements governed by the
θ-Criterion:
θ-Criterion
ach argument bears one and only one θ-role, and each θ-role is assigned to one
E
and only one argument. (Chomsky 1981, p. 36).
This account dovetailed neatly with the semantic analysis of predicates and selection
that was prevalent at the time, and indeed remains a standard one in textbooks on
formal semantics.16
4.1.1. The Standard Semantic Analysis

Under the standard semantics, predicates denote n-ary relations on individuals, nomi-
nals (and other arguments) denote the individuals related, and semantic composition
is the process of putting the two together. An example is the analysis of (30) as (31a)
(ignoring tense), where the mapping of sentence parts to semantic contributions is as in
(31b), and where semantic composition is a matter of applying the predicate function
sequentially to its arguments in parallel to the syntactic composition of the tree (31c).
(30) John gave Mary Fido.

(31) The Standard Analysis
a. (j,m,f)
b. gave z y x[ (x,y,z)]
John j
Mary m
Fido f
c. vP (j,m,f)
John v′ x[ (x,m,f)] j
gave VP y x[ (x,y,f)] m
V′ z y x[ (x,y,z)] f
gave Fido
This analysis provides a natural semantic interpretation for Government and Binding–
style notions of θ-theory. Thus we may understand the syntactic idea that give assigns
three θ-roles in terms of the semantic idea that give’ is a ternary function. We may
understand the syntactic requirement that a verb assign all its θ-roles in a well-formed
derivation in terms of the semantic requirement that the corresponding function be com-
bined with all its arguments in order to map to truth or falsity. And we may understand
the requirement that all potential argument expressions bear a θ-role in terms of the
need for the corresponding individuals to be the arguments of some predicate function
in order to be integrated into semantic composition.
These correlations emerge clearly in attempts from the time to spell out θ-roles
and θ-role assignment in a formal way. For example, Higginbotham (1985) proposes
that predicates be associated with a “θ-grid,” represented as a numbered sequence
<1>, <1,2> or <1,2,3>, according to whether the predicate is intransitive, transitive,
or ditransitive, respectively. As predicates combine with arguments, positions are
“starred” in the grid, representing θ-role assignment (32a). This procedure corre-
lates transparently with function-argument application as represented in the standard
semantics (32b):17
(32) a. b.
VP x[ (x,m)]
<1,2*>
y x[ (x,y)] m
V DP
kiss Mary
<1,2>
4.1.2. Neo-Davidsonian Semantic Composition
Given a representation like (31a), consisting of four semantic constituents for a sentence
consisting of four words, it’s hard to envision a syntax-semantics mapping different
from that in (31b). Things become more interesting, however, with neo-Davidsonian
approaches to sentence semantics. Under the latter, (30) receives the representation in
(33), where give’ does not relate the individuals m, j, and f directly but rather denotes a
unary predicate of events to which the arguments are related by means of binary rela-
tions like Agent, Goal, and Theme.
(33) e[ give’(e) & Agent(e,j) & Goal(e,m) & Theme(e,f)]
One possible syntax-semantics mapping for such a representation is (34), which I

will call the “P Analysis” since it was urged by Parsons (1991). Here again, nomi-
nals denote bare individuals, and the V-meaning simply folds into itself all the extra
semantic structure present in (33), including the relations that the individuals par-
ticipate in. Semantic composition proceeds as before, by function-argument applica-
tion (34b):
(34) The P Analysis
a. gave z y x e[ (e) & Agent(e,x) & Goal(e,y) & Theme(e,z)]
John j
Mary m
Fido f
b. vP e[give’(e) & Ag(e,j) & Gl(e,m) & Th(e,f)]
John v x e[give’(e) & Ag(e,x) & Gl(e, m) & Th(e,f)] j
gave VP y x e[give’(e) & Ag(e,x) & Gl(e,y) & Th(e, f)] m

Mary V′ z y x e[give’(e) & Ag(e,x) & Gl(e,y) & Th(e,z)] f
gave Fido
On the P Analysis, event structure represents a sublexical or “subatomic” analysis of

predicate meaning: the first line of (34a) is simply a more articulated version of the
corresponding line in (31b), with the extra structure inert. The semantic interpreta-
tion of θ-roles and the θ-Criterion from the Standard Analysis is thus preserved intact
in (34).
But the P Analysis is not the only one conceivable. Consider the radical alternative in
(35), which derives from proposals by Krifka (1992) and which I will call the “K Analy-
sis.” The basic syntax-semantics mapping is shown in (35a), with semantic composition
as in (35b), involving predicate conjunction in the first three steps and existential closure
in the last.18
(35) The K Analysis
a. gave e[ (e)]
John e[Ag(e,j)]
Mary e[Gl(e,m)]
Fido e[Th(e,f)]
b. vP e[ (e) & Ag(e,j) & Gl(e,j) & Th(e,f)]

|
John v′ e[ (e) & Ag(e,j) & Gl(e,m) & Th(e,f)]
gave VP e[ (e) & Gl(e,m) & Th(e,f)] e[Ag(e,j)]
V′ e[ (e) & Th(e,f)] e[Gl(e,m)]
gave Fido e[ (e)] e[Th(e,f)]
Although the K Analysis derives the same semantic result for sentence (30), it does so
via a very different route. And the difference has dramatic consequences for our under-
standing of θ-theory and its relation to semantics.
Under both the Standard Analysis and the P Analysis, bearing three θ-roles syntactically
can be correlated with taking three arguments semantically, where these arguments are the
sort that names might be expected to provide—individuals like John, Mary, or Fido. But
in (35a) give corresponds to a unary predicate, not a ternary one. Furthermore, its sole
argument is an event, and not an individual like John, Mary, or Fido. The K Analysis thus
provides no semantic correlate for the idea that give bears three θ-roles associated with the
argument nominals in (30). Put differently, if there is selection between the predicate and
its arguments, the K Analysis implies this is a wholly syntactic matter, not a semantic one.
Consider also the implications for the θ-Criterion, which captures the unacceptabil-
ity of (36) and (37) by saying that demand and put bear unassigned θ-roles in the former,
and that Alice is without a θ-role in the latter:
(36) a. *Mary demands. (Demands what??)
b. *John put the book. (Put it where??)
(37) *Mary yawned Alice.
Both the Standard Analysis and the P Analysis can correlate this assessment with seman-
tics, providing a semantic basis for the θ-Criterion. Thus, they can hold that the func-
tions corresponding to demand and put have not been presented with the required
number of arguments in (36a,b) and hence cannot denote a truth-value. And they can
hold that the unary function corresponding to yawn, having combined with its sole
argument in (37), simply has no power to combine with another.19
Again, under the K Analysis such reasoning is forfeit. Whatever sense of incomplete-
ness may attach to (36a,b), it cannot reflect semantics. The K Analysis provides both
of these expressions with coherent semantic compositions, yielding truth-evaluable
results (38).
(38) a. e[demanding’(e) & Ag( e,m)] b. e[putting’(e) & Ag( e,j)& Th(e,tb)]
e[demanding’(e) & Ag(e, m)] e[putting’(e) & Ag(e, j) & Th(e,tb)]

e[demanding’(e)] e[Ag(e,m)] e[putting’(e) & Tg(e, tb)] e[Ag(e,j)]
e[putting’(e)] e[Th(e,tb)]
Likewise whatever unacceptability the “extra argument” may introduce in (37), it cannot
be semantic in nature. This expression also has a coherent semantic composition under
the K Analysis, yielding a fully truth-evaluable result (39).
(39) e[yawning’(e) Th(e,a) & Ag(e,m)]
e[yawning’(e) & Th(e,a) & Ag(e,m)]

e[yawning’(e) & Th(e,a)] e[Ag(e,m)]
e[yawning’(e)] e[Th(e,a)]
The K Analysis thus implies that insofar as the θ-Criterion is correct, it must be a purely
syntactic condition governing a purely syntactic notion of selection.
The radical conclusions of the K Analysis for selection apply not only to lexical
heads like verbs but to functional heads as well. As noted earlier, Kratzer (1996) sug-
gests that a “little v” (which she terms a “voice head”) supplies the semantic θ-relation
of the subject/external argument, which itself denotes a bare individual (40a). Similarly,
Pylkkänen (2002, 2008) takes an Applicative (Appl) head to supply the θ-relation for
applied arguments in applicative constructions (40b). And many people have assumed
that prepositions like to supply the θ-relation borne by their objects (40c).
(40) a. vP b. c. PP
ApplP
v′ to Mary
m Mary x e[Gl(e,x)] m
m
x e[Ag(e,x)] Appl VP
x e[Benef(e,x)]
These proposals find no support in the K Analysis. If argument expressions come with
their thematic relations, then functional heads like v, Appl, or P cannot be the semantic
source of them (41). Functional heads cannot be responsible for “introducing argu-
ments” in any semantic sense.
(41) a. vP b. c. PP
ApplP
v′ to Mary
e[Ag(e,m)] Mary e[Gl(e,m)]
v e[Benef(e,m)]
Appl VP
At most, the selectional relation between a functional head (v, Appl, P) and its specifier
or complement must be a syntactic one.
4.2. θ-Features and Structure Projection

The K Analysis stands in sharp contrast to the Standard Analysis and its neo-Davidsonian
counterpart, the P Analysis, in encouraging a purely syntactic view of selection. Indeed,
this account is sufficiently at odds with current theorizing to make it an interesting ques-
tion as to what view of θ-roles, θ-theory and structure building might result from its
adoption and implementation. I explore this question below. The basic idea I pursue,
following Hornstein (1999) and in conformity with the severe restrictions on techni-
cal machinery imposed by the Minimalist Program, is that θ-roles are formal features,
borne by predicates as part of their lexical entries, assigned to argument expressions in
the numeration and coordinated through checking/agreement relations in the course of
syntactic derivation.
In working out these notions, I will assume the theory of features developed by
Pesetsky and Torrego (2004), wherein features come in four varieties, according to
whether they are interpretable/uninterpretable and valued/unvalued (42).
(42) INTERPRETABLE UNINTERPRETABLE

VALUED iFval uFval
UNVALUED iF uF
Under Pesetsky and Torrego’s account, two occurrences of a feature may undergo agree-
ment, producing two instances of what formally becomes a single feature. The precise
principles are stated in (43) (= (5) from Pesetsky and Torrego 2004, 4):
(43) Agree (feature-sharing version)
(i) An unvalued feature F (a probe) on a head H at syntactic location α (Fα) scans
its c- command domain for another instance of F (a goal) at location β (Fβ) with
which to agree.
(ii) Replace Fα with Fβ, so that the same feather is present in both locations.
Drawing on conventions from Head Driven Phrase Structure Grammar, Pesetsky and
Torrego adopt the notation in (44), where features are written with square brackets
and where those that have undergone agreement are written with a shared index in the
brackets.
(44) … uF[ ] … uF[ ] … ⇒ AGREE ⇒ … uF[3] … uF[3] …
Under basic assumptions in the Minimalist Program, only features that are both
interpretable and valued are legible to the interfaces. Within the Pesetsky and Tor-
rego framework, features like (45a–d) will thus be legible. In (45a), the single occur-
rence of F is both interpretable and valued. In (45b–d) distinct instances of F are
interpretable and valued, but F itself counts as interpretable and valued since its
distinct occurrences have undergone agreement and have thus become instances of
a single feature.
(45) a. iFval[ ]
b. iF[n] … uFval[n]
c. iF[n] … uF[n] … uFval[n]
d. iF[n] … uF[n] … uF[n] … uFval[n]
By contrast, none of the features in (46) is interface legible. In (46a,b) F is valued but not
interpretable. In (46c,d) F is interpretable but not valued. And (46e) contains only unvalued
and uninterpretable occurrences of F, since the latter have not undergone agreement:
(46) a. uFval[ ]
b. uF[n] … uFval[n]
c. iF[ ]
d. iF[n] … uF[n]
e. iF[ ] … uFval[ ]
To summarize the basic picture in this theory of features, then, an interpretable occur-
rence of F will need to agree with a valued occurrence of F, a valued occurrence of F will
need to agree with an interpretable occurrence of F, and an uninterpretable, unvalued
occurrence of F will need to agree with both.
4.2.1. Predicate–Object
Putting aside features that are iFval[ ]—both interpretable and valued—let us now con-
sider θ-features in Pesetsky and Torrego’s terms. Assume that we have features [ag(ent)],
[th(eme)], [g(oa)l], and so on, which are borne by both predicates and their arguments.
Assume, furthermore, that selection between a predicate and an object amounts to
θ-feature agreement (47).
(47) VP
kiss Mary
[TH [ ]] AGREE [TH[ ]]
The questions that arise at this point are the following: Which elements of (47) carry
interpretable θ-features? Which elements of (47) carry valued θ-features? And which
elements of (47) carry uninterpretable/unvalued θ-features?
As we saw, under both the Standard Analysis and the P Analysis it makes sense to
associate θ-roles (and hence θ-features) with predicate meaning: bearing θ-roles correlates
with argument structure. Hence, under both the Standard Analysis and the P Analysis it
makes sense to think of θ-features as interpretable on the predicate. But as we also saw,
the K Analysis implies something quite different. On the latter, argument nominals come
with their associated semantic relations; the latter are not contributed by the predicate
or any functional head. Thus, on the K Analysis, the opposite conclusion is the natural
one; that is, θ-features are interpretable on argument expressions and not on predicates.
If we accept the K Analysis, then, Mary in (47) should bear ith[ ], which will be interpre-
table at the interface as λe[Th(e,m)]. And given [th] as interpretable on Mary in (47), it’s
natural to view [th] as valued on kiss. The picture is therefore as in (48): [ith[ ]] on Mary
probes [uthval[ ]] on kiss, and they agree. This yields a Logical Form–legible feature.
(48) VP
kiss Mary
[uTHval[1]] [iTH[1]]
AGREE!
If we define External Merge as Merge accompanied by θ-feature agreement, then (48)

instantiates External Merge.
4.2.2. Subject–Predicate
Suppose we now wish to merge structure (48) with the nominal John as its subject.
Reasoning as above, John should bear an interpretable θ-feature under the K Analysis,
here [ag]. If kiss is taken to bear a valued [ag] feature, things proceed as above. John
combines by External Merge (49).
(49) VP
John V′
[iAG[2]]
kiss Mary
AGREE! [uAGval[2]] [iTH[1]]
[uTHval[1]]
This produces a transitive VP of the sort generated under the original shell structure theory.
Consider, however, the following “thought experiment”. Suppose that predicates can-
not bear more than a single valued θ-feature. More precisely, suppose that sets composed of
features of the same kind are treated like single features insofar as only one value is permit-
ted to them. Suppose, further, that in the default case [th] is the valued feature in any set of
θ-features that contains it—a special status for themes. If kiss bears an unvalued occurrence
of [ag], John can still externally merge, and θ-feature agreement can still occur (50):
(50) VP
John V′
[iAG[2]]
kiss Mary
AGREE! [uAG[2]] [iTH[1]]
[uTHval[1]]
But this derivation now yields an [ag] feature that is unvalued, and unvalued features are
not interface legible. Hence (50) itself will not constitute an interface-legible syntactic object.
Continuing our thought experiment, suppose that when we select a numeration con-
taining a V, we’re allowed to co-select a “light v” with the following properties:
Light v
ears a strong V feature, an EPP feature, and a valued occurrence of a θ-feature

B
that is unvalued on the V.
Since [ag] is unvalued on kiss this permits a numeration that includes a light v carrying
[uagval[ ]]. We can merge this element with (48). The strong V feature on v attracts the
lexical verb, raising it. Little v and V may then agree on [ag] as shown in (51).
(51) vP
v VP
AGREE! v kiss kiss Mary
[uAGval[2]]-[uAG[2]] [uAG[ ]] [iTH[1]]
[uTHval[1]] [uTHval[1]]
John can now merge with this structure, satisfying v’s EPP feature. The unvalued [iag[ ]]
feature on John probes the valued [uagval[2]] feature on v and agrees (52).
(52) vP
John v′
[iAG[2]]
AGREE! v VP
v kiss kiss Mary

[uAGval[2]] [uAG[2]] [uAG[ ]] [iTH[1]]
The result is now an LF-legible feature and an interface-legible structure.

This derivation reconstructs the “little v” of Chomsky (1993) in a purely formal role—
that is, not as an empty causative, nor as any sort of content-full head responsible for
semantically “introducing” the external argument, but rather as the bearer of a valued
but uninterpretable formal feature—a θ-feature. On the line of reasoning suggested above,
little v might be seen as arising from the unavailability to the grammar of a simplex object
like (53a), with two (or more) valued θ-features. In response, the second value is permitted
(as it were) to be “spun off” as a separate verbal head, bound to its parent by a strong V
feature (53b). The grammar subsequently constructs the legitimate complex object (53c),
equivalent to the original simplex object that was proscribed (53a):
(53) a. * V b. V v c. v
[u 1val[ ]] [u 1val[ ]] [u 2val[ ]]
[u 2val[ ]] [u 2[ ]] Strong V v V
[u 2val[n]] [u 1val[ ]]
[u 2[n]]
For reasons that will become clear later, I will refer to little v’s that bear valued (unin-
terpretable) θ-features as voice heads.
A natural question arises at this point, namely, how we know to merge arguments
in the order shown in (52). What determines merger of themes before agents? In the
original shell theory of Larson (1988), argument order was determined by the thematic
hierarchy given in (10a) above. I will continue to adopt that basic idea here, recasting
it in featural terms. Specifically I assume the hierarchy of θ-features in (54a), and the
general constraint in (54b) on the order of agreement in sets of features:
(54) a. Hierarchy of -Features: [AG] > [TH] > [GL] > [LOC] > …
b. Constraint: A feature in a set subject to a hierarchy cannot undergo agreement
before a lower - ranked feature in that set does.

In (51), agreement with [ag] in the set {[ag], [th]} for kiss occurs only after [th]
undergoes agreement. The constraint in (54b) is thus satisfied for this derivation.
By contrast, a derivation with kiss attempting to merge an [ag]-bearing item first
(Mary or v) would violate the constraint in (54b). Hence the order of composition
in (52).
We might note one further point concerning valued features and voice heads. The
proposals offered above allow for unergative verbs like sneeze bearing a single, valued
[ag] feature, and hence for a licit syntactic derivation of John sneeze not involving little
v (55a). This structure departs from the influential view of Chomsky (1995) that the
agent role always involves little v projection (55b) (cf. (28b)):
(55) a. VP b. vP
John sneeze John v′

[iAG[2]] [uAGval[2]] ``
v sneeze
AGREE!
Chomsky’s view can be incorporated into the present analysis if it can be made to follow
that lexical heads are simply unable to bear a valued [ag] feature. This would then force
a derivation as in (56), where v is required to value [ag]:20
(56) vP
John v′
[iAG[2]]
AGREE! v sneeze
[uAG[ ]]
v sneeze
[uAGval[2]] [uAG[2]]
I will not attempt to decide between (55a) and (56), or speculate on what principles
might preclude lexical Vs from bearing valued [ag] in the latter case,21 but will leave
both open as possibilities at this point.
4.2.3. Predicate-Oblique Argument

Consider now ditransitive give, supposing it to bear the set of θ-features in (57):
(57) give {[uAG[ ]], [uTHval[ ]] [uGL[ ]]}
This looks traditional in some respects: give bears three θ-features, corresponding to an
agent, a theme, and a goal. But notice again that none of these θ-features is interpretable.
Their presence with give is purely syntactic. Furthermore, only one of the features is
valued. [ag] and [gl] are both uninterpretable and unvalued and hence must come into
agreement with occurrences of these features that are interpretable and valued, on pain
of interface illegibility.
Under the hierarchy of θ-features, the goal nominal must merge first. But merging
the goal directly yields the situation encountered earlier in (50): we get an interpretable
but unvalued goal feature (58). What is the alternative?
(58) VP
give Mary
[uAG[ ]] [iGL[1]]
[uTHval[ ]]
[uGL[1]] UNVALUED!
We have seen that neither P nor little v is semantically responsible for introducing argu-
ments. Suppose, however, that P, like little v, can bear valued θ-features. To bearing
a valued [gl] feature can then merge externally with Mary, with the two undergoing
agreement (59a). And PP can in turn externally merge with the verb, with the unvalued
[gl] feature on give probing and agreeing with the valued [gl] feature on to (59b):
(59) a. PP b. VP
to Mary give PP
[uGLval[1]] [iGL[1]] [uAG[ ]]
[uTHval[ ]] to Mary
AGREE! [uGL[1]] [uGLval[1]] [iGL[1]]
AGREE!
The rest now proceeds as before. Since [th] is valued on give, the theme nominal (Fido)
can merge directly and unproblematically with (59b). We then merge agentive little v,
which triggers raising of give and agreement with its goal feature. Finally, the agent
nominal (John) externally merges (60):
(60) vP
John v′
[iAG[3]]
v VP
v give Fido V′
[uAGval[3]] [uAG[3]] [iTH[2]]
[uTHval[2]] give PP
AGREE! [uGL[1]] [uAG[ ]]
[uTHval[2]] to Mary
This derivation reconstructs the function of P in a purely formal role—that is, not as a
θ-relation-contributing head as in (40c), but again as the bearer of a valued but uninter-
pretable θ-feature. Notice that it also establishes an interesting derivational connection
between prepositions and voice heads: both execute the function of valuing θ-features.
This suggests an approach to voice alternations, where oblique and nonoblique forms
alternate, that I will explore further in the introduction to part 2, where I discuss passive
and dative alternations.
4.2.4. Multiple Oblique Arguments and Recursion

As a final exercise, reconsider our earlier example (18)/(20), repeated below as (61).
We said buy might be taken to involve four arguments here: an agent (John), a theme
(Fido), a source phrase (from Mary), and what I’ll simply call a “remuneration phrase”
(for $200).
(61) John bought Fido from Mary for $200

Suppose the relevant θ-features are ranked [ag] > [th] > [src] > [rem]. Under the rea-
soning sketched above, the lowest VP can be constructed by External Merge as in (62).
The two prepositions bear valued θ-features and merge with nominals bearing the cor-
responding interpretable θ-features. The PP bearing [rem] merges first with buy, since
[rem] is (by assumption) ranked lowest on the hierarchy of features. The PP bearing
[src] then merges with the structure buy for $200.
(62) VP
PP V′
from Mary buy PP

[uSRCval[2]] [iSRC[2]] [uAG[ ]]
[uTHval[ ]] for $200
AGREE! [uSRC[2]] [uREMval[1]] [iREM[1]]
[uREM[1]] AGREE!
Note that in the merge of the second PP, agreement between the interpretable source
feature on Mary and its uninterpretable, unvalued counterpart on buy appears to violate
the c-command condition on probe-goal relations. I must assume here (and hereafter)
that PP structure quite generally does not “count” for agreement relations, at least with
pure thematic prepositions like the ones involved (see also Williams 1994). As has fre-
quently been noted, this assumption seems independently required for binding-theoretic
relations as well (cf. a letter [from Mary] [about/to herself]).
Consider now the addition of the theme nominal. Conceptually, the simplest option
would be to merge Fido directly with the structure in (62), with the interpretable theme
feature on Fido agreeing with its valued counterpart on buy:
(63) VP
Fido V′
[iTH[3]]
PP V′
from Mary buy PP

[iSRC[2]] [uAG[ ]]
[uTHval[3]] for $200
AGREE? [uSRC[2]]
Note, however, that such an analysis would crucially require the θ-feature agreement
relation to extend across another element—the source phrase from Mary—that also
stands in a θ-feature agreement relation to buy and occurs closer to V than the theme.
Such a situation is in fact forbidden under standard assumptions about agreement.
More formally, if F is a class of related formal features (Case features, φ-features, etc.),
agreement falls under the following Minimality Constraint:
Minimality Constraint on Feature Agreement
For any α, β, and γ, α cannot agree with a feature from F on γ across a closer β that
also agrees with a feature from F on γ.22
Assuming θ-features constitute a class F to which minimality is sensitive, this con-

straint will exclude external merger of an additional specifier as a means by which verbs
can project an additional argument as in (63) since agreement with the head will be
impossible. How then can further structure building proceed?
In answer to this, I will assume an extension of the light v mechanism introduced
in (53) above. Specifically, I will suppose that in a Numeration containing V we are
permitted to co-select not only a “thematic” light v, one bearing a strong V feature, an
EPP feature, and a valued θ-feature, but also a “pure” light v, one bearing only the first
two, and no θ-feature at all. More formally:
Light v
Bears a strong V feature and an EPP feature and and may bear a valued occurrence
of a θ-feature that is unvalued on the V.
The new possibility represented by this change is shown schematically in (64) (cf. 53).
Thus, in the Numeration V may “spin off” a pure light v, bound to it by a strong V feature
(64a), with which it must reintegrate later in the derivation (64b):
(64) a. V v b. v
[u 1 [ ]] Strong V
[u 2val[ ]] v V
[u 1 [ ]]
[u 2val[ ]]
Applied to the present case, this means the Numeration for (61) can include a pure
light v, co-selected with buy. Merging this v with (62) and raising V through v’s strong
V feature yields (65a). The theme can now externally merge as the specifier of little v,
with θ-agreement as in (65b) and no violation of minimality. V-Raising resolves the
intervention problem encountered in (63):
(65) a. vP
v VP
buy v PP V′
[uAG [ ]]
[uTHval[ ]] from Mary buy PP
[uAG [ ]]
b. vP
Fido v′
[iTH[3]]
v VP
buy v PP V′
[uAG [ ]]
[uTHval[3]] from Mary buy PP
AGREE! [uAG [ ]]
The reminder of the derivation now proceeds as before: via merger of a thematic little
voice head and V-Raising, followed by projection of the agent nominal (66).
(66) vP
John v′
[iAG[4]]
v vP
v v Fido v′
[uAGval[4]]
buy v v VP
[uAG[4]]
buy v PP V′
[uAG[ ]]
AGREE! from Mary buy PP
[uAG[ ]]
for $200
As with P, I assume that little v presents no barrier to c-command, permitting apparent

“upward” agreement in (66) between the unvalued [ag] feature on buy and the valued
occurrence of [ag] on little v.23 The higher agent nominal (John) agrees with the the-
matic voice head.
4.3. Properties of the Revised Account

The general picture of structure building sketched above, grounded in the neo-Davidsonian
K Analysis and its implementation through θ-features and agreement, yields basic shell
structure with no appeal to X-bar theory, but also without appeal to either full or partial
lexical decomposition. As in the original shell theory, light v makes its appearance on
purely formal, structural grounds, here having to do with distribution of θ-features, the
nature of agreement, and minimality constraints on the agreement relation. Light ele-
ments like v have the effect of allowing the derivation to “separate” specifier arguments
in such a way that their agreement relations to the head never cross (67):
(67) xP
x′
x xP
x x x′
X x x XP
X x X′
X
The resulting account appears to faithfully preserve all the essential properties of the
original shell structure analysis and its relations to Right Wrap. Given its basic account
of structure building, the analysis clearly preserves the generalized asymmetry between
arguments of a predicate. For any two arguments α and β, either α asymmetrically
c-commands β or the converse (67).
The possibility of “pure” light v’s also preserves the recursivity of the original
account. We’ve already seen the possibility of derivations involving two little v’s, one a
voice head bearing [ag] and one a pure light element. In fact, the use of the latter can
in principle be indefinitely extended. Suppose, for instance, that [loc] were added to
our list of θ-features and ranked hierarchically beneath [rem]. Suppose further that buy
in (68a) were analyzed as bearing an unvalued instance of [loc]. Then there would be
no barrier to extending our analysis of (61) to (68a). Adding two pure light v’s to the
Numeration, and merging them in the lowest stages, would permit the right-descending
derivation shown in (68b), involving recursive little v:
(68) a. John bought [Fido] [from Mary] [for $200] [in NYC].
b. vP
John v′
v vP
v v Fido v′
v v v vP
buy v v v v′
buy v from Mary v VP
buy v V′
for $200 buy PP
in NYC
Larson (2004) analyzes sentence-final adverbial “adjuncts” precisely as oblique

complements combining with V in this way, before any other arguments. Such items
will, accordingly, always invoke recursive small v structures whenever they com-
bine with a ditransitive, or whenever more than one such adjunct combines with a
transitive.24
Finally, the revised shell theory preserves the original property of category neutral-
ity. The derivations offered above involve only two components beyond the standard
ones in syntactic theory: (i) a ranked set of θ-features and (ii) the availability of light
heads, either pure or bearing valued θ-features. Neither of these two is tied essentially to
the verbal domain. We have already observed, following Jackendoff (1977) that adjec-
tives like dependent appear to involve the same θ-roles as the verb depend (recall 22).
Assuming that thematic and pure “little a’s” are available on parallel grounds to little
v’s, derivations like (69) will be available with minimal elaboration:
(69) VP
be aP
Max a′
a AP
dependent a PP A′
on Mary dependent PP
for help
Similarly, under the proposals in Larson (1991a), where quantifiers, including determin-
ers and degree elements, are analyzed as θ-feature-bearing items, and where relative and
comparative clauses are analyzed as inner complements of D and Deg (respectively),
Right Wrap derivations with nominals like (21b) above can be reconstructed as in (70a),
where we invoke a little d element that attaches to a projection of D and attracts the
“big D” determiner. Similarly for comparative structures like taller than John grew (as
in Mary grew taller than John grew) in (70b), where -er (more, less, as, etc.) is analyzed
as a degree quantifier:
(70) a. dP b. degP
Pro d′ Pro
d DP deg DegP
the d man D′ -er deg tall
the CP -er CP
that John saw than John grew
5. CONCLUSION
In this General Introduction I have reviewed some of the historical and conceptual
background of shell theory, several of the theoretical problems that it encountered,
and attempts to resolve them involving appeals to lexical decomposition. I also devel-
oped an alternative, based on an approach to neo-Davidsonian semantics that I called
the “K Analysis.” The latter affords a radically syntactic view of selection, wherein
the θ-features resident on predicates, prepositions, and voice heads are never semanti-
cally interpretable on them, and hence where these items never serve to “introduce
arguments” in any semantic sense. I have demonstrated how this new account, unlike
the decompositional approaches, preserves the core properties of shell theory, allowing
strikingly similar derivations free of the problems noted for its original counterpart. A
keystone in this new analysis is the account of little v as a light element, either bearing
a single θ-feature (voice heads) or no θ-feature at all (“pure” little v).
In the part introductions that follow, I will assume the theory developed above as
background and show how it allows us to reconstruct some of the key uses to which
shell theory was originally put, including the analysis of dative voice alternations, com-
plex predicates, and the projection of nominal structures. In each case, as I hope to
show, substantial improvements result, both in the depth of understanding of the phe-
nomena in question and in the empirical “reach” of the account.
NOTES
1. These ideas are developed more fully in Larson (in preparation).

2. Bach (1979) explicitly acknowledges the connection of his own proposals to those of
Chomsky (1955/1975).
3. In fact, composition of this kind had already been suggested by Chomsky (1955/1975) and
developed within early versions of generative grammar, such as Fillmore (1965).
4. For more discussion, see the introduction to part 3.
5. Unary function structure might be “derivatively semantic,” however. Consider the framework
of Heim and Kratzer (1998), wherein semantic composition proceeds by function-argument
application applied to binary syntactic trees. In such a framework, unary function structure
might be viewed as the product of these assumptions; in effect, the binary structure of trees
must be projected into the semantic functions that interpret them, in order for the latter to
be able to interpret the former. Heim and Kratzer invite potential confusion on this point
in their discussion of lexical prominence relations and their relation to syntax. They write:
“Our system of type-driven interpretation principles implies a rather strong claim about the
linking of the verb’s arguments to syntactic positions. The lexically determined prominence
relations must be preserved in the syntax.” (p.54, my emphasis). In fact, there is no such
implication, at least if “preserved” is read to imply that prominence relations are determined
by verb meaning, and those relations then determine the way in which the corresponding
arguments combine in syntax. Again, it is equally plausible within Heim and Kratzer’s own
system to view the syntax as determining prominence relations among arguments, with the
latter (derivatively) determining the semantic functions that interpret the syntax.
6. It is important not to confuse the unary function structure assumed by Montague with the
requirement that arguments of a predicate be ordered. The latter is semantic structure in the
sense intended above. All treatments of formal languages must assume that relations have a
way of distinguishing their arguments—what Russell (1903) called the “direction” of a rela-
tion Russell (1903). We must distinguish the two arguments of love(x,y), for example, since
it matters truth-conditionally who is the lover (x) and who/what is the loved (y). We thus
assume that arguments of a predicate come in an order, express them as an ordered pair, and
so on. But this is independent of unary function structure. Even assuming that the arguments
of love’ combine one at a time, we still need to say which of its arguments combines first, and
which combines second. Thus, whether the function is structured is independent of the direc-
tion of the relation: the latter must always be independently assumed, but not the former.
7. This trend also occurred equally within Generalized Phrase Structure Grammar, as it devel-
oped into modern Head Driven Phrase Structure Grammar (Pollard and Sag 1994).
8. See the introduction to part 2 for more discussion of complement asymmetries.
9. Tentative evidence for the structural asymmetry between the PPs in (18) might be taken
from the fact that the for- and from-phrases seem to show superiority effects (ia,b) and an
asymmetry in quantificational binding possibilities (iia,b):
(i) a. From whom did you buy a dog for how much money?
b. For how much money did you buy a dog from whom?
(ii) a. Mary bought a dog from every owner for a sum of his choosing.
b. Mary bought a dog from its bidder for every sum bid.
10. Cf. also dependent/depend on Mary to help him, from Jackendoff (1977), where the comple-
ments are PP and CP.
11. See Lin (2001) for some discussion of these concepts.
12. See McCawley (1968, 1970) for the original statement of these ideas.
13. For a sustained, detailed and rigorous attempt to work out the Generative Semantics
Program see Seuren (1996), which engages these issues directly.
14. Some of the properties noted here are captured in other shell-like theories. For example,
the theory of Bowers (1993), which invokes a Pred head analogous in position to the upper
empty v/V, shows category-neutrality, given that its base notion of predication is not verb
specific and extends across the range of lexical categories; the Pred element is not similarly
recursive, however, since predication is relevant only to the highest argument, the subject.
Likewise, the “Cascade” Theory of Pesetsky (1995) allows for recursion and rightward
descent, although it does not invoke any single recursive element equivalent to v/V in shell
theory in order to achieve this.
15. The proposals that follow are developed further in the other introductory sections of this
volume, and especially in Larson (in preparation).
16. See Larson and Segal (1995), Heim and Kratzer (1998), and Chierchia and McConnell-
Ginet (2000).
17. Higginbotham’s proposal can in fact be seen as a version of Categorial Grammar, like that
adopted in Montague (1974), in which the syntactic category of an item directly encodes its
combinatory possibilities. Thus, corresponding to a “grid” <1,2>, Montague (1974) assigns
a category t/e/e. And corresponding to Higginbotham’s “starring” operation, Montague
(1974) employs a cancelation operation, so that combining an expression of category t/e/e
with one of category e yields an expression of category t/e, and so on.
18. The analysis in Krifka (1992) differs from what I am calling the “K Analysis” in important
details; nonetheless, the basic idea of associating θ-relations with arguments, and not with
predicates, and the syntactic view of selection that this implies are present in Krifka (1992).
19. This comes very close to Frege’s original view of predicates as “incomplete” or “unsatu-
rated” expressions containing a fixed number of “open positions” that require arguments
to complete, saturate, or fill them and thereby express a complete thought.
20. An alternative way of securing the presence of little v with unergatives is to take them
to involve an overt or implicit cognate object, so that John sneezed is underlyingly John
sneezed a sneeze. Unergatives then reduce to the general case of transitives like kiss, with no
special stipulation about [ag] required.
21. It is interesting in this connection that languages often appear to develop morphologically
independent light verbs associated with agentivity, like do (English), suru (Japanese), and ha
(Korean). It’s possible that this tendency to expressly atomize [ag] and give it its own special
form is linked to the nonavailability of valued [ag] on other Vs.
22. I assume the standard definition according to which β is closer to γ than α, just in case α
asymmetrically c-commands β, and β asymmetrically c-commands γ. Note that this defini-
tion entails that the remuneration PP for $200 is not closer to buy than the source PP from
Mary in (62), and hence does not interrupt the latter’s agreement relation to V.
23. Alternatively one might assume that [iag [ ]] on Mary agrees with both [ag] features below
it, and that the little v sister of buy presents no minimality barrier to this.
24. For more on this, see the introduction to part 2, on complex predicates.
REFERENCES
Adger, D. (2003) Core Syntax: A Minimalist Approach. New York: Oxford University Press.
Bach, E. (1979) “Control in Montague Grammar,” Linguistic Inquiry 10: 515–531.
Barwise, J. and R. Cooper (1981) “Generalized Quantifiers and Natural Language,” Linguistics
and Philosophy 4: 159–219.
Bowers, J. (1993) “The Syntax of Predication,” Linguistic Inquiry 24:591–656.
Chierchia, Gennaro and Sally McConnell-Ginet (2000) Meaning and Grammar: An Introduction
to Semantics. Cambridge, MA: MIT Press.
Chomsky, N. (1955/1975) The Logical Structure of Linguistic Theory. New York: Plenum Press.
——— (1981) Lectures on Government and Binding. Dordrecht: Foris.
——— (1986) Knowledge of Language. New York: Praeger.
——— (1991) Some notes on economy of derivation and representation. Reprinted in Chomsky
(1995).
——— (1993) “A Minimalist Program for Linguistic Theory,” in K. Hale and S. J. Keyser, eds.,
The View from Building 20 (1-53). Cambridge, MA: MIT Press.
——— (1994) “Bare Phrase Structure,” MIT Occasional Papers in Linguistics. 5 Cambridge, MA:
MIT Department of Linguistics and Philosophy, MITWPL.
——— (1995) The Minimalist Program. Cambridge, MA: MIT Press.
Couquaux, D. (1981) “French Predication and Linguistic Theory,” in R. May and J. Koster, eds.,
Levels of Syntactic Representation (pp. 33–64). Dordrecht: Foris.
Dowty, D. (1978) “Governed Rules as Lexical Transformations in Montague Grammar,” Linguistic
Inquiry 9: 393–426.
——— (1979) Word Meaning and Montague Grammar. Dordrecht: Kluwer.
——— (1982) “Grammatical Relations and Montague Grammar,” in P. Jacobson and G. Pullum,
eds., The Nature of Syntactic Representation (pp. 79–130). Dordrecht: Reidel.
Emonds, J. (1976) A Transformational Approach to English Syntax: Root, Structure-Preserving,
and Local Transformations. New York: Academic Press.
——— (1980) “Word Order and Generative Grammar,” Journal of Linguistic Research 1: 33–54.
Fillmore, C. (1965) Indirect Objects and the Ordering of Transformations. The Hague: Mouton.
Harley, H. and R. Folli (2007) “Causation, Obligation, and Argument Structure: On the Nature
of Little v,” Linguistic Inquiry 38: 197–238.
Heim, I, and A. Kratzer (1998) Semantics in generative grammar. Malden, MA: Wiley-Blackwell.
Higginbotham, James (1985) On semantics. Linguistic Inquiry 16: 547–593.
Hornstein, N. (1999) “Movement and Control,” Linguistic Inquiry 30: 69–96.
Jackendoff, R. (1977) X-bar Theory: A Study of Phrase Structure. Cambridge, MA: MIT Press.
Jacobson, P. (1983) “Phrase Structure and Grammatical Relations,” LSA Forum Lecture, Linguistic
Society of America Annual Meeting, Minneapolis, MN.
——— (1984) “Verb-Particle Constructions in Phrase Structure Grammar,” invited colloquium
presentation, University of Pennsylvania, Philadelphia. December 1984.
——— (1987) “Phrase Structure, Grammatical Relations and Discontinuous Constituents,” in G.
Huck and A. Ojeda, eds., Syntax and Semantics 20: Discontinuous Constituency (pp. 27–69).
New York: Academic Press.
Kayne, R. (1994) The Antisymmetry of Syntax. Cambridge, MA: MIT Press.
Keenan, E. and Y. Stavi (1983) “A Semantic Characterization of Natural Language Determiners,”
Linguistics and Philosophy 9: 253–326.
Kratzer, A. (1996) “Severing the External Argument from Its Verb,” in J. Rooryck and L. Zaring,
eds., Phrase Structure and the Lexicon (pp. 109–137). Dordrecht: Kluwer.
Krifka, M. (1992) “Thematic Relations as Links between Nominal Reference and Temporal Con-
stitution,” in I. A. Sag and A. Szabolcsi, eds., Lexical Matters (pp. 29–54). Stanford, CA: CSLI
Publications.
Larson, R. (1988) “On the Double Object Construction,” Linguistic Inquiry 19: 335–391.
——— (1989) “Light Predicate Raising,” MIT Lexicon Project Working papers 27. Cambridge,
MA: MIT Republished in this volume.
——— (1990) “Double Objects Revisited: Reply to Jackendoff,” Linguistic Inquiry 21: 589–632.
——— (1991a) “The Projection of DP (and DegP),” unpublished manuscript, Stony Brook Uni-
versity Stony Brook, NY.
——— (1991b) “Promise and the Theory of Control,” Linguistic Inquiry 22: 103–139.
——— (1991c) “Some Issues in Verb Serialization,” in C. Lefebvre, ed., Verb Serialization
(pp. 185–208). John Benjamins.
——— (2004) “On Sentence-Final Adverbs and ‘Scope,’” in M. Wolf and K. Moulton, eds., Pro-
ceedings of NELS 34. (pp. 23–44) Amherst, MA: GLSA, University of Massachusetts-Amherst.
——— (in prep) VP and DP.
Larson, Richard and Gabriel Segal (1995) Knowledge of meaning Cambridge, MA: MIT Press.
Lin, T.-H. (2001) Light Verb Syntax and the Theory of Phrase Structure. Doctoral dissertation,
University of California, Irvine.
McCawley, J. D. (1968) “Lexical Insertion in a Transformational Grammar without Deep Struc-
ture,” in Bill J. Darden, Charles-James N. Bailey, and Alice Davison, eds., Papers from the Fourth
Regional Meeting, Chicago Linguistic Society (pp. 71–80). Chicago: Linguistics Department,
University of Chicago.
——— (1970) “English as a VSO-Language,” Language 46: 286–299.
Montague, Richard. (1974) Formal Philosophy. New Haven, CT: Yale University Press.
Parsons, Terence. (1991) Events in the Semantics of English, Cambridge, MA: MIT Press.
Pesetsky, D. (1995) Zero Syntax. Cambridge, MA: MIT Press.
Pesetsky, D. and E. Torrego (2004) “The Syntax of Valuation and the Interpretability of Features,”
unpublished manuscript, MIT and University of Massachusetts, Boston.
Pylkkänen, Llina (2002) “Introducing Arguments.” Ph.D. thesis, MIT.
——— (2008) Introducing Arguments. Cambridge, MA: MIT Press.
Pollard, C. and I. Sag (1994) Head-Driven Phrase Structure Grammar. Chicago: University of
Chicago Press.
Rosenbaum, P. (1970). “A Principle Governing Deletion in English Sentential Complementation,” in
R. Jacobs and P. Rosenbaum (eds.), Readings in English Transformational Grammar (pp. 20–29).
Waltham, MA: Ginn.
Russell, B. (1903). Principles of mathematics. London: Routledge.
Seuren, P. (1996) Semantic Syntax. New York: Wiley-Blackwell.
Smith, C. (1964) “Determiners and Relative Clauses in Generative Grammar,” Language 40:
37–52.
Stowell, T. (1978) “What Was There before there Was There?” in D. Farkas, Wesley M. Jacobsen,
and Karol W. Todrys, eds., Papers from the Fourteenth Regional Meeting of the Chicago Lin-
guistics Society (pp. 457–471). Chicago: University of Chicago.
——— (1981) Origins of Phrase Structure. Doctoral dissertation, Massachusetts Institute of Tech-
nology, Cambridge, MA.
Williams, Edwin (1994) Thematic Structure in Syntax. Cambridge, MA: MIT Press.
Datives: Background
Shell theory was first developed to provide an analysis of dative constructions. In 1985,
as part of a workshop conducted at the Linguistic Society of America Summer Institute
in New York City, I became interested in the issue of control in constructions involving
the English verb promise. It seemed to me that a correct understanding of why and how
this verb (and several like it) violates the otherwise reliable Minimal Distance Principle of
Rosenbaum (1967) must involve its status as a “Dative Shifting verb”—roughly speak-
ing, a verb allowing its theme and goal arguments to invert in surface form. Upon further
research, I discovered that others had made the same connection between subject control
and Dative Shift, notably Bowers (1973) and later Culicover and Wilkins (1984), but,
equally, that none of these authors had an explicit theory of the operation. Indeed, the idea
of a derivational relation between prepositional datives and double object constructions
had, for all intents and purposes, been abandoned within transformational grammar after
the work of Oehrle (1976).1 One of my core motivations for investigating datives and
Dative Shift was to provide an explicit analysis of control relations with promise.
A second stimulus was a short paper circulating at the same Linguistic Society of
America Institute workshop concerning anaphora in double object constructions. The
authors, Andrew Barss and Howard Lasnik (1986), drew attention to domain asym-
metries like those in (1) and (2) in PP and double object datives, and noted, in this con-
nection, a simple puzzle: whereas the asymmetries in (1a–d) can be explained without
recourse to linearity, by appeal to the PP node in (3a), the asymmetries in (2) cannot—at
least not on the simplest structural hypothesis available at the time (3b). In the latter
NP1 and NP2 are structurally symmetric:
(1) a. Mary showed John to himself/*Mary showed himself to John.
b. Mary sent every check to its owner/*Mary sent his paycheck to every worker.
c. Mary gave nothing to anyone/*Mary gave anything to no one.
d. John sent each girl to the other’s parents/
*John sent the other’s daughter to the each parent.
(2) a. Mary showed John himself/*Mary showed himself John.
b. Mary sent every worker his paycheck/*Mary sent its owner every paycheck.
c. Mary gave no one anything/*Mary gave anyone nothing.
d. John sent each parent the other’s daughter/

*John sent the other’s parent each daughter.
(3) a. VP b. VP
V NP1 PP V NP1 NP2
show John P NP2 showed John himself
to himself
The conclusion, evidently, is that either the domain relations present in (1) and (2) are
sensitive to linear order, requiring reference to c-command plus precedence in describing
what is in the domain of what, or else the structure in (3b) is wrong (and quite possibly
that in (3a) as well). I was struck by this puzzle, in part because I thought I already knew
the answer to it.
As discussed in the General Introduction, derivations for prepositional datives in
Montague Grammar assumed an asymmetry between verbal arguments deeper than
that reflected in (3a). In (4a) the direct object (Fido) is derivationally superior to the
prepositional object (Mary), quite independently of structure introduced by PP. Dowty
(1978) had proposed an analysis within this framework in which such argument asym-
metry was also present in double object derivations (4b).2
(4) a. John give Fido to Mary b. John give Mary Fido
John give Fido to Mary John give Mary Fido
Fido give to Mary Mary give Fido
give to Mary give Fido
A translation of this approach into current phrase structural terms promised a solution
to the Barss and Lasnik puzzles, a way of thinking about argument asymmetry not rep-
resented in generative grammar at the time.
Montague Grammar–style derivations also suggested an attractive way of updating
“Dative Shift” and reintroducing it in a modern transformational context. Derivations
like (4a,b), when rendered as VP shell structures, had the effect of inducing a struc-
tural, c-command asymmetry between the theme and goal arguments comparable to
that existing between an agent and a theme (5a). Hence a natural analogy in thinking
about a Dative Shift derivation that would invert theme and goal was the well-known
passive derivation, which inverts agent and theme.3 Passive raises the theme to an empty,
subject-like A position, with the agent phrase realized as an adjunct (5b). Reasoning in
parallel, Dative Shift might raise the goal to an empty, object-like A position, with the
theme phrase realized as an adjunct (5c).
(5) a. [vP John [v′ write [VP [DP a letter] [v′ write [PP to Mary]]]]]
Asymmetric c-command
b. [VP a letter [v′[v′ written a letter ] [PP by John ]]]

Passive
c. [vP John [v′ write [VP Mary [v′[v write Mary ] [DP a letter]]]]]
Dative Shift
The analogy between passives and double object constructions seemed attractive on
various grounds. Beyond word order, a salient difference between PP datives and double
object constructions is the absence in the latter of to, a governed preposition having largely
the status of Case marking. The “disappearance” of to under Dative Shift might thus be
analogized to the suppression of accusative Case under Passive. Another property distin-
guishing PP datives and double objects is the inaccessibility of the theme to A-movement (cf.
A letter was written to Mary vs. ?*A letter was written Mary). This might be ascribed to the
adjunct status of the theme after Dative Shift, comparable to the adjunct status of the agent
after Passive. Furthermore, the presence of overt applicative morphology in many languages
showing dative alternations seemed comparable to the morphology found in passives, and
indeed in the Africanist descriptive tradition passive and applicative morphology are rou-
tinely grouped together as “voice” inflection. Finally, the inversion of arguments effected by
Dative Shift allowed for a recapturing of the insights of Bowers (1973) regarding control
with promise. If the control infinitive is analyzed as the underlying theme (denoting what is
promised), then the subject will in fact be the closer controller in PP dative forms like (6a,b):
(6) a. John promised to leave to Mary.

AGENT THEME GOAL
b. [vP John [v′ promise [VP [CP PRO to leave] [v′ promise [PP to Mary]]]]]
Control
It will also be the closer controller in the underlying form of (7a), even if Dative Shift
ultimately raises the goal to a position higher than theme before Spell-Out:
(7) a. John promised Mary to leave.

AGENT GOAL THEME
b. [vP John [v′ promise [VP Mary [v′ [V promise Mary ] [CP PRO to leave ]]]]]
Control + Dative Shift
Thus if control relations are established in underlying form, the subject control nature
of promise could also follow the Minimal Distance Principle. These themes were devel-
oped and discussed fully in Larson (1988) and Larson (1991).
1. PROJECTING DATIVES
The work described above was part of a general reawakening of theoretical interest in
the dative alternation that began in the late 1980s and has continued unabated to the
present. But while interest in dative structures has been strong, the same cannot be said
for derivational approaches to them. Apart from a small number of studies, includ-
ing Arregi (2003), Baker (1988, 1996a), Bleam (2002), Demonte (1995), den Dikken
(1995), Harada and Larson (2007, 2009), Larson (1988), Oba (2002), Ormazabal and
Romero (2010), and Zushi (1992), the dominant approaches in this area have been non-
derivational. Overwhelmingly, PP datives and double object datives have been analyzed
as independent constructions, separately projected.
1.1. What Kind of Information Is Relevant to Projection?

The claim that two constructions are projected separately is the claim that elements
of the two constructions differ, or differ in properties, in ways that are relevant to
projection itself. It is important to be clear about what properties are relevant in evaluat-
ing and justifying a claim of this kind. For example, it has long been recognized that PP
dative and double object pairs like (8a,b) differ subtly insofar as the latter (8b) implies
more strongly than the former (8a) that the children actually learned French. Similarly,
Kayne (1975) observes that (9b) seems to imply more strongly than (9a) that the baby
in question has already been delivered or is in existence. Accordingly, (9b) is less natural
than (9a) uttered by a pregnant woman or a woman planning to become so.
(8) a. I taught French to those children.
b. I taught those children French.
(9) a. I knitted this sweater for our baby.
b. I knitted our baby this sweater.
Nonetheless, such “extra” elements of meaning in the double object form seem to be
at most conversational implicatures given their unproblematic cancellability in simple
continuations like (10a,b):
(10) a. I taught those children French for a whole year. But the final exam results
showed the little brats hadn’t learned a damn thing.
b. I knitted our baby this sweater. I just hope it fits him when he’s born.
Theories of the pragmatics-grammar interface routinely assume that the former is inter-
pretive, not generative: that contributions like conversational implicatures are calcu-
lated from the syntax and semantics rather than determining syntactic projection in the
first place. And indeed such a view looks plausible for pairs like (8) and (9). Occurrence
of a nominal in direct object position often correlates with a reading of affectedness for
its referent, 4 and hence a speaker, in choosing (8b)/(9b) over (8a)/(9a), might well be
taken to implicate that the referent of the indirect object is affected by the action—that
the children were affected by the teaching (e.g., by learning what was taught), or that the
baby is affected by the knitting and, accordingly, in a position to be affected (e.g., by
being in existence already). If this is correct, then contrasts like those in (8) and (9) are
not relevant to the question of dative projection, and are in fact compatible with both
derivational and nonderivational views, the extra element of meaning in the double
object form being “read off the surface,” rather than figuring in its projection.
In general, differences of projection among verbs have been widely assumed to be
traceable either to semantic notions associated with relations that individuals or entities
can bear to the event or state that the verb describes—so-called thematic relations or
θ-roles—or, in Generative Semantics–style models, to core predicates with which such
notions are associated. To give an example, in Larson (1990) it is proposed that the
two blame constructions in (11a,b) are projected separately and are not derivationally
related. The suggestion is that blame, in its (11a) use, is fundamentally a dative verb
parallel to thank and give thanks in (12). By contrast, blame, in its (11b) use, is funda-
mentally a locative verb parallel to put and put blame in (13).
(11) a. Job blamed God for his troubles.
b. Job blamed his troubles on God.

(12) a. Job thanked God for his blessings.
b. Job gave God thanks/gave thanks to God for his blessings.
(13) a. Job put his troubles on God.
b. Job put the blame for his troubles on God.
Assuming dative and locative verbs to involve distinct sets of thematic roles, or distinct
decompositions, we can justify separate projections for these structures. For example,
(11a,b) might be given derivations as in (14a,b) (respectively), employing a theory
of projection like that sketched in the General Introduction, and where the relevant
θ-features are indicated.
(14) a. vP b. vP
Job v′ Job v′
iAGENT iAGENT
v VP v VP
blame v God V′ blame v DP V′
iGOAL
blame PP PP
his troubles blame
iTHEME
for his troubles on God
iOBLIQUE iLOC
Alternatively, we might appeal to lexical decompositions like (15a,b) (respectively),

where the dative sense of blame is analyzed in terms of caused possession, and where
the locative sense is analyzed in terms of caused motion:
(15) a. VP b. VP
DP V′ DP V′
Job V VP John V VP
CAUSE DP V′ CAUSE DP V′
God V DP blame for V PP

his troubles
HAVE blame for GO on God
his troubles
On the former analysis, the verb blame is assigned two distinct lexically specified θ-sets.
On the latter analysis, surface blame spells out two distinct sets of merged predicates:
‘cause-to-have/get-blame’ and ‘cause-blame-to-go’. And so on.
1.2. Separating Double Object Constructions from Oblique Locatives

Separation of projections for the two senses of blame in (11a,b) seems relatively straight-
forward given differences in word order, alternations in preposition choice, and the
distinct semantic intuitions reflected in the distinct paraphrases in (12) and (13). The
situation becomes more complex, however, with the dative alternation. Many analysts
have assumed that reasoning analogous to that with blame yields a parallel case for
separate projection with double object and PP dative structures. For example, Pinker
(1989), Goldberg (1995), Harley (1995, 2002), Krifka (1999), Richards (2001), and
Beck and Johnson (2004), among many others, have assumed that double object verbs
encode caused possession whereas PP dative verbs encode caused motion or caused
location.
The case for this analysis is most persuasive with verbs of transmission and ballis-
tic motion like send and throw. Consider (16a,b) analyzed as in (17a,b) (respectively)
where we put aside questions of how exactly the merged abstract predicates are spelled
out appropriately as send.
(16) a. John sent Mary a telegram.
b. John sent a telegram to Mary.
(17) a. VP b. VP
DP V′ DP V′
John V VP John V VP
CAUSE DP V′ CAUSE DP V′
Mary V DP a telegram V PP
HAVE a telegram GO to Mary

send = CAUSE to HAVE (by sending) send = CAUSE to GO (by sending)
In (17a) Mary is represented as a possessor/recipient, whereas in (17b) Mary is repre-

sented as a goal or target of motion.
Beyond the attractive simplicity of mapping one meaning (possession/motion) to one
form (double object/oblique), the analysis in (17) has both intuition and many distribu-
tional and semantic facts to recommend it. On reflection, it surely does seem that one
can conceive of send in two rather different ways: (i) as describing a means of bringing
something into someone’s possession—by sending it, as opposed to handing or throw-
ing it, or (ii) as describing a manner of moving an object from one point to another. Fur-
thermore, this divergence in sense appears to correctly track distributional differences
in the two forms. For example, the class of possessors is plausibly restricted to animates
like persons—only people can get or receive things—whereas the class of motion targets
has no similar conceptual restriction. This leads us to expect that nonanimates will be
able to occur in PP dative constructions as the object of to but not in the corresponding
double object form. This expectation is confirmed by well-known contrasts like (18a,b).
Although the oblique construction is fully acceptable, the double object form (18b) is
excluded unless Lisbon is assigned some special, anthropomorphic interpretation.
(18) a. Pilar sent a telegram to Lisbon.
b. *Pilar sent Lisbon a telegram.

This general line of reasoning appears extensible to a wide range of cases. Along with
examples (18a,b), in which the to-object is not a potential recipient/possessor, there are
examples with verbs like distribute, disperse, donate, gave away, gave out, release and lose
that encode release or loss of possession but no target possessor or recipient (19a,b) and
(20). Here too double object forms are blocked. Consider also examples expressing physi-
cal or metaphorical motion along a path, and arguably no notion of possession at all. Again
these are available in the PP dative but not in the double object construction (21)–(22).
(19) a. John distributed/dispersed/donated/gave away/gave out money (to/among those
charities).
b. *John distributed/dispersed/donated/gave away/gave out those charities money.
(20) a. John lost money to the stock market.
b. *John lost the stock market money. (on the (20a) reading)
(21) a. John dragged the chair (over) to Bill.
b. *John dragged Bill (cover) the chair.
(22) a. John conveyed the visitors/the point to Bill.
b. *John conveyed Bill the visitors/the point.
Rappaport Hovav and Levin (2008) make the additional point that path modifiers
such as halfway/all the way are available with verbs like send and throw in their oblique
PP form (23a), but never in their double object form (23b).
(23) a. John threw a ball (halfway/all the way) to Mary.
b. John threw (*halfway/*all the way) Mary a ball (*halfway/*all the way).
This provides further support for a view of the former as motion constructions.
Finally, many analysts have drawn attention to the fact that, just as there are discon-
tinuous caused-motion idioms in the PP dative form with no double object counterparts
(24a–d) (from Larson 1988), there appear to be discontinuous caused-possession idioms
in the double object form with no PP dative counterparts (25a–c).
(24) a. Lasorda sent his starting pitcher to the showers.
(cf. *Lasorda sent the showers his starting pitcher.)
b. Mary took Felix to the cleaners.
(cf. *Mary took the cleaners Felix.)
c. Felix threw Oscar to the wolves.
(cf. *Felix threw the wolves Oscar.)
d. Max carries such behavior to extremes.
(cf. *Max carries extremes such behavior.)

(25) a. John gave Mary the creeps.
(cf. *?John the creeps to Mary.)
b. John gave Mary flak.
(cf. *?John gave flak to Mary.)
c. John gave Mary the boot.
(cf. *?John gave the boot to Mary.)
Under the widely held view that phrases understood as semantic units are projected
into structure as syntactic units, the surface-discontinuous items in (24) and (25)—if
they are truly idioms—would require distinct representations as initial constituents in
underlying form (26a,b):
(26) a. VP b. VP
DP V′ DP V′
Mary V DP Oscar V PP
give the creeps throw to the wolves
idiomatic unit idiomatic unit

DOUBLE OBJECT IDIOMS OBLIQUE DATIVE IDIOMS
The existence of example pairs like (24) and (25) thus provides strong potential evi-
dence for projecting possessive double object forms independently of motional PP dative
forms.
1.3. Separating Oblique Locatives from Oblique Possessives

Facts like those just rehearsed make a powerful case for granting separate projectional
status to double object constructions understood as encoding caused possession, and
PP dative constructions understood as encoding caused motion/location. Accepting this
conclusion entails that neither of the two constructions should be derived from the
other, whether by Dative Shift, or any other such syntactic means. Suppose we do accept
the conclusion. Does this then settle the question of dative projection altogether? And
does it close the space for a Dative Shift analysis? In fact, it does not.
Rappaport Hovav and Levin (2008) argue persuasively that the separation repre-
sented by (17a,b), while correct so far as it goes, does not go far enough, and that a
further separation must be made, namely, between PP dative constructions understood
as encoding caused-motion/location and PP dative constructions understood as encod-
ing caused possession. In other words, they argue that the oblique dative form is polyse-
mous. In addition to encoding caused motion/location, it also expresses an independent
caused-possession sense. In effect, under their view, structure (17b) should be further
divided into two cases, which we might represent with decompositional structures as in
(27b,c), and where an abstract version of belong represents the underlying predicate in
the oblique possessive construction (27b).
(27) a. VP b. VP c. VP
DP V′ DP V′ DP V′
John V VP John V VP John V VP
CAUSE DP V′ CAUSE DP V′ CAUSE DP V′
Mary V DP a telegram V PP a telegram V PP
HAVE a telegram BELONG to Mary GO to Lisbon

CAUSED POSSESSION CAUSED POSSESSION CAUSED MOTION/LOCATION
Corresponding to the threefold division of structures in (27), Rappaport Hovav and Levin
(2008) propose a threefold division of verbs. Thus, there are verbs like give, lend, and
hand, which exclusively encode caused possession. There are verbs like drag and convey,
which exclusively encode caused motion. And there are verbs like throw and send, which
are polysemous between the two. As indicated in (27), the oblique dative construction is
compatible with both caused motion and caused possession, whereas the double object
construction encodes only the latter in English. It follows from this that drag can occur only
in the oblique construction and only with the caused-motion reading (28a); give may occur
in either the oblique or the double object construction, but only with a caused-possession
meaning (29); and throw and send may occur in the PP construction with either meaning,
but in the double object construction with only a single meaning (30):
(28) a. John dragged the chair to Bill. caused motion
b. *John dragged Bill the chair.

(29) a. John gave a book to Mary. caused possession
b. John gave Mary a book. caused possession
(30) a. John threw a ball to Mary. caused motion or caused possession

b. John threw Mary a book. caused possession
In support of their view, Rappaport Hovav and Levin (2008) observe that just as
double object give constructions entail a corresponding possessive (31a,b) and hence
yield a contradiction when this entailment is denied (31c), the very same is true with
oblique give constructions (32a–c). Likewise, just as double object give resists nonani-
mate indirect objects, so does oblique give (31d)/(32d).
(31) a. John gave Mary the book.
b. Mary had/got the book.
c. #John gave Mary the book, but she never got/received it.
d. #Pilar gave Lisbon the book.

(32) a. John gave the book to Mary.
b. Mary had/got the book.
c. #John gave the book to Mary, but she never got/received it.
d. #Pilar gave the book to Lisbon.

Conversely, whereas verbs of caused motion allow their to-PPs to co-occur with path
modifiers like halfway ((23a) and (33a)), give and similar verbs reject such modifiers
even in the oblique form (33b); in other words, they pattern identically with double
object verbs, which uniformly reject such modifiers (33c):
(33) a. John dragged/threw/slid/sent the chair halfway to Bill.
b. *John gave/handed/lent the ball halfway to Bill.

c. *John gave/handed/lent/threw/slid/sent Bill the chair halfway.
Rappaport Hovav and Levin’s distinction can also be detected with interrogatives;
to-PPs can answer some where-questions with the verbs identified as expressing caused
motion, but not with verbs expressing caused possession (34). Where-questions are also
never felicitously answered with double object forms of verbs that can express both
meanings (35):
(34) a. Where did you deposit/lose/*give your money?
(cf. I deposited it in /to my checking account, I lost it in/to the stock market.)
b. Where did John drag/throw/slide/send/ship the chair?
(cf. John dragged/threw/slid/sent/shipped the chair (over) to Bill.)
(35) a. Where did John send/ship/mail the chair?
b. ?*John sent/shipped/mailed Bill the chair.
The general idea that verbs with oblique frames can express distinct possession and loca-
tion senses seems hard to resist given patterns found elsewhere in the grammar of English.
The verb belong, mentioned above, is a case in point. Belong shows a clear locative meaning
in examples like (36a–e), selecting a range of locative PPs. In this use its sense is something
like ‘object or entity X is suitably, customarily, or properly situated at location Y’.
(36) a. This chair belongs in the living room/halfway along that wall/there.
b. A: Where does this chair belong?
B: It belongs in there/*to Mary.
c. John belongs in prison.
d Memories belong in the past.
e. This information belongs in the public domain.
But, equally, belong exhibits a clearly distinct possessive meaning in examples like
(37a–d), where the dative to is not locative/directional (37b), where the possessed can
be material or abstract (37c), and where the notion of possession embraces not only
ownership but the part-whole relation as well (37d). In broad respects, then, belong
resembles verbs like throw and send under the proposals of Rappaport Hovav and Levin
(2008) insofar as it encodes both location and possession senses.5 And, as in the case of
throw and send, belong’s two senses are both projected in an oblique frame.
(37) a. This chair belongs to Mary.
b. A: To whom does this chair belong?
B: It belongs *in there/(*halfway) to Mary.
c. The book/fault/responsibility/credit belongs to John.
d. This piece belongs to that puzzle.
Given the clear polysemy in oblique dyadic forms like belong, it seems quite reasonable
to conjecture a similar one in oblique triadic forms like throw and send.
In discussing examples like (33)–(35), Rappaport Hovav and Levin (2008) consider
polysemy only with respect to verbs, but not the preposition to. We might nonetheless
ask whether the distinction between verbs taking a to-PP is accompanied by a distinc-
tion in the preposition itself. With constructions expressing caused motion, to appears
to contribute the notion of “goal of motion along a path.” This meaning is absent from
give + PP, and to itself is absent in double object constructions; hence this suggests that
to makes no semantic contribution in caused-possession frames and is present purely
for grammatical reasons (e.g., Case marking). This would in turn imply two to’s in
English: a true directional/allative P and a semantically empty Case marker. Monose-
mous caused-motion verbs (drag) would presumably select only the former, monose-
mous caused-possession verbs (give) would select only the latter, and polysemous verbs
(throw) would select either, depending on their sense.
Direct evidence for such a view can be found in other languages. In European Por-
tuguese the verb enviar ‘send’ can take a PP headed by either a or para ‘to’ when the
PP-object is both a potential goal of motion and a potential possessor (38a). However,
a is excluded when the PP-object denotes a nonrecipient (38b). Furthermore, the verb
dar ‘give’ selects only a (38c), whereas the verb empurrar ‘push’ selects only para (38d).6
(38) a. O João enviou uma carta à Mary/para a Mary.
the John sent a letter to Mary
‘John sent a letter to Mary’.
b. O João enviou uma carta *a Lisboa/para a Lisboa.
the John sent a letter to Lisbon
‘John sent a letter to Lisbon’.
c. O João deu um livro à Mary/*para a Mary.
the John gave a book to Mary
‘John gave a book to Mary’.
d. O João empurraram/arrastaram a mesa *à Mary/para perto da Mary.
the John pushed/dragged a table to Mary/to the vicinity of Mary
‘John pushed/dragged a table to Mary’.
A natural interpretation of these facts is that a and para realize the two separate senses
of English to hypothesized above. Thus a is a dative preposition in ditransitives and
compatible only with caused possession, whereas para is an allative and compatible
only with caused motion. Verbs that are unambiguous in sense (‘give’ and ‘drag’) will
allow only one of these forms. Verbs that are ambiguous will allow either, up to disam-
biguation.7
There is also interesting evidence for dative-related prepositional ambiguity to be
found in English. Consider first the Spanish example (39a). As discussed by Ormaza-
bal and Romero (2010), this sentence is ambiguous between two quite different senses.
On the simple locative interpretation, Mary placed some collection of legs (chair legs,
mannequin legs, frog legs, etc.) upon the table—on its surface. On the second, “assembly”
interpretation, Mary attached the legs to the table, making the former part of the latter.
Examples like (39b), having no pragmatically natural assembly interpretation, exhibit
only the simple locative meaning.
(39) a. Mary puso las patas en la mesa.
Mary put the legs on the table
‘Mary placed the legs upon the table.’ (locative meaning)
‘Mary attached the table’s legs.’ (assembly meaning)
b. Mary puso las ranas en la mesa.
Mary put the frogs on the table
‘Mary placed the frogs upon the table.’ (locative meaning)
‘#Mary attached the table’s frogs.’ (assembly meaning)
Interestingly, Spanish can replace the locative preposition en with dative a (adding
a clitic). In this construction, however, the sentence becomes unambiguous, having
only the assembly meaning (40a); the dative equivalent to (39b) is thus unacceptable
(40b):
(40) a. Mary le puso las patas a la mesa.
Mary Cl-3Dat put the legs to the table
‘Mary attached the table’s legs.’ (assembly meaning only)
b. #Mary puso las ranas a la mesa.
Mary put the frogs to the table
‘#Mary attached the table’s frogs.’ (assembly meaning only)
The Spanish examples do not reveal whether the ambiguity in (39a) arises from the
poner + P combination or from the P (en) alone; however, the English data in (41) appear
to clarify the point. Note that like Spanish (39a), English (41a) shows both the locative
and assembly interpretations. But observe further that this polysemy is preserved even
in simple copular constructions like (41b), where the verb put is absent, strongly sug-
gesting that the polysemy inheres in the preposition itself. Familiar conjunction tests like
(41c) support a diagnosis of ambiguity and not vagueness: the unambiguous locative
reading of on in the first conjunct forces a locative reading of on in the second conjunct,
as expected of a genuine ambiguity (Zwicky and Sadock 1975):
(41) a. John put the wheels on the truck. (ambiguous)
b. The wheels are on the truck. (ambiguous)
c. The shipping containers are on the truck and the wheels are too. (unambiguous)
The conclusion thus seems inescapable that dative alternations involve not merely
alternation in verbal properties, but concomitant alternation in prepositional features as
well. Constructions encoding caused motion deploy an allative P (42a,c), whereas those
encoding caused possession employ a special dative P (42b,c), often derived from an
allative or locative, and potentially employed elsewhere in the grammar with directional
or locative meaning, but nonetheless distinct in this usage.8
(42) a. John dragged the chair to Bill. caused motion Allative P
b. John gave a book to Mary. caused possession Dative P
c. John threw a ball to Mary. caused motion or Allative P

caused possession Dative P
Taking stock now, our situation appears to be the following. We have seen strong
evidence that double object constructions and oblique dative constructions encoding
caused motion/location are projected separately, with no derivational relations between
them. At the same time, we have convincing evidence for the existence of a separate set
of oblique dative constructions encoding caused possession very like that encoded by
double object constructions. Given this latter result, the question of syntactic related-
ness arises again. Even with oblique locatives separated off and accorded independent
projectional status, we may still ask: what is the relation between the two forms of the
possessive dative—oblique possessives and double object possessives? Are these deriva-
tionally related, or are they projected separately? This question is plainly more challeng-
ing than the one we encountered earlier in (11). If both forms encode possession, what
differences underwrite separate projection?
1.4. Dative Idioms Again

One approach to the question of derivation might be to return to a data set mentioned
earlier, namely, idioms. Recall that idioms represent a potential probe into underlying
structure on the view that what is understood as a semantic unit must be projected as a syn-
tactic unit. If we could find distinct idioms corresponding to the two structures involved,
this would provide powerful support for the view that those structures are independently
required, even in the absence of a theory of what separate projection is based on.
Although straightforward in theory, the results of the last section complicate this
test in practice. For example, reconsider (24a–d) (repeated below), examples of oblique
idioms with no double object variants:
(cf. *Lasorda sent the showers his starting pitcher.)
b. Mary took Felix to the cleaners.
(cf. *Mary took the cleaners Felix.)
(cf. *Felix threw the wolves Oscar.)
(cf. *Max carries extremes such behavior.)
We have noted, following Rappaport Hovav and Levin (2008), that triadic PP structures
with to encode two quite different meanings: caused possession and caused motion.
Since we are now considering relations among caused-possession structures, only
oblique idioms with the former sense will be relevant to our considerations. However,
all of the idioms in (24) are strongly associated with caused motion ― in no case is the
object of to (the showers, the cleaners, the wolves, extremes) understood as a possessor.
Hence examples like (24a–d), although arguing for the independent status of locative
projection, are simply irrelevant to the question of relatedness between the possessive
forms and shed no light on the issue.
Evidently, to apply our test, the relevant idiom data must involve verbs that strictly
encode caused possession. The verb give is such a form. As we saw, give is compatible
only with a caused-possession meaning, and never caused motion. Hence we might look
to see whether there are idioms occurring uniquely in the double object form of give
(43a), and idioms occurring uniquely in the oblique form of give (43b):
(43) a. VP b. VP
DP V′ DP V′
… V DP … V PP
give …XYZ… give to …XYZ…
idiomatic unit idiomatic unit

DOUBLE OBJECT GIVE IDIOMS OBLIQUE GIVE IDIOMS
If we could find such idioms, we would have a potent argument for independent projec-
tion, even lacking a theory of what properties those projections recruit.
1.4.1. Oblique Give Idioms?

Green (1974, p 179) offers the boldfaced items in (44ai–di) as give idioms in the oblique
form. Her assessment seems correct. The meanings of the constructions appear unpre-
dictable from the meanings of their parts. Furthermore, we note that no double object
variant is possible with any of these forms (44aii–dii).
(44) a. i. Mao’s silence has given rise to an absurd rumor.
ii. *Mao’s silence has given an absurd rumor rise.
b. i. Sarah gave birth to a son.
ii. *Sarah gave a son birth.
c. i. Activism gave way to apathy.
ii. *Activism gave apathy way.
d. i. John gave rein to his feelings.
ii. *John gave his feelings rein.

Nonetheless, (44a–d) are plainly not the sort of oblique idioms we are looking for. We
are seeking instances of (43b), where give and the oblique PP form a semantic unit. But
here it is give and its adjacent nominal (rise, birth, way, rein) that do so.9
Indeed, it’s not clear that (44ai–di) instantiate structure (43b) at all. Note that the
nominal portion of these forms appears inseparable from the verb. Not only do these
forms lack a double object equivalent; they also do not passivize (45ai–di), nor is the
postverbal nominal extractable (45aii–dii):
(45) a. i. *Rise has been given to an absurd rumor (by Mao’s silence).
ii. What has Mao’s silence given to an absurd rumor? Ans.: *Rise.
b. i. *Birth was given to a son (by Sarah).
ii. What did Sarah give to a son? Ans.: *Birth.
c. i. *Way was given to apathy (by activism).
ii. What did activism give to apathy? Ans.: *Way.
d. i. *Rein was given to his feelings (by John).
ii. What did John give to his feelings? Ans.: *Rein.
The diagnosis of this behavior offered in Larson (1988) is that give and its nominal
constitute a compound form, where N has essentially been incorporated into V (46):10
(46) VP
DP V′
α V PP
V N to β
give rise
birth
way
rein
On this view, we are not dealing with give per se, but rather with a series of distinct
compound verbs (give-rise, give-birth, etc.), each with its own specific semantics. Both
passivization and extraction would be ruled out, but the issue of projection into a
double object form would be moot as well.11
To the best of my knowledge, there exist no oblique give idioms of the (43b) sort,
where V and PP constitute an idiomatic unit. The caused-possession construction
appears to be regular in its PP form.
1.4.2. Double Object Give Idioms?

Consider now the set of expressions noted in (25) above, expanded further in (47ai–fi).
These appear to involve give and a discontinuous theme. All are marginal to outright
unacceptable in the oblique construction (47aii–fii).
(47) a. i. John gave Mary a strange feeling/the creeps.
ii. ?*John gave a strange feeling/the creeps to Mary.
b. i. John gave Mary a kiss/a kick/a shove.
ii. ?*John gave a kiss/a kick/a shove to Mary.
c. i. John gave Mary flak.
ii. ?*John gave flak to Mary.
d. i. John gave Mary a new idea/a ride.
ii. ?*John gave a new idea/a ride to Mary.
e. i. John gave Mary a hand.
(cf. John applauded Mary and John assisted Mary)
ii. ?*John gave a hand to Mary.
f. i. Dr. Jones gave Mary a new arm (using advanced surgical procedures).
ii. ?*Dr. Jones gave a new arm to Mary.
Green (1974, p. 226) refers to such constructions as “non-literal give expressions,”

presumably based on the fact that, unlike in John gave Mary a book, no literal transfer
of possession occurs from one individual to another. The key question is whether some
of these expressions constitute idioms.
The claim that examples in (47ai–fi) (or some subset of them) are idioms is the
claim that they are noncompositional, that is, that their meanings cannot be calculated
from meanings specifiable for their parts and the structure in which they appear—either
because their parts have no appropriate meanings or because the calculation itself is
irregular, and unpredictable from its input.12 To get clear on this claim and what it
entails, consider first the examples in (48a–d):
(48) a. That dive gave me the bends.
b. The view gave me vertigo.
c. Students gave me backtalk.
d. Alice gave me a (cheery) wave.
e. My landlord gave me two weeks’ notice.
On the analysis we are pursuing, give, in both its double object and oblique forms,
means ‘cause to possess or receive’, where what is possessed or received may include
not only material objects but also nonmaterial entities such as physical or psychologi-
cal states (48a,b), actions (48c), gestures (48d), and communications (48e). Regarding
the boldfaced nominal elements in (48), Webster’s Seventh New Collegiate Dictionary
provides the subentries in (49a–e) (respectively) for each:
(49) a. bend n … 3: pl but sing or pl in construction caisson disease.
caisson disease n: a sometimes fatal disorder marked by neuralgic pains and paralysis

and caused by too rapid decrease in air pressure after a stay in compressed atmosphere
—also called bends.
vertigo n 1a: a disordered state in which the individual or his environs seem to whirl
b.
dizzily …
backtalk n 1a: an impudent, insolent or argumentative reply.

c.
wave n … 4: a sweep of hand or arm or of some object held in the hand used as a signal
d.
or greeting.
n
otice n … 1a … (3): notification by one of the parties to an agreement or relation of
e.
intention of terminating it at a specified time.
Assume now the informal composition rule in (50), which maps from give and its two
DP arguments to the caused-possession meaning:
(50) [vp give DP1 DP2 ]  ‘cause DP1 to have/get DP2’
Taking (49) and (50) together, we predict the results in (51a–e) for the VPs in (48a–e)
(adjusting for tense and suppressing irrelevant details):
(51) a. g ave me the bends: ‘caused me to have a disorder marked by neuralgic pains and
paralysis and caused by too rapid decrease in air pressure after a stay in compressed
atmosphere’
gave me vertigo: ‘caused me to have a disordered state in which I or my environs

b.
seemed to whirl dizzily’
gave me backtalk: ‘caused me to get an impudent, insolent or argumentative reply’

c.
gave me a wave: ‘caused me to get a sweep of hand or arm or of some object held in
d.
the hand used as a signal or greeting’
e. gave me two weeks’ notice: ‘caused me to have notification as the party to an
agreement or relation of intention of terminating it in two weeks’ time’
In my judgment (and others’ too) these are in fact the meanings of the expressions in
question. If so, (48a–e) are shown to be compositional and therefore nonidiomatic: their
meanings result from the appropriate, separately specified meanings of their parts (give,
bend, vertigo, etc.) together with the regular meaning of the [VP give DP1 DP2 ] structure
in which they appear.
Consider now the examples in (52a–e), discussed in Larson (1988) and routinely cited
as clear instances of dative idioms of the double object form (43a) (Pesetsky 1995; Richards
2001; Harley 2002). Are these in fact noncompositional and idiomatic, unlike (48a–e)?
(52) a. The Count gave me the creeps.
b. John gave Mary flak (about/during her presentation).
c. His boss gave Max the boot.
d. Moe gave Larry the finger.
e. Alice gave me a piece of her mind.
Again assume that give, in both its double object and oblique forms, means ‘cause to
possess/receive’, where what’s possessed or received may include physical or psychologi-
cal states, actions, gestures, communications, and so on. Again we consult entries in
Webster’s Seventh New Collegiate Dictionary (WSNCD) and The American Heritage
Dictionary of the English Language (AHDEL) for the boldfaced nominals in (52),
extracting the results in (53):13
(53) a. creep n 1: a movement of or like creeping 2: a distressing sensation like that caused by
the creeping of insects over one’s flesh; esp a feeling of apprehension or horror—usu.
used in the plural. (WSNCD)
flak n … 2 Informal a. Excessive or abusive criticism. b. Dissention, opposition.

b.
(AHDEL)
boot n…6 Brit a blow delivered by or as if by a booted foot: kick; also : a rude
c.
discharge or dismissal. (WSNCD)
finger n … 6 an obscene gesture of defiance or derision made by pointing or jabbing the

d.
middle finger upward. (AHDEL)
piece n … -idioms a piece of (one’s) mind Frank and severe criticism; censure.
e.
(AHDEL)
And again, under our informal composition rule (50), we compute meanings for the VPs
in (52a–e), obtaining the results in (54) (adjusting for tense and suppressing irrelevant
details):
(54) a. gave me the creeps: ‘caused me to have a feeling of apprehension or horror’
gave Mary flak: ‘caused Mary to have/get excessive or abusive criticism’

b.
gave Max the boot: ‘caused Max to have/get a rude dismissal’

c.
gave Larry the finger: ‘caused Larry to have/get an obscene gesture of defiance or
d.
derision made by pointing or jabbing the middle finger upward’
gave me a piece of her mind: ‘caused me to have frank and severe criticism; censure’
e.
If the claim of idiomaticity is correct for (52a–e), then these results should be wrong;
give me the creeps, for example, should not mean what (54a) says it means. But in fact
the results are not wrong. In my judgment (and that of others), (54a) does render the
meaning of the construction. If this is so, then (52a) is also compositional and not idi-
omatic. The same outcome, and the same conclusion, obtains for all of (52b–e).
Evidently, the compositional result we obtain with an example like The Count gave me
the creeps derives from our being able to assign a definite reference to the creeps, essen-
tially as the proper name for a certain psycho-physical state, comparable to the bends or
vertigo. This view, embraced by the authors of WSNCD and AHDEL, and by lexicog-
raphers generally, seems unavoidable given that the nominal element in give ~ the creeps
can be elaborated and modified in regular ways comparable to the bends, the measles and
so on (55a–c). Such behavior is not typical of idioms, which generally lose their figurative
senses in that context (56):
(55) a. The Count gave me an amazingly bad and violent case of the creeps/a case of the
creeps that lasted for weeks/creeps that I just couldn’t shake.
b. That dive gave me an amazingly bad and violent case of the bends/a case of the
bends that lasted for weeks/bends that damn near killed me.
c. My trip abroad gave me an amazingly bad and violent case of the measles/a case of
the measles that lasted for weeks/measles that covered my body.
(56) The thick fur flew/The shit hit the slowly whirling fan/John kicked the big bucket.
Consider also the fact that genuine idioms typically resist substitution of subparts, even
by elements of similar sense, on pain of reverting to literal meaning (57a–d). By con-
trast, give ~ the creeps freely permits substitution in its nominal portion, exhibiting a
remarkably broad range of variant forms with similar or related meanings (58). All result
by substitution of the nominal expression (the creeps). If the creeps is an independent
item making independent reference in the give construction, it is natural to expect related
expressions making similar contributions. On the idiom view, this behavior is mysterious.
(57) a. The cat got out of the bag/sack/tote/pouch.
b. The fur/hair flew.
c. The shit hit/struck/collided with the fan.
d. John kicked the bucket/pail.
(58) The Count gives me the creeps/the willies/the shivers/the shakes/the chills/
the jitters/goosebumps/gooseflesh/the fits/the heebee-jeebies/the screamin’ meemies.
These results appear compatible with the basic view of Larson (1988) that the sense of idi-
omaticity present in these cases resides strictly in the nominal. Like the blues and the blahs,
the creeps names a psychological state whose definition is vague and uncodified, the terms
for which typically derive by metonymic extension from its physical effects or results (cf.
the shivers, the shakes), by some form of sound symbolism (e.g., the heebee-jeebees), or by
both (e.g., the screamin’ meemies). As such, the reference can’t be predicted from the noun
form itself. But once reference is fixed via an entry like (53a), the interaction with give is
perfectly regular.
The above reasoning with give ~ the creeps appears to me quite general. Thus with
other putative double object idioms of this kind (59a)/(60a), the nominal elements also
receive entries yielding correct, compositional interpretations when combined with the
normal sense of give as caused possession. Furthermore, these nominals can often be
elaborated in regular ways (59b/60b), 14 and often show variants by substitution in the
nominal portion alone (59c)/(60c,d).
(59) a. John gave Mary flak (about/during her presentation).
b. John gave Mary a lot of flak/far more flak than anyone had expected.
c. John gave Mary flak/static/shit/crap/hassle/trouble.
(60) a. John gave Mary the boot.
b. John gave Mary the royal order of the boot.
c. We will all be for the boot by Friday.
d. John gave Mary the boot/the sack/the chop.
I conclude, following Larson (1988), that none of these cases constitute genuine double
object idioms.15 If so, there are no give idioms of either the oblique or double object
form in the caused-possession meaning, and hence no evidence for separate projection
from this source.16
1.4.3. Double Object Forms without Oblique Variants?

The above results undercut the view that (47ai–fi) (repeated below), or any subset of them,
constitute idioms, and hence undermine their use in an argument for separate projection
of double object datives. Nonetheless, they do leave a important distributional question.
As we observed, although nonliteral give expressions occur in double object structures,
their simple oblique variants are marginal to unacceptable. It thus seems there must be
some feature(s) fitting such expressions for use in one structure and not the other.
(47) a. i. John gave Mary a strange feeling/the creeps.
ii. ?*John gave a strange feeling/the creeps to Mary.
b. i. John gave Mary a kiss/a kick/a shove.
ii. ?*John gave a kiss/a kick/a shove to Mary.
c. i. John gave Mary flak.
ii. ?*John gave flak to Mary.
d. i. John gave Mary a new idea/a ride.
ii. ?*John gave a new idea/a ride to Mary.
e. i. John gave Mary a hand.
(cf. John applauded Mary and John assisted Mary)
ii. ?*John gave a hand to Mary.
f. i. Dr. Jones gave Mary a new arm (using advanced surgical procedures).
ii. ?*Dr. Jones gave a new to Mary.
Putting things another way, if both the double object and oblique forms of give
univocally express caused possession, as proposed, and if there is no idiomatic-
ity in these constructions, as just demonstrated, then what is the source of their
divergence?
Discussion in this domain has often contented itself with contrasts like (47a–f),
drawing the conclusion that nonliteral give constructions are simply unavailable in the
oblique form. Recent corpus research recorded in Rappaport Hovav and Levin (2008)
has shown that this assumption is factually incorrect and that such examples occur
far more frequently than is typically assumed. And indeed Green (1974) observes data
showing that the situation is more complicated, and more interesting, than is usually
acknowledged. Green notes that in a number of contexts, including relativization (61),
question formation (62), and cases where the to-PP is “heavy” (63), examples of these
constructions in the oblique form do in fact become available to many speakers (includ-
ing myself):
(61) a. The infection that Martha gave to John nearly killed him.
b. The shove that Liz gave to Richard shocked us all.
c. The flak that Joan gave to Ed utterly decimated him.
d. The idea that Bill gave to Sue caused her to rewrite her thesis.
e. The ride Mack gave to Ellen ended in disaster. (= (28a–e) in Green 1974, p. 177)
(62) a. What kind of infection did Martha give to John?
b. How hard a shove did Liz give to Richard anyway?
c. What sort of flak did Joan give to Ed?
d. Which idea did Bill give to Sue?
e. How long a ride did Mack give to Ellen ? (= (29a–e) in Green 1974, p. 177)
(63) a. The Snopes brats gave the mumps [to every single kid who lived within two blocks
of them].
b. He threatened to give a beating [to anyone who attempted to reveal where they had
been].
c. We gave the peace sign [to all of the American soldiers we saw].
d. They gave an idea [to all of us who had read the assignments faithfully].
e. We gave a ride [to each of the crying, ragged, long-neglected children].
(= (25a–g) in Green 1974, pp. 175–176)
Green also notes the interesting example (64a), in which the nominal in a nonliteral give
construction is anaphorically resumed in an oblique form conjunct. Note that although
the definite pronoun it is acceptable with the appropriate reference in the oblique con-
struction (64a), the corresponding full nominal is not (64b).
(64) a. Mary gave John the flu/the finger/the answer, and then she gave it to Bill.
b. ?*Then she gave the flu/the finger/the answer to Bill.
(cf. (4–6) in Green 1974, p. 170)
In reflecting on these facts, Green makes the intriguing suggestion that the constraints
operative in (61)–(64) are “surface realizational” in nature. Observe that although the
second conjunct of (64a) is acceptable in the oblique form, the corresponding double
object form (65a) is ruled out by a general constraint forbidding definite pronouns from
final position with double objects (65b,c).17 In effect, the oblique form is available in a
circumstance where the double object form is independently excluded.
(65) a. *Mary gave John the flu/the finger/the answer, and then she gave Bill it.
b. They gave it to John.
c. *They gave John it.
A similar point emerges with (63). English (like many languages) typically favors “heavy”
material in final position for prosodic, processing, and/or informational reasons. How-
ever, this is not possible for the counterparts of (63) in the otherwise preferred double
object form. In the double object form, heavy material will not be final (66a–e). Fur-
thermore, this material cannot be repositioned finally given the general unavailability of
“Heavy NP Shift” with double objects (67a–e). The upshot is thus again that the oblique
form becomes available (63a–e) where the double object form (66a–e) or derivatives of it
(67a–e) are either disfavored or outright excluded by independent constraints.
(66) a. The Snopes brats gave [every single kid who lived within two blocks of them]
the mumps.

b. He threatened to give [anyone who attempted to reveal where they had been]
a beating.

c. We gave [all of the American soldiers we saw] the peace sign.
d. They gave [all of us who had read the assignments faithfully] an idea.
e. We gave [each of the crying, ragged, long-neglected children] a ride.
(= (23a–g) in Green 1974, pp. 174–175)
(67) a. *The Snopes brats gave the mumps [every single kid who lived within two blocks
of them].
b. *He threatened to give a beating [anyone who attempted to reveal where they had
been].
c. *We gave the peace sign [all of the American soldiers we saw].
d. *They gave an idea [all of us who had read the assignments faithfully].
e. *We gave a ride [each of the crying, ragged, long-neglected children].
(= (24a–g) in Green 1974, p. 175)
Finally, consider the examples in (61)–(62). In these cases, as Green puts it (1974), extraction
of the theme “has left the clause containing give bereft of any indication that it is restricted
in regard to dative movement” (p. 177). Green’s idea appears to be that with the nominal
element (infection, shove, flak, idea, ride) that is the hallmark of this class of datives removed
from the PF string of the lower CP, the speaker receives no surface evidence from within
CP alone that he/she is dealing with a give construction that otherwise requires the double
object form; hence the oblique form is allowed.18 In effect, because the surface string doesn’t
explicitly show that one is dealing with a proscribed oblique form, one is allowed to use it.
1.4.4. Languages without Double Object Forms

Green’s general outlook has interesting implications for languages said to lack the double
object construction altogether. If both the double object and the oblique dative form encode
the same meaning of caused possession, and if the restrictions observed in (47) are in fact
the product of surface constraints suspended when the double object form is unavailable (or
when the relevant restrictions cannot be locally detected), then we might expect construc-
tions counterpart to (47a–f) to emerge unhindered in the oblique form. In effect, because the
double object form is simply a variant that “wins” over the oblique form whenever avail-
able, its general unavailability would allow its oblique counterpart to surface.
European Portuguese is a potential example. As noted earlier, European Portuguese
exhibits oblique dative forms, with caused possession versus caused motion indicated
by the form of the preposition—a versus para, respectively (68a–c):
(68) a. O João deu um livro à Mary/*para a Mary
the John gave a book to Mary
‘John gave a book to Mary’
b. O João enviou uma carta à Mary/para a Mary
‘John sent a letter to Mary’

c. O João enviou uma carta *a Lisboa/para a Lisboa.
the John sent a letter to Lisbon

‘John sent a letter to Lisbon.’
However, European Portuguese exhibits no double accusative form (69a), and although
it permits inverted PP order (69b), it does not appear to contain the Spanish-like clitic
dative structures that have been argued to be the equivalent of English double object
constructions (69c,d) (Demonte 1995; Bleam 2003; Cuervo 2003):
(69) a. *O João deu a Mary um livro.
the John gave the Mary a book
‘John gave Mary a book.’
b. O João deu à Mary um livro
the John gave to-the Mary a book
‘John sent to Mary a letter.’

c. Susana lei mandó a Pabloi una carta.
Susana Cl-3Dat sent to Pablo a letter
‘Susana sent a letter to Paul.’ (Spanish)
d. Susana lei mandó una carta a Pabloi. (Spanish)
Thus, in surface form, European Portuguese datives—whether expressing caused motion

or caused possession—seem to be restricted to the oblique construction.
Given this situation, it is then interesting to observe that European Portuguese exhib-
its a wide variety of nonliteral give constructions equivalent to those in (47) in its sole
oblique dative form (70).19
(70) a. A Mary deu banho ao João.
the Mary gave bath to-the John
‘Mary gave John a bath.’
(cf. ?*Maria gave a bath to John.)
b. A Mary deu uma pintadela à grade.
the Mary gave a quick paint to-the fence
‘Mary gave the fence a quick painting.’
(cf. ?*Mary gave a quick painting to the fence.)
c. A Mary deu uma olhadela ao livro. / deu um empurrão ao Pedro.
the Mary gave a look to-the book gave a push to-the Peter
‘Mary gave the book a look/gave Peter a push.’
d. A Mary deu uma mão ao Pedro.
the Mary gave a hand to-the Peter
‘Mary gave Peter a hand.’
e. A Mary não dá bola ao Pedro.
the Mary not give boll to-the Peter.
‘Mary doesn’t give a shit about Peter.’
f. Ele deu a volta à Mary.

he gave the turn to-the Mary
‘He convinced Mary.’
g. O Pedro deu para trás à Mary.
the Peter gave for back to-the Mary
‘Peter contradicted Mary, didn’t support Mary.’

h. Isso dá jeito à Mary.
that gives hand to-the Mary
‘That comes in handy to Mary’
i. De repente, deu uma coisa à Mary e começou a insultar toda a gente.
of sudden gave one thing to-the Mary and she started to insult all the people
‘Suddenly, something hit Mary [in the figurative sense], and she started to insult
everybody.’
One way of interpreting these data might thus be to say that, given the lack of an inde-
pendent double object construction (either by separate projection or syntactic derivation),
the features of meaning/use that suit the relevant meanings to this form in English cannot
be encoded in a separate form in European Portuguese, and hence are realized uniformly in
the oblique. In effect, European Portuguese gives us a glimpse of what English would be like
with realization constraints lifted uniformly, and not just in special contexts like (61)–(64).20
This line of argument is a general one extending well beyond triadic constructions. Con-
sider languages like Korean and Japanese, which have oblique dative + copula constructions
as their favored structures for expressing possession. Here too some meanings counterpart
to Green’s nonliteral give expressions are expressed in these oblique forms (71)–(72):
(71) a. Chulsu-ege cek-i itta. (Korean)
Chulsu-dat book-nom be
‘Chulsu has a book.’
b. Chulsu-ege (coeun) sangkak-i itta.
Chulsu-dat good idea-nom be
‘Chulsu has a good idea.’
c. Chulsu-ege munjae-ga itta.
Chulsu-dat problem-nom be
‘Chulsu has a problem.’
d. Chulsu-ege yel-i itta.
Chulsu-dat fever-nom be
‘Chulsu has a fever.’

(72) a. John-ni wa kangae-ga aru. (Japanese)
John-dat top idea-nom be
‘John has an idea.’
b. John-ni wa netu-ga aru.
John-dat top fever-nom be
‘John has fever.’

Under an analysis of give as expressing caused possession, the possibility of such oblique
possession constructions suggests that the unavailability of their give counterparts in
English cannot reflect any deep incompatibility between these meanings and oblique
expression, but rather arises from something more surface-oriented, in line with the
suggestion by Green (1974).21
1.5. Recasting the Problem Area: Applicative Constructions

The picture we have arrived at this point, through consideration of English and a number of
primarily European languages, is basically the one in (27), where the double object construc-
tion is associated with one meaning (caused possession) and the oblique form is associated
with at least two (caused possession, caused motion/location). We’ve seen solid evidence
to think caused-possession and caused-motion constructions are projected separately. On
the other hand, we’ve seen no such evidence for separate projection of the two caused-
possession structures themselves. On reflection, these results raise at least two questions.
First, and most simply, what is the relation between the two caused-possession struc-
tures (73a,b)? If they are separately projected, what are the differing items (or properties
of items) that underwrite this?
(73) a. VP b. VP
DP Vʹ DP Vʹ
John V VP John V VP
←??→
send DP Vʹ send DP Vʹ
Mary V DP a telegram V PP
send a telegram send to Mary
CAUSED POSSESSION CAUSED POSSESSION
Second, what ties the double object construction exclusively to caused possession? If
the oblique form can encode both concepts, why can’t the double object form do so as
well? Why can’t we have the double object counterparts of oblique locatives (74a)? If
the latter cannot be projected, what items (or properties of items) are lacking for this?
(74) a. * VP b. VP
DP Vʹ DP Vʹ
John V VP John V VP
←??→
send DP Vʹ send DP Vʹ
Lisbon V DP a telegram V PP
send a telegram send to Lisbon
CAUSED MOTION/LOCATION CAUSED MOTION/LOCATION

Research on languages with applicative constructions appears to offer crucial insight
into both of these questions, and in so doing dramatically expands and recasts our per-
spective on the issue of dative projection.
Applicative constructions have been vigorously studied within the framework of
Relational Grammar (see Aissen 1983; Allen and Frantz 1983; Chung 1976; Dryer
1983; Gerdts 1988; Gibson 1992; Kimenyi 1980; and Perlmutter and Postal 1983b,
among others), and have been more recently brought to theoretical attention in the
Extended Standard Theory by the work of Marantz (1984, 1993) and Baker (1988),
and in the framework of Lexical Functional Grammar by Alsina and Mchombo
(1993) and Bresnan and Moshi (1993).22 Examples of applicative constructions
include (75a) and (76) from Chichewa (Baker 1988). The nominal immediately fol-
lowing the verb (nkhandwe, mbidzi, mfumu) is referred to as the applied object or
primary object; the other nominal (msampha, nsapato, mtsuko) is called the base
object or secondary object.
(75) a. applied obj base obj
Mbidzi zi-na-perek-er-a nkhandwe msampha.
zebras sp-pst-hand-app-asp fox trap
‘The zebras handed the fox the trap.’ (= (3a) in Baker 1988, p. 229)
b. Mbidzi zi-na-perek-a msampha kwa nkhandwe.
zebras sp-pst-hand-asp trap to fox
‘The zebras handed the trap to the fox.’ (= (2a) in Baker 1988, p. 229)
(76) Kalulu a-na-gul-ir-a mbidzi nsapato.
hare sp-pst-buy-app-asp zebras shoes
‘The hare bought the zebras shoes’. (= (46a) in Baker 1988, p. 240)
(77) Mavuto a-na-umb-ir-a mfumu mtsuko.
Mavuto sp-pst-mold-app-asp chief waterpot
‘Mavuto molded the chief the waterpot.’ (= (141a) in Baker 1988, p. 290)
Intensive empirical study has shown constructions like (75a), (76), and (77) to exhibit
properties directly parallel to English double object constructions, and indeed the pair
in (75) plainly resembles the English to-dative alternation in word order and meaning.
But the parallels go deeper. Thus, in English the first object in a double object structure
is accessible to Passive, but not to A-bar extraction (78a,b), whereas the inverse holds
with the second object (79a,b):
(78) a. John was bought __ those shoes.
b. *The shoes were bought John __.
(79) a. [What John bought Mary __] was those shoes.
b. *[Who John bought __ those shoes] was Mary.

Applicative constructions show that same pattern. The applied object is accessible to
passive (80a) and cannot be A-bar extracted (81a) By contrast, the base object shows
the opposite constellation (80b)/(81b):23
(80) a. Mbidzi zi-na-gul-ir-idw-a nsapato (ndi kalulu).
zebras sp-pst-buy-app-pass-asp shoes by hare
‘The zebras were bought shoes (by the hare).’ (= (46b) in Baker 1988, p. 240)
b. *Nsapato zi-na-gul-ir-idw-a mbidzi (ndi kalulu).
shoes sp-pst-buy-app-pass-asp zebras by hare
‘The shoes were bought for zebras (by the hare).’ (= (47) in Baker 1988, p. 240)
(81) a. *Iyi ndiyo mfumu imene ndi-ku-ganiz-a Mavuto a-na-umb-ir-a mtsuko.
this is chief which 1sS-pres-think-asp Mavuto sp-pst-mold-app-asp waterpot
‘This is the chief that I think Mavuto molded the waterpot for.’
(= (141b) in Baker 1988, p. 290)
b. *Uwu ndiyo mtsuko umene ndi-ku-ganiz-a Mavuto a-na-umb-ir-a mfumu.
this is waterpot which 1sS-pres-think-asp Mavuto sp-pst-mold-app-asp chief
‘This is the waterpot that I think Mavuto molded for the chief.’
(= (143) in Baker 1988, p. 290)
Examples like (75a), (76), and (77) thus seem directly comparable to the English datives
that are their glosses.
1.5.1. The Semantic Range of Applicative Constructions

Given the evident connection between English double object datives and applicatives,
the latter are of special interest in exhibiting two things that the former do not. The first
concerns semantic range.
As discussed above, English double objects are restricted to cases like (75a) and (76),
where the applied object has the thematic status of recipient-goal; double object form
strictly encodes caused possession. This is seen not only in the blunt impossibility of
(82a) as a counterpart to an oblique locative like (82b) but also in other, more subtle
ways. For example, the oblique object in an English for-dative can bear a number of
participant roles beyond recipient-goal (83ai–iii). Only the latter survives, however,
when the construction shifts to double object form (83bi–iii).
(82) a. *John put the table the key.
b. John put the key on the table.
(83) a. John baked the cake for Mary.
i. ‘John baked the cake for Mary to have.’
ii. ‘John baked the cake as a favor to Mary/at her request.’
iii. ‘John baked the cake in Mary’s place, as a substitute for her.’
b. John baked Mary the cake.
i. ‘John baked the cake for Mary to have.’
ii. #‘John baked the cake as a favor to Mary/at her request.’
iii. #‘John baked the cake in Mary’s place, as a substitute for her.’
Importantly, applicative constructions do not share this restriction, cross-linguisti-

cally exhibiting a much wider range of senses than what is found in English. Thus,
along with benefactives like (76) and (77), there are also benefactives like (84a,b)
involving readings that are missing in English. The Halkomelem benefactive applica-
tive in (84a) has the ‘favor’ or ‘substitutive’ sense that disappears in (83b); and the
Chichewa (84b) has a malefactive/source reading that is entirely unavailable with the
English for-structure:
(84) a. ni θә́y -əɬc -θ -ámʔš -əs ʔə kwθə nə-snә́xwəɬ
aux fix -app -tr -1O -3a obl Det 1pos-canoe
‘He fixed my canoe for me.’ (= (2.71) in Farrell 2005, p. 85)
b. Kambuku a-na-b-er-a mkango njinga
leopard sp-pst-steal-app-asp lion bicycle
‘The leopard stole the bicycle from/on the lion.’ (= (28b) in Baker 1988, p. 240)
Beyond these there are also instrumental applicatives, like Chichewa (85a), and locative/
directional applicatives, like Kinyarwanda (86a) and (87a), all of which may be com-
pared with the corresponding obliques (85b), (86b), and (87b), respectively.
(85) a. Fisi a-na-dul-ir-a mpeni chingwe.
hyena sp-pst-cut-app-asp knife rope
‘The hyena cut the rope with the knife.’ (= (22b) in Baker 1988, p. 238)
b. Fisi a-na-dul-a chingwe ndi mpeni.
hyena sp-pst-cut-asp rope with knife
‘The hyena cut the rope with the knife.’ (= (22a) in Baker 1988, p. 238)
(86) a. Umwaana y-a-taa-ye-mo amaazi igitabo.
child sp-pst-throw-asp-app water book
‘The child threw the book into the water.’ (= (26b) in Baker 1988, p. 239)
b. Umwaana y-a-taa-ye igitabo mu maazi.
child sp-pst-throw-asp book in water
‘The child threw the book into the water.’ (= (26a) in Baker 1988, p. 238)
(87) a. Umugore y-oohere-je-ho –iisoko umubooyi.
woman sp-send-asp-app market cook
‘The woman sent the cook to the market.’ (= (27b) in Baker 1988, p. 239)
b. Umugore y-oohere-je umubooyi kw’ –iisoko.
woman sp-send-asp cook to market
‘The woman sent the cook to the market.’ (= (27a) in Baker 1988, p. 239)
Even more exotic applicative types have been reported in the literature. Farrell (2005)
(citing Gerdts 1988) mentions Halkomelem “stimulus” applicatives like (88a), which con-
trast with the oblique (88b). Baker (1988) (citing Kimenyi 1980) also notes the existence
of Kinyarwanda manner applicatives (89a) and Chichewa reason applicatives (90).
(88) a. ni cən c’ә́q’ -méʔ kwθə sqwəmә́y.
aux 1subj astonished -app Det dog
‘I was astonished at the dog.’ (= (2.72d) in Farrell 2005, p. 86)
b. ni cən c’ә́q’ ʔə k θə sxә́k’ s.

w w
aux 1subj astonished obl Det carving
‘I was astonished at his carving.’ (= (2.72a) in Farrell 2005, p. 86)
(89) a. Umugabo a-ra-som-an-a ibaruwa ibyiishiimo.
man sp-pres-read-app-asp letter joy
‘The man is reading a letter with joy.’ (= (ib) in Baker 1988, p. 468)
b. Umugabo a-ra-som-a ibaruwa n’ -iibyiishiimo.
man sp-pres-read-asp letter with joy
‘The man is reading a letter with joy.’ (= (ia) in Baker 1988, p. 467)
(90) Nsima iyi ndi-ku-dy-er-a njala.
cornmeal this 1sS-pres-eat-app-asp hunger
‘I am eating this cornmeal form/because of hunger.’ (= (ii) in Baker 1988, p. 468)
These results suggest a rather surprising answer to our question raised above
in connection with (74a,b): what ties the double object construction exclusively
to caused possession? We see that in fact nothing ties the double object form
exclusively to caused possession—nothing universal at any rate. When we move
beyond English, the tie breaks down, and, indeed, the locative applicative struc-
ture in (87a) corresponds quite closely to the “missing” English structure (74a)
that we inquired about earlier. More generally, the broad semantic range of appli-
catives indicates that a conclusion parallel to that drawn by Rappaport Hovav and
Levin (2008) for oblique constructions is also appropriate for double objects, now
understood as applicatives: just as oblique structure cannot be tied to a specific
sense like caused location, so double object structure cannot be tied to a particular
sense like caused possession. The absence of a richer set of double object forms
in English evidently reflects language-specific facts about its particular grammar
but not any sort of “deep,” Universal Grammer–imposed correspondence between
meaning and form.
1.5.2. The Morphological Form of Applicative Constructions

Applicative constructions also offer a sharper formulation of our question raised
above in connection with (73a,b): what is the relation between the two English
caused-possession structures? We now see that this is part of a much broader puzzle
about the relation between two expression frames: oblique and applicative. And we
see something else as well. A notable fact about all the applicative constructions
listed above is the presence of a specific verbal affix, glossed APP, that accompanies
the form. Baker (1988) suggests that this element is a constant, and hence present
even when “hidden” by null suppletion. Thus, in the Chichewa pairs in (91) and (92),
the verb tumiz ‘send’ shows the normal alternation between applicative (91a) and
oblique (91b) frames. The pair in (92), involving the verb pats ‘give’, shows a very
similar alternation, except that in its applicative frame (92a) no actual applicative
morpheme appears.
(91) a. Ngombe zi-na-tumiz-ir-a mbuzi mitolo ya udzu.
cows sp-pres-send-app-asp goats bundles of grass
‘The cows sent the goats bundles of grass.’ (= (121b) in Baker 1988, p. 281)
b. Ngombe zi-na-tumiz-a mitolo ya udzu kwa mbuzi.
cows sp-pres-send-asp bundles of grass to goats
‘The cows send bundles of grass to the goats.’ (= (121a) in Baker 1988, p. 280)
(92) a. Joni a-na-pats-a amai ake nthochi.
John sp-pres-give-asp mother his bananas
‘John gave his mother the bananas.’ (= (122b) in Baker 1988, p. 281)
b. Joni a-na-pats-a nthochi kwa amai ake.
John sp-pres-give-asp bananas to mother his
‘John gave the bananas to his mother.’ (= (122a) in Baker 1988, p. 281)
Baker argues, plausibly, that given the productive status of applicative formation in
Chichewa and the parallel behavior of (91a) and (92a) across a wide range of phenomena,
the verb complex in the former should be analyzed, in effect, as a-na-pats-Ø-a ‘sp-pres-give-
app-asp’, where Ø is a null applicative morpheme. This then suggests a parallel account of
English double object constructions. Given their close distributional parallels with applica-
tive constructions, it is natural to conclude that they too contain an applicative morpheme—
more generally that the latter is in fact a universal ingredient in constructions of this kind.
We thus arrive at what seems the appropriate general form of the question we began
with in this section: the question of dative projection and the dative alternation. Double
object form in English we now see to be a species of applicative form, and applicative
form we see to be capable, in the general case, of encoding all the same meanings as
oblique form, hence allowing systematic applicative-oblique alternations of the kind
noted above. The appropriate questions therefore become: What is the nature of appli-
cative-oblique alternation? How are the two constructions related?
(93) Applicative Oblique

Form: α V-app β γ ←??→ α V γ [p β]
Senses: caused possession caused possession
benefactive/malefactive/ benefactive/malefactive/
substitutive substitutive
instrumental instrumental
caused motion/location caused motion/location
stimulative stimulative
manner manner
reason reason
Is one form more basic and the other derived from it, or are both forms equally basic,
in which case, given the apparent identity of senses across the two constructions, what
elements are responsible for the separation?
2. RELATING APPLICATIVES AND OBLIQUES
The most detailed and systematic account of the applicative-oblique alternation that I
am aware of is that of Baker (1988), building crucially on the work of Marantz (1984).
Baker proposes a simple but elegant derivational approach. The table in (93) suggests
that, apart from word order, the chief difference between applicative and oblique forms
is the presence of the applicative morpheme -APP in the former and of an oblique
preposition P in the latter. Baker (following Marantz 1984) proposes, in essence, that
these are the same element, specifically that -APP has the status of an affixal oblique
preposition. Applicative constructions result by incorporating this preposition into V,
as shown in (94a):
(94) a. Applicative Construction b. Oblique Construction

VP VP
V PP NP V NP PP
V P P NP shoes buy shoes P NP
buy-APP -APP Mary for Mary
Baker argues that these structures, interacting with universal principles, can pre-
dict the differing syntactic properties of applicative versus oblique constructions,
for example, the passivization and extraction facts noted in (78)–(81). He fur-
thermore shows that variation in the lexical inventory of affixal Ps and their
morphological properties, taken together with differences in Case (structural vs.
inherent) and conditions of Case assignment by verbs, can be recruited to explain
the differences we find among applicative constructions, both within languages
and across them.
2.1. What Counts as a Derivational Analysis?

Without rehearsing the specific virtues or deficits of Baker’s analysis, 24 I wish to draw
attention to a feature of it that is significant from a general point of view: like all deriva-
tional accounts, it claims that the basic compositional semantic structure of an applicative-
oblique pair is given by one of its members—here the second, oblique member—and that
the understanding of both constructions issues from this single source. The force of this
basic point has not, in my opinion, been fully appreciated in modern proposals purport-
ing to offer “nonderivational” approaches to the applicative-oblique alternation, but in
fact seems crucial in appreciating what is truly required of such a proposal.
To clarify what’s at stake, consider two approaches to the dative alternation within
the framework of classical Montague Grammar: one in the style of Partee (1973), which
gives an explicit syntactic treatment, and the other by Dowty (1978), which treats
Dative Shift semantically.
2.1.1. Syntactic Dative Shift

Partee (1973) explores how classic transformational relationships might be integrated
into the syntactic component of a Montague Grammar for English. To do this, Partee
allows syntactic composition rules to build, not merely word strings as in Montague
(1974), but labeled bracketings that encode structural relationships. This permits the
formulation of syntactic rules like Passive (95), which operate on these bracketings in a
way standard for the times:25
(95) Passive: If φ ∈ Pt and φ has the form [t [T α] [IV [TV β] [T δ] γ ]],
then F101(φ) ∈ Pt , where F101(φ) is: [t [T δ] [IV is EN [TV β] [by [T α]] γ ]
Translation Rule: identity mapping
The passive operation figures in a derivation like (96), where translations of the relevant
expressions into intensional logic (IL) are given at the right:
(96) Derivation Translation
[t [T John] [IV is EN [TV kiss] [by [T Mary]]]], F101 kiss’(j)(m)

| ← (Passive)
[t [T Mary] [IV [TV kiss] [T John]]], F4 kiss’(j)(m)
[T Mary] [IV [TV kiss] [T John]], F5 …
[TV kiss] [T John]
Note that by expressing the active-passive relationship as a syntactic one, the semantic
translation of the Passive rule can be kept to the simplest possible mapping: identity.
This analysis makes the explicit claim that the semantics of passives is understood
through the corresponding active; the passive has no independent semantics of its own.
A similar Partee-style approach is available for the dative alternation, for example,
by a rule like (97):
(97) Dative Shift: If φ ∈ PIV and φ has the form [IV [DTV β] [T α] [to [T δ]] γ ],
then F102(φ) ∈ PIV , where F102(φ) is: [IV [TV β] [T δ] [T α] γ ]
Translation Rule: identity mapping

This operation occurs in the derivation in (98), where I ignore the details below the VP
(IV) level, and where translations of relevant expressions are given at the right:
[t [T John] [IV [DTV give] [T Mary] [T Fido]]], F4 give’(m)(f)(j)
[t John] [IV [DTV give] [T Mary] [T Fido]], F102 give’(m)(f)
| ← (Dative Shift)
[IV [DTV give] [T Fido] [to [T Mary]]] give’(m)(f)

… …
Again, by expressing the oblique–double object relationship as a syntactic one, the
semantic translation of the Dative Shift rule is kept to the simple identity mapping. This
approach makes a claim about the understanding of double object forms parallel to that
of passives, that is, that a double object construction is understood through the corre-
sponding oblique. Double object datives have no independent semantics of their own.
2.1.2. Semantic Dative Shift

An alterative approach is that of Dowty (1978), whose analysis of the oblique dative
form is illustrated by the derivation tree in (99), where translations of relevant expres-
sions are given at the right:
John gives Fido to Mary, t, F4 give’(m)(f)(j)
John, T give Fido to Mary, IV, F5 give’(m)(f)
Fido, T give to Mary, TV, F22 give’(m)
give, DTV Mary, T

Like Partee, Dowty takes the oblique form to represent the base form of composition
for a ditransitive verb like give. Here the goal argument (Mary) combines first, followed
by the theme (Fido) and the agent (John). The syntactic operations merging the first and
the last are simple concatenations (right and left, respectively). The operation (F5) used
to merge the theme with the complex transitive expression give to Mary is the familiar
Right Wrap operation discussed in the General Introduction.
Consider now Dowty’s analysis of the double object form for the same sentence (100). The
notation “Dative Shift” in the derivation portion is Dowty’s own (cf. Dowty 1978, p. 400):26
John gives Mary Fido, t, F4 λx[give’(m)(f)(x)] ≡ give’(m)(f)(j)
John, T give Mary Fido, IV, F5 λzλx[give’(z)(f)(x)](m) ≡ λx[give’(m)(f)(x)]
Mary, T give Fido, TV, F22 λyλzλx[give’(z)(y)(x)](f) ≡ λzλx[give’(z)(f)(x)]
give, TTV Fido, T λyλzλx[give’(z)(y)(x)]

(Dative Shift) →
give, DTV give’
From a syntactic point of view, this derivation matches (99) in employing the same
sequence of merge operations, applied in the same order, proceeding from bottom to
top: Right Concatenation, Right Wrap, Left Concatenation. However, these opera-
tions apply to the theme and goal arguments taken in the opposite order from (99).
That the ultimate semantic outcome is the same as (99) results from a category-
changing rule that Dowty (1978) labels “Dative Shift.” The latter applies to the
lexical verb give in (100), changing it from a DTV-expression to a TTV-expression,
and changing its translation from give’ to λyλzλx[give’(z)(y)(x)]. The latter inverts
the semantic order of combination between the theme and goal arguments by means
of lambda abstraction. But note that the element give’ appearing within the complex
lambda expression is the very same one that appears in (99). Thus Dowty’s analysis,
while not appealing to a syntactic operation that shifts the theme and goal expres-
sions, appeals to a semantic operation that has the same effect vis-à-vis the entities
those expressions denote: it inverts their order of composition. Accordingly, just as
the Partee-style syntactic approach derives the double object form from the oblique
form, so Dowty’s semantic approach derives the interpretation of the predicate used
in the double object derivation from the interpretation of the predicate used in the
oblique derivation.
It’s worth noting that Dowty’s semantic analysis, which makes use of category-
changing rules, could as easily be expressed in terms of an argument-inverting func-
tor Ø with null phonology that would take a ditransitive verb (DTV) like give as an
argument and yield an expression give-Ø of category TTV. A translation for Ø is given
in (101), where ℛ is a variable over expressions of the type of DTV’s. This would
produce the same argument inversion that was achieved by the category-changing
rule in (100):27
John give-Ø-s Mary Fido, t, F4 λx[give’(m)(f)(x)] ≡ give’(m)(f)(j)
John, T give-Ø Mary Fido, IV, F5 λzλx[give’(z)(f)(x)](m) ≡ λx[give’(m)(f)(x)]
Mary, T give-Ø Fido, TV, F22 λyλzλx[give’(z)(y)(x)](f) ≡ λzλx[give’(z)(f)(x)]
give-Ø, TTV Fido, T λℛλyλzλx[ℛ(z)(y)(x)] (give’) ≡

λyλzλx[give’(z)(y)(x)]
Ø, APP give, DTV give’
This is in fact a natural way to interpret the contribution of applicative morphemes
within Dowty’s approach: as argument-inverting operators similar to those figuring in
his treatment of Passive; hence we might label the argument-inverting functor “APP”
to signify this connection.
2.1.3. How Many “Computational Engines”?

The Partee-style syntactic approach and the Dowty-style semantic approach plainly
constitute different analyses of the dative/applicative alternation. But, as we’ve noted,
they also share a core commitment to the oblique form as basic, and derive the double
object form from it. The syntactic approach does so by deriving double object expres-
sions from oblique form expressions by means of a transformation. In this it closely
resembles Baker’s (1988) account, simply deploying a different rule to make the connec-
tion: Dative Shift versus Preposition Incorporation. The semantic approach also takes
the oblique form as basic, deriving double object verb interpretations from oblique form
verb interpretations by means of a category-changing rule (or applicative functor ele-
ment). In this sense, the semantic account is as much committed to Dative Shift as the
syntactic account. Neither is nonderivational in a deep sense insofar as neither involves
projection of the alternation from independent sets of formatives.
Choosing between a syntactic approach and a semantic approach would evidently
hinge on detailed predictions the two make regarding specific phenomena. But there is
a clear sense, I believe, in which we do want to choose between the two approaches and
not combine elements from both.
A Partee-style syntactic account, in its pure form, can be taken as instantiating the
view that semantic composition is inherently invariant and simple, with apparent devia-
tions from compositionality arising through computational processes in the syntactic
component. When expressions are “not where they should be” for composition, it’s not
because anything has changed in the semantics; it’s because items have been moved or
otherwise rearranged in the syntax. The upside of syntactic computational complexity is
semantic computational simplicity, including a translation rule for Dative Shift and Pas-
sive as trivial as the identity mapping. Baker’s (1988) Preposition Incorporation analysis
inherits these same properties: the syntactic complexity of (94a) would be essentially
semantically vacuous; those aspects of semantics bearing on argument structure would
be constant for (94a) and (94b), deriving specifically from the latter.
The Dowty semantic account can be taken as instantiating the opposite view: that syn-
tactic composition is inherently invariant and simple, with deviations from composition-
ality arising through computational processes in the semantic component. Again, when
expressions are not where they should be for composition, it’s not because they have been
rearranged or moved from a more semantically transparent position; it’s because the inter-
pretations of the parts have changed. Thus in (99) and (100), as we saw, the same simple set
of merge operations apply in the same sequence to expressions taken in inverse order. The
ultimate semantic result is nonetheless the same because the semantic combining structure
of the predicate had been computationally altered in the double object case.
It seems clear that, ideally, we would like to be able to confine computational com-
plexity to a single component. In the suggestive phrasing of Marantz (1997), we would
like to be able to rely on a “single computational engine.” In a theory like Dowty’s,
in which the “engine” is semantic, this means analyzing productive and semiproduc-
tive alternations and grammatical function–changing operations like passive, middle
formation, unaccusativity, unspecified object deletion, raising to subject, and raising
to object as semantic processes, not involving movement or any sort of departure from
surface syntax. In the Partee-style theory, the conclusion is the opposite: that we try
to see complexity in the syntax-semantics map as arising in the syntactic side, with a
genuinely simple compositional semantics underneath. Plainly, what we do not want is
a theory making indiscriminate use of computational complexity in both components,
for example, a theory that maintains a syntactic analysis of passive, in which subject
arguments are analyzed as having moved, and a theory of the dative alternation making
use of category-changing rules or argument-inverting operators like those in (100) or
(101). Such a theory would be fundamentally unprincipled, and undesirable.
2.2. “Nonderivational” Analyses of Double Object Constructions

With these remarks in mind, let us now consider two current nonderivational approaches
to dative projection: the applicative analyses of Marantz (1993) and Pylkkänen (2002,
2008), and the neo–Generative Semantics analysis of Harley (2002).
2.2.1. The Applicative Analysis

Marantz (1993) offers structure (102) as an analysis of the English double object form
John bought Mary a book. The structure resembles that of Larson (1988) in invoking
VP shells, but differs importantly in its view of the higher V -head. Marantz (1993)
analyzes the latter neither as an empty position nor as a null causative, but rather as an
applicative element (appl), in explicit recognition of the tie between the double object
and the applicative construction discussed above.28
(102) IP
NP Iʹ
John I VP
NP Vʹ
Mary V VP
APPL NP Vʹ
a book V X
buy
Marantz (1993) takes this structure as entirely independent of the corresponding oblique
form, viewing appl (contra Marantz 1984 and Baker 1988) as nonprepositional and unre-
lated to the P element appearing in oblique constructions. Indeed, in its discussion of appli-
catives, Marantz (1993) makes virtually no reference to oblique constructions at all.
What then is the theory of projection underlying (102)? In fact, although Marantz
discusses a variety of interesting syntactic consequences for this structure, his remarks
on its projection are entirely programmatic and metaphorical. Marantz (1993) offers
the general view that specifiers of applicative heads (Mary in (102)) are affected objects,
but provides neither a definition of affectedness nor tests for detecting its presence
beyond occurrence in the applicative construction itself.29 As discussed earlier, silence
on this point is not benign. Affectedness, as the term is normally understood, appears
to be at most an implicature in double object constructions, and one readily canceled in
discourse. The first conjunct of (103a) is in double object form, for example; nonetheless,
the continuation in the second conjunct is unproblematic. Furthermore, given the con-
tinuation, it’s difficult to grasp what sense of “affectedness” might attach to the indirect
object Mary—it’s hard to see how Mary could be affected by a book she didn’t receive
and may indeed have never even known about. It is likewise difficult to see what sense
of affectedness might be implied in (103a) that is not also present in (103b) and would
be sufficient to underwrite their separate projection in syntax.
(103) a. John bought Mary a book but never gave it to her.
b. John bought a book for Mary but never gave it to her.
Beyond reference to affectedness, Marantz (1993) adverts to the roles that various kinds
of items (instruments, themes, beneficiaries, etc.) might play in a “natural theory of
event composition”: what elements are or are not central to events, and hence included
or not included in various tree parts. But an explicit theory is never provided, to say
nothing of a means for deducing its syntactic consequences. As such the enterprise seems
largely circular, with the cognitive metaphysics of events deduced from the tree struc-
tures that are assumed in the analysis, rather than the other way around.
More recently, Pylkkänen (2002, 2008) has attempted to supply the missing theory
of projection for Marantz-style structures, providing an explicit compositional seman-
tics for the account. The results are instructive. Developing a suggestion by Marantz
(1993), Pylkkänen argues that there are in fact two kinds of applicative constructions,
which she terms “high applicatives” and “low applicatives.” The general idea can be
illustrated via the two English for-phrases in (104a). The nominal in the inner for-PP
denotes the intended recipient of the pie-baking and can also appear as an applied object
in an English double object structure (104b). The nominal in the outer for-PP denotes
the intended beneficiary of the pie-baking, with the “favor” and/or “substitutive” read-
ing discussed earlier (recall (83) and (84)). Nominals in this second kind of for-PP can-
not appear as applied objects in English (104c).
(104) a. John baked a pie [for Mary] [for Susan].
b. John baked Mary a pie [for Susan]. (same meaning as a.)
c. *John baked Susan a pie [for Mary]. (same meaning as a.)
In traditional approaches to phrase structure, the ordering of the two for-PPs would be
accounted for as a difference in attachment height (105); hence we might refer to them
as “high for-phrases” versus “low for-phrases”:
(105) VP
VP PP
V NP PP for Susan ← High beneficiary for-phrase
bake a pie for Mary ← Low recipient for-phrase
Although English disallows high beneficiaries in the double object form, in other
languages, as we have noted, the counterpart applicative constructions are possible.
Pylkkänen (2002, 2008) offers the Chaga example in (106a) (cited from Bresnan and
Moshi 1993) as a case in point and gives the Marantz-style tree in (106b) as an illustra-
tion of its semantic composition:
(106) a. N-ä-ï-lyì-í-à m-kà k-élyá.
FOC-1SG-PRES-eat-APPL-FV 1-wife 7-food
‘He is eating food for his wife’.
b. VoiceP
He λxλe[eating(e) & Agent(e,x) & Theme(e,food) & Benefactive(e,wife)]
Voice λe[eating(e) & Theme(e,food) & Benefactive(e,wife)]

λxλe[Agent(e,x)]
wife λxλe[eating(e) & Theme(e,food) & Benefactive(e,wife)]
Appl λe[eating(e) & Theme(e,food)]

λxλe[Benefactive (e,x)]
eat food
λxλe[eating(e) & Theme(e,x)]
A notable aspect of this derivation is its crucial use of a semantic operation from
Kratzer (1996) called “Event Identification” (EI). EI is used in putting the Voice and
Appl(icative) heads together with their complements. EI is neither function application
nor simple conjunction, but rather a special, stipulated combination of the two, cre-
ated specifically for combining thematic role–bearing heads with their event-denoting
complements:
(107) Event Identification (Kratzer 1996, p. 122)
f g → h λxλe[f(x)(e) & g(e)]
<e,<s,t>> <s,t> <e,<s,t>>
In prose, this operation puts a thematic relation together with an event-denoting expression
by saturating the internal argument of the first, conjoining the result with the second, and
then lambda-abstracting over the internal argument of the thematic role, allowing it to
combine “anti-compositionally” with an argument to come. In other words, EI does with
thematic relations like Agent, Benefactive, and so on essentially what Dowty’s relation-
inverting Dative Shift operation does with give: it “twists” their basic composition order.
To convince oneself of this assessment—that EI is simply a relation-inverting device
needed to reconcile the basic compositionality of thematic relations like Agent and
Benefactive with trees like (106b)—consider the interpretation of a tree containing a
high beneficiary for-phrase of the sort discussed above.
(108) a. John [ate food for his wife].
b. λe[eating(e) & Theme(e,food) & Benefactive(e,wife)]
λe[eating(e) & Theme(e,food)] λe[Benefactive(e,wife)]
eat food for wife

λxλe[eating(e) & Theme(e,x)] λxλe[Benefactive (e,x)]
(108b) is the oblique English counterpart of the Chaga high applicative. It involves
the same semantic elements as the phrasal sister to Voice in (106b), simply combin-
ing them in a different order. Note crucially, however, that in contrast to (106b),
(108b) requires nothing beyond function application and standard predicate con-
junction to secure the same result—no stipulated, relation-inverting operation of
“EI” is needed. In a clear sense then, (108b), the oblique form, could be said to
represent the “natural order of composition” for the elements yielding the benefac-
tive reading of the high applicative. Pylkkänen’s treatment of high applicatives can
thus be viewed as derivational: it requires Dowty-like relation inversion applied to
the predicates responsible for the oblique derivation in order to achieve the correct
semantic outcome.
A similar result holds for Pylkkänen’s (2002, 2008) treatment of low applicatives,
although there are additional complications. Pylkkänen advocates an analysis of recipi-
ent possessives and sources in which, as she puts it, “low applied arguments bear no
semantic relation to the verb whatsoever: they bear only a transfer of possession rela-
tion to the direct object” (2008, p14 ).30 (109b) displays the semantic analysis that
Pylkkänen recommends for (109a) (ignoring irrelevant details):
(109) a. John wrote Mary that letter.
b.
∃e[writing(e) & Agent(e,John) & Theme(e,that_letter) &
to-the-possession-of(that_letter,Mary)]
Here the referent of Mary is not related to the event quantification by means of any
binary thematic relation such as Goal; rather it is related directly to the referent of the
theme argument by means of to-the-possession-of(x,y). This semantics figures impor-
tantly in Pylkkänen’s account of the differences between high and low applicatives. As
we’ve seen, high applicatives can combine at a relatively high point in the derivation
since they only require access to the event variable of the verb. Low applicatives, by
contrast, must combine earlier in the derivation, in the local vicinity of the direct object
to which they are related.31
Pylkkänen (2008) proposes the interpretations for low applicative morphemes in
(110) and the derivation for the low applicative English sentence Mary bought John a
book shown in (111):
(110) a. Low-appl-to (Recipient applicative):
λxλyλf<e,<s,t>>λe[ f(e,x) & Theme(e,x) & to-the-possession-of(x,y)]
b. Low-appl-from (Source applicative):
λxλyλf<e,<s,t>>λe[ f(e,x) & Theme(e,x) & from-the-possession-of(x,y)]
(111) VoiceP
Mary λxλe[buying(e) & Agent(e,x) & Theme(e,book) & to-the-possession-of(John,book)]
Voice λe[buying(e) & Theme(e,book) & to-the-possession-of(John,book)]

λxλe[Agent(e,x)]
buy λf<e,<s,t>>λe[ f (e,x) & Theme(e,book) & to-the-possession-of(book,John)]
John λyλf<e,<s,t>>λe[ f(e,x) & Theme(e,book) & to-the-possession-of(book,y)]
Appl the book

λxλyλf<e,<s,t>>λe[ f(e,x) & Theme(e,x) & to-the-possession-of(x,y)]
The derivation in (111) is peculiar in several respects. Note, for example, that the syn-
tactic head-complement relation between buy and [John appl the book] conflicts with
the semantic functor-argument relation involved. According to (111), it is the phrase
[John appl the book] that semantically selects the head buy, not the other way around.
More relevant for our purposes, however, is the argument abstraction order in the
translation of the low applicative morpheme. In the relation to-the-possession-of(x,y), y
is represented as the internal argument and x is represented as the external argument.32
However, abstraction on these variables in (110a) occurs in the opposite order: λxλy . . .
The translation of the low applicative morpheme thus doesn’t merely relate the theme
and goal arguments; it inverts their basic composition order with respect to the predi-
cate to-the-possession-of(x,y).
That relation inversion is an artifact of the applicative structure can once again
be seen by considering a counterpart oblique derivation, holding other elements
of Pylkkänen’s syntax and semantics constant. Observe that in (112a) argument
abstraction with for is in the order λyλx—the basic composition order of the to-
the-possession-of(x,y) relation itself (and similarly for from in (112b)). This allows
the derivation in (113), which derives the same result as Pylkkänen’s (111) without
relation inversion:
(112) a. for (Recipient preposition):
λyλxλf<e,<s,t>>λe[ f(e,x) & Theme(e,x) & to-the-possession-of(x,y)]
from (Source preposition):

b.
λyλxλf<e,<s,t>>λe[ f(e,x) & Theme(e,x) & from-the-possession-of(x,y)]
(113) VoiceP
Mary λxλe[buying(e) & Agent(e,x) & Theme(e,book) & to-the-possession-of(book,John)]
Voice λe[buying(e) & Theme(e,book) & to-the-possession-of(book,John)]

λxλe[Agent(e,x)]
buy λf<e,<s,t>>λe[ f(e, book) & Theme(e,book) & to-the-possession-of(book,John)]
the book λxλf<e,<s,t>>λe[ f(e,x) & Theme(e,x) & to-the-possession-of(x,John)]
for John
λyλxλf<e,<s,t>>λe[ f (e,x) & Theme(e,x) & to-the-possession-of(x,y)]
Pylkkänen appears to be aware of compositionality problem imposed by her syntax.

She observes explicitly that in I wrote John a letter, the surface c-command properties
don’t track the compositional semantics, giving the diagrams in (114a,b) to represent
the basic conundrum (= (14a,b) in Pylkkänen 2008, p. 17). Accepting (114a) as syn-
tactically basic, but needing (114b) for her semantics, she appeals to computational
complexity (relation inversion) in the semantics to reconcile the two:33
(114) a. RIGHT C-COMMAND PROPERTIES b. RIGHT SEMANTIC RELATION
John a letter
APPL a letter APPL John
But of course Pylkkänen’s solution is not the only one possible. (114a) could be ana-
lyzed as syntactically derived from an oblique, which, as we’ve seen, represents the
natural composition order even under Pylkkänen’s own semantics. Indeed, this move
would be natural given Pylkkänen’s appeals to standard A-movement operations like
Passive in other parts of her analysis, and given the desideratum noted above that we
would like a theory that deploys a single computational engine, assigning computa-
tional complexity to the syntax or the semantics, but not to both. Pylkkänen (2002,
2008) does not do this, however, producing a hybrid theory that can be described
not only as semantically derivational but also as undesirable in the sense discussed
earlier.
Modern applicative approaches to double objects are thus seen to have the fol-
lowing properties. To the extent that they succeed in ignoring oblique-applicative
correspondences of the kind noted in (93), they are unable to supply a precise and
substantial theory of projection, one based on more than undefined notions of affect-
edness or gestures at a “natural theory of event composition” still awaiting discovery
(Marantz 1993). To the extent that they do provide a precise and substantial theory
of projection and semantic composition, this is seen to be parasitic on the seman-
tics of the oblique domain—a derivational theory in the sense of Dowty (1978). To
obtain the proper semantic results for her applicative structures, Pylkkänen (2002,
2008) must rely on relation-inverting operators whose inputs are the basis of oblique
form semantic composition. The resulting theory is one that spreads computational
complexity into both the syntactic and semantic components, an undesirable result,
as we have noted.
These negative assessments of modern applicative approaches apply strictly to the
projection and/or interpretation of their structures, and do not preempt the essential
correctness of those structures themselves. As I will suggest below, it is possible to recast
Marantz’s basic proposal within a derivational theory that preserves many of its con-
siderable virtues and allows a semantics along the general kind Pylkkänen (2002, 2008)
suggests, but without appeal to relational inversion.
2.2.2. Neo–Generative Semantics

Consider now the approach of Harley (1995, 2002), developing ideas by Pesetsky
(1995). The basic idea is a variant of the neo–Generative Semantics analysis already
discussed in relation to (17), where a higher causative V embeds a small clause express-
ing possession versus motion. An innovation in the Pesetsky and Harley accounts is their
analysis of the small clause as of category PP, not VP (115a,b).
(115) a. vP b. vP
DP Vʹ DP Vʹ
John V PP John V PP
CAUSE DP Pʹ CAUSE DP P
Mary P DP a letter P PP
PHAVE a letter PLOC to Mary
In support of separating double object and PP dative structures in this way, Harley
cites evidence of the sort reviewed above. She discusses differences in affectedness
(8)–(9), the nonavailability of double object counterparts to PP datives with nonani-
mate to-objects (18a,b), the fact that caused-motion idioms have no double object
counterparts (24a–d), and the existence of double object idioms (section 1.4.2). Some
of these arguments are based on simple misanalysis, as we have seen. Thus differences
of affectedness were seen to be implicatures and not a plausible basis for projection;
furthermore, so-called double object idioms were demonstrated to be no such thing.
However, the central deficiency in Harley’s analysis is the one observed by Rappaport
Hovav and Levin (2008), namely, its view (shared by many other researchers) that PP
datives uniformly encode caused motion. We have seen compelling evidence that this
claim is simply incorrect: that PP datives also encode caused possession. This is particu-
larly evident in the case of give, which shows none of the properties of caused-motion
forms in its oblique frame.
Nonetheless, although these considerations undermine the specific arguments Harley
(2002) offers for separate projection of double object and oblique dative structures, they
do not necessarily preempt the possibility of separate projection. Earlier, following the
conclusions of Rappaport Hovav and Levin (2008), we suggested the picture in (27),
where the oblique dative form encodes both caused possession and caused motion. We
used the English verb belong to express the oblique possession frame. A similar view
might be recruited within the general Harley-Pesetsky approach. That is, we might
potentially appeal to a third, prepositional small clause head PBELONG, as in (116b).
(116) a. VP b. VP c. VP
DP Vʹ DP Vʹ DP Vʹ
John V PP John V PP John V PP
CAUSE DP Pʹ CAUSE DP Pʹ CAUSE DP Pʹ
Mary P DP a letter PP a letter DP

P P
PHAVE a letter PBELONG to Mary PLOC to Lisbon
CAUSED POSSESSION CAUSED POSSESSION CAUSED MOTION/LOCATION
Arguments for the separate projection of oblique dative and double object form in verbs
of caused possession would then need to demonstrate the independent existence of PHAVE
and PBELONG and elucidate their specific properties.
It is unclear whether such arguments can be given, and indeed the challenges seem
formidable. For example, it does not appear possible to investigate PHAVE and PBELONG
by reference to the overt English verbs have and belong. English belong, as a posses-
sive, is not counterpart to have, but rather to verbs like possess and own, which refer
specifically to ownership (117a) and do not express any of the extended senses of have
(117b,c).
(117) a. A: Who owns this car?
B: It belongs to me.
b. Mary had/#owned/#possessed an idea/the flu/a terrible time/a baby/a bath/a stroke…
c. #An idea/The flu/A terrible time/A baby/A bath/A stroke … belonged to Mary.
Overt English belong is thus clearly not Pbelong.

Languages like Korean appear to show transitive-oblique possession pairs in which
the ownership sense and the extended senses of English have are equally expressed
(118)–(121).34
(118) a.
Chulsu-ege cek-i itta.
Chulsu-dat book-nom be
b. Chulsu-ga cek-ul kajigo-itta.
Chulsu-nom book-acc have-cop
(119) a. Chulsu-ege (coeun) sangkak-i itta.

Chulsu-dat good idea-nom be
b. Chulsu-ga (coeun) sangkak-ul kajigo-itta.
Chulsu-nom good idea-acc have-cop
(120) a. Chulsu-ege munjae-ga itta.

Chulsu-dat problem-nom be
b. Chulsu-ga munjae-ul kajigo-itta.
Chulsu-nom problem-acc have-cop
c. Chulsu-ga sahui-sanghwal-e munjae-ul kajigo-itta.
Chulsu-nom social-life-loc problem-acc have-cop
‘Chulsu has a problem with his social life.’
(121) a. Chulsu-ege yel-i itta.

Chulsu-dat fever-nom be
b. Chulsu-ga yel-ul kajigo-itta.
Chulsu-nom fever-acc have-cop
c. Chulsu-ga simjang-e yel-ul kajigo-itta.
Chulsu-nom heart-loc fever-acc have-cop
‘Chulsu has a fever in his heart.’
Nonetheless, the differences between these two constructions seem to be mainly mat-
ters of information structure, and not something on which projection might be taken
to rest. The Korean speaker providing (118)–(121) suggested that in the copular “a”
examples, the dative possessor had a topicalized sense not found in the “b” sentences.
Thus (119a) might be a natural response to a request for advice on a difficult situation.
The speaker, having nothing himself to suggest, utters (119a) to convey that Chulsu (in
contrast to others) has an idea. This sense does not attach to (119b). Conversely, the
transitive forms were perceived to convey specificity in the accusative-marked nominal.
(120b) and (121b) were judged odd as expressing problems of indeterminate nature or
an unspecified illness marked by elevated temperature. Preferable were examples like
(120c) and (121c), where the problem and illness are made concrete.
Justifying a PHAVE-PBELONG pair seems but one special case of a much broader challenge
raised by applicative constructions for the Harley-Pesetsky program. Accepting applica-
tives to represent the general form of the double object construction, a Harley-Pesetsky
separate projection theory appears committed to finding symmetric preposition pairs, not
just for possession (PHAVE/PBELONG), but indeed for all of the relations underlying applica-
tive-oblique alternations. That is, it would need to identify distinct prepositional small
clause (SC) heads for all the applicative and oblique constructions surveyed above:
(122)
Appl SC head Oblique SC head Sense
a. PHAVE ⇔ PBELONG possession
b. P?? ⇔ PLOC motion/location
c. P?? ⇔ PFOR benefactive/malefactive/substitutive
d. P?? ⇔ PWITH instrumental
e. P?? ⇔ PIN? manner
f. P?? ⇔ PBECAUSE reason
The prospects seem dubious, not merely given the empirical challenge of arguing for the
relevant pairs, but because the small clause analysis itself seems inadequate to the range of
cases that applicatives instantiate. Consider, for example, the applicative-oblique instru-
mental pair in (123a,b), where I use English words even though English (unlike Chichewa)
doesn’t permit instrumental double objects:
(123) a. vP b. vP
DP Vʹ DP Vʹ
John V PP John V PP
CAUSE DP Pʹ CAUSE DP Pʹ
the key P DP the lock P Pʹ

the rock the vase
P?? the lock PWITH with the key
the vase with the rock
*John opened the key the lock John opened the lock with the key
*John broke the rock the vase John broke the vase with the rock
A Harley-style account appears committed to finding a single, general, predicational rela-
tion counterpart to PHAVE underlying (123a), and a single, general, predicational relation
counterpart to PBELONG underlying (123b). What these relations might be, however, is quite
mysterious, to me at any rate.35
The difficulty encountered here reflects, I believe, a fundamental semantic error in the
neo–Generative Semantics/small clause approach to double object constructions in the
first place, that is, the widespread idea that in an example like (124a) the indirect object
(Mary) stands in a predicational relationship to the direct object (a book), justifying a
semantic (and hence syntactic) representation like (124b).
(124) a. John gave Mary that book.
b. [John cause [Mary have that book]]
This view has some initial plausibility when attention is restricted to datives/benefac-
tives and locatives. But it does not appear sustainable across the full range of comple-
ment relations that applicatives exhibit.36
The commitment to small clauses at the base of the neo–Generative Semantics pro-
gram is not inescapable or forced on us. Alternative semantic conceptions are not only
possible but arguably superior. Compare (124b) to the neo-Davidsonian event-semantic
representation in (125b):
(125) a. John gave Mary that book.
b.
∃e[giving(e) & Agent(e,John) & Theme(e,that_book) & Goal(e,Mary)]
In the latter, the referents of the direct and indirect objects bear no direct predicational
relationships to each other at all. Rather, their connection is secured via the giving event
e, to which each argument is independently related through a binary thematic role
(Agent, Theme, Goal). In the Davidsonian semantic program, such radical “separation”
of arguments from the predicate and from each other has been argued to be crucial for
obtaining a correct semantics of plurality and conjunction (Schein 1993, forthcom-
ing; Pietroski 2005), and for analyzing many other empirical phenomena (see Parsons
1991). This semantics is also directly extensible to instrumental cases like (123a,b) for
which the predicational small clause account yields no counterpart analysis:
(126) a. ∃e[opening(e) & Agent(e,John) & Theme(e,the_lock) & Instrument(e,the_key)]
b.
∃e[breaking(e) & Agent(e,John) & Theme(e,the_vase) & Instrument(e,the_rock)]
I conclude that nonderivational accounts of the dative alternation in neo–Generative

Semantics bear a considerably greater explanatory burden than their exponents have
realized. At the narrowest, they owe an account of PHAVE and PBELONG; in what ways do
these two possessive relations differ semantically/thematically, and how do these dif-
ferences support independent projection? They owe an account of how their general
small clause analysis can be extended across the range of applicative structures of
which the English double object construction appears to be but one parochial instance
(or a justification for not applying their analysis to applicatives generally). And, at
the broadest level, they owe a defense of the small clause semantics against modern
neo-Davidsonian views that provide no basis for their core predicational assumptions
and offer convincing arguments against them. At present, these burdens appear to me
quite substantial.
2.3. The Derived Status of Double Object Forms

The considerations reviewed above provide no secure basis for a nonderivational
approach to the applicative-oblique alternation—no properties of obliques versus appli-
catives that would support their separate projection. We might ask, then, whether there
are arguments to the contrary: are there arguments in favor of derivation?
2.3.1. Informational Restrictedness

One general pattern in derived forms is that they show informational restrictedness
vis-à-vis underived ones. Active-passive pairs, for example, are known to differ in
this way with respect to the encoding of old and new information. Ward and Birner
(2004) observe that, in general, the derived subject of a passive must not represent
newer information in the discourse than the NP in the by-phrase. Thus in (127a) the
passive subject he represents old information as compared to the by-phrase object
Ivan Allen Jr., and the sentence is felicitous. By contrast, in (127b) the passive sub-
ject the mayor represents new information as compared to the by-phrase object him
(= Ivan Allen Jr.), and the sentence is infelicitous. (127c) shows that this constraint
does not govern an active sentence. Having the pronoun later is acceptable in an
underived sentence.
(127) a. The mayor’s present term of office expires January 1.

He will be succeeded by Ivan Allen Jr.
b. Ivan Allen Jr. will take office January 1.
#The mayor will be succeeded by him.
c. Ivan Allen Jr. will take office January 1.

The mayor will swear him in.
A similar difference appears to hold between PP datives and double object constructions
(128a,b). Thus in (128a) the indirect object him represents old information as compared
to the direct object a medal, and the sentence is felicitous. By contrast, in (128b) the
indirect object John represents new information as compared to the direct object one
(= a medal), and the sentence is infelicitous.37 (128c) and the sentences in parentheses in
(128a,b) show that this constraint does not govern oblique forms. Having the pronoun
later is acceptable.
(128) a. John will be honored at the ceremony.

The mayor will be awarding him a medal.
(cf. The mayor will be awarding a medal to him.)
b. Medals will be given at the ceremony.
#The mayor will be awarding John one (of them).
(cf. The mayor will be awarding one (of them) to John.)
c. Short speeches will be given at the ceremony.

The mayor will mention John in several of them.
Similar effects have been reported for other languages. For example, it has been
argued persuasively that Japanese dative ditransitives in the word order [VP dp-dat
dp-acc V] are double object constructions (Zushi 1992; Miyagawa and Tsujioka
2004; Harada and Larson 2009). Kaiser and Nakanishi (2001) report that whereas
the acc-dat is unrestricted in that either argument can be specific/nonspecific or rep-
resent old or new information, the dat-acc order is constrained. Specifically, the
dat argument must be specific or represent old information. Kaiser and Nakanishi
demonstrate this by means of test frames like that given (in English) in (129), where
“S” “IO,” and “DO” stand for “subject,” “indirect object,” and “direct object,”
respectively. Japanese permits deletion of an argument introduced previously into the
discourse. Thus in (129a,b) the occurrence of IO in the while-clause licenses the null
anaphoric element (∅IO) in the main clause.
(129) a. DO-IO: While I have heard [that S1 DO IO V], it seems [that S2 DO ∅IO V].
IO-DO: While I have heard [that S1 IO DO V], it seems [that S2 ∅IO DO V].
b.
In the usual way, the specific/nonspecific status of the antecedent determines how the
null anaphoric element is understood. Thus, if speakers judge ∅IO to refer to the same
individual(s) as in the while-clause, IO is specific or represents old information. If speak-
ers judge that ∅IO can refer to different individual(s), then IO is not informationally
restricted in this way.
Kaiser and Nakanishi (2001) tested this paradigm for Japanese with the examples
in (130):
(130) a. DO-IO
Taroo-ga Hanako-o [Penn-no gakusei]-ni syookaisita
Taro-nom Hanako-acc [Penn-gen student]-dat introduced
to kiita kedo Jiroo-mo Hanako-o ∅IO syookaisita-rasiiyo.
COMP heard while Jiro-too Hanako-acc introduced-seem
‘While (I) have heard that Taro introduced Hanako to a Penn student,
it seems that Jiro introduced Hanako (to a Penn student), too.’
IO-DO
b.
Taroo-ga [Penn-no gakusei]-ni Hanako-o syookaisita
Taro-nom [Penn-gen student]-dat Hanako-acc introduced
to kiita kedo Jiroo-mo ∅IO Hanako-o syookaisita-rasiiyo.
COMP heard while Jiro-too Hanako-acc introduced-seem
‘While (I) have heard that Taro introduced to a Penn student Hanako, it seems
that Jiro introduced (to a Penn student) Hanako, too.’
The results were revealing. In (130a), it was judged that ∅IO could refer to the same
Penn student or to a different one, but with (130b), ∅IO was overwhelmingly taken to
refer to the same student. In other words, IO was judged to represent old information.
Kaiser and Nakanishi (2001) also tested this paradigm on cases where the direct
object was elliptical in the second clause (∅DO), and where word order between IO and
DO was permuted (131). Their Japanese test examples are repeated in (132):
(131) a. IO-DO: While I have heard [that S1 IO DO V], it seems [that S2 IO Ø do V].
DO-IO: While I have heard [that S1 DO IO V], it seems [that S2 Ø do IO V].
b.
(132) a. IO-DO
Taroo-ga Hanako-ni [gengogaku-no hon]-o miseta
Taro-nom Hanako-dat [linguistics-gen book]-acc showed
to kiita kedo Jiroo-mo Hanako-ni Ø do miseta-rasiiyo.
COMP heard while Jiro-too Hanako-dat showed-seem
‘While (I) have heard that Taro showed to Hanako a book on linguistics,
it seems that Jiro showed to Hanako (a book on linguistics), too.’
DO-IO
b.
Taroo-ga [gengogaku-no hon]-o Hanako-ni miseta
Taro-nom [linguistics-gen book]-acc Hanako-dat showed
to kiita kedo Jiroo-mo Ø do Hanako-ni miseta-rasiiyo.
COMP heard while Jiro-too Hanako-dat showed-seem
‘While (I) have heard that Taro showed a book on linguistics to Hanako, it
seems that Jiro showed (a book on linguistics) to Hanako, too.’
With an elliptical direct object, the results were surprisingly different from the previ-
ous ones. In both (132a,b), speakers indicated no strong preferences as to whether ∅DO
could be the same or different—old or new information. Both interpretations were
equally available. Kaiser and Nakanishi (2001) interpret their results as follows: since
the DO-IO (i.e., ACC-DAT) order is the informationally less restricted, pragmatically
more neutral order, it in fact represents the basic order, despite speaker intuitions to the
contrary and despite standard textbook presentations. They go on to suggest that the
IO-DO (i.e., DAT-ACC) order of Japanese derives by movement, insofar as it shows
much the same specificity and old-information effects as those induced by scrambling
and object shift.
Research on applicative languages appears to yield the same picture. In two stud-
ies of discourse function in Salish applicatives, Gerdts and Kiyosawa (2005a, 2005b)
demonstrate the persistent interpretive accompaniment of applicative formation to be
discourse prominence: “Either the outcome of the action affecting the object is central
to the story or the applied object itself is highly topical . . . The function of applica-
tives thus parallels the function of passives, which are used in many languages to place
a patient that is more central than the agent into subject position” (p. 24). Donohue
(2001), in a study of applicatives in the Austronesian language Tukang Besi, echoes this
result: “especially for goals, the appearance in an applicative construction is associated
with a higher degree of topicality and prominence in the narrative” (p. 254). Finally,
Peterson (2007), surveying applicatives in a sample of 55 languages, draws a similar
conclusion: “the primary motivation or purpose behind the use of applicative construc-
tions [is] the indication of high topicality status, and any more clearly morphosyntactic
function correlates with indication of an argument’s relative high topicality” (p. 84).
It thus appears that, like double object structures, applicatives are associated with a
form of informational restrictedness—what Gerdts and Kiyosawa refer to as an extra
interpretive “kick.”38 Specifically, the applied object is associated with old information/
high topicality.
2.3.2. Quantificational Restrictedness

A second point of restriction in double object forms is the so-called scope freezing effect,
first suggested to the author by David Lebeaux as a parallel to well-known facts about
passives. Chomsky (1957) notes that passivization typically affects interpretation of
quantifiers so that (133a) may be true while (133b) is false “under the normal inter-
pretation of these sentences—e.g., if one person in the room knows only French and
German, and another only Spanish and Italian” (p. 101):39
(133) a. Everyone in this room knows at least two languages. ∀ > 2; 2 > ∀
b. At least two languages are known by everyone in this room. 2 > ∀; ??∀ > 2
Lebeaux observed that an analogous, and if anything stronger, contrast appears to hold
for PP dative and double object structures. For many people, (134a) may be true in
circumstances where (134b) is false, for example, where each language in question is
taught to different pairs of individuals.
(134) a. John taught every language to two persons. ∀ > 2; 2 > ∀
b. John taught two persons every language. 2 > ∀; *∀ > 2
The double object form seems to “freeze” the scope of two quantified object nominals
whereas the oblique form does not.
As noted in Larson (1990), the scope freezing effect is interesting in that it does not
involve a fixed low scope for the second, outer quantifier, but rather a restriction on the
relative scopes of the goal-theme pair. Thus double object structures show Antecedent-
Contained Deletion in an outer quantified object (135a,b) (= (iia,b) in Larson 1990,
fn. 10):
(135) a. John gave someone [everything that Bill did [VP e ]].
b. Max wants to give someone [everything that you do [VP e ]].
On standard accounts of the Antecedent-Contained Deletion phenomenon (Sag 1976;

May 1985; Larson and May 1990), correct reconstruction of the elliptical VP requires
the quantified nominal containing it to receive scope at least as broad as the VP serving
as the reconstruction source. In the case of (135a) this entails that everything that Bill
did [VP e ] must receive scope at least as wide as the matrix VP in order to get its per-
ceived interpretation (136a). In the case of (135b) two scopes must be available given
the two possible interpretations of the sentence. There must be an embedded scope pos-
sible for everything that you do [VP e ] in view of the interpretation (136bi), and there
must be a matrix scope available in view of (136bii):
(136) a. ‘John gave someone everything that Bill gave him.’
b. i. ‘Max wants to give someone everything that you give him.’
ii. ‘Max wants to give someone everything that you want to give him.’
Importantly, again as noted in Larson (1990), the quantified goal phrase someone
must in all cases be understood with scope over the theme, whatever the latter’s scope.
Hence the representation for (135a) must be approximately as in (137a). And the rep-
resentations corresponding to the two readings of (135b) must be approximately as
in (137bi,ii) (where the empty VP and the VP serving as its reconstruction source are
indicated in boldface).
(137) a. [someone]i [everything that Bill do [VP e]]j John [VP gave ti tj ].
b. i. Max wants [[someone]i [everything that you do [VP e]]j PRO to [VP give ti tj ]].
ii. [someone]i [everything that you do [VP e ]]j [Max [VP wants PRO to give ti tj ]].
Thus, no matter where everything that you do [VP e ] occcurs for VP reconstruction, some-
one must occur higher.
Larson (1990) observes a similar effect with respect to intensionality phenomena.
The verb promise in (138) creates an intensional context in its scope. (138a) (= (iii)
in Larson 1990, fn. 10) allows both quantifiers to be read de dicto (in the Q1-Q2
order) (139a), the first to be read de re and the second de dicto (139b), or both to
be read de re (139c). What it does not allow, however, is for the second quantifier
to be read de re without the first being read de re as well, maintaining wide scope
over it (139d).40
(138) a. I promised to rent someone every apartment in the building.
b. [someone]i [every apartment in the building]j [I promise PRO to rent ti tj ].
(139) a De re De dicto
Q1, Q2 P (I made the following general promise: I will rent all the
apartments to a single individual)
b. Q1, Q2 P (There is some individual whom I promised to rent all the
apartments to)
c. Q1, Q2 P (There is some individual such that for each of the
apartments I promised to rent it to him/her)
d. Q2 Q1 X (For each of the apartments I promised to rent it to someone)
Once again the relative order of objects imposed by the double object form is preserved
under scope dislocation from the initial positions.
Bruening (2001) draws an interesting empirical comparison between scope preser-
vation in the double object phenomenon and a word order preservation phenomenon
found with object shift. He notes the paradigm in (140) (= (42) in Bruening 2001, 251)
from Icelandic, citing Collins and Thráinsson (1996):
(140) a. Ég lána ekki Maríu bækurnar.
I lend not Maria the books
‘I do not lend Maria the books.’
b. Ég lána Maríu ekki __ bækurnar.

c. ?Ég lána Maríu bækurnar ekki __ __.

d. *Ég lána bækurnar ekki Maríu __.

Thus, from the basic order in (140a), the goal phrase can shift leftward over a negation
or adverb at the VP edge (140b), or both goal and theme can shift leftward (with varying
degrees of degradation depending on the speaker). What is not possible, however, is for
a theme to shift over a goal, leaving the latter behind. The relative order goal-theme is
thus frozen. This phenomenon is strikingly reminiscent of the scope freezing phenom-
enon, as Bruening notes, and suggests a common explanation.
3. DERIVATIONAL APPROACHES TO THE DATIVE ALTERNATION
Having reviewed the challenges facing nonderivational approaches to the dative

alternation, now understood as one instance of the broader applicative-oblique alterna-
tion, and having seen some indications of derived status for double object/applicative
variants, it’s reasonable to consider alternative, derivational approaches in this domain.
3.1. Dative Shift and OBL2 Advancement

The idea of a derivational relation between English prepositional and double object datives
dates back to the earliest work in generative grammar. Chomsky (1955/1975, p. 493),
discussing question formation with example (141a) and observing its unavailability in
(141b) versus (141c), remarks: “This suggests there is a sentence more elementary than
[141a], namely, [141d], and that [141a] is derived from [141d] by a transformation Tα.”
(141) a. The teacher gave him several books. (= (235a) in Chomsky 1975, 492)
b. Whom did the teacher give the books? (= (239a) in Chomsky 1975, 493)
c. Whom did the teacher give the books to? (= (240a) in Chomsky 1975, 493)
d. The teacher gave several books to him. (= (241) in Chomsky 1975, 492)
Chomsky continues: “It seems to be true in general that for sentences of the form NP1-
V-NP2-to/for-NP3 there is a related form NP1-V-NP3-NP2 . . . We must then require that the
who-question transformation not apply to any Tα–transform, in order to eliminate [141b].”
The proposed transformation of “Dative Shift” or “Dative Movement” was subsequently
investigated by Fillmore (1965) and Green (1974) in major works, and by Emonds (1972),
Jackendoff and Culicover (1971), and Fischer (1972), among many others. It was also fre-
quently discussed in textbooks of the period (Burt 1971; Akamajian and Heny 1976; Baker
1977; Culicover 1976; Jacobs and Rosenbaum 1968; Keyser and Postal 1976).
The derivational proposals in Larson (1988, 1990, 1991) had this tradition as back-
ground, along with results from Relational Grammar (RG), which had begun to appear
widely in the 1980s (Perlmutter and Postal 1983a; Perlmutter and Rosen 1984). RG
offered a novel and appealing theory of sentence structure whose primitives were not
the usual configurational ones of linear precedence and dominance but instead gram-
matical relations like subject (the “1-relation”), object (“2-relation”), indirect object
(“3-relation”), and so on. In place of underlying representations that assigned items to
initial structural positions, RG assigned items to initial grammatical relations (“initial 1,”
“initial 2,” etc.). And in place of derivations wherein items changed structural positions,
RG offered a theory in which they changed grammatical relations. Thus whereas the
standard view of Passive involved its moving an NP from a position directly dominated
by VP to one directly dominated by S, RG analyzed Passive simply as changing the gram-
matical relation of a phrase from object to subject—“2 to 1,” as it was put in RG terms.
RG and configurational theories tracked each other to some extent. Tree structures,
particularly those with binary branching, induced hierarchical asymmetry among
phrases, making it possible to talk of one phrase as being “higher” or “more structur-
ally prominent” than another. Principles of the Extended Standard Theory entailed that
movement must be upward in the general case, from a structurally lower to a struc-
turally higher position. Similarly, RG assumed a strict “Relational Hierarchy” among
grammatical relations 1 > 2 > 3 > . . . and took relational changes to be “upward” in the
general case. Thus there could be “2 to 1 Advancement” but not (on standard thinking)
“1 to 2 Retreat.” There were also theoretical issues for both frameworks that arose in
surprisingly parallel ways. In RG it was debated whether advancement from 3 (indirect
object) to 1 (subject) was possible without intermediate status as a 2 (direct object). In
configurational theories, parallel questions arose as to whether a lower phrase could
raise to subject position across an intermediate argument that c-commanded it.
At the same time, there were intriguing differences between the two approaches that
invited exploration. For example, the Extended Standard Theory analyzed the brack-
eted elements in (142) as simple nonthematic “adjuncts,” and their boldfaced nominals
as inaccessible to A-movement operations (Passive, Raising).
(142) Col. Mustard killed the victim [for Prof. Plum] [in the conservatory] [with the rope].
By contrast, RG viewed such nominals as bearing “oblique grammatical relations” like

Benefactive (ben), Locative (loc), and Instrumental (instr), allowing the possibility
of relation-changing rules like ben  2, loc  2, instr  2, and so on. The latter
were not mere theoretical possibilities but were in fact argued to exist in languages
with applicative constructions like those reviewed above. Relational Grammarians
were the first (to my knowledge) to offer sophisticated theoretical analyses of appli-
catives, and indeed of many interesting constructions in languages that, at the time,
were unexplored within the Extended Standard Theory. Most significantly for my own
thinking, RG drew the connection between applicative-oblique alternations and tradi-
tional Dative Shift, and revealed its depth. One of the puzzles of Dative Movement as
a transformation in English is its confinement to a semantically limited class of predi-
cates, as well as apparent exceptions to its application. Thus, unlike English Passive,
which applies to virtually any transitive predicate, English Dative Movement does not
apply to all ditransitives (e.g., not to put on its locative interpretation41) but is instead
confined to give-type verbs; it also seems to show lexical exceptions even within that
class (e.g., give vs. donate). Such considerations, especially after the work of Oehrle
(1976), convinced many working in the Extended Standard Theory that Dative Move-
ment should be abandoned as a transformation. But among Relational Grammarians,
the remarkable breadth and productivity of applicative formation in other languages
was known, and its status as a rule of grammar widely accepted. From that wider
perspective, concluding that there is no rule of OBL  2 based on English datives, and
ignoring Kinyarwanda applicatives, was counterpart to concluding that there is no rule
of 2  1 based on Korean passives (which are semantically limited to active agentives
and show exceptions even within that class) and ignoring passives in English.
3.2. OBL  2 as A-Movement?

Bringing OBL  2 into the Extended Standard Theory from RG seemed desirable, much

like importing Right Wrap from classical Montague Grammar. But, as in the latter case,
doing so involved serious challenges. Indirect objects in double object constructions
plainly occupy A(rgument)-positions, as demonstrated by standard tests (binding, con-
trol, etc.). Hence a derivational theory of double object forms must involve movement
to an A-position (A-movement). Within the Extended Standard Theory of the time,
A-movement was canonically exemplified by Passive and Raising, which involved dis-
placement from a Caseless, thematic position (subject, object) to a Case-marked, non-
thematic position. For a derivational theory of double objects, all of these assumptions
were problematic.
On any theory taking the double object form to derive literally from an underlying to-
dative structure, indirect objects will originate in PP-object position, which is not Caseless:
(143) John gave Mary a book [to Mary]

True, English does permit A-movement of PP-objects with pseudo-passives (e.g., The
bed was slept in by George Washington), but this fact only highlights the further ques-
tion of why the dative preposition “disappears” in the double object structure. If Dative
Shift is like pseudo-passive formation, then where has to gone?
Still more problematic was the issue of target position. Under the natural view
that movement is possible only to an empty site, Dative Shift requires the equivalent
of a vacant object position. At the time, however, object positions were assumed to
be licensed strictly by the thematic requirements of a predicate. Furthermore, the so-
called Projection Principle required all thematic positions to be occupied at D-Structure
(Chomsky 1981). These points conspired to exclude empty object positions. The sole
potential target for A-movement was subject position, whose existence was ensured by
a special “Extended Projection Principle” (EPP), independent of thematic requirements.
In the case where subject position was nonthematic, it became available as a target
for movement. A-movement thus reduced to displacement from a thematic subject or
object position to a nonthematic subject position. Passive, Subject-to-Subject Raising,
and Unaccusative and Middle formation were admitted, but Subject-to-Object Raising
and Dative Shift were not.
With regard to source position and the disappearance of to, Larson (1988) suggested
a generalization of the view, standard at the time, that passive morphology on a verb
“absorbs” the Case normally assigned to its object. Since to is plainly a governed preposi-
tion in the dative construction (no other P is possible), and since it appears to make no
independent semantic contribution, its status essentially reduces to that of Case marking.
Larson (1988) proposed that, like a passive verb, a Dative Shift verb could essentially
“absorb” the Case of its goal argument by absorbing the governed preposition itself. This
both eliminated to and made its complement Caseless, as required for A-movement.42
The issue of target position invoked an RG-inspired response. Early versions of
transformational grammar took passive sentences to arise directly from corresponding
actives. One version of this idea, advanced by Chomsky (1970), was that Passive divided
into separate operations of Agent Postposing and Object Preposing. The first vacated
the subject position, making it available for movement (144b); the second established
its new occupant (144c):
(144) a. John kissed Mary
b. ___ (was) kissed Mary [by John] Agent Postposing
c. Mary (was) kissed ___ [by John] Object Preposing
RG contained its own version of this scenario. Under the so-called Stratal Uniqueness
Law, grammatical relations were allowed at most a single bearer at any one derivational
stage (or “stratum”). Thus, in a single clause, there could not be two concurrent 1s, two
concurrent 2s, and so on. This meant that in order for a phrase bearing a grammatical
relation n + 1 to advance to n status, the erstwhile bearer of n must be “demoted,” mak-
ing room for the new relation bearer. In RG this was accomplished by allowing bearers
to demote to “chomeur” status, a special grammatical relation the bearing of which
made an item inaccessible to further relation changing rules.
Larson (1988) translated chomeur status in RG as adjunct status in the Extended Stan-
dard Theory, 43 and attempted to provide a nonthematic landing site for movement by
means of X-bar theory. Specifically, suppose that given the structure in (145a), where α has
an additional thematic role to assign, we are allowed to project, not only the standard X-bar
configuration in (145b), where the additional argument ZP is realized as a specifier, but also
(145c) where ZP is realized as an adjunct, and an empty spec position is generated as a pure
matter of X-bar theory. The latter is a plausible counterpart to spontaneous demotion to
chomeur in the RG approach; in the Extended Standard Theory it represents an extension
of the EPP insofar as a position is made available on purely structural grounds:
(145) a. X′ b. XP or c. XP
X YP ZP X′ e X′
α X YP X′ ZP
α X YP
Standard Projection Adjunct Projection (“Demotion”)

The empty position in (145c) is nonthematic, the relevant θ-role having been assigned to
the “demoted” adjunct ZP; e is thus a potential target for A-movement. Under suitable
assumptions about Case or proper government, and the unavailability of the adjunct,
YP can access the empty position, raising over an argument that was initially projected
above it. This scenario was suggested to underlie the VPs of double object forms (144a),
of with/of variants of spray-load verbs (146b), and also of passives (146), as discussed
in Larson (1991):44
(146) a. VP b. VP c. VP
NP V′ NP V′ NP V′
Mary V′ NP the truck V′ PP Mary V′ PP
V NP Fido V NP with hay V NP by John

give Mary load the truck kissed Mary
Dative Shift Spray-Load Alternation Passive
3.3. Problems and Questions

Larson (1988, 1990, 1991) explores a variety of consequences of these proposals for
to-and for-datives, for indirect passives and psych verbs, and for control. There are also
suggestions about how to analyze apparent constraints on Dative Shift. Again, since
these points are discussed in the papers themselves, it is more useful to focus here on
some of the problems and significant questions inherent in the approach.
As noted above, and explicitly acknowledged in the papers themselves, a core idea in
this approach was to analogize Dative Shift to Passive given that both involve inversion
of local grammatical relations (2-to-1 and 3-to-2). But in fact the analogy was difficult
to carry beyond a basic level. The suggestion that to-suppression in Dative Shift be
seen as a version of Case suppression in Passive was perhaps reasonable in 1988 given
the poor understanding of the latter. However, subsequent attempts to make the latter
clearer (e.g., Baker, Johnson, and Roberts 1989) did not suggest any natural extension
to datives. Furthermore, it was unclear whether the analysis could extend naturally to
English for-dative alternations like (147a,b):
(147) a. John baked a cake for Mary.
b. John baked Mary a cake.
c. John baked a cake.
To bring this pair under the same account offered for give, the for-PP in (147a) must be
analyzed as an argument of bake, and for itself must be analyzed as a governed, seman-
tically redundant preposition, equivalent to Case marking. Only as such could it be
“absorbed” by the verb. This in turn plainly requires us to postulate a different bake in
(147a,b) versus (147c) where the for-PP is absent: bake in (147a,b) must be ditransitive,
encoding the notion of beneficiary. This view was defended and elaborated in Larson
(1990) in response to criticisms by Jackendoff (1990), but its status is clearly less secure
than the counterpart proposal for to-datives.
The question of for-datives was part of the broader challenge of extending the analysis
from English Dative Shift to the more exotic OBL  2 advancements found in applicative
languages. At the time, dative PPs with give-type verbs and locative PPs with put were
analyzed as arguments (148a), but, in general, phrases like the bracketed ones in (142)
were regarded as occupying nonargumental “adjunct” positions (148b). A-movement of
NP from such sites was illicit under prevailing views. How, therefore, could one provide
a derivational account of applicatives like (84a) from Halkomelem (repeated below as
(148c)), whose source must presumably involve a high benefactive adjunct like (148b)?
(148) a. VP b. VP
V NP PP VP PP
give the box to Mary V NP for me

put in the car
fix my canoe
c. ni θəә́y -əɬc -θ -ámʔš -əs ʔə kwθə nə-snә́xwəɬ
aux fix -app -tr -1o -3a obl Det 1pos-canoe
‘He fixed my canoe for me.’
As noted earlier, RG analyzed the boldfaced NPs in (142), not as objects of adjunct
PPs, but rather as terms bearing oblique grammatical relations in the clause. Larson
(1988) adopted this assumption too, taking the relevant PPs to be uniformly inner ver-
bal complements, and not outer adjuncts (149):
(149) a. VP b. VP
NP V′ NP V′
the box V PP my canoe V PP
give to Mary fix for me

put in the car
As discussed in the next section’s introduction, this move essentially effaces the struc-
tural distinction between arguments and adjuncts, and was widely viewed with skepti-
cism, despite Chomsky (1986) having laid the groundwork for such an approach with his
theory of lexical marking (“L-marking”) relations between a head and its complements.
Finally, we should note that Larson’s (1988) analysis of the landing site for Dative
Movement was heavily dependent on particular theoretical machinery. As discussed
above, provision of an empty landing site was achieved via a specific X-bar projection
rule—Adjunct Projection/Argument Demotion—that built (150b) directly from (150a),
crucially with no intermediate structure along the lines of (150c):
(150) a. V′ b. VP c. V′
V NP e V′ V′ NP
give Mary V′ NP V NP Fido
V NP Fido give Mary

give Mary
If the latter were allowed, then there is no clear reason why the remaining agent phrase
couldn’t simply be projected into SpecVP position, yielding (151). The latter derives the
correct word order for the double object form but gets the domain asymmetries wrong,
as Barss and Lasnik (1986) had shown; it allows no possibility of A-movement of the
indirect object to correct them.
(151) VP
NP V′
John V′ NP
V NP Fido
give Mary
In order to preclude this result, (150b) needed to be built with its empty position in a
single step, with the agent to be added later, in a higher VP shell. Whatever one’s views
of this proposal—as sheer stipulation or a plausible extension of EPP-type reasoning
into the VP—subsequent developments made them moot. Chomsky’s (1994) convincing
rejection of an independent, templatic X-bar theory in favor of a strictly compositional,
“minimalist” account of structure building simply ruled out proposals of this kind.
3.4. Other Derivational Accounts

Since the late 1980s, derivational approaches to the dative alternation have been com-
paratively few in number. Nonetheless, several have been offered. Baker (1988), in work
that appeared simultaneously with Larson (1988), proposed a derivational analysis of
English datives based on his account of applicative constructions discussed earlier (cf.
(94)). The former, like the latter, were taken by Baker to involve P incorporation from
an underlying oblique, where the preposition is phonetically null in this case (152):
(152) a. VP b. VP
V PP NP V NP PP
V P P NP Fido give Fido P NP
give to to Mary to Mary
Double Object Construction Oblique Construction
Baker (1996b) subsequently judged such “flat” structures to be inadequate, particularly

(152a), given its failure to capture the basic goal-theme asymmetries in double object
constructions observed by Barss and Lasnik (1986). Baker (1996a) argued, in particular,
for the more articulated pair of structures in (153), where the syntax is now uniformly
right-descending, where the theme phrase (the ring) occupies the same position in both
structures, and where the goal argument in the double object form raises to the specifier
position of an AspP projection interpolated within VP:45
(153) a. VP b. VP
NP V′ NP V′
John V AspP John V AspP
pass NP Asp′ pass NP Asp′
Mary Asp VP e Asp VP
pass NP V′ pass NP V′
the ring V NP the ring V PP
pass Mary pass to Mary
Double Object Construction Oblique Construction
Under the assumption that the specifier of AspP is an A-position, this proposal is
free of special X-bar machinery, unlike the Adjunct Projection/Argument Demotion
analysis of Larson (1988).46 At the same time, the account serves to highlight some
interesting new concerns. Current Minimalist Theory (Chomsky 2000, 2001) frames
its standard movement scenario in terms of a higher functional head probing for, and
agreeing with, a feature on a lower, phrasal “goal.” Agreement allows the functional
head to activate its EPP or “edge” feature, attracting the target phrase to its specifier
position. Applied to (153a), this view would see Asp as probing for, and agreeing with,
a feature on the indirect object Mary, attracting the latter to its Spec. The question
arises as to exactly what feature is involved in this agree relationship. More pressing,
however, is the issue of locality. Probe-goal relationships are widely assumed to obey
minimality, which forbids a probe from agreeing across a closer potential agreement
target. A derivation like (153a) would seem to involve Asp probing for, and agreeing
with, the lower, more distant goal phrase (Mary) across the higher and closer theme
(the ring), violating minimality. Locality is indeed a general issue for derivational
theories of double objects, in which a lower argument raises across another, drawn by
a higher, agreeing probe.
One technical solution to the minimality problem would appeal to the notion of
equidistance, introduced by Chomsky (1993) in his account of object shift. Under
the technical definition, raising of the verbal head (pass) to Asp in (153a) renders the
Spec of AspP equidistant from the Spec and complement positions of the lower V. The
direct and indirect objects thus both become equally accessible to Spec. An interesting,
and less stipulative, alternative is provided by Zushi (1992), who argues that standard
Japanese dative structures like (154a), which superficially resemble English PP datives,
are in fact the equivalent of English double object constructions, deriving essentially
as in (154b):47
(154) a. John-ga Mary-ni hon-o ageta.
John-nom Mary-dat book-acc gave
‘John gave Mary a book.’

b. VP c.
NP V′
John-ga VP V′
V′ ageta VP
PP
Mary-ni NP V′ NP V
hon-o PP V hon-o PP V
Mary-ni ageta Mary-ni ageta

[iΣ] [uΣ, EPP]
Scrambling
Zushi proposes that the goal phrase achieves its higher position by a form of A-scrambling
(scrambling to an A-position), conceived of as a kind of inner topicalization. On Zushi’s
view this movement is necessitated by the demands of Case theory on the goal nominal
(Mary-ni). Zushi’s appeal to scrambling is noteworthy given modern accounts that ana-
lyze scrambling as involving, not as a featural relation between a higher probe and a lower
goal, but rather one between a head and its complement. On the analysis of scrambling in
Ko (2005), a verb bearing an uninterpretable scrambling feature (uΣ) and an EPP feature
can agree with an interpretable instance of Σ (iΣ) on a low complement (e.g., Mary-ni). V’s
EPP feature then allows the latter to raise to the VP edge (154c). On this kind of theory,
the relevant feature for agreement (Σ) is one borne by the verbal head itself, and not by
a higher functional element, and no issue of intervention by the theme (hon-o) arises.
Indeed, the higher theme argument is unavailable for scrambling since it lies outside the
c-command domain of the V head, and thus outside the domain of Σ-agreement.48
Finally, we might briefly consider the derivational theories of den Dikken (1995) and
Oba (2002), which both attempt a principled connection between double object and
possessive constructions based on a derivational view of the latter.49 Thus Oba (2002)
draws on the work of Freeze (1992), who proposes that possessive have constructions like
Mary has a car derive from underlying copular constructions involving be and a locative
preposition, represented as to in (155).50 The copula is analyzed as an unaccusative or
raising predicate, selecting a PP containing two arguments: a theme (a car) and a location
(to Mary) (155a). The surface form is derived by incorporating to into be and by raising
the goal to subject position. The combination of to + be spells out as have (155b):
(155) a. IP b. IP
NP I′ NP I′
I vP Mary I vP
vBE VP vBE VP
V PP V vBE V PP
be NP P′ to be to be NP P′
a car P NP a car P NP
have
to Mary to Mary
Oba (2002) proposes essentially the same account of double object constructions, view-
ing them, in effect, as embedding the derived possessive vP structure under a higher
causative verb, realized syntactically as vCAUSE (156a). The preposition once again incor-
porates into a copula; the goal (Mary) raises to VP specifier position; finally, the derived
possessive (to + be), raises onto vCAUSE. The combination of causative little v plus derived
possessive spells out as the double verb (here represented as give) (156b):
(156) a. vP b. vP
NP V′ NP V′
John vCAUSE VP John vCAUSE VP
NP VP V vCAUSE NP VP
V PP V
to be Mary PP
be NP P′
to be NP P′
give
a car P NP
a car P NP
to Mary to
Evidently, this account faces questions faced by derivational accounts generally: What
licenses the position that is targeted by the raised goal (here, SpecVP)? What features
drive this movement? How is minimality/locality respected in raising the goal across
the theme? At the same time, Oba (2002) and den Dikken (1995) draw attention to a
new and important broader point, that is, that the question of a derivational relation
between oblique and ditransitive datives plausibly reduces to the larger question of a
derivational relation between oblique and transitive possessives. If transitive have-type
possessives derive from copular constructions in which a goal raises across a theme
(155), and if double object constructions are essentially “causativized have possessives”
(as many have assumed), then it is difficult to resist the conclusion that double object
constructions must involve raising of a goal across a theme as well. Put differently, if the
movement analysis in (155) is correct, then this surely constitutes Dative Shift in its most
basic form, with that observed in give-type constructions derivative upon it.
Den Dikken’s and Oba’s derivational proposals also have interesting implications
for the analysis of applicative constructions, which were earlier seen to admit a range
of semantic relations. Taken at its most general, the approach might be seen to imply
that all applicative structures have underlying an unaccusative derivation at their root.
Thus recall the Chichewa instrumental applicative-oblique alternation noted in (85),
repeated below as (157a,b).
(157) a. Fisi a-na-dul-a chingwe ndi mpeni
hyena sp-pst-cut-asp rope with knife
‘The hyena cut the rope with the knife.’ (= (22a) in Baker 1988, p. 238)
b. Fisi a-na-dul-ir-a mpeni chingwe.
hyena sp-pst-cut-app-asp knife rope
‘The hyena cut the rope with the knife.’ (= (22b) in Baker 1988, p. 238)
English permits only the counterpart of (157a), the oblique contruction (cf. (158a));
the double object counterpart of (157b) is ungrammatical (cf. (158b)):
(158) a. John opened [the safety deposit box] [with his key].
b. *John opened [his key] [the safety deposit box].
Nonetheless, English does exhibit instrumental oblique-transitive alternations of the

sort in (159a,b). Extending the approach of Freeze (1992), in which an oblique advances
over a theme, (159a) might be assigned the basic structure in (160a), and (159b) might
be taken to involve an unaccusative derivation along the lines of (160b):
(159) a. The safety deposit box opened with John’s key.
b. John’s key opened the safety deposit box.
(160) a. b.
IP IP
NP I′ NP I′
I vP J′s key I vP
v? VP v? VP
V PP V v? V PP
open NP P′ open with open with NP P′
the sdb P NP the sdb P NP

open
with J′s key with J′s key
Following den Dikken’s and Oba’s accounts, the latter might in turn be offered as the basis
of instrumental applicative alternations, with a derivation like (160b) embedded under
a higher causative, equivalent to (156b). This view has the intriguing consequence that
double object instrumentals, although plainly forbidden in the surface grammar of English,
might nonetheless not be “exotic” to it after all. If the core alternation (159) is available in
the grammar of English, then under the logic pursued above, the absence of the ditransitive
variant (159b) must arise from some specific, and perhaps relatively superficial, aspect of
the derivation counterpart to (156b)—for example, the interaction between vCAUSE and the
little v? involved with instrumentals, or the availability of Case checking by the v? head.
4. VOICE ALTERNATION
I will now offer an updated analysis of Dative Shift based on the updated shell
analysis sketched in the General Introduction, incorporating some elements of the
derivational theories discussed above. The key idea I will pursue is that the oblique-
applicative alternation is not the product of any decompositional semantic relations
between predicates but in fact results from a formal option that is present in the
fundamental mechanisms of syntactic structure building itself, and that is the basis
of voice alternation generally, including not only dative/applicative alternations but
passives as well.51
4.1. Little v and P

In the General Introduction we considered structures like (161) involving three argu-
ments. Under unique valuation, lexical V may bear at most a single valued (uninter-
pretable) θ-feature, which we took to be the theme feature ([th]) as the default case.
Because give can value at most one θ-feature, all other valuation must occur through
independent elements. The preposition to was taken to value the goal feature ([gl])
under agreement with Mary, and a little v voice head was taken to value the agent fea-
ture ([ag]) under agreement with John:
(161) vP
John v′
[iAG[3]]
v VP
v give Fido V′
[uTHval[2]] give PP
[uTHval[2]] to Mary
AGREE!
As we noted, this picture establishes a certain equivalence between v and P insofar as

both are available to value θ-features:
(162) v P
[uθval[ ]] ⇔ [uθval[ ]]
This in turn suggests alternative derivational possibilities in which this equivalence is

exploited. Logically speaking, and other things being equal, we might expect an alter-
native derivation in which P values the agent feature rather than little v, and we might
expect a derivation in which little v values the goal feature instead of P.
4.2. Passives
We observed earlier that the simple external merge of an agent phrase John with a transi-
tive VP like kiss Mary will not yield an interface-legible object since the [ag] feature will be
unvalued (163a). Our response was to introduce a voice head bearing valued [ag] (163b):
(163) a. VP b. vP
John V' John v'

[iAG[2]] [iAG[2]]
kiss Mary v VP
[uAG[2]] [iTH[1]]
AGREE! [uTHval[1]] v kiss kiss Mary
[uAGval[2]] [uAG[2]] [uAG[]] [iTH[1]]
AGREE! [uTHval[1]]
Consider an alternative possibility. The preposition by is associated with agency. Sup-
pose we take by to bear a valued [ag] feature. Then, in place of (163b), we might pro-
ceed as in (164), first merging John with by, and then merging the result with kiss Mary,
in both cases under agreement with [ag]:52
(164) a. PP b. VP
by John PP VP
[uAGval[1]] [iAG[1]]
by John kiss Mary
AGREE! [uAGval[1]] [iAG[1]] [uAG[]] [iTH[1]]
AGREE! [uTHval[1]]
In (164b) all θ-features are properly valued, yielding interface legibility. Nonetheless,
the resulting sentence is not well-formed. By John kisses Mary is not a licit alternative
to the transitive John kisses Mary.
I suggest that the derivation (164) is excluded on grounds independent of θ-theory—
specifically, for reasons of Case. Suppose following Chomsky (1995) that higher T bears
a nom feature and that little v voice head bears an acc feature. In a typical transitive
structure like (165), Mary will be local to T and John will be local to v. Agreement for
Case is thus straightforward:
(165) TP
T vP
John v′
NOM [iAG[2]]
v VP
v kiss kiss Mary

ACC
The same is not true, however, with (166), the equivalent structure for (164). Since PP
intervenes between T and Mary, T finds no Case goal to agree with. At the same time,
there is no other local Case probe available to Mary since the little v voice head is absent.
The derivation therefore fails on Case-theoretic grounds.
(166) TP
T VP
PP V′
NOM X
by John kiss Mary
Consider now an alternative possibility as a “thought experiment.” Suppose that the
[th]-feature on the lexical verb could somehow be “devalued” so that the set of θ-features
for kiss became {[uag[ ]], [uth[ ]]}. Merge with Mary would then yield an interpretable
but unvalued feature (167a), but since [th] is unvalued on kiss, we would have the option
of co-selecting a voice head bearing a valued [th] feature. Merging it with VP and raising
V yields (167b).
(167) a. VP
PP V′
by John kiss Mary

[uAGval[2]] [iAG[2]] [uAG[2]] [iTH[1]]
[uTH[1]] UNVALUED!
b. vP
v VP
v kiss PP V′
[uTHval[1]] [uAG[2]]
[uTH[1]] by John kiss Mary
AGREE! [uAGval[2]] [iAG[2]] [uAG[2]] [iTH[1]]
[uTH[1]]
Now note that in the latter, kiss agrees with Mary on the theme feature [th]. Further-
more, little v agrees with kiss on the theme feature [th]. It follows that little v agrees
with Mary on the theme feature [th]. Since Mary and little v agree, the latter can acti-
vate its EPP feature, raising the theme to its Spec across the agentive by-phrase (168):
(168) vP
Mary v′
[iTH[1]]
v VP
v kiss PP V′
[uTHval[1]] [uAG[2]]
[uTH[1]] by John kiss Mary
[uAGval[2]] [iAG[2]] [uAG[2]] [iTH[1]]
[uTH[1]]
The transitivity of agreement and the fact that Mary has already undergone agreement
with kiss thus conspire to circumvent the minimality constraint that would otherwise
forbid a higher probe (here v) from agreeing with and raising a lower goal (here Mary)
across an intervening phrase of the same featural type (here by John).
In the resulting structure (168), the Case agreement problems identified earlier in (165) no
longer occur. Mary is now accessible to higher T and can raise further to SpecTP, and so on
(169). The agent John can be assumed to undergo Case agreement with by in the usual way:
(169)
TP
T vP
Mary v′
NOM
v VP
v kiss PP V′
by John kiss Mary
I suggest this derivation as a general analysis of passive constructions. For concreteness

we may take passive morphology to have the effect of “devaluing” the [th]-feature on
the lexical verb on which it occurs (170):53
(170) Passive: V ⇒ V-en
[uthval[ ]] [uth[ ]]
… …
The properties of a “passivized” verb will then interact with Case theory, together with
the equivalence of little v and P as sources of θ-feature valuation, to yield the results
given above. In fact, we will predict the following complete distributional array:54
(171) a. John kisses Mary.
b. *By John kisses Mary.
c. Mary is kissed by John.
d. *John is kissed Mary.
e. *Mary is kissed John.

Unaccusative verbs like arrive might be analyzed in a similar way, that is, as verbs with
no agent feature and a lexically unvalued theme. Derivation would then proceed essen-
tially as with passives, with valuation of [th] through a voice head (172):55
(172) vP
Mary v′
[iTH[1]]
v VP
v arrive arrive Mary

[uTHval[1]] [uTH[1]] [uTH[1]] [iTH[1]]
Summarizing, then, the general alternation between little v and P offers a simple but
attractive picture of passive voice alternations and related unaccusative constructions.
4.3. Applicative Shift (“A-Shift”)

In our discussion of oblique ditransitive constructions in the General Introduction we
noted that directly merging the goal (Mary) with give would yield an unvalued [gl]
feature. Recall (58), repeated below as (173):
(173) VP
give Mary
[uAG[ ]] [iGL[1]]
[uTHval[ ]]
[uGL[1]]
UNVALUED!
Our solution was to value the low goal by means of P (to). Recall (59), repeated below
as (174).
(174) a. PP b. VP
to Mary give PP
[uGLval[1]] [iGL[1]] [uAG[ ]]
[uTHval[ ]] to Mary
AGREE!
Suppose that instead of proceeding as in (174), we were to continue from (173) by merg-
ing the theme Fido and then merging a co-selected little v voice head bearing valued
[gl]. V raises and v-V agree (175).
(175) vP
v VP
v give Fido V′
[uGLval[1]] [uAG[ ]] [iTH[2]]
[uTHval[2]] give Mary
AGREE! [uGL[1]] [uAG[ ]] [iGL[1]
[uTHval[2]]
[uGL[1]]
Notice that the transitivity of agreement we observed earlier with passives obtains here
as well. Give agrees with Mary on the goal feature [gl], and little v agrees with give on
[gl]. Hence little v agrees with Mary on [gl]. Since Mary and little v agree, the latter can
activate its EPP feature, raising the goal to its Spec across the theme argument (176).
(176) vP
Mary v′
[iGL[1]
v VP
v give Fido V′
[uGLval[1]] [uAG[ ]] iTH[2]]
[uGL[1]] [uAG[ ]] [iGL[1]
[uTHval[2]]
[uGL[1]]
Transitivity of agreement thus once again allows us to circumvent minimality. Little v is
not required to probe past Fido for agreement with Mary. Instead, agreement is estab-
lished through the raised verb.
The structure in (176) can now merge with a little v bearing valued [ag]. The lower
verbal complex raises and agrees with v (177a). The agent John then merges, agreeing
on the [ag] feature (177b):
(177) a. vP
v vP
v v Mary v'
[uAGval[3]] [iGL[1]
v give v VP
[uTHval[2]] v give Fido V'
[uGL[1]] [uAGval[1]] [uAG[1]] [iTH[2]]
[uGL[1]] [uAG[ ]] [iGL[1]
[uTHval[2]]
[uGL[1]]
b. vP
John v'
[iAG[[3]]
v vP
v v Mary v′
[uAG[val[3]] [iGL[1]
v give v VP
AGREE! [uAG[val[1]] [uAG[3]]
[uTHval[2]] v give Fido V′
[uGL[1]] [uGLval[1]] [uAG[1]] [iTH[2]]
[uGL[1]] [uAG[ ]] [iGL[1]
[uTHval[2]]
[uGL[1]]
I propose (175)–(177) as a general derivation for applicative structures, including

English double object constructions. The movement in (176) might thus be called Appli-
cative Shift or “A-Shift.”56 The crucial difference between the A-Shift derivation and the
oblique derivation lies in how we value the goal feature on give. In the oblique structure
(174b), this is done through P. In the applicative structure it is done through little v. This
difference entails a corresponding difference in final structure position. The ability of P
to value [gl] permits the goal to remain low, where the thematic hierarchy dictates its
merge position. In (175)–(177), Mary also merges low. But because little v, valued for
[gl] and bearing EPP feature, merges higher, Mary must follow it to a higher position,
crossing the theme.
4.4. Observations and Comparisons

The A-Shift analysis preserves some aspects of the Larson (1998) account. Thus it
retains the RG-inspired connection between double objects/applicatives and passives
insofar as both involve inversion of an initially lower argument and an initially higher
one. Furthermore, this inversion is to an A-position—the specifier position of a verbal
head (little v). The A-Shift analysis also maintains the RG view that oblique phrases,
at least those of circumstantial content (goal, location, manner, duration, instrument,
cause/purpose), constitute low inner complements of the verb and not simple adjuncts.
Hence movement from these positions continues to count as A-movement.
At the same time, the two accounts diverge in significant ways. As noted earlier,
Larson (1988) locates the source of voice alternation in X-bar theory, specifically, in
whether projection of an additional argument (ZP) from the structure in (178a) fol-
lows the standard template in (178b) or the special “adjunct projection” template in
(178c):
(178) a. X′ b. XP c. XP
X YP ZP X′ e X′
α X YP X′ ZP
α X YP
Projection via the standard template yields oblique datives, oblique locatives, and active
transitives as in (179a–c), respectively, in which arguments occupy positions dictated by
the thematic hierarchy:
(179) a. VP b. VP c. VP
Fido V PP hay V PP John V NP
give to Mary load on the truck kiss Mary
Projection via the adjunct template yields double object datives, with/of-locative forms,
and passives (180a–c), in which lower arguments raise from their original merge site to
the higher empty specifier position in (178c):
(180) a. b. VP c. VP
VP
Mary V NP the truck V′ PP Mary V′ PP
V NP Fido V NP with hay V NP by John
give Mary load the truck kissed Mary
In all of the latter, the outer argument has the status of an adjunct, equivalent to an RG
chomeur.
In the A-Shift analysis, voice alternation has a very different source. In place of stipu-
lated X-bar machinery, we appeal to the equivalence of little v and P in the valuing of
θ-features (181) (which repeats (162)):
(181) v P
[uθval[ ]] ⇔ [uθval[ ]]
The P-option derives oblique datives and locatives, with roughly the same structures as
before. Arguments occupy positions dictated by the thematic hierarchy (182a,b):
(182) a. VP b. VP
Fido V′ hay V′
give PP load PP
to Mary on the truck
The little v–option derives double object datives and with/of-locatives (183a,b), in
which an argument undergoes inversion:
(183) a. vP b. vP
Mary v′ DP v′
give-v VP the truck load-v VP
Fido V′ PP V′
give Mary with hay load the truck
Passives present a “mixed” picture. Like oblique constructions, passives employ P (by)
to value a θ-feature ([ag]) that is unvalued on V. This PP occurs, moreover, in a position
dictated by the thematic hierarchy (184a). But like double objects and with/of-locatives,
passives involve V-Raising and inversion of a lower argument (184b). Active transitives
(184c) present a mixed picture as well. Like oblique datives and locatives, their argu-
ments appear in positions dictated by the thematic hierarchy with no inversion. But like
for double objects and with/of-locatives, valuation of the nontheme θ-feature occurs via
little v and not P.
(184) a. VP b. c.
vP vP
PP V′ Mary v′ John v′
by John kissed Mary kissed-v VP kiss-v VP
PP V′ kiss Mary
by John kissed
Thus, beyond dispensing with the special X-bar machinery of Larson (1988), the
A-Shift account also significantly reanalyzes the relations among obliques, double
objects, passives, and actives. Note furthermore that in the structures involving argu-
ment inversion ((183a,b) and (184b)), the lowest argument no longer has the status
of an adjunct but is instead a normal specifier under the A-Shift account, projected
by External Merge.
4.4.1. Which Form Is “Basic”?

The A-Shift account also diverges importantly from Larson (1988) on the issue of which
form—oblique or double object/applicative—is more “basic” in the dative alternation.
Larson (1988) embodies the view of Chomsky (1955/1975) and many subsequent authors
that the oblique construction is the basic one. Oblique constructions project by means stan-
dard X-bar theory with their arguments appearing in the order prescribed by the thematic
hierarchy and licensed smoothly by Case theory. Oblique structure is thus straightforward
and unproblematic. Double object constructions, by contrast, require a special, noncom-
positional X-bar template to ensure a landing site for movement, as well as an extended
notion of “Case absorption” to remove the dative to that otherwise might be expected.
As we’ve seen, the A-Shift analysis makes no appeal to X-bar templates. Oblique con-
structions and double objects/applicatives both employ the same structure building opera-
tions. A-Shift also makes no appeal to Case absorption in explaining why double object
forms lack dative to. The latter is simply not part of the numerations that build double
object forms in the first place, its role being taken by little v. And although double object
forms do involve inversion, and an argument thematically “out of place,” the factors that
produce this are all universal in nature. Little v voice heads are universally available, enter-
ing the numeration as an option of Universal Grammar. Likewise, θ-features EPP features,
and agreement are universally given. By contrast, oblique form derivations make essential
use of language-specific lexical elements—prepositions—whose semantic content must be
acquired by the learner and mapped to compatible θ-features. We saw earlier that dative
prepositions often alternate with, and are sometimes phonetically identical to, allative prep-
ositions indicating path of motion. On the A-Shift analysis allative and dative Ps would bear
distinct θ-features, whose distribution the language learner would be required to work out.
Thus, although the double object form does involve inversion of an argument from a posi-
tion that is occupied by PP in the oblique form, it is not clear this implies a more “basic”
or “less complex” status for the latter under A-Shift. Put somewhat differently, although P
and v are equivalent in bearing valued θ-features, they are not equivalent tout court. Little
v’s are “pure” bearers of θ-features, with no independent semantic content and no indepen-
dent lexical status. Prepositions, by contrast, are “impure” bearers, whose θ-features are
mixed with independent lexical content. This mixture must be learned and coordinated with
the properties of the constructions in which they occur. Oblique constructions thus always
involve an additional lexical content that double object/applicative forms do not.
4.4.2. Relations to Other Theories

The A-Shift analysis compares in interesting ways with the accounts cited earlier, deriva-
tional and nonderivational. The final structure in (177) resembles that of Baker (1996a) in
overall geometry. In both trees the theme occupies essentially the same position in double
object and oblique forms (cf. (153) above), and in both a functional projection that is the
target of movement intervenes between an upper VP containing the agent and a lower VP
containing the theme. Baker takes this projection to be AspP, whereas the A-Shift account
identifies it as vP, treating double object/applicative formation as a strictly VP-internal
matter. This difference is relevant to the minimality issue arising for derivational accounts.
As we saw, in the A-Shift analysis, the transitive featural relation between little v, lexi-
cal V, and the lower goal phrase allows an inversion analysis that incurs no minimality
violation. Lexical V establishes its featural relation to the goal before combining with the
theme in a way reminiscent of the scrambling analysis (Zushi 1992 and Ko 2005). Since
verbal θ-features are plausibly confined to verbal categories (v/V), this general approach
does not seem available to the AspP analysis; there are no obvious grounds for identifying
Asp as a potential bearer of [gl].
We might also note that the general oblique-applicative alternation proposed here
appears capable of capturing the connections made by den Dikken (1995) and Oba
(2002) between ditransitive and transitive/oblique possessives. Larson (in preparation)
argues that oblique-copular possessives and have possessives are related by the same
A-Shift mechanism postulated for applicatives, and that this too represents a form of
voice alternation. Thus oblique-copular possessives are analyzed as deriving analo-
gously to prepositional datives, with P supplying the valued θ-feature. Compare (185a)
with belong to the relevant derivational stage with give (185b).
(185) a. b. VP
VP
Fido V′ Fido V′
belong PP give PP
[uTHval[ ]] [uAG[ ]]
[uGL[1]] to Mary [uTHval[ ]] to Mary
[uGLval[1]] [iGL[1]] [uGL[1]] [uGLval[1]] [iGL[1]]
AGREES! AGREES! AGREES! AGREES!
By contrast, have possessives receive an applicative derivation, counterpart to that of

double objects, where the valued θ-feature is resident on a little v voice head. Compare
(186a) to the relevant derivational stage with give (186b):57
(186) a. vP
Mary v′
[iGL[1]]
v VP
have v Fido V′
[uTHval[2]] [uGLval[1]] [iTH[2]]
[uGL[1]] have Mary
[uTHval[2]] [iGL[1]]
[uGL[1]]
b. vP
Mary v′
[iGL[1]]
v VP
give v Fido V′
[uAG[ ]] [uGLval[1]] [iTH[2]]
[uGL[1]] [uAG[ ]] [iGL[1]]
[uTHval[2]]
[uGL[1]]
4.4.3. Nondative Applicatives
The A-Shift analysis appears to offer a sufficiently general account of projection so as to
span the range of applicative-oblique relations observed earlier in (93). That is, A-Shift
can plausibly be offered as the derivational bridge between the two structures, answer-
ing the implicit question of relationship “←??→” that was posed above:
(187) Applicative Oblique
Form: α V-app β γ ←A-Shift→ α V γ [P β]

Sense: caused possession caused possession
benefactive/malefactive/ benefactive/malefactive/
substitutive substitutive
instrumental instrumental
caused motion/location caused motion/location
stimulative stimulative
manner manner
reason reason
Thus, a locative applicative equivalent to John put Fido there can be assigned the analy-
sis in (188), where the relevant θ-feature is now [loc] rather than [gl], but where the
derivation is identical to (177b) in all other respects:58
(188) vP
John v′
[iAG[3]]
v vP
v v there v′
[uAGval[3]] [iLOC[1]]
put v v VP
[uAG[3]]
... put v Fido V′
[uAG[ ]] [uLOCval[1]] [iTH[2]]
[uTHval[2]] put there
[uLOC[1]] [uAG[ ]] [iLOC[1]]
[uTHval[2]]
[uLOC[1]]
The relationship between the oblique and applicative derivations expressed by the
A-Shift analysis is thus an extremely general one.
As a final point of comparison, we may note the striking parallelism between the
configurations derived under the A-Shift analysis and the structure discussed earlier
from Marantz (1993) in his analysis of the double object structures as applicative con-
structions (102), repeated below with some minor changes in (189a). Apart from node
labels, virtually the only difference between (189a) and (189b) is that the latter identifies
the “mystery element” X that is left unspecified and virtually undiscussed in Marantz
(1993). Under the A-Shift account, X is revealed as none other than the trace of the
raised goal phrase, initially projected into low position according to the thematic hier-
archy, and raised from that position by the EPP feature on the little v voice head—the
equivalent of Marantz’s appl in the current approach.
(189) a. IP b. vP
NP I′ John v′
John I VP v vP
NP V′ Mary V′
Mary V VP v VP
APPL NP V′ Fido V′
Fido V X give Mary

give
Marantz (1993) A-Shift
Indeed, the convergence with the work of Marantz extends further. The oblique/voice head
alternation posited in the A-Shift analysis is strongly reminiscent of Marantz’s original (1984)
approach to applicatives, where the oblique derivation alternates with one in which an appli-
cative affix, morphologically merged with the main verb, contributes essentially the same
derivational content as an oblique P, but without literal incorporation in the sense of Baker
(1988). The A-Shift analysis preserves this basic picture, identifying the shared content pro-
vided by P and the verbal element as θ-features. A-Shift might thus best be regarded as a
development of Marantz’s (1993) analysis in derivational terms, providing the missing theory
of projection for it, and specifying some missing elements, but also incorporating original
insights on the relation between oblique and applicative derivation from Marantz (1984).
4.5. Case
Double object constructions are known to raise interesting questions for the theory of Case,
in part because of their apparent violation of the “one Case assigner–one Case assignee”
relation that is widely assumed to govern nominal arguments, and in part because of the
puzzling status of dative Case itself. Case assignment has also been viewed as a plausible
source of cross-linguistic variation in the availability of double object constructions. In
this final section I will briefly review some basic issues and sketch some tentative proposals
in connection with the A-Shift approach to applicatives/double objects outlined above.
4.5.1. Case Assignment with Double Objects

The two postverbal DPs in English double object constructions both appear to bear
accusative Case. Under current theories (e.g., Chomsky 2000, 2001) in which v is the
sole locus of the structural accusative feature and carries a single feature-checking
possibility, an account of English double objects where both DPs trace their Case to a
single little v appears problematic (190).
(190) v V DP1 DP2 Two Structural Accusatives??

What then are the Cases assigned to the two DPs in double object constructions, and
what is their source?
A wide range of responses have appeared in the literature. Baker (1988) and Larson
(1988) suggest that accusative Case be divided into distinct structural and inherent vari-
eties, with the former arising from v and the latter arising from V, and with one variety
assigned to each DP (191a). Emonds (1993), Pesetsky (1995), and Harley (2002) propose
a covert prepositional Case assigner for the outer object, coupled with structural accusative
Case for the inner object (191b). Czepluch (1982) and den Dikken (1995) argue for essen-
tially the converse: a covert prepositional Case assigner for the inner object, with structural
accusative Case assigned to the outer object (191c). Anagnostopoulou (2003) offers yet
another analysis, in which the outer object receives structural accusative Case from v and
the inner object receives inherent genitive/dative Case from an applicative head (191d).
(191) a. v V DP1 DP2 One Structural Accusative + One Inherent Accusative
b. v V DP1 [ P DP2 ] One Structural Accusative + One P Case
c. v V [ P DP1 ] DP2 One P Case + One Structural Accusative
d. v Appl DP1 DP2 One Inherent Case + One Structural Accusative
Although these proposals are widely regarded as competitors, in fact it seems unlikely
that any single one of them can accommodate the full range of double object forma-
tions, especially once the latter are assimilated into the broader class of applicatives.
Indeed, all may ultimately be needed to cover the range of cross-linguistic variation. For
example, because English prepositions govern accusative Case (Kayne 1984), English
double object constructions are potentially analyzable as involving covert prepositions
along the lines of (192b) or (192c):
(192) a. Mary gave John a present.
b. Mary gave John [PP[P Ø ] a present].
c. Mary gave [PP[P Ø ] John] a present.
But this view does not appear sustainable for double accusative objects in a language
like Korean, which exhibits morphological accusative, but whose postpositions do not
govern structural Cases (193):
(193) Minhee-ga Chulsu-lul cek-lul cueyatta.
Minhee-nom Chulsu-acc book-acc gave
‘Minhee gave Chulsu a book.’
Korean would thus seem to require an analysis in which a genuine structural accusative
(and not an adpositionally assigned one) is available to both theme and goal.
On the other hand, the analyses in (191b) and (191c) do seem attractive for other lan-
guages, or even subclasses of constructions within them. Farrell (2005), citing Gerdts (1988),
discusses Halkomelem applicative forms like (194a,b), where ∂s ‘RECIP’ and ∂ ɬ ‘ben’ are
applicative morphemes. Note that in these examples, the outer theme argument is uniformly
governed by an overt prepositional element ʔ∂ glossed ‘OBL’ (“oblique”).
(194) a. ni ʔám-ǝs-t-ǝs kwθǝ sqw ǝméyʔ [ ʔǝ kwθǝ sθ’ám ].
aux give-recip-tr-3a Det dog obl Det bone
‘He gave the dog the bone.’
b. ni q’wél-ǝɬ-c-t-ǝs ɬǝ sɬéni [ ʔǝ kwθǝ sǝpilíl ].
aux bake-ben-tr-3a Det woman obl Det bread
‘He baked bread for the woman.’
Larson (1990) (following Schneider-Zioga 1988) proposes that the class of English
spray-load examples like (195a,b) are also (in effect) double object/applicatives.59 Here
also the outer object is governed by a preposition (with, of ). On this proposal English
would have both accusative outer objects (with give, send, etc.) and oblique outer
objects (with load, empty, etc.)
(195) a. They loaded the truck [with hay].
b. They emptied the tank [of water].
Both Halkomelem and English are therefore candidates for an analysis as in (191b).
It can likewise be argued that the pattern in (191c) is instantiated, for instance, by Span-
ish in clitic-doubled forms like (196a) in the V-[a DP]-DP word order. Demonte (1995) and
Bleam (2003), among others, argue that such examples show clear diagnostics of double
object forms, including scope freezing behavior, restriction to animates, and so on.
(196) a. Susana le dió [a Juan ] un libro.
Susana Cl.him gave to John a book
‘Susana gave John a book.’
b. Susana envió un libro a Juan/a Paris.
Susana sent a book to John/to Paris
‘Susana sent a book to John/to Paris.’
The element a in (196a) is arguably identical, or closely related, to the prepositional ele-
ment a also found in allatives like (196b). If so, then (196a) would appear to exemplify
analysis (191c).
Finally, Anagnostopoulou (2003) argues that (191d) is instantiated by Modern
Greek.60 (197) is a typical Modern Greek double object construction, where the goal
phrase (tu Giani) is inflected for genitive/dative Case and the theme (to vivlio) is inflected
for accusative:
(197) (Tu) edhosa tu Giani to vivlio.
cl-gen gave-1sg the John-gen the book-acc
‘I gave John the book.’ (= (18) in Anagnostopoulou 2003, 15)

Anagnostopoulou argues persuasively that accusative on the outer theme object
cannot be any form of inherent Case, given that inherent accusative shows definite-
ness restrictions in Modern Greek that are not present with the theme object of a
double object construction. Moreover, the goal and theme cannot be governed by
prepositions given that PP-objects cannot be clitic-doubled in Modern Greek (198a),
whereas the theme and goal in Modern Greek double object constructions can be
optionally doubled (198b). These considerations (among others) eliminate propos-
als (191a–c).
(198) a. *(Tu) edhosa to vivlio s-ton Giani
cl.gen gave-1sg the book-acc to-the John
‘I gave John the book.’ (= (22) in Anagnastopolou 2003 p. 17)
b. (Tu) (to) edhose tu Petru to vivlio i Maria
cl.gen cl.acc gave-3sg the Peter-gen the book-acc the Maria-nom
‘Mary gave Peter the book.’ (= (99) in Anagnastopolou 2003 p. 63)
Similar conclusions appear natural for Romanian constructions like (199a,b) and for
Japanese dative constructions like (200).
(199) a. Mihaela (ii) trimite o scrisoare la Maria.
Mihaela cl.dat sends a letter to Mary
b. Mihaela (ii) trimite Mariei o scrisoare.
Mihaela cl.dat sends Mary.dat a letter
Mihaela send a letter to Mary.
(200) Taroo-ga Hanako-ni hon-o agatta.
Taroo-nom Hanako-dat book-acc gave
‘Taroo gave Hanako a book.’
As in MG, Romanian dative constructions show both prepositional (199a) and

Case-inflected (199b) variants, where the Case form of the prepositional object is
distinct from dative.61 Dative in (199b) thus does not appear to arise from a covert
version of la. And, as discussed by Zushi (1992), Miyagawa and Tsujioka (2004),
and Harada and Larson (2009), although Japanese (200) superficially resembles a
prepositional dative form, its properties in terms of binding and scope are directly
comparable to a double object construction. Thus the ni-marked goal appears
to asymmetrically c-command the o-marked theme; ni- and o-marked quantifier
phrases are frozen with respect to scope; ni-marked goals show the behavior of
Case-marked nominals, not PPs, with respect to Q-float, and so on. Ni-marking thus
resembles some form of inherent Case in such examples, unassociated with PP. MG,
Romanian, and Japanese therefore all appear to be candidates for an analysis as in
(190d). Hence each of the possibilities in (190) and (191) appears to derive some
measure of empirical support.
4.5.2. The Puzzle of Dative Case
Another interesting aspect of Case in double object constructions is the nature of
dative Case itself. As has often been observed, dative seems to straddle the divide
between structural Cases like nominative and accusative, which are unassociated with
particular θ-features and hence compatible with any, and inherent Cases like loca-
tive and instrumental, which, in languages that possess them, are closely linked to a
specific semantics.
In many languages, dative-marked arguments can appear in subject position, for
example, under Passive, where they manifest subject-like properties. Thus in Japanese,
as discussed in Larson (1988) drawing on data from Shimizu (1975), dative ni-marking
can be suppressed under Passive just like structural accusative, with the goal phrase
appearing as a nominative subject; compare (201a,b) and (202a,b).62 (202b) cannot
be analyzed as deriving from a double accusative intermediate stage comparable to
English, where Hanako would bear accusative Case, since Japanese flatly excludes
double accusative marking in such cases (202c):
(201) a. Taroo-ga Hanako-o tataku.
Taroo-nom Hanako-acc hit
‘Taro hits Hanako.’
Hanako-ga Taroo-ni
b. tatakareru.
Hanako-nom Taroo-dat hit-pass

‘Hanako is hit by Taro.’
(202) a. Taroo-ga Hanako-ni kotzutsumi-o okuru.
Taroo-nom Hanako-dat package-acc send
‘Taro sends Hanako a package.’
b. Hanako-ga Taroo-ni kotzutsumi-o okurareru.
Hanako-nom Taroo-dat package-acc send-pass
‘Hanako is sent a package by Taro.’
c. *Taroo-ga Hanako-o kotzutsumi-o okuru.
Taroo-nom Hanako-acc package-acc send
‘Taro sends Hanako a package.’
In the Romanian and Spanish clitic-doubled datives discussed above, the dative/a-
marked goal phrase can also be promoted to subject; however, unlike in Japanese,
oblique marking must be retained on pain of ill-formedness (203a,b)/(204a,b):
(203) a. Mariei i-a fost trimisa o scrisoare.
Maria.dat cl.dat-has been sent.fem.sg a letter
‘Maria has been sent a letter.’

b. *Maria a fost trimisa o scrisoare.
Maria.nom has been sent.fem.sg a letter
‘Maria has been sent a letter.’
(204) a. A Juan
le fue dado
el premio.
to John cl.dat was given the prize

‘John was given the prize.’
b. *Juan fue dado el premio.
John was given

the prize
Syntactically, dative-marked nominals thus both resemble and depart from structur-
ally Case-marked elements: on the one hand, they are accessed by syntactic opera-
tions affecting the latter and come to occupy the same positions; sometimes dative
can even be suppressed like a structural Case, as with Japanese direct passives. At the
same time they often behave like an inherent Case in disallowing their inflection to
be absent even while undergoing A-movement operations, as in the case of Romanian
and Spanish.
Semantically, dative also seems to straddle the fence between structural and accu-
sative Cases. Like an inherent Case, dative is often associated with a certain class of
semantic roles including goal and experiencer. On the other hand, this class is much
wider than that of typical inherent Cases, and in various languages may include posses-
sion, location, instrument, and indeed a wide range of oblique functions (see Blake 2001
for a useful survey and discussion).
4.5.3. Case Features

I would now like to sketch an approach to Case in double object/applicative construc-
tions, drawing once again on the general account of features proposed by Pesetsky and
Torrego (2004). We noted earlier that in this analysis, features come in four varieties,
according to whether they are interpretable/uninterpretable and valued/unvalued. This
general picture was represented as in (205):
(205)
INTERPRETABLE UNINTERPRETABLE
Valued iFval uFval
Unvalued iF uF
Having adopted this analysis for θ-features, it is natural to assume it for Case features
as well. I will assume that categories traditionally analyzed as assigning Case (T, v, P)
are the bearers of interpretable, but unvalued, Case features. Correlatively, I will assume
that categories traditionally analyzed as receiving Case (DP) are the bearers of valued,
but uninterpretable, Case features. Specifically, I will assume that the D heads of argu-
ment DPs bear the latter. Finally, I will assume that other Case-inflected items in the
nominal phrase (nouns, attributive adjectives, etc.) are the usual bearers of uninterpre-
table, unvalued versions of Case features. Typical agreement scenarios will thus include
those in (206a–c), where the c-command relationships proceed left-to-right and where
“OBL” is a cover term for oblique Case assigned by P63.
(206) a. T ... [DP D NP ]
[inom[1]] [unomval[1]] [[unom[1]]
b. v ... [DP D NP ]
[iacc[1]] [uaccval[1]] [uacc[1]]
c. P ... [DP D NP ]
[iobl[1]] [uoblval[1]] [uobl[1]]
In (206a), D, bearing uninterpretable, valued nom, is c-commanded by an NP bearing an

uninterpretable, unvalued instance of the same feature. The two can therefore undergo
agreement. Subsequently, a merged T bearing an interpretable, unvalued nom Case feature
comes to c-command the DP headed by D. These two can also undergo agreement, yield-
ing an interpretable, valued feature with several occurrences. Similarly for the scenarios in
(206b,c). All these yield interface-legible objects. With these points in mind, let us consider
now some of the Case-assignment paradigms considered above in (190) and (191).
One Structural Accusative + One P Case

The simplest paradigm is the one in (191b), repeated below as (207a), involving one
structural accusative Case and one prepositional Case. This pattern was suggested to
govern Halkomelem applicatives and the with/of variants in English spray-load alter-
nations. (207b) offers a tentative analysis of the latter for a vP portion of the English
sentence John loaded the truck with hay:
(207) a.
v V DP1 [ P DP2]
b.
vP
v vP
v v DP v′
[uAGval[ ]]
load v [iACC[1]] the truck v VP
[uACCval[1]] [uACC[1]]
load v PP V′
with hay load DP

[iACC[2]] [uACCval[2]]
the truck
[uACCval[1]] [uACC[1]]
Following Kayne (1984) I assume that objects of English prepositions generally bear
accusative Case; furthermore, following Chomsky (2000, 2001), I assume that structural
accusative Case on objects is associated with agentive little v. In the current framework,
these views are executed as follows: the locative DP the truck is initially projected into
lowest position in accordance with the thematic hierarchy; D (the) bears a valued, unin-
terpretable acc feature with which its sister noun (truck) agrees. This DP subsequently
raises to the specifier of the loc-bearing little v voice head; from that point it is accessible
to the higher little v voice head bearing [ag], whose interpretable [acc] feature undergoes
agreement with it. Within PP itself, the preposition with bears an interpretable, unvalued
acc feature, whereas its DP object hay bears an uninterpretable, valued acc feature;
these two undergo agreement. Case relations are thus directly accommodated.
Two (Non-Prepositional) Accusative Cases I

In the current system, multiple accusative arguments that are unassociated with covert
prepositions continue to raise a question. Consider (208a). Since valued instances of
features do not probe for agreement, it follows that the two DPs shown will not undergo
agree on their [uaccval[ ]] features. This means that when a higher v bearing [iacc[ ]]
is merged and undergoes agreement with DP1, probing will terminate under minimality,
without the lower instance of [uaccval[ ]] on DP2 becoming part of an interpretable
feature. This will yield an object that is not legible at Logical Form. Raising DP2 to a
position above DP1 as in (208b) will not evade the problem. Now the higher DP2 inter-
cepts the probe from v, preventing DP1 from becoming part of an interpretable feature:
(208) a.
v
[iACC[1]] DP1
[uACCval[1]] DP2
[uACCval[ ]]
b.
v
[iACC[1]] DP2
[uACCval[1]] DP1
[uACCval[ ]] DP2
[uACCval[ ]]
How then can multiple accusativity in English and Korean be accounted for?
Chomsky (1995) proposes that structural accusative Case is assigned by the little v
element responsible for the agentive θ-role. In our terms, a little v voice head bearing
a valued (unintepretable) θ-feature ([ag]) is associated with an interpretable (unvalued)
Case feature ([acc]). Suppose this association of Case and θ-role generalizes, so that a little
v element bearing a valued (unintepretable) [gl] feature can also carry an interpretable
(unvalued) [acc]-feature. Then, in an A-Shift derivation like (177b), partially repeated
below as (209) with irrelevant details suppressed, there will be two sources of structural
Case in the structure: an accusative Case feature associated with the higher agentive voice
head, and an accusative Case feature associated with the lower goal voice head:
(209) v′
v vP
v v Mary v′
[uAGval[ ]] [uACCval[2]]
give v [iACC[2]] v VP
…
give v
[uGLval[1]] Fido V′
[iACC[1]] [uACCval[1]]
…
give Mary
The Case requirements of the two DPs are therefore met by the presence of two Case
assigners (in the informal sense of the latter).
The availability of two accusative Cases allows for the familiar fact that English can
passivize the inner object in a double object structure (210):
(210) Mary was given Fido
On the usual view that Passive involves suppression of accusative Case (here, suppres-
sion of an accusative Case feature) in concert with neutralization of the agent θ-role,
this circumstance in (209) will still leave a structural accusative Case feature available
to the outer object, correctly allowing for (210).
The picture in (209) also suggests a way of excluding A-Shift derivations like the locative
one discussed earlier in (188). We might take this limitation to reflect constraints on which
voice heads are associable with a structural Case feature. Thus, if little v valued for LOC
cannot bear interpretable ACC, then (188) will be straightforwardly excluded on Case-
theoretic grounds: the lower DP will have no way of coming into agreement with an inter-
pretable ACC feature.64 This will essentially limit English locative applicative alternations
to the form in (191b), where the outer object has an independent prepositional Case source.
Two (Non-Prepositional) Accusative Cases II

The proposal in (209) represents a plausible approach to accusative Case assignment in
English double object constructions, one that maintains the usual bi-unique relationship
between Case assigners (items interpretable for Case but unvalued for it) and argument
Case assignees (items valued for Case but uninterpretable for it). Nonetheless, other,
more radical alternatives are possible and arguably necessary. We noted that Korean
allows multiple accusative marking in double object constructions; (212a) (= (193))
alternates with a dative example like (212b).
(211) a. Minhee-ga Chulsu-lul cek-lul cueyatta.

Minhee-nom Chulsu-acc book-acc gave
b. Minhee-ga Chulsu-eykey cek-lul cueyatta.
Minhee-nom Chulsu-dat book-acc gave
In fact, give-type constructions are only one place where Korean exhibits multiple
accusativity. Others include so-called possessor raising structures, in which accusa-
tive alternates with genitive (212a,b), measure and frequency adverbial constructions
(213a,b), and quantifier constructions in the order NP-Q (214a,b).
(212) Possessor Raising

a. Mary-ka John-ul pal-ul cha-ess-ta.
Mary-nom John-acc foot-acc kick-pst-decl
‘Mary kicked John’s foot.’
b. Mary-ka John-uy pal-ul cha-ess-ta.
Mary-nom John-gen foot-acc kick-pst-decl
‘Mary kicked John’s foot.’
(213) Measure and Frequency Adverbials

a. Mary-ka maykcwu-lul twu sikan-ul masi-ess-ta.
Mary-nom beer-acc two hour-acc drink-pst-decl
‘Mary drank beer for two hours.’
b. Minhee-ga John-lul twu pen-ul cha-ess-ta.
Minhee-nom John-acc two time-acc kick-pst-decl
‘Minhee kicked John twice.’
(214) Quantifier Float Constructions

a. Mary-ka twu haksayng-tul-ul sohwanhay-ss-ta.
Mary-nom two student-pl-acc call-pst-decl
‘Mary called two students.’
b. Mary-ka haksayng-tul-ul twu-ul sohwanhay-ss-ta.
Mary-nom student-pl-acc two-acc call-pst-decl
‘Mary called two students.’
These constructions exhibit interesting properties. As discussed extensively by Cho (2000),

possessor raising structures are recursive so that (212a), for example, can be elaborated as
in (215). Furthermore, multiple accusative constructions may be combined, so that cases
like (216) are possible. Finally, and perhaps most remarkably, sequences of accusatives may
apparently invert under passive, yielding corresponding sequences of nominatives (217):
(215) Mary-ka John-ul pal-ul kkup-ul

cha-ess-ta.
Mary-nom John-acc foot-acc end-acc kick-post-decl
‘Mary kicked the end of John’s foot.’ (= (31a) in Cho 2000, p. 96)
(216) Mary-ka haksayng-tul-ul twu-ul tali-lul twu pen-ul cha-ess-ta.
Mary-nom student-pl-acc two-acc leg-acc two time-acc kick-pst-decl
‘Mary kicked two students’ legs twice’
(217) a. John-i pal-i kkup-i cha-i-ess-ta.
John-nom foot-nom end-nom kick-pass-pst-decl
‘The end of John’s foot was kicked.’ (= (35a) in Cho 2000, p. 102)
Haksayng-tul-i twu-i
b. tali-tul-i twu pen-i cha-i-ess-ta.
student-pl-nom two-nom leg-pl-nom two time-nom kick-pass-pst-decl
‘Two students’ legs were kicked twice.’
The latter is particularly striking because, as Cho (2000) notes, the acc-marked items
in these examples appear to constitute nonconstituents—that is, separate phrases. This
is obvious on semantic grounds with object-adverb sequences like (213a,b) but holds
with the other cases as well. Thus the acc-acc sequence in (212a) can be interrupted by
adverbs (218a), in contrast to the gen-acc sequence in (212b); cf. (218b).65
(218) a. Mary-ka (seykey) John-ul (seykey) tali-lul (seykey) cha-ess-ta.
Mary-nom hard John-acc hard leg-acc hard kick-pst-decl
‘Mary kicked John’s leg hard.’
b. Mary-ka (seykey) John-uy (*seykey) tali-lul (seykey) cha-ess-ta.
Mary-nom hard John-gen hard leg-acc hard kick-pst-decl
‘Mary kicked John’s leg hard.’
These results raise the interesting question of how multiple nominatives like (217a,b)
can be licensed in the absence of multiple sources of nominative Case, and how this
possibility relates back to multiple accusativity.
Interestingly, the Pesetsky and Torrego feature system makes available a radical option
that has not (to my knowledge) been exploited before in this context. As observed above,
Case assigners correspond here to items interpretable for Case but unvalued for it; Case
assignees correspond to items valued for Case but uninterpretable for it. This leaves open
at least one more relevant possibility: items bearing a Case feature that is both uninter-
pretable and unvalued. To see the interest of this possibility, reconsider the two situations
schematized in (208), repeated below in (219). Notice that if DP1 in (219a) were allowed
to bear an uninterpretable and unvalued accusative Case feature, then it would be able to
undergo agreement both with the lower, valued instance of the [acc] feature on DP2 and
with the higher, interpretable instance of [acc] on v, yielding an interface-legible object.
Likewise, notice that if DP2 bearing an uninterpretable and unvalued accusative Case
feature is initially positioned below DP1 bearing a valued instance of this feature, but
DP2 subsequently moves above DP1, then the same positive outcome will occur (219b):
(219) a.
v
[iACC[1]] DP1
[uACC[1]] DP2
[uACCval[1]]
b.
v
[iACC[1]] DP2
[uACC[1]] DP1
[uACCval[1]] DP2
[uACC[ ]]
Movement
On this analysis, a single Case is “shared” among arguments. More precisely, the
lower argument is valued for Case, whereas the higher argument obtains its value by
agreement.
This picture is familiar from other contexts. With nominals like (220) from
Icelandic (Kester 1996), it is traditional to describe the noun kennigɑr ‘theories’
as inflected for feminine plural nominative, and the determiner and adjectives as
obtaining their inflection by agreement. This way of speaking implies that inflection
is somehow “real” on N, but derivative on the other elements, present simply as
“concord” (221).
(220) a. allαr þessαr þrjár nyju
all.fem.pl.nom these.fem.pl.nom three.fem.pl.nom new.fem.pl.nom
kennigαr
theories.fem.pl.nom
‘all these three new theories’
(221)
D
AP
AP
AP N
Concordial
“Real”
Inflection
Inflection
In Larson (in preparation), concordial inflection in nominals is analyzed through unin-

terpretable/unvalued features in the Pesetsky and Torrego (2004) system. In essence,
then, (219a,b) represents an extension of this “concordial” analysis to the verb phrase:
in both instances the higher DP stands in a relation of concord with the lower one with
respect to Case.
A concordial analysis is attractive for multiple accusativity in Korean. Notice, for
example, that there is no barrier in principle to the number of agreeing accusatives
above the lowest, valued one (222), just as there is no barrier in principle to the number
of inflected modifiers that can stack above N in (221):
(222)
v
DP [iACC[1]]
[uACC[1]] DP
[uACC[1]] DP
[uACC[1]] DP V
[uACCval[1]]
“Concordial “Valued
ACCs” ACC”
Observe another, subtler point as well. On the account of movement from Chomsky
(2000, 2001) discussed above, the EPP feature on a functional head can raise a phrase
to its specifier position, but that phrase must first undergo agreement with the head.
In the case of passives, the relevant functional head is widely assumed to be T, and the
relevant feature on T is arguably Case—specifically [nom]. Assuming a pre-movement
structure roughly as in (222), an interesting question now arises. On the Pesetsky and
Torrego system, agreement is taken to result in a single feature with multiple instances.
Hence when we analyze T in (223a) as attracting DP bearing the feature nom, with
which T agrees, an ambiguity arises. Does T attract/raise an instance of nom, or does
it raise the feature itself, in all its instances? If the latter, then the entire sequence of
nom-marked DPs might be seen as raised simultaneously, despite its nonconstituent
status (223b):66
(223) a.
T
v [iNOM[ ]]
DP
[uNOM[1]] DP
[uNOM[1]] DP
[uNOM[1]] DP V
[uNOMval[1]]
“Concordial “Valued
NOMs” NOM”
b.
DP
[uNOM[1]] DP
[uNOM[1]] DP
[uNOM[1]] DP
[uNOMval[1]]
T
v [iNOM[1]]
DP
[uNOM[1]] DP
[uNOM[1]] DP
[uNOM[1]] DP V
[uNOMval[1]]
Pesetsky and Torrego’s account of features thus offers a potentially novel analysis of
multiple accusativity, and associated multiple nominativity under Passive.
One P Case + One Structural Accusative?

The concordial account also affords an interesting reanalysis of the Case paradigm
in (191c), repeated below as (224a), potentially exemplified by Spanish examples like
(196a), repeated below as (224b):
(224) a. v V [ P DP1 ] DP2 One P Case + One Structural Accusative
b. Susana le dió [ a Juan ] un libro.
Susana cl.him gave to John a book
‘Susana gave John a book.’
Spanish is well known to exhibit a “differential object marker” (dom) a with ani-
mate objects of transitive verbs (225a,b) (=(1a,c) resp. from Brugè and Brugger 1996
p. 3); this item is homophonous with the dative preposition a and derives from it
historically.
(225) a. Esta mañana he visto *(a) Juan/la hermana de Maria.
this morning I-have seen dom Juan/the sister of Maria
‘This morning I saw Juan/the sister of Maria’. (= (1a) in Brugè and Brugger 1996)
b. Esta mañana he visto (*a) la nueva iglesia.
this morning I-have seen dom the new church
‘This morning I saw the new church.’ (= (lc) in Brugè and Brugger 1996)
Differential object markers represent something of a puzzle for Case theory given their
occurrence in environments where accusative Case would seem to be already available.
If the dom itself carries an interpretable accusative Case feature (like a preposition), and
agrees with the animate direct object, then what becomes of the interpretable accusative
Case feature associated with little v (226a)? And if dom does not agree for Case with the
animate direct object, how does the accusative Case feature on little v probe past DOM
to the direct object, in apparent violation of minimality (226b)?
(226) a. [ v V [ DOM DP ]
[iACC[ ]] [iACC[1]] [iACC[1]]
??
b. [ v V [ DOM DP ]
[iACC[1]] [iACC[1]]
??
One potentially attractive idea is that DOM is a concordial element—specifically, an ele-

ment of category K differing from Ps precisely in bearing an uninterpretable, unvalued
instance of acc. This entails that in an example like (225a), a will essentially be “transpar-
ent” to accusative Case: agreement will pass through it from little v to the object DP:67
(227)
v VP
[iACC[1]]
visto KP
a Juan
[uACC[1]] [uACCval[1]]
One might extend this picture to the analysis of Spanish double object forms like (224b),
roughly along the lines sketched earlier in (219b). Suppose, extending ideas by Demonte
(1995), Bleam (2000), and Cuervo (2003), that the doubled clitic occupies the position
of the voice head, and that the KP recipient goal is initially projected in low position,
following the dictates of the thematic hierarchy, subsequently raising to the specifier of
the clitic/voice phrase (228).
(228)
v vP
KP v′
a Juan le VP
DP V′
un libro dió KP
a Juan
movement
An account parallel to that of Korean multiple accusativity now becomes possible, where
Case to the lower/outer object is established by concord as in (229). KP bears the [uacc[1]]
feature as a copy from its K head (a). This feature can undergo agreement with the [uacc[
]] feature on the lower DP (un libro) (arrow marked “1”). When little v is combined, it
undergoes agreement with the [uaccl[1]] features on both KP and K (arrow marked “2”).
The result is a single interpretable acc feature associated with two valued instances. Again,
uninterpretable, unvalued features are the glue that binds all the feature instances together:
(229)
v vP
[iACC[1]]
KP v′
[uACC[1]]
le VP
(2)
a Juan
[uACC[1]] [uACCval[1]] DP V′
un libro dió KP
[uACCval[1]] a Juan
[uACC[1]] [uACCval[1]]
(1)
This general concordial analysis of Spanish datives might be extended to a broader

range of structures including unaccusatives and psych-verb constructions, where con-
cordiality is also permitted with respect to nominative.68
One Inherent Case + One Structural Accusative?

Finally, we may consider the patterns repeated below from Modern Greek (197), Romanian
(199b), and Japanese (200), offered as potential examples of inherent Case combined with
structural accusative.
(230) a. (Tu) edhosa tu Giani to vivlio. Modern Greek

cl-gen gave-lsg the Gianis-gen the book-acc
‘I gave John the book.’
b. Mihaela (ii) trimite Mariei o scrisoare. Romanian
Mihaela cl.dat sends Mary-dat a letter
Mihaela sends Mary a letter.

c. Taroo-ga Hanako-ni hon-o agetta.
Taroo-nom Hanako-dat book-acc gave Japanese
‘Taroo gave Hanako a book.’
In the present account, where θ-roles are features, the notion of an inherent Case—one
associated with a particular θ-role θ (or set of them {θ1, . . . θn})—is straightforward to
define. A Case feature CASE will be an inherent one iff for any DP α, α bears [uCASEval[ ]]
implies α bears [iθ[ ]] for some θ-feature [θ] (or α bears one of [iθ1[ ]], . . ., [iθn[ ]]). In other
words, an inherent Case feature is one such that bearing a valued instance of it implies
bearing an interpretable version of a specific θ-feature. To illustrate schematically, consider
an instrumental preposition P and an object bearing the inherent instrumental Case feature
[instr] (231):
(231) PP
P DP
[uθINSTval[1]] [iθINST[1]]
[iINSTR[2]] [uINSTRval[2]]
Under our definition, if DP bears a valued instance of [instr] it must also bear an
interpretable instance of the associated θ-feature, here [θINST]. This requirement is met
in (231). By contrast, bearing a valued instance of a structural Case feature exercises no
constraints regarding θ-features. This then constitutes the difference between inherent
and structural Case.
In fact, the assessment of genitive/dative morphology in (230a–c) as true inherent Case
along the lines just sketched is difficult to reconcile with current theoretical assumptions. Sup-
pose, following Anagnostopoulou (2003), we regard little v bearing gl (= Anagnastopolou’s
appl) as agreeing for inherent genitive Case with tu Giani in (232a) through the Spec-head
relation. Then little v bearing agent must probe for accusative Case on the lower theme
across the higher goal (232a). This constitutes a clear minimality violation69. On the other
hand, if little v bearing gl probes for accusative Case on the lower theme, following the
scenario suggested above for English, then what probes inherent Case on tu Giani? Little
v bearing agent is not an inherent Case probe (232b):
(232) a. vP
… v′
v vP (= APPLP)
[iACC[3]]
tu Giani v′
[iGL[1]]
[uGENval[2]] v VP
[uGLval[1]]
[iGEN[2]] to vivlio V′
?? [uACCval[3]]
…edhosa…
b. vP
??
…edhosa…

b. vP
… v′
v vP (= APPLP)
[iACC[3]]
tu Giani v′
?? [iGL[1]]
[uGENval[3]] v VP
[uGLval[1]]
[iACC[2]] to vivlio V′
[uACCval[2]]
…edhosa…
Neither scenario is plausible.

Harada and Larson (2007, 2009) propose that examples like (230c) (and by exten-
sion (230a,b)) involve the concordial scheme discussed above. In brief, dative in VP and
genitive in DP are suggested to have a dual status in grammar: as inherent Cases asso-
ciated with particular thematic features and as invariant concordial/adjectival Cases.
Larson (in preparation) proposes that English DPs exhibit this dualism in the alterna-
tion in (233), where the PP object position is associated with inherent genitive Case
associated with of, and where prenominal position is one associated with adjectival/
concordial elements.
(233) a. an old friend [of John’s]
b. John’s old friend
Harada and Larson extend this dualism to dative -ni in the Japanese VP, which is
taken to alternate between a postposition and an invariant concordial Case marker.
In brief, an argument goal ni-phrase like Hanako-ni is assumed to be merged in the
low V-complement position bearing a concordial uacc feature (233). In this posi-
tion Hanako-ni is unable to undergo agreement with the higher, valued accusative
theme but can scramble to the VP edge, essentially following Zushi (1992). In this
position, beween the interpretable iacc feature on little v and the valued uaccval
feature on hon-o, it can undergo agreement. One Case is thus distributed between
two arguments:
(234) vP
Taroo-ga v′
v VP
[iACC[1]]
Hanako-ni VP
[uACC[1]]
hon-o V′
[uACCval[1]]
Hanako-ni agetta
[uACC[ ]]
Scrambling
In essence, then, this scenario is identical to the one proposed for Korean. The difference
between the two languages is that Korean concordial Case features are realized morphologi-
cally in the form of their valued counterparts whereas Japanese -ni is analyzed here as mor-
phologically invariant no matter what Case feature is involved. Thus a Japanese ni-marked
phrase can also be concordial with a nominative as well as with an accusative, as Harada
and Larson (2009) discuss. This view appears to be extensible to the examples in MG and
Romanian as well; dative morphology might be seen as marking concordial status.
5. SUMMARY
Work on the dative alternation over more than two decades has dramatically deep-
ened our knowledge of the construction and its wider connections in grammar. New
comparative linguistic data, particularly from the realm of applicative constructions,
has considerably clarified the empirical burdens of explanation in this area. The dative
alternation, which initially presents itself as a small and idiosyncratic corner of English
grammar, plausibly takes its place as part of the broader applicative-oblique alternation
observed in many world languages. The semantic breadth of this alternation, together
with its robust regularity in many languages, has powerful implications for analysis.
In this introduction I have examined basic aspects of syntactic projection and have
argued that although nonderivational approaches to the applicative-oblique alternation
are widely assumed in current theorizing, the evidential basis for them is considerably
weaker than has been supposed, and the challenges to them considerably greater than
has been generally appreciated. Indeed, there has been considerable unclarity about
what constitutes a nonderivational approach in the first place. To the best of my knowl-
edge, discussion of independent projection has confined itself almost entirely to pos-
sessives and related forms (benefactives). The prospects for nonderivational analyses
beyond this narrow domain seem to me rather dim, for the reasons discussed.
By contrast, derivational approaches, although a decidedly minority position in cur-
rent thinking, seem to me to have retained their interesting implications for the theory
of phrase structure and to offer robust possibilities for further development. Here I have
briefly tried to show how the updated approach to shell structure sketched in the General
Introduction can be extended to yield a natural, derivational account of applicative-oblique
alternation, viewed as arising from a simple option in structure building: introduction of
valued θ-features via P versus little v. As with shell theory generally, although the exact
mechanisms for projecting structures have changed substantially, many of the basic results
of the derivational approach to datives in Larson (1988) and the other papers in this sec-
tion appear to be preserved intact. The phenomena that this approach relates, and the basic
explanatory connections it suggests, still seem to me interesting and worthy of pursuit.
NOTES
1. Research on datives did, however, continue very actively in other frameworks, especially
Relational Grammar, which was an important source of empirical results and insights for
my own work. See sections 1.5 and 3.1.
2. For further discussion of Dowty’s proposals see section 2.1.2.
3 This analogy, I later discovered, was one that had occurred to many, particularly within the
framework of Relational Grammar. See also the discussion in Farrell (2005) with respect to
applicative constructions.
4. This might be analyzed as a matter of scope: that nominals either generated in or moved
to object position (but not lower) come into the scope of a higher element, for example, an
aspectual head, that determines this reading; see MacDonald (2009).
5.
Belong is not the only English verb showing an oblique locative/possessive alternation of this
kind. The verb go is analogous in pairs like (ia,b):
(i) a. John goes/went to that grocery store.
b. This piece goes/went to that puzzle.
The first use of go is plainly motional/locative, but the second use is possessive, with mean-
ing very similar to (37d) with belong. Possessive uses of go like those of English are regularly
found in other languages, for example, Spanish.
6. I am grateful to Pilar Barbosa (p.c.) for discussion of the data in (38).
7. One complicating factor in this otherwise simple picture is that both a and para are available
as allative prepositions with simple motion verbs (i):
(i) O João foi a Lisboa/para a Lisboa.
the John went to Lisbon
‘John went to Lisbon.’
P. Barbosa reports that the senses are slightly different in the two cases; specifically, use of a
implies that John went to Lisbon and is necessarily coming back (the journey is brief); use
of para does not.
8. This conclusion has familiar parallels in languages like Japanese and Korean, where dative
elements are often suggested to have dual status as (structural) Case markers and as
contentful postpositions. Thus in Japanese, dative -ni is widely taken to function both as
a Case marker with give-type verbs (ia) and as a directional postposition with verbs of
motion (ib). Evidence for the Case-marker status of the former versus. the postpositional
status of the latter is that the first -ni supports quantifier float (Q-float) whereas the sec-
ond does not (iia,b). Further evidence for the dual status of -ni might be derived from the
fact that Japanese contains an independent allative postposition -e ‘to’ (iiia), which also
resists Q-float (iiib). A simple conclusion might be that Japanese contains both a dative
Case particle -ni and an allative postposition -ni, the latter equivalent to the unambiguous
postposition -e.
(i) a. Hanako-ga inu-ni esa-o yatta.
Hanako-nom dog-dat food-Acc gave
‘Hanako gave food to the/some dog(s).’
b. Hanako-ga honya-ni arui-te-ikimashita.
Hanako-nom bookstore-dat walk-te-went
‘Hanako walked to the/some bookstore(s).’
(ii) a. Hanako-ga inu-ni sanbiki esa-o yatta.
Hanako-nom dog-dat three food-acc gave
‘Hanako gave food to three dogs.’
b. ??Hanako-ga honya-ni sangen arui-te-ikimashita.
Hanako-nom bookstore-dat three walk-te-went
‘Hanako walked to three bookstores.’
(iii) a. Hanako-ga honya-e arui-te-ikimashita.
Hanako-nom bookstore-to walk-te-went
‘Hanako walked to the/some bookstore(s).’
b. *Hanako-ga honya-e sangen arui-te-ikimashita
Hanako-nom bookstore-to three walk-te-went
Hanako walked to three bookstores.
An interesting complication is the fact that for numbers of Japanese speakers, -e is also
compatible with unambiguous caused-possession verbs like ageru and yaru (‘give’), with an
only slightly reduced level of acceptability (iva), even though Q-float remains unacceptable
(ivb). Note also that judgments of unacceptability differ subtly in (iib)/(iiib). Some speakers
report that even when understood allatively, -ni sanctions Q-float more readily than -e.
(iv) a. ?Hanako-ga inu-e esa-o yatta.
Hanako-nom dog-to food-acc gave
‘Hanako gave food to the/some dog(s).’
b. *Hanako-ga inu-e sanbiki esa-o yatta.
Hanako-nom dog-to three food-acc gave
‘Hanako gave food to three dogs.’
It thus appears that even if dative -ni is ambiguous between a Case marker and an alla-
tive postposition, the allative -e itself has the option of being understood much like a dative
marker. Furthermore, even when it is understood as an allative (iib)/(iiib), -ni is less postposi-
tion-like than -e. This matter plainly deserves further study, which cannot be attempted here.
9. See Green (1974), Appendix IV, p. 224, for an extensive list of such dative idioms.
10. Further evidence for the incorporated nature of these nominals is their obligatorily bare,
unmodified form, strongly suggesting status as a simple N (ia):
(i) a. *Mao’s silence gave quick rise to an absurd rumor.
(cf. Mao’s silence quickly gave rise to an absurd rumor.)
b. ??Sarah gave recent birth to a son.
(cf. Sarah recently gave birth to a son.)
c. *Activism gave immediate way to apathy.
(cf. Activism immediately gave way to apathy.)
d. *John gave some/far too much rein to his feelings.
In this respect give rise, give birth idioms diverge from those of the familiar pay attention/keep
tabs sort, which do allow modification (iia,b) and are susceptible to A-movement (iiia,b):
(ii) a. John paid some/a great deal of/far too much attention to the proposal’s problems.
b. John kept close/very close/very few tabs on Mary’s activities.
(iii) a. Some/A great deal of/Far too much attention was paid to the proposal’s problems.
b. Close/Very close/Very few tabs were kept on Mary’s activities.
This behavior suggests that the nominal expressions in the latter class are indeed indepen-
dent elements in argument position, in contrast with those in the dative class.
11. This analysis seems superior to the suggestion of Richards (2001) that forms like (44a–d)
simply correspond to underlying caused-locative idioms in the sense of Harley (1995, 2002).
There is no clear sense in which rise goes to a rumor location, that reins go to feelings, and
so on. Indeed, these forms barely seem to embed a recognizable ‘give’ meaning at all.
12. This definition is standard; thus Radford (1997): “We can define idioms as expressions . . .
which have an idiosyncratic meaning that is not a purely componential function of their
individual parts” (p. 159).
13. It’s worth noting that the WSNCD citation for creep recorded in (53a) makes no mention
of give, implicitly confirming that this sense of creep can be characterized independently of
the verb. Likewise, a piece of (one’s) mind is listed by AHDEL as an idiom under the entry
for piece, and not under the entry for give.
14. (60) shows interesting additional properties. As the WSNCD entry (53c) indicates, the
intended sense of boot borrows from British English, and (in my experience) American
speakers who use this expression know its provenance. As far as I know, all elaborations
(60b) and substitutions (60d) of this expression are also borrowings from British English,
and also known to be. This suggests that when such an expression is borrowed, its “foreign”
nature continues to govern its distribution.
15. Harley (1995, 2002) and Richards (2001) cite the purported dative idioms as part of an argu-
ment that inferences between give and get like (ia,b), noted in Larson (1988), are syntactically
grounded. Both authors assume a syntactic decomposition of give, essentially as cause-have (ic):
(i) a. The Count gave Mary the book/the creeps.
b. Mary got the book/the creeps.
c. The Count cause [Mary have the book/the creeps]
Harley (1995, 2002) adopts an unaccusative analysis of get, an idea first proposed by
Haegeman (1985). On this view, the get-structure in (ib) derives by raising the subject of
have (iib); Richards suggests a nonraising structure involving a higher predicate become,
where become-have presumably spells out as get (iib):
(ii) a. Mary get [ ____ have the book/the creeps]
b. Mary become [have the book/the creeps]

A simple issue for both syntactic accounts of the relation between (ia,b) is the counterpart
relation between (iiia,b); we noted that the oblique form entails the corresponding posses-
sive just as the double object form does:
(iii) a. The Count gave the book to Mary.
b. Mary got the book.
Two responses seem possible. One is to embrace a derivational view of the dative alter-
nation wherein the relation between (iiia) and (iiib) is mediated by an intervening double
object stage; the analysis then proceeds as in (iia) or (iib). The other is to appeal to a
nonsyntactic, inferential mechanism to relate (iiia,b)—for example, some form of meaning
postulate. Both Harley (1995, 2002) and Richards (2001) explicitly reject a derivational
account of the dative alternation, and so appear committed to the latter. But this then
entails a serious redundancy in their views. If an inferential mechanism is required for the
relation between (iiia) and (iiib), then it is available for the relation between (ia) and (ib) as
well. Indeed, given the main results in this section—that forms like give ~ the creeps are not
idioms at all but are instead fully compositional—a purely semantic account of the relation
between give ~ the creeps and get ~ the creeps becomes both possible and desirable.
16. Levin (pers. comm.) notes other purported cases of double object idioms not involving give,
such as (ia–c):
(i) a. Mary showed John the ropes.
b. John promised Mary the moon.
c. Max read John the riot act.
Here again there are reasons to doubt either that verbal idiomaticity is involved or that
the construction is confined to double object form in the first place. Regarding (ia) we again
note that the nominal part routinely receives an independent dictionary definition as in (iia)
from AHDEL and occurs with a variety of other verbs outside double object form (iib):
(ii) a. 7. ropes Informal Specialized procedures or details
b. learn the ropes, know the ropes, teach X the ropes
Regarding (ib) it seems dubious that this expression is an idiom; instead, its communica-
tive effect is based on its obvious violation of the preconditions for commissive performa-
tives like promise. In brief, given that the sun, moon, and stars are plainly something that
cannot be promised, hearers will derive the appropriate “make an unreasonable/unfulfill-
able promise” meaning as an implicature calculated from the speech act conditions for
promising. Finally, regarding (ic) it is possible that the latter is indeed a true idiom, but
equally there is no evidence that the latter is confined to double object form. AHDEL lists
read the riot act as an idiom under its entry for Riot Act. But the example it supplies is in
the oblique form:
(iii) Riot Act: . . .
-Idiom: read the riot act To warn or reprimand energetically or forcefully: The teacher
read the riot act to the rowdy class.
The entry then goes on with a section of Word History whose very first sentence contains
an oblique.
Word History The riot act has been read to far more people than the disturbers of the
(iv)
peace the Riot Act
was intended to control . . .
So although read the riot act may indeed be a true idiom, it plainly alternates.
17. This constraint was first observed and discussed by Ross (1967).
18. Alternatively, in modern minimalist terminology, we might say that the constraint operative
in (47a–f) is sensitive to PF and must be imposed within the local phase.
19. I am grateful to Pilar Barbosa (personal communication) for the examples in (70) and
enlightening discussion of them.
20. Alternative analyses are of course possible. Regarding the Spanish examples in (69c,d),
Bleam (2003), Cuervo (2003), and De Pedro Munilla (2004) have proposed that the former
(ia) is in fact a double object construction with the latter derived from it by optional move-
ment (ib). The word order similarity between the latter and a true oblique construction is
thus an illusion for these authors:
(i) a. Susana lei mandó a Pabloi una carta.
Susana Cl-3Dat sent to Pablo a letter
‘Susana sent a letter to Paul.’
b. Susana lei mandó [una carta] a Pabloi ___
One could envision a similar account of European Portuguese, taking it to contain dative
clitics counterpart to those in Spanish, but obligatorily null, and a movement operation
counterpart to that in Spanish, but obligatorily applied (iia). Examples like (70d) would
then be strictly equivalent to English double objects, and what European Portuguese lacked
would not be the double object construction but rather a true oblique give frame (iib):
(ii) a. O João Cl enviou [uma carta] à Mary ___
the John her.Cl sent a letter to Mary

‘John sent a letter to Mary.’
b. A Mary Cl deu [uma mão] ao Pedro ___
‘Mary gave Peter a hand.’

While I have no decisive refutation of such a proposal, it seems at least suspicious that its ele-
ments conspire to make European Portuguese datives surface-indistinguishable, up to preposition
choice, from true oblique forms, which must be postulated independently in the language (iii):
(iii) O João enviou uma carta à Mary/para a Mary
‘John sent a letter to Mary’
Indeed, given the fact (mentioned in note 8) that a ‘to’ has directional/locative uses else-
where in European Portuguese, the movement postulated in (ii) has the effect of erasing
the only evidence for an underlying double object construction apart from meaning. Until
further evidence is provided for the key assumptions—the presence of null clitics and an
obligatory fronting rule—such an account of European Portuguese appears largely ad hoc.
See Ordóñez (1998) and Belletti and Shlonsky (1995) for arguments that structures like (ia)
in Romance are in fact the derived ones, arising by fronting of a PP, and are not the base
structures, contra Bleam (2003), Cuervo (2003), and De Pedro Munilla (2004).
21. Harley (1995, 2002) proposes that Japanese possessive constructions like (72) are not in fact
oblique copular constructions but rather have-constructions with a quirky, dative-marked
subject. She notes that the dative-marked possessor c-commands the nominative-marked
possessum to its right, as expected of the possessor in a have-type construction (i).
(i) a. [Taroo-to Hanako]i-ni otagaii-e-no ai-ga ar-u.
Taro-and Hanako-dat each.other-to-gen love-nom be-pres
‘Taro and Hanako have love for each other.’
b. *[Otagai]i-ni [Taroo-to Hanako]i-e-no ai-ga ar-u.
each.other-dat Taro-and Hanako-to-gen love-nom be-pres
Lit. ‘Each other have love for Taro and Hanako.’
Dative-marked nominals in Japanese possessor constructions resemble those in Japanese
ditransitive dative constructions. The ni-marked goal c-commands the accusative-marked
theme to its right (ii):
(ii) a. Hanako-ga [karera]i-ni [otagaii-no sensei]-o syookaisi-ta.
Hanako-nom they-dat each.other’s teacher-acc introduced
‘Hanako introduced [them] to [each other’s teacher].’
b. Hanako-ga [otagaii-no sensei]-ni [karera]i-o syookaisi-ta.
Hanako-nom each. other’s teacher-dat them-acc introduced
‘*Hanako introduced [each other’s teacher] to [them].’
Japanese ditransitive datives have been analyzed as double object constructions (Zushi
1992; Miyagawa and Tsujioka 2004). Given Harley’s (1995, 2002) account of double
objects as causatives of have-small clauses, the convergence of patterning in (i) and (ii)
appears significant. However, Harada and Larson (2009) show that the facts in (i) and (ii)
are only part of the pattern. Japanese locatives, also marked with -ni, show the very same
distribution. Thus, in transitive locatives, the dative-marked location phrase c-commands
the nominative-marked locatum to its right (iiia–c). And in Japanese ditransitive locatives,
the dative-marked location phrase c-commands the accusative-marked theme (iva–c).
(iii) a. Tokyo-ni John-ga iru
Tokyo-dat John-nom be-pres
‘John is in Tokyo.’
b. [10-izyoo-no meekaa-no tihoo-no koozyooi]-ni [sokoi-no syuyoo-seihin]-ga ar-u.
10-more-gen maker-gen countryside-gen factory-dat it-gen main-product-nom
be-pres
Lit. ‘Their main products are at the factories in the countryside of more than 10 makers.’
c. *[S okoi-no tihoo-no koozyoo]-ni [10-izyoo-no meekaai-no syuyoo-seihin]-ga ar-u.
it-gen countryside-gen factory-dat 10-more-gen maker-gen main-
product-nom be-pres
Lit. ‘More than 10 maker’s main products are at its factory in the countryside.’
(iv) a. John-ga nidai-ni hon-o nose-ru
John-nom cart-dat book-acc load-pres
‘John loads the books on the cart.’
b. Yamada-sensei-ga [Taroo-to Hanako]i-no tukue-no-ue-ni
Yamada-professor-nom Taro-and Hanako-gen desk-gen-over-dat
otagaii-no tooan-o oi-ta.
each.other-gen answer.sheet-acc put-pst
‘Professor Yamada put each otheri’s answer sheets on [Taro and Hanako]i’s desks.’
c. *Yamada-sensei-ga [otagaii-no tukue-no-ue]-ni
Yamada-professor-nom each.other-gen desk-gen-over-dat
[[Taroo-to Hanako]i-no tooan]-o oi-ta.
Taro-and Hanako-gen answer.sheet acc put-pst
‘Professor Yamada put [Taro and Hanako]i’s answer sheets on each otheri’s desks.’
No Harley-style small clause account appears available for the locative examples. Harada
and Larson (2009) analyze both Japanese possessives and locatives (transitive and ditransi-
tive) as true oblique constructions whose ni-phrases have been A-scrambled to the left, from
which position they can c-command the possessum/locatum (v).
(v) a. … DP-ni … DP-o … DP-ni (datives, transitive locatives)
b. … DP-ni … DP-ga … DP-ni (intransitive locatives, possessives)
This analysis is shown to explain a wide variety of facts about both constructions. See
Harada and Larson (2007) and below for details.
22. For a useful descriptive discussion of applicative constructions, see Farrell (2005).
23. These properties hold of so-called asymmetric applicative constructions, found in many
applicative languages, in which the applied object shows direct object–like properties and
the base object does not. Some languages exhibit symmetric applicative constructions in
which the two nominals both exhibit direct object–like properties. I will not address the
latter here but see Baker (1988), Bresnan and Moshi (1993), Jeong (2007), Marantz (1993),
McGinnis (2001), and Peterson (2007) for discussion.
24. See Garrett (1990) for a critical appraisal of some elements of the approach, as well as Alsina
and Mchombo (1993) and Bresnan and Moshi (1993).
25. (95) simplifies Partee’s own rule of Passive (2.3.2, p. 66) for purposes of present discussion.
26. In (100) I simplify translations into IL to eliminate intensions and other irrelevant complexi-
ties; see Dowty (1978) for the suppressed details.
27. Again, this translation suppresses intensions and other complexities that are irrelevant for
present purposes.
28. The item X in (102), whose presence is necessary to confer specifier status on the theme a
book, is left undiscussed by Marantz (1993). See section 4 for a proposal.
29. This point is particularly striking in Marantz’s (1993) discussion of his diagram (52), which
purports to display “the ‘flow of affectedness’ among affected arguments in an event”
(p. 144). The placement of arguments in the diagram is explicitly acknowledged to derive
from the ordering of arguments observed empirically in double object constructions. Since
the double object construction is, in effect, being used to define the relevant notion of
affectedness, the latter cannot then be invoked, without circularity, to explain projection of
double object structures.
30. This is in fact a version of the small clause analysis that neo–Generative Semanticists
endorse; see section 2.2.2 below for discussion.
31. Larson (2010) shows that Pylkkänen’s uncoupling of the indirect object argument from the
event structure of the verb has the negative consequence of allowing a large family of unde-
sirable entailments. In particular, under Pylkkänen’s semantics (ia) incorrectly entails (ib):
(i) a. John wrote that letter and Bill gave Mary that letter.
b. John wrote Mary that letter.
I put aside this issue below to focus on the compositional issues.
32. Another oddness in Pylkkänen’s semantics is the duplication of the Theme relation, which
is contributed both by the verb buy and by the applicative morpheme; the redundancy is
eliminated (presumably by logical equivalence) when the interpretations verb and its sister
phrase are combined, but Pylkkänen does not comment on this point.
33. (114b) is not quite identical to Pylkkänen’s (14b) insofar as Pylkkänen (2008) represents a
letter as a nonconstituent. The reason for this is not discussed or commented on.
34. I am grateful to Hee sung Im (personal communication) for discussion of the Korean data
in (118)–(121), some of which are taken from Im (2007).
35. In effect, what we are asking for with respect to (123a) is the single, abstract, prepositional
relation that Harley-Pesetsky would assign to the monotransitive pairs in (i), stripping off
the specific meanings of open and break:
(i) a. The key opened the lock.
b. The rock broke the vase.
Similarly for (123b) and the monotransitive pair in (ii):
(ii) a. The lock opens with this key.
b. ?The vase (will) break with this rock.
36. One potential response for neo–Generative Semanticists might be to divide the class of appli-
catives, offering their account only for datives/benefactives and locatives. Note, however,
that Marantz (1993, p. 125) allows for instrumental objects to be quite low in structure,
below themes, so the division required does not appear to be a natural one from the stand-
point of applicative typology.
37. As is well known, the infelicity of cases like (128b) gets even sharper with definite reference by the
final pronoun (ia); however, if both items are of equal familiarity, the deviation is ameliorated:
(i) a. The Congressional Medal of Honor will be conferred at the ceremony.
#The President will be awarding John it.
b. I received the Congressional Medal of Honor at the ceremony yesterday
?The President gave me it.
38. Gerdts and Kiyosawa (2005b) use the phrase “semantic kick,” but their discussion plainly
indicates the latter to be pragmatic, not semantic.
39. I am grateful to Benjamin Bruening (personal communication) for this reference.
40. As cited in Larson (1990), Schneider-Zioga (1988) shows that the same scope freezing effect
shows up with spray-load pairs like (i)–(iii):
(i) a. The worker loaded one box on every truck.
b. The worker loaded one truck with every box.
(ii) a. Max sprayed some slogan on every wall.
b. Max sprayed some wall with every slogan.
(iii) a. I drained one chemical from every tank.
b. I drained one tank of every chemical.
Bruening (2001) gives examples parallel to (135) and (138) involving the spray-load alter-
nations, as well as a number of additional phenomena supporting the conclusion drawn in
Larson (1990), that is, that the scope freezing phenomenon involves the relative scope of the
two postverbal quantifiers and not the absolute scope of the outer one. Surprisingly, however,
Bruening (2001) attributes to Larson the view of scope freezing that the facts in (135) and
(138) were cited to defeat, that is, that scope freezing is a matter of the absolute confinement
of the outer quantifer. This unfortunate error has led to a certain amount of confusion about
the source of the basic data in (135) and (138) and the key theoretical conclusion that they
yield (relative vs. absolute scope); see, for example, Authier and Reed (2003).
41. Put has a dative meaning in Jagers put the question to Pip, and Dickens (1861) gives dou-
ble object examples with put in this usage: “Put yourself this question.” This usage seems
largely obsolete now.
42. Pseudo-passive might then be looked at as a subcase where the case of a locative P is absorbed
by the verb that selects PP but where the P itself, being semantically nonredundant, is not
absorbed.
43. This was in fact an obvious move given the commonplace observation in RG that chomeurs
were often realized as adjuncts.
44. Stranding analyses of “floated quantifiers” (Sportiche 1988) and the existence of examples
like (ia) strongly suggested that the derived subject moves through a preverbal, VP-internal
position on its way to the specifier of IP/TP
(i) a. The workers were all fired by Mary.
b. The workers were [all ___] fired all the workers by Mary.
A derivation as in (146c) permitted this; a “direct-to-subject” raising analysis did not.

45. Baker attributes this idea to L. Travis. Baker’s analysis was adopted and developed by Bleam
(2000) for Spanish, following influential work by Demonte (1995).
46. Baker (1996a) does not directly address the absence of the preposition to in this deriva-
tion, but adverts (in fn. 12) to other work. Presumably, a (null) preposition-incorporation
account remains available.
47. Zushi (1992) postulates an intermediate step not shown in (154b), whereby the PP first
moves into the VP specifier position before moving to the edge of VP.
48. Zushi’s basic proposal is developed further in Harada and Larson (2009).
49. In my discussion below I use the simpler presentation of this idea by Oba (2002). Nonethe-
less, the fundamental idea is clear in den Dikken (1995).
50. (155) is Oba’s minor update of Freeze (1992) to include little v. Freeze’s own work develops
ideas from Benveniste (1971) on the derived status of transitive have-type possessives.
51. This analysis is developed more fully in Larson (in preparation).
52. As noted in the General Introduction this requires us to assume that [iag[ ]] can c-command
out of PP, but this assumption seems independently plausible on binding-theoretic grounds;
cf. a letter by John to himself).
53. This “rule” is plausibly the by-product of more basic considerations. For example, if pas-
sive morphology simply participializes/deverbalizes the verb on which it occurs, and if
nonverbal elements cannot bear valued θ-features generally, then (170) will follow as a
derived result.
54. The string in (171e) has in fact two derivational sources, both of which fail. One derivation
merges John without little agentive v and crashes due both an unvalued [ag]-feature and the
lack of Case agreement for John. The second derivation merges little agentive v but still crashes
for lack of a Case probe to agree with John. See Larson (in preparation) for further details.
55. These proposals entail that both passive and unaccusative subjects pass through a vP Spec
position on their way to TP Spec. This seems correct given preverbal position as a potential
site for Q-stranding in both constructions:
(i) a. The boys (both) will (both) have (both) been (both) fired by now.
b. The boys (both) will (both) have (both) arrived by now.
56. See Ormazabal and Romero (2010) for a derivational approach somewhat similar in spirit
to that offered here.
57. This essentially updates the analysis of Benveniste (1971) and Freeze (1992).
58. Of course the direct availability of structures like (188) immediately raises the question of
why locative double objects are not freely available in English, and indeed in all languages.
I offer a tentative suggestion below, based on Case.
59. See also Bruening (2001) for development of this view.
60. See also Czepluch (1982), Hellan (1990) and Holmberg and Platzack (1995).
61. Unlike MG, however, Romanian allows clitic doubling in both PP and dative-inflected con-
structions. We should note that the PP and dative-inflected constructions represent two
separate dialects of Romanian and not an alternation with a single one. I am grateful to
Susana Huidobro (personal communication) for insightful discussion of these data.
62. See also Anagnostopoulou (2003) for further discussion.
63. In English, the Case feature associated with P is not obl but rather acc.
64. Applicatives are known to exhibit implicational relationships typologically: if a language
has locative applicatives, it will almost certainly have recipient-goal/benefactive applica-
tives, but not vice versa. It’s tempting to view this distribution in terms of the thematic hier-
archy. Suppose that if little v can bear {uθ1val, iacc}, then little v can bear {uθ2val, iacc}, for
any θ2 higher on the hierarchy than θ1. Then any language having a locative voice head that
licenses accusative should also allow a recipient-goal voice head that licenses accusative,
assuming θRGOAL > θLOC; the converse will not hold, however.
65. Similarly for the more elaborate versions of these cases like (215); these also can be inter-
rupted by adverbs:
(i) Mary-ka (seykey) John-ul (seykey) pal-ul (seykey) kkup-ul (seykey) cha-ess-ta.
Mary-nom hard John-acc hard foot-acc hard end-acc hard kick-pst-decl
‘Mary kicked the end of John’s foot hard.’
(cf. (31a) in Cho 2000, p. 96)
66. See Larson (in preparation) for further discussion.
67. This proposal can be seen as technically spelling out the idea in Kayne (1984) that preposi-
tions (or preposition-like elements) can, in certain, instances “transmit” accusative Case
from a verbal source. A similar analysis also suggests itself as an account of familiar pseudo-
passives like The bed was slept in: P might be analyzed as concordial in this instance, so that
suppressing the accusative feature on little v would have the effect of suppressing accusative
Case on in’s object. See Larson (in preparation) for details.
68. An alternative possibility for Case on the lower/outer object (un libro) is that it arises from
the voice head, as suggested earlier for English (cf. 209). This proposal is not straightfor-
ward, however, given that the clitic also co-occurs in integral relations structures discussed
earlier, where the outer object is governed by a preposition and no issue of le checking Case
arises. Recall (40a), repeated below as (i)):
(i) María le puso las patas a la mesa.
María Cl-3Dat put the legs to the table
‘Mary attached the table’s legs.1’
69. So far as I can tell Anagnastopolou (2003) simply doesn’t address this issue, a surprising
silence given her focus on locality.
REFERENCES
Aissen, J. (1983) “Indirect Object Advancement in Tzotzil,” in D. Perlmutter and P. Postal, eds.,
Studies in Relational Grammar 1 (pp. 272–302). Chicago: University of Chicago Press.
Akamajian, A. and F. Heny (1976) An Introduction to the Principles of Transformational Syntax.
Cambridge, MA: MIT Press.
Allen, B. and D. Frantz (1983) “Advancements and Verb Agreement in Southern Tiwa,” in D. Perl-
mutter and P. Postal, eds., Studies in Relational Grammar 1 (pp. 303–314). Chicago: University.
of Chicago Press.
Alsina, A. and S. Mchombo (1993) “Object Asymmetries and the Chichewa Applicative Construc-
tion,” in S. Mchombo, ed., Theoretical Aspects of Bantu Grammar (pp. 17–45). Stanford, CA:
CSLI Publications.
Anagnostopoulou, E. (2003) The Syntax of Ditransitives. Berlin: Mouton de Gruyter.
Arregi, E. (2003) “Dative Alternations in Basque,” unpublished ms., University of the Basque
Country, Vittoria, Spain.
Authier, J.-M. and L. Reed (2003) “Quantifier Scope and the Structure of Faire-par,” in A. T.
Perez-Leroux and Y. Roberge, eds., Romance Linguistics: Theory and Acquisition (pp. 19–32).
Amsterdam: John Benjamins.
Baker, C. L. (1977) Introduction to Generative-Transformational Syntax. New York: Prentice
Hall.
Baker, M. (1988) Incorporation. Chicago: University of Chicago Press.
——— (1996a) “On the Structural Position of Themes and Goals,” in J. Rooryck and L. Zaring,
eds., Phrase Structure and the Lexicon (pp. 7–34). Dordrecht: Kluwer.
——— (1996b) The Polysynthesis Parameter. Chicago: University of Chicago Press.
Baker, M., K. Johnson, and I. Roberts (1989) “Passive Arguments Raised,” Linguistic Inquiry 20:
219–251.
Barss, A. and H. Lasnik (1986) “A Note on Anaphora and Double Objects,” Linguistic Inquiry
17: 347–354.
Beck, S. and K. Johnson (2003) “Double Objects Again,” Linguistic Inquiry 34: 97–123.
Belletti, A. and U. Shlonsky (1995) “The Order of Verbal Complements: A Comparative Study,”
Natural Language and Linguistic Theory 13: 489–526.
Benveniste, E. (1971) Problems in General Linguistics. Coral Gables, FL: University of Miami
Press.
Blake, B. (2001) Case. Cambridge: Cambridge University Press.
Bleam, T. (2000) Leísta Spanish and the Syntax of Clitic Doubling. Doctoral dissertation, Univer-
sity of Delaware, Newark.
Bleam, T. (2003) “Properties of the Double Object Construction in Spanish,” in R. Núñez-
Cedeño, L. López, and R. Cameron, eds., A Romance Perspective on Language Knowledge
and Use. Selected Papers from the 31st Linguistic Symposium on Romance Languages (LSRL)
(pp. 233–252). Amsterdam: John Benjamins.
Bowers, J. (1973) Grammatical Relations. Doctoral dissertaion, MIT, Cambridge, MA.
Bresnan, J. and L. Moshi (1993) “Object Asymmetries in Comparative Bantu Syntax,” in S. Mchombo,
ed., Theoretical Aspects of Bantu Grammar (pp. 47–91). Stanford, CA: CSLI Publications.
Bruening, B. (2001) “QR Obeys Superiority: Frozen Scope and ACD,” Linguistic Inquiry 32:
233–273.
Brugè, L. and G. Brugger (1996) “On the Accusative á in Spanish,” Probus 8: 1–51.
Burt, M. (1971) From Deep to Surface Structure: An Introduction to Transformational Syntax.
New York: Harper & Row.
Cho, S. (2000) Three Forms of Case Agreement in Korean. Doctoral dissertation, Stony Brook
University, Stony Brook, NY.
Chomsky, N. (1955/1975) The Logical Structure of Linguistic Theory. New York: Plenum.
——— (1957) Syntactic Structures. The Hague: Mouton.
——— (1970) “Remarks on Nominalization,” in: R. Jacobs and P. Rosenbaum, eds., Readings in
English Transformational Grammar (pp. 184–221) Waltham, MA: Ginn.
——— (1981) Lectures on Goverement and Binding Theory. Dordrecht: Foris.
Chomsky, Noam (1986) Knowledge of Language. New York: Praeger.
——— (1993) “A Minimalist Program for Linguistic Theory,” in K. Hale and S.J. Keyser, eds., The
View from Building 20 (pp. 1-52). Cambridge, MA: MIT Press.
——— (1994) “Bare Phrase Structure,” MIT Occasional Papers in Linguistics. Cambridge, MA:
——— (1995) The Minimalist Program. Cambridge, MA: MIT Press.
——— (2000) “Minimalist Inquiries,” in R. Martin, D. Michaels, and J. Uriagereka, eds., Step
by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik (pp. 89–155). Cambridge,
MA: MIT Press.
——— (2001) “Derivation by Phase,” In M. Kenstowicz, ed., Ken Hale: A Life in Language (pp. 1–54).
Chung, S. (1976) “An Object-Creating Rule in Bahasa Indonesian,” Linguistic Inquiry 7: 41–87.
Cuervo, M. (2003) Datives at Large. Doctoral dissertation, MIT, Cambridge, MA.
Culicover, P. (1976) Syntax. New York: Academic Press.
Culicover, P. and W. Wilkins (1984) Locality in Linguistic Theory. New York: Academic Press.
Czepluch, H (1982) “Case Theory and the Dative Construction,” Linguistic Review 2: 1–38.
Demonte, V. (1995) “Dative Alternation in Spanish,” Probus 7: 5–30.
den Dikken, M. (1995) Particles. Oxford: Oxford University Press.
De Pedro Munilla, M. (2004) “Dative Doubling Structures in Spanish: Are They Double Object Con-
structions?” in V. Chand et al, eds., WCCFL 23: Proceedings of the 23rd West Coast Conference
on Formal Linguistics (pp. 168–181). Somerville, MA: Cascadilla Press.
Dickens, Charles (1861) Great Expectations. London: Chapman and Hall.
Donohue, M. (2001) “Coding Choices in Argument Structure: Austronesian Applicatives in Texts,”
Studies in Language 25: 217–254.
Dowty, D. (1978) “Governed Transformations as Lexical Rules in a Montague Grammar,” Linguistic
Inquiry 9: 393–426.
Dryer, M. (1983) “Indirect Objects in Kinyarawanda Revisited,” in D. Perlmutter and P. Postal,
eds., Studies in Relational Grammar 1 (pp. 129–140). Chicago: Unive. of Chicago Press.
Emonds, J. (1972) “Evidence That Indirect Object Movement Is a Structure-Preserving Rule,”
Foundations of Language 8: 546–561.
Emonds, J. (1993) “Projecting Indirect Objects,” Linguistic Review 10: 211–263.
Farrell, P. (2005) Grammmatical Relations. New York: Oxford University Press.
Fillmore, C. (1965) Indirect Object Constructions in English and the Ordering of Transformations.
The Hague: Mouton.
Fischer, S. (1972) The Acquisition of Verb-Particle and Dative Constructions. Doctoral dissertation,
MIT Cambridge, MA.
Freeze, R. (1992) “Existentials and Other Locatives,” Language 68: 553–595.
Garrett, A. (1990) “Applicatives and Preposition Incorporation,” In K. Dziwirek, P. Farrell, and
E. Mejías-Bikandi, eds., Grammatical Relations: A Cross-Theoretical Perspective (pp. 183–198).
Stanford, CA: Center for the Study of Language and Information.
Gerdts, D. (1988) Object and Absolutive in Halkomelem Salish. New York: Garland.
Gerdts, D. B. and K. Kiyosawa (2005a) “Discourse Functions of Salish Applicatives,” in Papers
for the 40th International Conference on Salish and Neighbouring Languages, J. C. Brown,
M. Kiyota and T. Peterson, eds., UBCWPL 16 (pp. 98–124). Vancouver, BC: Department of
Linguistics, University of British Columbia.
Gerdts, D. B. and K. Kiyosawa (2005b) “The Function of Salish Applicatives,” Solveiga Armoskaite
and James J. Thompson, eds., Proceedings of the 10th Workshop on the Structure and Constituency
of Languages of the Americas, UBCWPL 17 (pp. 84–94). Vancouver, BC: Department of Linguis-
tics, University of British Columbia.
Gibson, J. (1992) Clause Union in Chamorro and in Universal Grammar. New York: Garland.
Goldberg, A. (1995) Constructions: A Construction Grammar Approach to Argument Structure.
Chicago: University of Chicago Press.
Green, G. (1974) Semantics and Syntactic Regularity. Bloomington: Indiana University Press.
Haegeman, L. (1985) The Get-Passive and Burzio’s Generalization,” Lingua 66: 53–77.
Harada, N. and R. K. Larson (2007) “Datives in Japanese,” presentation at Mayfest 2007, May
11–12, University of Maryland, College Park, MD.
——— (2009) “Datives in Japanese,” in R. Shibagaki and R. Vermeulen, eds., Proceedings of the
5th Workshop on Altaic Formal Linguistics (WAFL5). MIT Working Papers in Linguistics 54
(pp. 3–17). Cambridge, MA: MIT Working Papers in Linguistics.
Harley, H. (1995) Subjects, Events and Licensing. Doctoral dissertation, MIT, Cambridge, MA.
——— (2002) “Possession and the Double-Object Construction,” in P. Pica nd J. Rooryck, eds.,
Linguistic Variation Yearbook 2 (pp. 31–70) Amsterdam: John Benjamins.
Hellan, L. (1990) “The Phrasal Nature of Double Object Clusters,” in W. Abraham, W.
Kostmeijer, and E. Reuland, eds., Issues in Germanic Syntax (pp. 67–92). Berlin: Mouton
de Gruyter.
Holmberg, A. and C. Platzack (1995) The Role of Inflection in Scandinavian Syntax. Oxford:
Oxford University Press.
Im, Hees sung (2007) “Korean Possessive Structures.” Unpublished Honors Thesis, Stonybrook
University, Stony Brook, New York.
Jackendoff, R. (1990) “On Larson’s Treatment of the Double Object Construction,” Linguistic
Inquiry 21: 427–456.
Jackendoff, R. and P. Culicover (1971) “A Reconsideration of Dative Movement,” Foundation of
Language 7: 397–412.
Jacobs, R. and P. Rosenbaum (1968) English Transformational Grammar. Waltham, MA: Blaisdell.
Jeong, Y. (2007) Applicatives: Structure and Interpretation from a Minimalist Perspective. Amsterdam:
John Benjamins.
Kaiser, E. and K. Nakanishi (2001) “Syntax-Pragmatics Interactions: Scrambling in Japanese
and Finnish Ditransitive Constructions,” unpublished manuscript, University of Pennsylvania
Philadelphia.
Kayne, R. (1975) French Syntax. Cambridge, MA: MIT Press.
——— (1984) Connectedness and Binary Branching. Dordrecht: Foris.
Kester, E.-P. 1996. The Nature of Adjectival Inflection. Utrecht: Onderzoeksinstituut voor Taal
en Spraak.
Keyser, S. J. and P. Postal (1976) Beginning English Grammar. New York: Harper & Row.
Kimenyi, Alexandre (1980) Relational Grammar of Kinyarwanda. Berkeley: University of California
Press.
Ko, H. (2005) Syntactic Edges and Linearization. Doctoral dissertation, MIT, Cambridge MA.
Kratzer, A. (1996) “Severing the External Argument from Its Verb,” in J. Rooryck and L. Zaring,
eds.,: Phrase Structure and the Lexicon (pp. 109–137). Dordrecht: Kluwer.
Krifka, M. (1999) “Manner in Dative Alternation,” WCCFL 18: Proceedings of the 18th west
coast conference on formal linguistics. Sonya Bird, Andrew Carnie, Jason D. Haugen, and Peter
Norquest, eds. (pp. 260-271). Somerville, MA: Cascadilla Press.
Larson, R. K. (1988) “On the Double Object Construction,” Linguistic Inquiry 19: 335–391.
——— (1991) “Promise and the Theory of Control,” Linguistic Inquiry 22: 103–139.
——— (2010) “On Pylkkänen’s Semantics for Low Applicatives,” Linguistic Inquiry 41: 701–704.
——— (in preparation) DP and VP.
Larson, R. K. and R. May (1990) “Antecedent Containment or Vacuous Movement: Reply to
Baltin.” Linguistic Inquiry 21: 103–122.
MacDonald, J. (2009) The Syntactic Nature of Inner Aspect: A Minimalist Perspective. Amsterdam:
John Benjamins.
Marantz, A. (1984) On the Nature of Grammatical Relations. Cambridge, MA: MIT Press.
——— (1993) “Implications of Asymmetries in Double Object Constructions,” in S. Mchombo, ed.,
Theoretical Aspects of Bantu Grammar (pp. 113–150) CSLI: Stanford.
Marantz, A. (1997) “No Escape from Syntax: Don’t Try Morphological Analysis in the Privacy of
Your Own Lexicon,” in A. Dimitriadis, ed., Proceedings of the 27th annual Penn Linguistics
Colloquium. Philadelphia: Pennsylvania Working Papers in Linguistics.
May, R. (1985) Logical Form: Its Structure and Derivation. Cambridge, MA: MIT Press.
McGinnis, M. (2001) “Variation in the Phase Structure of Applicatives,” Linguistic Variation
Yearbook 1: 105–146.
Miyagawa, S. and T. Tsujioka (2004) “Argument Structure and Ditransitive Verbs in Japanese,”
Journal of East Asian Linguistics 13: 1–38.
Montague, R. (1974) Formal Philosophy. New Haven, CT: Yale University Press.
Oba, Y. (2002) “The Double Object Construction and the Extraction of the Indirect Object,”
Linguistics Analysis 32: 40–71.
Oehrle, R. T. (1976) The Grammatical Status of the English Dative Alternation. Doctoral disserta-
tion, MIT, Cambridge MA.
Ordóñez, F. (1998) “Post-Verbal Asymmetries in Spanish,” Natural Language and Linguistic
Theory 16: 313–346.
Ormazabal, J. and J. Romero (2010) “The Derivation of Dative alternations,” in M. Duguine,
S. Huidobro, and N. Madariaga, eds., Argument Structure and Syntactic Relations from a
Cross-Linguistic Perspective (pp. 203–232) Amsterdam: John Benjamins.
Parsons, T. (1991) Events in the Semantics of English. Cambridge, MA: MIT Press.
Partee, B. (1973) “Some Transformational Extensions of Montague Grammar,” Journal of Philo-
sophical Logic 2: 509–534.
Perlmutter, D. and P. Postal (1983a) “The Relational Succession Law,” in D. Perlmutter and P. Postal,
eds., Studies in Relational Grammar 1 (pp. 30–80). Chicago: University of Chicago Press.
Perlmutter, D. and P. Postal (1983b) Studies in Relational Grammar 1. Chicago: University. of
Chicago Press.
Perlmutter, D. and C. Rosen (1984) Studies in Relational Grammar 2. Chicago: University. of
Chicago Press.
Pesetsky, D. (1995) Zero Syntax: Experiencers and Cascades. Cambridge, MA: MIT Press.
Peterson, D. (2007) Applicative Constructions. Oxford: Oxford University Press.
Pietroski, P. (2005) Events and Semantic Architecture. Oxford: Oxford University Press.
Pinker, S. (1989) Learnability and Cognition: The Acquisition of Argument Structure. Cambridge,
MA: MIT Press.
Pylkkänen, L. (2002) Introducing Arguments. Doctoral dissertation, MIT, Cambridge, MA.
——— (2008) Introducing Arguments. Cambridge, MA: MIT Press.
Radford, A. (1997) Syntax: A Minimalist Introduction. New York: Cambridge University Press.
Rappaport Hovav, M. and B. Levin (2008) “The English Dative Alternation: The Case for Verb
Sensitivity,” Journal of Linguistics 44: 129–167.
Richards, N. (2001) “An Idiomatic Argument for Lexical Decomposition,” Linguistic Inquiry 32:
183–192.
Rosenbaum, P. (1967) The Grammar of English Predicate Complement Constructions. Cambridge,
MA: MIT Press.
Ross, J. (1967) Constraints on Variables in Syntax. Doctoral dissertation, MIT, Cambridge, MA.
Sag, I. (1976) Deletion and Logical Form. Doctoral dissertation, MIT, Cambridge, MA.
Schein, B. (1993) Plurals and Events. Cambridge, MA: MIT Press.
——— (forthcoming) Conjunction Reduction Redux. Cambridge, MA: MIT Press.
Schneider-Zioga, P. (1988) “Double Objects and Small Clauses,” unpublished manuscript, Univer-
sity of Southern California, Los Angeles, CA.
Shimizu, M. (1975) “Relational Grammar and Promotion Rules in Japanese,” in R. Grossman,
L. J. San, and T. Vance, eds., Proceedings of the Eleventh Regional Meeting of the Chicago
Linguistic Society (CLS 11). (pp. 529–535) Chicago: University of Chicago.
Snyder, W. and K. Stromswold (1997) “The Structure and Acquisition of English Dative Construc-
tions,” Linguistic Inquiry 28: 281–317.
Sportiche, D. (1988) “A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure,”
Linguistic Inquiry 19: 425–451.
Ward, G. and B. Birner (2004) “Information Structure and Non-Canonical Syntax,” in L. Horn
and G. Ward, eds., The Handbook of Pragmatics (pp. 153–174). Oxford: Blackwell.
Zushi, M. (1992) “The Syntax of Dative Constructions in Japanese,” unpublished manuscript,
McGill University, Montreal.
Zwicky, A. and J. Sadock (1975) “Ambiguity Tests and How to Fail Them,” in J. Kimball, ed.,
Syntax and Semantics 4. (pp. 1–36). New York: Academic Press.
1 On the Double Object Construction
Richard K. Larson
Barss and Lasnik (1986) discuss certain asymmetries in double object constructions,
such as (1a,b):
(1) a. John sent Mary a letter.
b. I promised Felix a new set of golf clubs.
They observe facts suggesting that the second NP (a letter, new set of golf clubs) is in
each case in the domain of the first, but not vice versa. These results conflict with stan-
dard views about the syntax of double object sentences and appear to raise problems for
the view that “α is in the domain of β” should be explicated in terms of the structural
notion of c-command.
In this article I present an analysis of the double object construction that implements
a proposal about dative structure first suggested by Chomsky (1955/1975). According
to this view, a simple dative like John sent a letter to Mary derives from an underlying
form in which the verb and its indirect object make up a constituent that excludes the
direct object. The specific proposal adopted here is that dative complement construc-
tions like John sent a letter to Mary involve an underlying clauselike VP whose “sub-
ject” is a letter and whose “object” is (to) Mary (2a); this inner constituent is obscured
at S-Structure by an operation of V-Raising (2b):
(2) a. John [VP a letter [V′ send to Mary]]
b. John send [ VP a letter [V′ t to Mary]]
With this view of dative complementation, double objects can be syntactically derived
by a modern form of Dative Shift. In particular, they can be produced by applying the
familiar operations responsible for passive sentences within VP. The former indirect
object (Mary) becomes a derived VP “subject,” and the former direct object (a letter)
assumes adjunct status within V′. As I show, the resulting structure accounts for Barss
and Lasnik’s facts straightforwardly in terms of c-command and provides insight into
various other properties of the double object structure as well.
After briefly reviewing Barss and Lasnik’s observations in section 1, I introduce the
account of dative complementation adopted here in section 2. In section 3 I present a
derivational account of double objects that identifies Dative Shift as Passive, and I show
that apparent surface differences between the two operations (morphological marking,
Case assignment, and so on) are independently explainable. In section 4 I argue for
the connection between Passive and Dative Shift with data from indirect passives and
psych-verb constructions, and in sections 5 and 6 I examine English-internal and cross-
linguistic constraints on Dative Shift. Finally, I conclude, in section 7, with a discussion
of the VP complementation structures that play a central role in this account.
1. ASYMMETRIES OF SYNTACTIC DOMAIN
Barss and Lasnik (1986) point out a number of important asymmetries in the behavior
of the two objects in double object constructions. All involve phenomena in which
constituent structure relations-specifically, c-command-have been assumed to play a
central role. Thus, reflexives and reciprocals (anaphors) must be c-commanded by their
antecedents. Double object structures show an asymmetry with respect to the licensing
of anaphors:1
(3) a. I showed Mary herself.
*I showed herself Mary.
A quantifier must c-command a pronoun at S-Structure if it is to bind it. Double objects

show asymmetries regarding quantifier-pronoun binding possibilities:
(3) b. I gave every workeri hisi paycheck.
*I gave itsi owner every paychecki.
A wh-phrase c-commanded at D-Structure by an NP containing a pronoun cannot be

moved over that NP if wh- and the pronoun are coreferential. This is the so-called weak
crossover effect. Double objects show weak crossover asymmetries:
(3) c. Which mani did you send hisi paycheck?
*Whosei pay did you send hisi mother?
A wh-phrase cannot in general be moved over another wh-phrase that c-commands it

(in other words, is “superior” to it) in underlying representation. Double objects show
an asymmetry in superiority effects:
(3) d. Who did you give which paycheck?
*Which paycheck did you give who?
Constructions of the form each . . . the other, as in Each man saw the other or Each man
saw the other’s friend, may have a reciprocal reading when and only when the each-
phrase c-commands the other-phrase. Double objects show asymmetries with respect to
the each . . . the other construction on its reciprocal reading:
(3) e. I showed each man the other’s socks.
*I showed the other’s friend each man.

On the Double Object Construction 139
Finally, negative polarity items must occur in the c-command domain of an “affec-
tive element” such as negation or a negative quantifier. Double objects show asym-
metries with respect to a negative polarity item such as any and a licensing affective
element:
(3) f. I showed no one anything.
*I showed anyone nothing.
If it is assumed that these phenomena do indeed involve c-command, then (3a–f)

all point to the same conclusion: in constructions involving a verb phrase of the form
V-NP-NP, the first NP c-commands the second, but not vice versa. As Barss and Lasnik
observe, this immediately casts doubt on the two most frequently assumed structures
for double objects:
(4) a. VP b. VP
V NP1 NP2 V′ NP2
V NP1
(4a) is the structure for double objects proposed by Oehrle (1976); (4b) is the one
proposed by Chomsky (1981). Under a definition of c-command based on first branch-
ing nodes (Reinhart 1979), NP1 and NP2 mutually c-command each other in (4a);
hence, this structure predicts no asymmetries in relations based solely on hierarchical
structure. In (4b) NP2 asymmetrically c-commands NP1, predicting that the latter is
in the domain of the former but not conversely. Both sets of predictions are strongly
contradicted by the facts in (3). Under a definition of c-command based on contain-
ment in maximal projections (Aoun and Sportiche 1983), NP1 and NP2 will mutually
c-command each other in both (4a) and (4b), predicting no asymmetries of syntactic
domain. Again, this prediction is falsified by the data in (3).2 Evidently one of two con-
clusions is possible: (a) the syntactic data noted above are not in fact to be explicated
by c-command alone; some other notions (such as linear precedence) must be invoked;
or (b) these facts are indeed structural and some configuration other than (4a) or (4b)
is involved.
The situation with double objects contrasts with that of standard oblique dative
structures. The asymmetries observed with V-NP-NP structures occur with V-NP-PP
structures as well:
(5) a. I presented/showed Mary to herself. (anaphor binding)
*I presented/showed herself to Mary.
b. I gave/sent every checki to itsi owner. (quantifier binding)
??I gave/sent hisi paycheck to every workeri.

c. Which checki did you send to itsi owner? (weak crossover)
*Which workeri did you send hisi check to?
d. Which check did you send to who? (superiority)
*Whom did you send which check to?
(*To whom did you send which check?)
e. I sent each boy to the other’s parents. (each . . . the other)
*I sent the other’s check to each boy.
f. I sent no presents to any of the children. (negative polarity items)
*I sent any of the packages to none of the children.
In the case of oblique datives, however, these results do not appear to raise any spe-
cial problems for c-command. The facts are accommodated smoothly, it seems, by
appealing to the structure introduced by PP. Suppose the VPs in (5) are as in (6a) or
(6b):
(6) a. VP b. VP
V NP1 PP V′ PP
P NP2 V NP1 P NP2
2
Kayne (1983a) and Czepluch (1982) argue that double objects involve empty PP structure:
(i) a. in VP
Then b.
(6a) NP1 asymmetrically VP c-commands NP2 under the definition of
c-command proposed by Reinhart (1979): NP2 is dominated by a branching node
(PP) notVdominating NP1. Similarly,
V PPin (6b)
NP2 NP1 asymmetrically c-commands NP2
under the definition of c-command given by Aoun and Sportiche (1983): NP2 is
contained PP in a maximal
NP2 projection
P (PP)
NP1 that fails to contain NP1. This illustrates
quite clearly why double objects present such a puzzle for syntactic analysis: if
complement asymmetry in standard datives is simply a matter of the structure
P NP1 e
introduced by PP, then why, in double object constructions, where such structure
is absent, do we not find symmetric behavior? This is what we expect, but it is not
what wee see.
(kayne) (Czepluch)
2. THE STRUCTURE OF DATIVES REVISITED
Chomsky (1955/1975) proposes an interesting alternative analysis of dative structures,

one that attributes asymmetry among complements to a source other than PP structure.
According to Chomsky, a sentence like (7a) is actually derived from a structure of the
form (7b) (by extraposition of the PP to him):3
(7) a. The teacher gave several books to him.
b. The teacher [gave to him] several books.
In (7b) the indirect object is in fact an “inner object” forming a constituent with the verb
that excludes the surface direct object. Here, as in (6a,b), there is an underlying asym-
metry between dative verb complements. The indirect object (NP2) is in the structural
domain of the direct object (NP1), but not conversely:
(8) VP
X NP1
gave to NP2
However, in (8) this asymmetry is not a matter of PP structure. Rather, it is introduced

by the branching node labeled X, presumably some projection of V.
The structure in (8) departs quite sharply from those in (6) under plausible assump-
tions about the relation between thematic and hierarchical structure. In (6a) the two
complements are structurally on a par: both are sister to V. We may take this to cor-
respond to the view that theme and goal are both assigned by V (the latter perhaps
with some contribution by the preposition). In (6b) the direct object alone is sister to V,
whereas the indirect object phrase is sister to the small verbal constituent V′. This we
can take to correspond to the view that the direct object alone receives a θ-role directly
from V, whereas the indirect object receives its role “compositionally” from V′. Struc-
ture (8) asserts what is in effect the inverse of (6b): here it is the indirect object that is
the direct argument of the verb, the object NP1 receiving a θ-role from the “phrasal
verb” give to him.
The position taken in Chomsky (1955/1975) can be supported, I believe, by argu-
ments parallel to those given in Marantz (1984) for the claim that it is VP that assigns
a θ-role to the matrix subject, and not simply V. Marantz observes that the predicate
expressed by a transitive verb + object regularly depends on the contribution of the
object, as shown by VPs like throw a baseball, throw support behind a candidate, throw
a boxing match. A similar phenomenon can be observed with datives. For example,
consider the pair (9a,b):
(9) a. Beethoven gave the Fifth Symphony to the world.
b. Beethoven gave the Fifth Symphony to his patron.
Giving an object to the world (to posterity, mankind, etc.) has a rather different
character from giving an object to an individual. In the first case we understand the
given object to be the Fifth Symphony qua composition; the transfer of possession is
metaphorical, so that (9a) is roughly synonymous with ‘Beethoven created the Fifth
Symphony’. In the second case we understand a physical object to be transferred-
perhaps a sheaf of papers on which the composition is transcribed. The exact semantic
role assigned to the direct object thus depends on the nature of the recipient appearing
in the goal phrase.
The idea that a verb and its outer complements can form a single thematic complex is
also supported by the existence of “discontinuous idioms” of the following kind (noted
in Emonds (1972)):

to the cleaners
b. Mary took Felix to task .
into consideration
Evidently, in (10a) the dative verb send assigns a thematic role to the object his starting
pitcher in concert with the complement phrase to the showers; similarly for (10b–d).
The possibility of such idioms is straightforward under the structure in (8), where the
indicated elements form an underlying constituent. It is quite unexpected under the
structures in (6), however, where V and the outer complement form no thematic
complex.
The argument from idiom data appears at first to be compromised by examples like
(11a–d) (pointed out to me by D. Pesetsky), which seem to involve verb + object idioms
(give x’s all, give hell, give the boot, give the creeps, show x’s cards) that assign a com-
positional role to the indirect object:
(11) a. Max gave his all to linguistics.
b. Alice gives hell to anyone who uses her training wheels.
c. Oscar will give the boot to any employee that shows up late.
d. The Count gives the creeps to anyone he’s around long enough.
e. Phyllis should show her cards to other group participants.
However, on closer inspection it is not clear that such examples really raise a prob-
lem. Note that the standard entailment X-give-Y-to-Z ⇒ Z-get-Y is preserved with the
examples (11a–d), and note further that under this entailment the original idiomatic
force is preserved:
(12) a. Linguistics gets [my all].
b. I caught/got [hell] from Alice.
c. Peter got [the boot].
d. Geez, you get [the creeps] just looking at him.

Similarly for (11e); alongside this example we find (12e), where again the idiomatic
force is preserved:
(12) e. Unwittingly, Alice tipped [her cards].
These results are unexpected on the view that verb + object is an idiomatic complex;
for example, if V were being understood idiomatically in (11a–d), there would surely
be no expectation that the entailment X-give-Y-to-Z ⇒ Z-get-Y would hold, as it
clearly does. What these facts suggest, then, is that contrary to initial impressions,
the idiomaticity in (11a–e) lies not in the verb + object combination but rather in the
object alone. That is, (11) and (12) suggest that one’s all, hell, the boot, and so on
are being treated by the grammar as rather strange sorts of objects that, because they
can be given, can be gotten as well. On this view, give and show do not in fact form
idiom complexes in (11) or (12); rather, they simply interact compositionally with a
semantically opaque NP.4
2.1. V-Raising
In analyzing the structure of double objects, I will adopt a version of Chomsky’s
(1955/1975) proposal, one deriving from work by Bach (1979), Dowty (1979), and
Jacobson (1983, 1987). The basic assumption is that the VP in a dative is as illustrated
in (13):5
(13) a. John sent a letter to Mary.
b. VP
SpecV′ V′
V VP
e NP V′
a letter V PP
send to Mary
According to (13b), the verb phrase underlying send a letter to Mary is a strictly binary
branching structure. The VP consists of an empty V taking a VP complement whose
specifier is a letter, whose head is send, and whose sole complement is the PP to Mary.
This structure may be understood intuitively as follows: send takes the complement to
Mary, forming a small predicate send-to-Mary as in Chomsky (1955/1975). The latter
is predicated of an “inner subject” a letter, forming a VP with clauselike structure: a
letter send to Mary. This VP is then in turn predicated of a subject like John to yield the
full sentence (13a).
Of course, John a letter send to Mary is not a well-formed sentence of English: the
verb must appear to the left of a letter. The central assumption here is that the cor-
rect surface form arises by movement of the verb send to the empty V position-that is,
head-to-head movement along lines discussed by Baker (1985) and Chomsky (1986a).
This movement leaves a trace in the original site and creates a sequence of coindexed V
positions:
(14) VP
SpecV′ V′
Vi VP
send NP V′
a letter Vi PP
t to Mary
V-Raising may be taken to follow from certain Case and agreement requirements holding
of Infl, V, and NP.6 Suppose, following the general proposals of Roberts (1985), that V
must ultimately head a projection governed by Infl in order to receive tense and agree-
ment information.7 Furthermore, suppose (following Stowell (1981), Travis (1985), and
Koopman (1986)) that Case is assigned under government, where the direction of gov-
ernment is rightward in English. In (13b) V is not the head of a projection governed by I.
Moreover, the NP a letter in the lower SpecV′ is not governed by the verb and so cannot
receive Case.8 V may be seen as raising in (14) to meet these joint requirements. In the
resulting configuration the VP headed by send is governed by Infl. Furthermore, V may
be plausibly analyzed as governing a letter: V is to the left of NP, and NP is the specifier
of a maximal projection sister to it; hence, send can assign Objective Case to a letter in
(14), as required.9
The situation posited here for VP in English is analogous to the situation widely
assumed for S in VSO languages.10 Under many proposals, the surface order of matrix
constituents in languages like Welsh, Irish, and Berber is derived by V-Raising; this
permits the subject NP to receive Case from V and permits the verb to obtain tense and
agreement, which, in VSO languages, appear to be located in Comp (Sproat 1985); rais-
ing also obscures the presence of an underlying VP (15a):
(15) a. b.
V S V VP
NP VP NP V′
V NP V NP
t t
In a similar way, the VP-internal raising assumed here allows Case, tense, and agree-
ment information to be assigned properly, while obscuring an underlying V′ (15b).
The domain of application is evidently different, but the motivation and effects are
the same.
2.2. Consequences
The raising analysis preserves the crucial feature of Chomsky’s (1955/1975) account
noted earlier, namely, that the direct object will c-command the oblique object quite
independently of the structure introduced by PP:
(16) VP
SpecV′ V′
V VP
a. show NP V′
b. gave
c. told
Max V PP
every dog
nothing
t to himself
to its owner
to anyone
In (16a) Max c-commands himself, but not conversely, hence the binding asymmetries
in (5a). Again, in (16b) the quantified object c-commands the indirect object, but not
conversely, hence the quantifier-pronoun asymmetries in (5b). Finally, in (16c) the nega-
tive polarity item is in the scope of the affective element nothing, but not vice versa,
hence the facts in (5f). The remaining cases, (5c–e), are analogous.
This analysis also provides a direct account of certain familiar but rather puzzling
facts about conjunction in datives. Note the acceptability of examples like (17a,b) where
a conjunction appears between the two sets of dative complements:
(17) a. John sent a letter to Mary and a book to Sue.
b. I gave five dollars to Maxwell and three dollars to Chris.
Given the usual assumption that conjunction unites constituents, such examples are
problematic for the structures in (6a,b) (see Sag et al. (1985) for discussion): a letter to
Mary and a book to Sue are not constituents. However, under the analysis suggested
here, datives like (17a) are understood straightforwardly:
(18) VP
SpecV′ V′
V VP
send VP and VP
NP V′ NP V′
a letter V PP a book V PP
e to Mary e to Sue
What we have in (17a,b) is a form of across-the-board movement—in particular, across-

the-board V-Raising. Here conjunction does indeed join constituents.11
The status of direct objects as “VP subjects” in this account may appear to raise
problems with respect to the binding theory. Under the usual definitions, anaphors
like himself must be bound in their minimal governing category, where the latter is
defined as the smallest maximal projection containing a subject and a lexical governor
(Chomsky 1981). This would seem to exclude a sentence like (19), where herself takes
as its antecedent, not the close VP subject (the task), but instead the distant IP subject
(Mary):
(19) Mary entrusted [VP the task [t to herself]]
In fact, such examples can be accommodated without jeopardizing the VP subject sta-
tus of direct objects. Chomsky (1986b) suggests an alternative definition of govern-
ing category—one involving not the notion “subject” but rather the notion “complete
functional complex” (CFC). Under this view, an anaphor α must be bound in its mini-
mal CFC—in the minimal domain containing α in which “all grammatical relations
compatible with its head are realized” (p. 169). In general, the two notions of governing
category—domain-of-a-subject and CFC—define identical domains; however, precisely
in the case at hand they diverge. Although herself is not bound in the domain of its
closest subject in (19), it is bound in the minimal CFC containing it, namely, in IP. The
latter is the domain in which all grammatical relations compatible with give are realized.
Hence, under the suggested reformulation of governing category in terms of CFC, the
anaphor does satisfy the binding theory. In view of this I will henceforth simply assume
the definition of governing category in Chomsky (1986b).12,13
2.3. V′ Reanalysis and “Heavy NP Shift”

The account of datives proposed above allows a novel approach to certain “movement
phenomena,” which I introduce here for later use. Consider sentences of the following kind:
(20) a. I gave to John everything that he demanded.
b. Max sent to me the longest letter anyone had ever seen.
Such examples have standardly been analyzed as deriving from more basic
dative configurations by a rule of “Heavy NP Shift,” which moves the object NP
rightward:
(20) a′. I gave t to John [everything that he demanded]
b′. Max sent t to me [the longest letter anyone had ever seen]
This rule appears to be conditioned (in an obscure way) by the relative phonological
“weights” of the object NP and the verbal complements that it moves over, hence the
name.
Once this analysis of datives is accepted, a very different account of these phenom-
ena becomes possible. Given the underlying structures of the datives in (20), we can
take these examples to arise, not by rightward movement of NP, but rather by left-
ward movement of a predicate phrase—that is, not as in (20a′,b′), but as in (20a′′,b′′):
(20) a″. I [gave to John] everything that he demanded t
b″. Max [sent to me] the longest letter anyone had ever seen t
Under this view, “Heavy NP Shift” is in reality a case of “Light Predicate Raising.”14
(21) a. VP
SpecV′ V′
V VP
e NP V′
everything V PP
that . . .
give to John
b. VP
SpecV′ V′
V VP
e NP V
everything V PP
that . . .
give to John
To make this idea precise, I introduce the following optional rule of V′ Reanalysis:
V′ Reanalysis
Let α be a phrase [v′ . . . ] whose θ-grid contains one undischarged internal θ-role.
Then a may be reanalyzed as [v . . . ].
(21) c. VP
SpecV′ V′
V VP
give NP V
to John
everything t
that . . .
This reanalysis rule allows any predicate with (exactly) one unsaturated internal θ-role
to be syntactically reconstrued as a complex lexical category—in effect, a complex tran-
sitive verb. Thus, consider the underlying VP of (20a), where the θ-grid for [v′ give to
John] contains one unsaturated internal argument (2la). If V′ Reanalysis does not apply,
then head-to-head movement of V will occur as above, raising give to the [v e] position
and yielding give everything that he demanded to John. On the other hand, if reanalysis
does apply, then the result is (21b). Raising now applies to the entire complex constitu-
ent give to John, yielding (20a), as shown in (21c). I will assume that, as a consequence
of reanalysis, the Case-assigning properties of the verb are inherited by the complex
predicate; hence, Case marking of the object proceeds as before.
It is natural to inquire about the rationale of a rule like V′ Reanalysis in the grammar.
I consider this issue in section 7.4, where I suggest that reanalysis follows from a certain
kind of “mismatch” between the ways in which θ-theory and X-bar theory encode the
notion “transitive predicate.” Here I will simply note that this analysis of “NP Shift”
phenomena, when carried through in a perfectly general way, has some very strong
consequences. Consider, for example, the implications of (22a–c) (the last example due
to Engdahl (1983)):
(22) a. I would consider foolish [anyone who leaves his doors unlocked].
b. I saw at the conference yesterday [everyone who believes in UFOs].
c. I offended by not recognizing immediately [my favorite uncle from Cleveland].
If “NP Shift” is in fact complex predicate raising, then (22a) entails that small
clause constructions like I consider John foolish must (contrary to recent
proposals) have an underlying VP in which the AP is sister to V, namely,
[vp John [v′ consider foolish]]. 15 Example (22b) requires that the complex predi-
cate see-at-the-conference-yesterday be available for raising. Accordingly, on this
account modifiers like at the conference and yesterday cannot be outermost adjuncts
(as is standardly assumed) but rather must be innermost complements. Finally,
(22c) implies (contrary to Chomsky (1982, 1986a) and much other recent work)
that the licensing of parasitic gaps does not (or need not) involve variables left by
matrix A-movement. Under a predicate raising analysis of “NP Shift,” the object NP
remains in situ at all times; since no variable is generated, some process other than
chain composition must be involved.16
3. THE STRUCTURE OF DOUBLE OBJECT CONSTRUCTIONS
With the account of dative constructions developed above we now return to double
object structures. I will argue that domain asymmetries and various other properties
of this construction can be explained under a derivational approach to double object
structures.
Work in the Extended Standard Theory over the last ten years has generally not
assumed a transformational relation between dative and double object constructions
(Baker (1985) is an exception). This is no doubt due in part to the unclear status
of “Dative Shift” in theories embracing very general operations like Move NP (or
Move α) (Chomsky 1981, 1986b).17 And in part too, well-known restrictions and
lack of full productivity in the dative-double object relation have led many to con-
clude that this relation must be lexical rather than transformational in character (see
Allerton (1978), Dowty (1978), Green (1974), Hawkins (1981) and Oehrle (1976)
for discussion).
Nonetheless, despite these problems, there remain clear reasons why one might want
to relate oblique dative and double object structures transformationally. First, although
the relation between the two shows irregularities in English, in other languages the rela-
tionship is quite systematic. In particular, in languages with so-called applicative con-
structions (see Marantz (1984) and Baker (1985) for discussion) oblique and double
object structures show a highly productive relation strongly suggestive of derivational
relatedness. This argues that transformational operations similar to Dative Shift must
be available in principle. Second, a derivational approach to the dative-double object
relation is clearly desirable under any strong theses about the relation between structure
and assignment of thematic roles. For example, Baker (1985) advances the following
hypothesis:
Uniformity of θ-Assignment Hypothesis
Identical thematic relationships are represented by identical structural relations

between the items at the level of D-Structure.
This proposal is attractive on conceptual grounds since it narrows considerably the

class of initial D-Structure realizations of a given set of thematic roles. Notice that it
also virtually forces a derivational account of the dative-double object relation, since the
thematic roles assigned in these constructions are identical.18
3.1. “Dative Shift” as Passive

These considerations establish a prima facie case for attempting to construe the dative
double object relation transformationally. The challenge, then, is to bring this deri-
vation within the scope of established theoretical principles and to constrain it in
appropriate ways. Recall the underlying VP of a typical dative like John sent a letter
to Mary:
(23) VP
NP V′
a letter V PP
send to Mary
The deep VP is clauselike, with the NPs a letter and Mary standing roughly in the
relation of subject and object. Suppose we strengthen this parallel by assuming that
the governed preposition to appearing in (23) has the status of (dative) Case marking,
analogous to that appearing on indirect objects in more highly inflected languages.19
Consider now the possibility of extending operations generally held to apply between
subjects and objects to structures like (23). In particular, consider the possibility of pas-
sive formation in the inner VP. Under familiar proposals, the derivation of passives
involves two central effects: withdrawal of Case from an object position, and sup-
pression of thematic role assignment to a subject position (see Burzio 1986; Chomsky
1981). This triggers NP Movement to subject position. The suppressed subject θ-role is
(optionally) realized by an adjunct phrase:
(24) IP
NPi I′
Mary I VP
was VP PP
V NPi by a snowball
hit e
Suppose we amend this account slightly in the following way: rather than assuming that
a subject θ-role is suppressed in passives, we will assume that it is assigned in a special
way—specifically, in an adjunct configuration:
Argument Demotion
If α is a θ-role assigned by Xi, then α may be assigned (up to optionality) to an

adjunct of Xi.
This modification leaves the analysis of (24) unchanged. The IP subject receives its
thematic role compositionally from VP; hence, when the subject θ-role is demoted in a
passive and is assigned to the by-phrase, the latter appears adjoined to VP.
Let us apply this amended view of Passive to send as it occurs in the inner VP
in (23). First, Passive absorbs the Case assigned to the indirect object. Assuming
that we can regard the preposition to governed by send as pure Case marking, this
amounts to saying that to is absorbed. Second, the θ-role assigned to the subject
of VP (the direct object role) undergoes demotion, reducing this position to non-
thematic status. Since the direct object receives its θ-role from V′, under Argument
Demotion this θ-role must be assigned to a V′ adjunct. Accordingly, the direct object
is realized as a V′ adjunct. The situation is thus as follows: the indirect object is
Caseless in its deep position, and the VP subject position is nonthematic (and hence
empty). In the usual way, then, the indirect object undergoes NP Movement to the
VP subject position (25). Finally, send raises into V-head position, assigning Case
rightward to the VP subject (26). This yields the S-Structure form for the VP in John
sent Mary a letter.
(25) VP
SpecV′ V′
V VP
e NPi V′
Mary V′ NP
V NPi a letter
send e
(26) VP
SpecV′ V′
V VP
send NPi V′
Mary V′ NP
V NPi a letter
t e
The inner VP in (25) is analogous to the passive in (24). In both instances an object
has been moved to subject position, with the former subject assuming adjunct status.
For convenience, let us give the name “Passive” to NP Movement like (24), which
promotes an argument to IP subject position, and the name “Dative Shift” to NP Move-
ment like (25), which promotes an argument to VP subject position. We will refer to the
suite of operations embracing both as “PASSIVE.”
It is important to note that although the proposed derivation of Dative Shift sen-
tences makes crucial appeal to internal argument positions that are empty at D-Structure,
such positions are in no sense freely admitted under this account. Empty subcategorized
positions are licensed strictly by dethematicization of a thematic position. The fundamen-
tal logic of the Projection Principle discussed in Chomsky (1981) thus continues to apply,
and analyses ruled out by this principle (such as Raising to Object, which involves an
athematic, empty internal argument position that is not produced by demotion) continue
to be excluded.
It should also be observed that the connection drawn here between Passive
and Dative Shift is quite similar to that made within the framework of Relational
Grammar (see Perlmutter (1983) and Perlmutter and Rosen (1984)). In Relational
Grammar both Passive and Dative Shift are standardly viewed as instances of a
single operation of advancement, which promotes argument phrases with respect
to their grammatical relations. Thus, Passive is viewed as “2 → 1 advancement”
and Dative Shift as “3 → 2 advancement,” where “1,” “2,” and “3” designate
the subject, direct object, and indirect object relations, respectively. Arguments
that are supplanted in their grammatical relation—the deep subject of a passive,
the deep direct object in a double object structure—assume the special status of
chomeurs and become unavailable for subsequent relation-changing operations. In
effect, what we have given here is a structural interpretation of the standard Rela-
tional Grammar analysis, recasting the notion “advancement” uniformly in terms
of Move NP and understanding chômeur status as θ-role assignment in an adjunct
configuration.20
3.2. Some Consequences

The structural relations that arise here from Dative Shift directly account for Barss
and Lasnik’s (1986) facts. As a result of NP Movement, the inner object in (25),
which is a V′ specifier, asymmetrically c-commands the outer object, which is a V′
adjunct. Hence, the asymmetries observed in (3) are straightforwardly assimilated to
those in (27):
(27) a. Every boyi was recommended by hisi mother.
*?Heri son was recommended by every motheri .
b. Which boyi was recommended by hisi mother?
*Whose motheri was heri son recommended by?
c. Who was recommended by who?
*Who was who recommended by?
d. Each boy was recommended by the other’s mother.
*The other boy was recommended by each mother.
e. No one was recommended by anyone.
*Anyone was recommended by no one.
This analysis also neatly predicts a certain “classic” fact regarding the interaction of
double objects with “Heavy NP Shift.” It is well known that a heavy inner object in a
double object construction cannot be “shifted” to the right periphery of S:21
(28) a. *John sent a letter [every musician in the orchestra].
b. *Max gave a book about roses [the tall man in the garden].
c. *Mary promised to win [some spectator in the grandstands].
Recall now that on the proposals outlined above, “Heavy NP Shift” results from a form
of reanalysis. Specifically, when a V′ constituent has one unsaturated internal argument
(in other words, has the thematic properties of a transitive verb), then that V′ may be
reconstrued as V and undergo V-Raising. It is the raising of a complex verbal constituent
that results in (the appearance of) “Heavy NP Shift.”
On this view, in order to derive an example like (28a), we would have to be able to
reanalyze the V′ indicated in (29):
(29) VP
SpecV′ V′
V VP
e NPi V′
every musician V′ NP
in the orchestra
V NPi a letter
send e
But this is not possible. The circled V′ simply does not have the required thematic struc-
ture for reanalysis. Send has, as we have assumed, two internal arguments. NP-trace
[npi e] saturates one of these, and the adjoined NP a letter saturates the other. As a result,
the indicated V′ will have no unsaturated internal arguments. Since send e a letter can-
not be reanalyzed as V, it cannot be raised as a unit, and so (28a) cannot be derived.
Under other approaches to double objects, the ungrammaticality of (28a) has typi-
cally been assimilated to the general ungrammaticality of A-movement from the inner
object position in English (Kayne 1983a; Whitney 1983; Ziv and Scheintuch 1979):
(30) a. *?John, Mary said that she gave a present.

b. *?Who did Mary say that she gave a present?
c. *John is tough to give a present.
Under the analysis of “Heavy NP Shift” adopted here, however, such an assimilation cannot
be made. Since no movement of NP is assumed and since the availability of Ā-movement is
independent of V′ Reanalysis, (28) and (30) must be given separate explanations.
There is evidence from Norwegian indicating that this separation is in fact correct.
Like English, Norwegian allows “heavy” NPs to appear at the right periphery of S (data
from Christensen (1987)):
(31) a. Vi har lånt [den interessante boken du nevnte] til Petter.
we have lent the interesting book you mentioned to Peter

b. Vi har lånt til Petter [den interessante boken du nevnte].
(32) a. Vi skal lese [den interessante boken du nevnte] i morgen.
we shall read the interesting book you mentioned tomorrow
b. Vi skal lese i morgen [den interessante boken du nevnte].
Norwegian departs from English, however, in more freely allowing Ā-movement of

inner objects in the double object construction ((33a) from Christensen (1987), (33b-d)
from A. Hestvik (personal communication)):
(33) a. Ingen studenter har vi lånt romaner.
no students have we lent books
b. Jon sa Marit at hun ga en presang.
John said Mary that she gave a present
‘John, Mary said that she gave a present.’
c. Hvem sa Marit at hun ga en presang?
who said Mary that she gave a present
‘Who did Mary say that she gave a present?’
d. ?Jon er vanskelig a gi en presang.
John is difficult to give a present
‘John is difficult to give a present.’
Now, significantly, despite this greater latitude in Ā-movement, the equivalents of

(28a–c) remain ill-formed in Norwegian:
(34) a. Vi har lånt [den hyggelige gutten du kjenner] en bok.
we have lent the nice boy you know a book
b. *Vi har lånt en bok [den hyggelige gutten du kjenner].
Ā-extraction and “Heavy NP Shift” of the inner object thus pattern differently in
Norwegian, supporting the idea that the constraints applying to them in English also
have different sources.22
3.3. Apparent Differences between Passive and Dative Shift

If we are to maintain that passives and double object structures are fundamentally the
same construction—that they arise by the same process operating in different domains
(IP and VP, respectively)—then certain obvious differences between the two must be
accounted for. For example, in passives the verb is marked with overt participial mor-
phology, -en, whereas in double object constructions the verb appears in its usual active
form. In passives the demoted adjunct phrase (the logical subject) may be suppressed
(35a); however, in double object structures the demoted adjunct phrase (the logical
direct object) cannot be suppressed (35b):
(35) a. Mary was hit (by a snowball).
b. John sent Mary *(a letter).
(where Mary is understood as goal in the latter). In passives the adjunct phrase must appear
with a Case-assigning preposition (by), whereas in double object structures both NPs at issue
show up as “bare accusatives.” Finally, the active-passive relation and the oblique-double
object relation differ greatly in productivity, with the latter being much more restricted. I
will postpone discussion of productivity until section 5; let us take the other points in turn.
3.3.1. Morphology and Subject Suppression

The absence of participial morphology and the nonsuppressibility of the “VP sub-
ject” in double object structures can, I believe, be explained under proposals by Jaeg-
gli (1986). Recall that on the usual view of passive, the subject θ-role is held to be
“absorbed” by passive morphology. Jaeggli proposes that this absorption be under-
stood in the following way: in a passive the subject θ-role is assigned to the (adjoined)
passive morpheme -en:
(36) V
V en
hit
Adapting ideas by Zubizarreta (1985), Jaeggli suggests that this θ-assignment possibility
arises from the special status of the subject θ-role in the lexical representation. Briefly,
since the IP subject position is not a subcategorized one, a θ-role assigned to this posi-
tion cannot be linked in the lexical representation to any particular set of categorial fea-
tures. As a result, the IP subject θ-role is free to be assigned to various phrases, including
full nominal phrases, and also to morphological elements like -en. Other thematic roles
(such as those assigned to objects) do not have this categorially “unlinked” character
and so must be assigned to full nominal arguments.
Under these proposals, differences of morphology and subject suppression in datives
versus passives now follow directly from the linked versus unlinked status of the relevant
subject position. Datives involve a demoted VP subject. Since the VP subject position is
subcategorized for, an object θ-role cannot be assigned to a bound morpheme equivalent
to -en. Rather, it must be assigned to a full NP. Accordingly, in double object construc-
tions a “passive morpheme” cannot appear, and so an NP theme argument must appear.
Passives involve a demoted IP subject. Since the IP subject position is not subcategorized
for, the subject θ-role can be assigned to -en, and hence a full nominal subject argument
need not be present.23,24
Note that since -en receives the subject θ-role in a passive, by-phrases have a purely
adjunct status on the above view. When a by-phrase appears, as in (24), this expression
is assumed to receive its thematic role through the -en morpheme—essentially, the by-
phrase “doubles” the subject θ-role. Although Jaeggli makes no commitment on this
point, I will assume that the position of the by-phrase follows the generalization stated
earlier: that is, since the by-phrase receives the subject θ-role (through -en), it is adjoined
to VP, the constituent that assigns the subject θ-role in the unpassivized case. The basic
structural parallelism between the outer object in a double object structure and the by-
phrase object in a passive is thus preserved, even though the latter is not the primary
“target” of the subject θ-role.25
3.3.2. Case Assignment to Double Objects

As in other accounts of double objects, this analysis must assume Case marking of the
outer NP in its adjoined site. The central questions to be answered are these: (a) How
precisely is this Case assigned? and (b) How is it that two “measures” of Objective Case
come to be assigned in a single configuration?
The answer that I suggest to (a) is that Case assignment to the outer NP is licensed
by V′ Reanalysis as introduced earlier. Consider the lowest V′ in (26). This phrase has
the structure [V′ t e], where t is the trace of V-Raising, and e is the trace of NP Movement.
Under familiar assumptions the moved NP receives its thematic role through e; hence,
θ-role assignment to the goal argument in (26) must occur through t. This in turn means
that e discharges one of the two internal arguments of the dative verb; hence, the lowest
V′ has exactly one unsaturated θ-role, that corresponding to the direct object (theme).
By earlier assumptions, V′ Reanalysis may now optionally recategorize this V′ as a V:
(37) VP
SpecV′ V′
V VP
send NPi V′
Mary V NP
V NPi a letter
t e
In this configuration the outer NP complement a letter is the sister of a complex predi-
cate whose category is V—a complex transitive verb. It is, in fact, in the canonical con-
figuration of direct objects: [V′ V NP]. It is this relation, I propose, that allows a letter
to receive Case.26
At this point question (b) becomes relevant: Why is it Objective Case that is assigned, and
what allows this Case to be assigned twice? To answer this, I will make a somewhat contro-
versial proposal: namely, that Objective Case is assigned twice in the double object construc-
tion because in fact this is the typical situation for Objective Case assignment. That is, I want
to suggest that quite generally in transitive structures two Objective Cases are involved—one
structural and one inherent—and that the double object construction is simply an instance
where the two Cases are “pulled apart” and assigned to different arguments.27
To make this precise, I introduce the following assumption about structural Objective
Case: V assigns Objective Case in the configuration [Infl Infl [vp V . . . ]]. Thus, when gov-
erned by Infl, V assigns Objective Case. One way to think of this is that Infl has its own
Objective Case that must be assigned through a “host” V. The conditions on this host are
the usual ones: V must govern and be adjacent to the Case recipient, and so on. I also assume
that a verb may assign an inherent Objective Case to its highest internal argument as a
purely lexical property. The conditions for inherent Case assignment are again that V govern
and be adjacent to NP. According to these proposals, then, in a simple transitive structure
like (38) kiss is governed by I and therefore assigns structural Objective Case to NP*:
(38) IP
NP I′
John I VP
V NP*
kissed Mary
Assuming then that kiss also determines Objective Case for its internal argument as a
lexical property, two Cases are assigned to Mary: one structural and one inherent.28
Next consider (37). In this structure the outer NP complement a letter is governed
by the complex V (just as Mary is governed by kiss in (38)). By assumptions, [v t e]
inherits the Case-assigning properties of its head. Hence, [v t e] can assign to a letter the
inherent objective Case associated with send. Correlatively, the inner NP complement
is governed by the raised V, which is in turn governed by I. Again by earlier assump-
tions, V assigns (structural) objective Case to Mary. Thus, both NPs in the double object
construction receive objective Case, satisfying the Case Filter: the inner NP receives
structural Case through Infl, and the outer NP receives inherent Case through v.
Under the present account Passive and Dative Shift must have the properties shown
in table 1. Since we are assuming that two objective Cases are assigned in a transi-
tive structure-one structural and one inherent-we must assume that Passive actually
suppresses two Cases to maintain the familiar Government-Binding account of NP
Movement. On the other hand, only inherent Case is affected by Dative Shift: to is
absorbed, but no structural Case is withdrawn.
Table 1 Properties of Passive and Dative Shift
Case Thematic Role
Passive Suppress inherent Demote subject 6-role

Suppress structural
Dative Shift Suppress inherent Demote subject 6-role
This constellation of effects can be obtained by assuming that whenever our general-
ized NP Movement operation applies, an inherent Case is suppressed, and that when-
ever V is affixed by participial morphology (-en), it can no longer “host” the assignment
of structural Case from Infl.29 The effect of this, descriptively, is that PASSIVE sup-
presses or suspends a Case in whatever domain it applies, where by the domain of NP
Movement I mean the set of distinct projections α1, α2, . . . , αn intervening between the
head and tail of the A-chain produced by movement. Thus, the domain of Dative Shift
is VP, and PASSIVE suppresses one inherent Case, the Case assigned within VP. On the
other hand, the domain of Passive includes both V and I projections (an NP is moved
out of VP into the IP specifier position). Correspondingly, Case is suppressed in both
domains-a structural Case and an inherent Case.
4. “INDIRECT” PASSIVES
The connection between Passive and Dative Shift proposed here has implications
for the analysis of “indirect” or dative passives. (39a) is a typical example of this
construction:
(39) a. Mary was sent a letter.
Under the Standard Theory (and other frameworks), such sentences are analyzed as
arising by a two-step process: Dative Shift applies to a simple dative (39b), yielding a
double object structure (39c); Passive then applies to the latter, yielding (39a):
(39) b. (Someone) sent a letter to Mary.
c. (Someone) sent Mary a letter.
On this view it is always a (derived) direct object that is promoted in dative passives.
Under the present account, an alternative derivation becomes possible. Suppose we
apply PASSIVE to the simple dative in (39b), withdrawing the Case (the preposition to)
from the indirect object just as with Dative Shift. But instead of demoting the θ-role of
the direct object, we demote the role assigned to the subject. Such a move observes the
basic correlation of Case and θ-role suppression (“Burzio’s Generalization”). The inner
object now moves directly to the subject position in order to receive Case, and the verb
raises as usual, yielding (39a):30
(40) IP
NPi I′
Mary I VP
was SpecV′ V′
V VP
sent NP V′
a letter V NPi
t e
Note that although the indirect object is promoted directly to subject position in (40),
with no intermediate double object structure involved, the direct object will have essen-
tially the same Case status as it would under a two-step derivation, where it would
appear as a V′ adjunct. Application of PASSIVE in (40) not only suppresses the (Dative)
Case of the moved NP but also blocks assignment of structural Objective Case to a let-
ter. As a result, the direct object will receive only the inherent Objective Case assigned
by send. It follows then that although a letter occupies its D-Structure position in (40),
it behaves exactly like the outer object in a double object structure with respect to Case.
Passives of dative structures like (39a) contrast with examples like (41) in which a
direct object has been passivized. The latter are generally quite marginal:
(41) ?*A letter was sent Mary.
This result is predicted by our analysis. (41) cannot be derived directly from (39b) by NP
Movement: Case assignment to the indirect object has been suppressed (to is absent),
but it is the direct object that appears in subject position. The only source for (41) is
through an application of NP Movement to the outer object in a double object structure,
leaving Mary behind in direct object position (irrelevant details suppressed):
(42) [A letter] was sent Mary t
Recall now that in a double object structure the (derived) direct object receives only
structural Case. But observe that since this passive involves -en morphology, structural
Case assignment is suppressed (sent will not “host” the assignment of Case from Infl).
Accordingly, Mary is Caseless in (41), and the sentence is ruled out.
Passivization of the outer object is often judged to improve somewhat when the inner
object is pronominal:
(43) ??A letter was given me by Mary.
Oehrle (1976) suggests that this difference derives from the familiar fact that pronouns,
unlike full NPs, are able to undergo cliticization. Note that constructions like (43) are
well-formed only if the pronoun bears weak stress:
’im
(44) a. A letter was given by Mary.
*HIM
b. I didn’t say that a letter was given to Fred by Mary,
*I said that a letter was given ME.
Suppose then that English pronouns like me have, as a marginal option, the possibility of
cliticizing onto an adjacent verb. Suppose further (following standard views) that clitics
occupy an Ā-position, where they are not subject to the Case Filter. Then in (41) Mary will
be stranded without Case, yielding a violation. On the other hand, in (43) me will have the
option of cliticizing onto given, escaping the Case Filter. Finally, in (44a,b) the presence of
contrastive stress can be taken to block cliticization, again producing a Case Filter violation.31
4.1. “3 → 1” Advancement?
A “direct” analysis for dative passives is clearly possible only if we concede to Passive
one of the central properties of Dative Shift, namely, the ability to suppress or absorb
the preposition to. Hence, dative passives are a potential strong source of evidence for
the claim that passives and double objects arise by the same operation. If it could be
shown that direct derivations do occur, this claim would be considerably strengthened.
Let us consider some empirical evidence.
4.1.1. Japanese and Ancient Greek

In English a direct derivation for dative passives is hard to establish, since every case
where we might claim a “one-step” analysis will also have a “two-step” account in
which the indirect object first moves to direct object position by Dative Shift. However, in
other languages—those cited in the Relational Grammar literature as showing “3 → 1”
advancement—the situation appears somewhat clearer. Consider the following paradigm
from Shimizu (1975) for Japanese:
(45) a. Taroo-ga Hanako-o tataku.
Taro Subj Hanako Obj hit
‘Taro hits Hanako.’
b. Hanako-ga Taroo ni tatakareru.
Hanako Subj Taro by hit-Pass
‘Hanako is hit by Taro.’

c. Taroo-ga Hanako ni kotzutsumi-o okuru.
Taro Subj Hanako to package Obj send
‘Taro sends a package to Hanako.’
d. Hanako-ga Taroo ni kotzutsumi-o okurareru.
Hanako Subj Taro by package Obj send-Pass
‘Hanako is sent a package by Taro.’
e. *Taroo-ga Hanako-o kotzutsumi-o okuru.
Taro Subj Hanako Obj package Obj send
(45a-c) are simple examples of active, passive, and dative constructions, respectively. As
(45d) shows, it is possible to construct a dative passive corresponding to (45c); however,
Japanese forbids the intermediate double object structure (45e) required on a two-step
derivation of the dative passive. On the basis of this, Shimizu suggests that the deriva-
tion of (45d) involves an advancement of Taroo directly from indirect object to subject
status, with no intervening direct object stage.
Similar facts are observed by Feldman (1978) for Ancient Greek. Feldman notes that
dative constructions like (46a) may be passivized as in (46b), with the indirect object
promoted to subject status. However, a double accusative structure like (46c) never
occurs in Ancient Greek:
(46) a. epitáksousi állo ti méìdzon hyymíìn

lay upon-3pl-fut other some greater-acc you-pl-dat
‘They will lay some other, greater [command] upon you.’
b. állo ti méìdzon epitakhtheésthe
lay upon-2pl-fut-pass
‘You will have some other, greater [command] laid upon you.’
c. *epitáksousi állo ti méìdzon hyyáàs
you-pl-acc
Accordingly, Feldman suggests an analysis of (46c) in which hyymiin is promoted

directly from indirect object to subject status. Languages like these appear to support
the thesis of “3 → 1” advancement in a direct way.32
4.1.2. Psych Verbs

A “3 → 1” analysis of dative passives also appears to be supported indirectly by the
analysis of other movement constructions. Under the proposals of Perlmutter (1978,
1983) and Burzio (1986), passives and unaccusatives bear an intimate relation to each
other and involve parallel forms of NP Movement:
(47) a. The boat was sunk t.
b. The boat sank t.

Thus, passive morphology induces sink to assign no Case to its internal object and no
θ-role to its subject. Unaccusative sink manifests essentially the same behavior as a mat-
ter of its basic lexical properties.
Suppose now that the derivation of dative passives proceeds as we have suggested,
where given may involve suppression of Case assignment to the indirect object and sup-
pression of θ-role assignment to the subject. Then we expect a parallel set of unaccusa-
tives. That is, we predict verbs X that, as a matter of their lexical properties, suppress
Case to an innermost object, inducing it to move to subject position:
(48) a. NP1 was give-en NP1 t
b. NP1 X NP1 t
Interestingly, there do seem to be predicates with the relevant properties. These are
so-called psych verbs—predicates like annoy, excite, frighten, worry, and please as they
occur in the following sentences:
(49) a. The exam worried Max.
b. Flies frequently annoy Sam.
c. Cleanliness pleases Felix greatly.
Belletti and Rizzi (1988) discuss the syntax of psych verbs. They cite various facts
from English and Italian suggesting that the underlying relations among the arguments
in such sentences are very different from what is implied by the surface arrangement
NP1–V–NP2. To summarize their results briefly:
(I) Surface subjects (NP1) of psych verbs behave like derived subjects on a variety of
tests in Italian. In contrast with subjects of normal transitive verbs (know, like, admire,
and so on), psych-verb subjects do not license anaphoric clitics, do not have an “arb”
interpretation (meaning ‘people’ or ‘one’), and cannot be embedded under causative
constructions sensitive to derived versus underived subject status. Furthermore, NP1
behaves as if it has attained subjecthood from a position lower than the surface object
(NP2). This is suggested by binding facts. As is well known, the subject of a psych verb
can contain a reflexive bound to the object NP (50a,b), something that is not possible
with genuine transitives (51a,b):
(50) a. [Nude pictures of herself] don’t offend Mary.
b. [Stories about himself] excite John.
(51) a. *[Nude pictures of herself] absolved Mary of the crime.
b. *[Stories about himself] don’t describe John very well.
Assuming the usual c-command condition on binding of anaphors, these facts suggest
that the surface subject in (50) is actually c-commanded by the object at some level.
(II) The surface object (NP2) behaves like a genuine object with respect to Case
marking. Pronominal objects of psych verbs in English and pronominal object clitics in
Italian show up in their accusative form. On the other hand, NP2 behaves like a subject
with respect to certain anaphoric phenomena. As discussed by Giorgi (1984), the Ital-
ian anaphoric possessive proprio may function as a long-distance anaphor, and it shows
the “subject orientation” typical of such elements. Correlatively, the surface object of a
psych verb can bind a long-distance proprio embedded within the surface subject:
(52) Chiunque dubita della propriai buona fede preoccupa Giannii.

‘Whoever doubts his own good faith worries Gianni.’
(Belletti and Rizzi 1988, (69b))
This suggests that at some level the psych-verb object is also a subject.33
Belletti and Rizzi (1988) account for these facts with the following structure:
(53) IP
NPi I′
pictures I VP
of himself
V′ NP*
V NPi Max
worry e
The basic idea here is that psych verbs are, in effect, unaccusatives with two internal
arguments.34 Worry fails to assign Case to one of these (the innermost, pictures of him-
self) and also fails to assign an external thematic role. This forces movement to subject
position, as shown. In the resulting structure the surface subject is a derived subject. Fur-
thermore, it attains its position from a site lower than the surface object. If we accept the
“anywhere” version of Principle A of the binding theory that Belletti and Rizzi advance,
this accounts for the facts in (I).35 Turning to the surface object, we find that it is in fact a
structural (VP) subject-indeed, the “most prominent” (that is, highest) θ-marked subject
in (53). This accounts for the long-distance anaphor facts in (52). Finally, the surface
object Max is also an object for purposes of Case assignment. Belletti and Rizzi suggest
that NP* receives an inherent Objective Case assigned by V′.36
Notice now that modulo the presence of V-Raising, the structural relations holding
among complements in (53) are identical to those holding in (40): the surface subject is
derived, and it attains subjecthood from a position c-commanded by the surface object.
Moreover, the surface object is both an object and a structural (VP) subject. Such an
analysis fills the “gap” in the paradigm of passive-unaccusative pairs in (47) and (48):
psych verbs become the unaccusative counterparts of dative passives under a “3 → 1”
analysis of the latter. We have both dative passives and “dative unaccusatives.”
5. CONSTRAINTS ON DATIVE MOVEMENT IN ENGLISH
As noted earlier, the oblique-double object alternation is not fully productive in English.
There are well-known verbs like donate and distribute that appear in the oblique dative
construction but have no double object counterpart (54a); and there are verbs like
envy and spare that occur in double object constructions with no well-formed oblique
“source” (54b) (the latter pair is from Dowty (1978)):
(54) a. John donated the money to charity.
*John donated charity the money.
b. The judge spared John the ordeal.
*The judge spared the ordeal to John.
Data like these have led a number of researchers to doubt the derivational connection
between oblique and double object forms (Allerton 1978; Dowty 1978; Oehrle 1976,
1983) and to analyze the relation as a lexical one holding between distinct verb entries.
On this view, verbs like give are assigned two lexical entries with identical semantic
content but distinct subcategorization frames: one that specifies a direct object and PP
complement, and a second that specifies two NP objects.
If we are to maintain a derivational analysis of double object structures, then clearly
we must give some account of the limitations on Dative Shift. We must find some way
of understanding why the latter cannot apply in certain instances (55a) but must apply
in others (55b):
(55) a. Oblique → *Double object
b. *Oblique → Double object
Furthermore, this account should obviously be of a general nature: constraints on Dative

Shift should be ones applying to other instances of Move α. To motivate an approach to
this issue, let us begin by considering the preposition to more closely.
5.1. To as “Case Marking”

The analysis of Dative Shift presented above makes crucial appeal to the idea that
in a VP like send a letter to Mary, to represents Case marking. It is this assump-
tion that permits us to assimilate Passive and Dative Shift by allowing us to view
suppression of to as suppression of Case. However, although a Case-marking view
of to is reasonable in examples like John sent a letter to Mary, it is not plausible in
other instances:
(56) a. Klaus flew his dirigible to Helgoland.
b. The spoon fell to the floor.
c. Oscar bowed to the queen.
d. The meeting ran from two o’clock to five.

In (56a–d) the goal phrases are all unselected adverbials, and the preposition appears to
make a genuine semantic contribution, specifying (roughly) the goal of “motion” along
some trajectory or path. There is evident notional relatedness between these and the
dative instances of to; however, to does not appear as mere Case marking in (56). These
facts thus raise a simple question: how can we maintain our analysis of Dative Shift and
still preserve the intuitive identity of to across its various uses?
In answer to this question I want to propose that to is in fact always contentful that
it is never mere Case marking, strictly speaking-but that in certain contexts (namely, in
V′s headed by Dative-Shifting verbs) its grammatical contribution effectively “reduces” to
Case marking and therefore can be suppressed under Passive. To make this precise, assume
that to is uniformly an independent preposition in English with its own role to assign.
Assume also that although verbs may select an indirect object argument, they cannot assign
this argument Case. From the second assumption it follows (under the Case Filter) that a V′
containing a dative verb and an indirect object α can be well-formed only if an independent
Case-assigning element like to appears: [v′ V [pp to α]]. Furthermore, from the first assump-
tion it follows that both V and P must independently assign thematic roles to α. Suppose
that give and to assign the following roles to the indirect object argument:
give: Beneficiary
Goal of motion along some path
to: Goal of motion along some path
Then the suite of θ-roles assigned by V subsumes the role assigned by to; hence, the
semantic contribution of the latter is redundant. This in tum means that in a V′ like
[v′ give to Mary] the grammatical contribution of to effectively reduces to the Case
marking it provides for Mary.
It is this sense, I suggest, in which to constitutes pure Case marking in dative con-
structions involving give, send, and so on. Although the preposition is not, strictly
speaking, without semantic content, this content is fully “recoverable” from the local
syntactic context—specifically, from the verb with which it co-occurs. We may now take
it that in such circumstances PASSIVE may absorb to as a Case marker, triggering Dative
Shift in the by now familiar way.
5.2. Oblique → *Double Object: Recoverability

Given this view of Case and θ-role assignment in V′, a simple approach now opens up
for the analysis of verbs like donate, contribute, and distribute—verbs that appear in
oblique dative constructions but resist Dative Shift:
(57) a. I donated money to charity.
*I donated charity money.
b. I distributed apples to the children.
*I distributed the children apples.
c. I contributed my time to the auction.
*I contributed the auction my time.

We can view the second member of each pair as violating (the equivalent of) “recover-
ability of deletion.”37
Suppose that unlike give, send, promise, and so forth, the verbs in (57) do not
specify their third argument as a goal—that is, suppose that donate assigns only the
Beneficiary role to its indirect object. Consider a V′ like donate to charity. Here V
and P are semantically compatible under the stated assignments; however, to is not
redundant—its grammatical contribution does not “reduce” to Case marking. In this
situation suppression of to would result in an unrecoverable loss of thematic informa-
tion associated with the preposition. Such suppression will thus be blocked by famil-
iar principles excluding nonrecoverable deletions. Dative Shift will be forbidden from
applying.
This analysis of why verbs forbid Dative Shift—namely, because they do not specify
the content of to in their thematic array-appears to be supported by evidence from the
behavior of certain verb-particle constructions. Consider the expressions give away and
give out noted in Green (1974). These compounds contain the Dative-Shifting verb
give together with a directional adverb particle indicating (roughly) centrifugal motion.
Superimposing the latter component on the meaning of give results in phrasal verbs
that preserve the notion of “beneficiary” from their stem but intuitively connote only
transfer of possession from a source, and not transfer of possession to a goal. Under our
account, we expect such verb-particle combinations, like simplex donate and distribute,
to be merely compatible with to and hence to resist Dative Shift. This expectation is
correct:38
(58) a. I gave away money to charity.
*I gave away charity money./*I gave charity away money.
b. I gave out apples to the children.
*I gave out the children apples./*I gave the children out apples.
The behavior of verb-particle compounds like give away and give out thus appears to
confirm the idea that ability to undergo Dative Shift depends crucially on the directional
content of the role assigned to the indirect object. When this content does not include
that specified by to, or when it is “overwritten” by an added directional adverbial ele-
ment, Dative Shift fails.
The general proposal that Dative Shift applies freely up to recoverability is consistent
with the observation of Marantz (1984) that although the dative alternation does not occur
with every verb form taking an oblique indirect object in English, there is nonetheless a
relativized sense in which it is fully productive: namely, so long as one remains within cer-
tain limits imposed by semantics, the alternation applies quite freely to any predicate taking
an indirect object. Marantz draws attention to this relativized productivity in connection
with the introduction of new verb forms. Consider a hypothetical verb shin meaning ‘to
kick with the shin’, as applied in a sentence like (59a) (Marantz 1984, p. 177):
(59) a. Elmer shinned the ball to me during soccer practice.
As Marantz points out, for such verbs, which involve directing an object with a body
part, any speaker accepting (59a) will also immediately accept its Dative-Shifted variant
(59b) despite the novelty of the form:
(59) b. Elmer shinned me the ball during soccer practice.
This result is expected on the current view: any verb falling within the appropriate semantic
class (one that assigns a role to its third argument that subsumes the role assigned by to)
will allow a recoverable suppression of to; hence, Dative Shift will apply freely.39
The recoverability hypothesis also suggests a simple approach to double object pro-
ductivity in other languages. In English, dative-type alternations are not available with
oblique instrumental or locative phrases, presumably because the relevant prepositional
content is not recoverable from V:
(60) a. I cut the salami with a knife.
*I cut a knife the salami.
b. John left his books on the sofa.
*John left the sofa his books.
However, in languages with so-called applicative constructions, alternations parallel

to (60a,b) do in fact occur. Consider the following data:
(61) a. i. Saja mem-bawa surat itu kepada Ali.
I Trans-bring letter the to Ali
‘I brought the letter to Ali.’
ii. Saja mem-bawa-kan Ali surat itu.
I trans-bring-app Ali letter the
‘I brought Ali the letter.’
(Bahasa Indonesia; Chung 1976)
b. i. Mereka men-dapat suatu pekerdjaan untuk anak-ku.
they trans-find a job for child-my
‘They found a job for my daughter.’
ii. Mereka men-dapat-kan anak-ku suatu pekerdjaan.
they trans-find-app child-my a job
‘They found my daughter a job.’
(Bahasa Indonesia; Chung 1976)

c. i. Fisi a-na-dul-a chigwe ndi mpeni.
hyena sp-past-cut-asp rope with knife
The hyena cut the rope with the knife.’
ii. Fisi a-na-dul-ir-a mpeni chigwe.
hyena sp-past-cut-app-asp knife rope

‘The hyena cut the rope with the knife.’
(Chichewa; Baker 1985)
d. i. Abaana b-iica-ye ku meeza.
Children sp-sit-asp on table
‘The children are sitting on the table.’
ii. Abaana b-iica-yc-ho meeza.
children sp-sit-asp-app table
‘The children are sitting on the table.’
(Kinyarwanda; Kimenyi 1980, cited in Baker 1985)
(61c,d) are parallel to the excluded English examples. Note especially the morphological
marking in the form of an “applied affix” (App) that appears on the verb.40
On our account, we can attribute the broadened scope for double object formation in
applicative languages directly to the presence of morphological marking on the verb. Sup-
pose applied constructions are derived via NP Movement analogously to English double
object forms, and that the applied affixes in languages like Bahasa Indonesia, Chichewa,
and Kinyarwanda are essentially “registration markers” for some particular role like
instrument or spatial location. When affixed to V, they specify its manner or location
role in such a way as to make the contribution of an overt preposition redundant. This
permits P to be absorbed as Case under “Dative” Shift without violating recoverability.
Productivity and morphological marking are thus directly linked.41
5.3. *Oblique → Double Object: Unaccusativity

Let us now tum to the second case in which the oblique–double object alternation may
fail: the case of verbs like spare that allow double objects but have no corresponding
oblique forms. Recall the general situation of a V′ containing a three-argument verb
and an indirect object argument α: [v′ Vα]. We have noted that since V cannot itself
assign Case to α, some semantically compatible preposition must occur. Suppose now
that the grammar simply contains no preposition compatible with the role(s) assigned
by V to α. That is, suppose that as a matter of the “semantic fields” carved out by the
prepositional system of the language, there is no appropriate P. Then, as a matter of the
lexical properties of V (specifically, its meaning), the relevant NP argument must always
fail to receive Case.
It is plausible to think that this situation might fall under a version of Burzio’s Gen-
eralization (Burzio 1986), which correlates Case and θ-role assignment possibilities. In
particular, we might take Burzio’s Generalization as implying in this instance that if a
verb does not permit Case assignment to the V′ object by virtue of its semantics ruling
out any potential prepositional Case assigner, then that verb cannot assign a θ-role to
a subject. Such situations would represent a rather special case of “unaccusativity”
(Perlmutter 1978, 1983).42
I want to propose that obligatory double object formation represents a case of unaccu-
sativity in this sense-that the relevant verbs are thematically incompatible with any potential
Case assigner and hence force their third argument to undergo movement. I illustrate this
proposal with the example of spare. The verb spare has the interesting semantic property
that its notional indirect object argument, although a beneficiary, is not, and cannot be, a
goal. In The judge spared John the ordeal John benefits by the action of the judge; however,
benefit accrues precisely because the ordeal in question does not go to John. Thus, to is
semantically incompatible with the role spare assigns to its third argument:
(62) a. *The judge spared the ordeal to John.
Other potential prepositions are unavailable as well. Note that although the indirect object of
spare is notionally not a goal, neither is it a source. In The judge spared John the ordeal the
ordeal in no way originates with John. Accordingly, spare rejects the preposition from, and
it contrasts with verbs like rob, whose notional indirect object (when it occurs) is a source:
(62) b. The judge robbed money from John.
c. *The judge spared heartache from John.
Note also that in constructions employing the preposition of to mark loss or nonpos-
session (such as I deprived John of his livelihood, Max relieved Oscar of his duties), the
preposition uniformly marks the theme (his livelihood, his duties) and not the beneficiary/
maleficiary (John). So of cannot “rescue” an oblique construction with spare:
(62) d. *The judge spared the ordeal of John.
Other cases are analogous. Thus, since spare effectively forbids Case assignment to its
indirect object, under Burzio’s Generalization it assigns no thematic role to a subject.
Hence, the third argument of spare obligatorily undergoes NP Movement, as required.43
Certain cases of obligatory double objects appear to involve factors beyond unaccu-
sativity in the sense discussed here. Consider additional examples of the following kind
drawn from Green (1974):
(63) a. Mary gave John a cold.
*Mary gave a cold to John.
b. Mary gave John a broken arm.
*Mary gave a broken arm to John.
c. Mary gave John a black eye.
*Mary gave a black eye to John.
(64) a. Mary gave John a bath.
*Mary gave a bath to John.
b. Mary gave John a kiss.
*Mary gave a kiss to John.
c. Mary gave John a punch in the nose.
*Mary gave a punch in the nose to John.

Intuitively, the oddness in the second example of each pair does seem to derive in part
from the difficulty of understanding the direct object as undergoing “motion” along
some path. A black eye, a broken arm, a bath, or a kiss in no sense travels from Mary to
John, which is to say that the compositional semantics of the VP is not fully compatible
with the direction/path component of meaning contributed by to.
Nonetheless, there appears to be an extra dimension of “affectedness” at work in (63)
and (64) that favors the double object versus the oblique construction. In all of (63) and
(64) there is a strong sense in which the deep indirect object (John) names the individual
affected by the action described in the clause, and in which the deep direct object (a cold,
a bath, and so on) does not. This is particularly clear in the second set of cases, where the
examples can be paraphrased by a simplex verb with an affected object:
(65) a. Mary bathed John.
b. Mary kissed John.
c. Mary punched John in the nose.
Now, as discussed in Tenny (1987), the surface direct object position is the canonical
site of affected objects. Hence, in addition to the effects of unaccusativity, the pre-
ferred status of the Dative-Shifted form in (63) and (64) may well reflect the strong
preference for having the notional affected argument in the appropriate structural
position.44
In closing this discussion of constraints on Dative Shift, it is worth pointing out that
our proposals will extend correctly to dative passive constructions. In particular, we can
maintain the “3 → 1” analysis of such constructions and still explain why, when a verb
fails to allow Dative Shift, it also fails to allow a dative passive (66a–d) (from Dowty
1978) and why, when a verb has only a Dative-Shifted form and no simple dative, it
permits only a dative passive (67a–d) (also from Dowty 1978):
(66) a. John donated the money to the foundation.
b. *John donated the foundation the money.
c. The money was donated to the foundation (by John).
d. *The foundation was donated the money (by John).
(67) a. *The judge kindly spared the ordeal to John.
b. The judge kindly spared John the ordeal.
c. *The ordeal was kindly spared to John (by the judge).
d. John was kindly spared the ordeal (by the judge).
On the present account, dative passive and double object derivations differ solely in
which θ-role is demoted—IP subject or VP subject, respectively. Since both involve
the same transformational operation, we expect the same constraints to apply. Thus,
if the preposition to is not recoverable from donate under the derivation responsible
for (66b), then it will not be recoverable under the derivation responsible for (66d).
Likewise, if spare cannot assign Case to John in conjunction with a preposition (67a),
then John will have to undergo movement whether it ends up in VP subject position
(67b) or in IP subject position (67d); in no case will some other argument be allowed
to be moved (67c).
6. TWO CROSS-LINGUISTIC QUESTIONS
The analysis of double objects proposed above raises certain natural questions when
facts from languages other than English are considered.45
I have suggested that the to appearing in datives like John gave a book to Mary
amounts to Case marking and that its disappearance in double object structures is
equivalent to the absorption of Case marking in passives. However, it is well known
that, quite generally, “true” Dative Case marking cannot be suppressed under Passive.
For example, in German the verb helfen imposes dative Case on its object (68a). How-
ever, this case cannot be absorbed under Passive (68b); rather, the dative argument
must remain internal and the passive surfaces as an impersonal construction (68c):
(68) a. Hans hilft ihm.
Hans helps him-dat
b. *Er-nom wurde geholfen.
he was helped
c. i. Es wurde ihm geholfen.
it was him-dat helped
ii. Ihm wurde geholfen.
him-dat was helped
‘He was helped.’
Similar facts obtain in Russian, as discussed by Freidin and Babby (1984). Although
Russian permits passive of accusative objects (69), passive of predicates that impose
dative (or other oblique cases) on their objects is not permitted (70):
(69) a. Ivan čitaet knigu.
Ivan-nom reads book-acc
‘Ivan is reading the book.’
b. Kniga čitaetsja (Ivanom).
book-nom is-being-read (Ivan-inst)
‘The book is being read by Ivan.’
(70) a. Rabotnik podražaet inostrannym metodam.
worker-nom copies foreign-dat methods-dat
‘The worker is copying foreign methods.’
b. *Inostrannye metody podražajutsja rabotnikom.
foreign-nom methods-nom are-copied worker-inst
‘Foreign methods are being copied by the worker.’

Freidin and Babby (1984) suggest that this result follows from the rather natural princi-
ple that lexical properties must be expressed. (70b) violates this principle since the lexi-
cal Case-marking property of podraf.aet ‘copies’—the fact that it assigns dative-fails to
be expressed. The first question is thus the following: How does our analysis of Dative
Shift square with the observation that cross-linguistically dative and other oblique cases
cannot be suppressed under Passive?
A second question concerns the fact that although Dative Shift is possible in English
and various Germanic languages (for instance, Dutch and Danish (71a,b)), this alter-
nation is not universally available. For example, French and various other Romance
languages forbid it (72a,b):
(71) Dutch
a. i. Zij gaf het boek aan de man.
she gave the book to the man
ii. Zij gaf de man het boek.
she gave the man the book
Danish (Herslund 1986)
b. i. Han sendte blomster til sin sekretær.
he sent flowers to his secretary
ii. Han sendte sin sekretær blomster.
he sent his secretary flowers
(72) French (Kayne 1983a)
a. i. Jean a donné un livre à Marie.
John has given a book to Mary
ii. *Jean a donné Marie un livre.
John has given Mary a book
Spanish
b. i. Juan dio un libro a Maria.
John gave a book to Mary
ii. *Juan dio Maria un libro.
John gave Mary a book
Since double object constructions are analogous to passives on the present account,
and since the Romance languages possess passive formations (Ces lettres ont ere ecrites
par mon frere ‘These letters were written by my brother’), we would like to know why
Dative Shift is unavailable in Romance.
To answer these questions, I will appeal to a proposal by Kayne (1981) regarding
why preposition stranding is possible in English but forbidden in many other languages,
including Romance languages like French and Spanish. Kayne suggests that the basic
property of English that permits preposition stranding is that its prepositions assign
Objective Case. This allows prepositions in English to be thematically reanalyzed with
the verb when a prepositional object is extracted, which in tum permits the trace of the
latter to be licensed under the Empty Category Principle (ECP). In languages like French
and Spanish, where prepositions assign Oblique Case, reanalysis is blocked owing to
Case conflict between V (an Objective Case assigner) and P (an Oblique Case assigner);
this results in an ECP violation whenever a prepositional object is extracted.
If we adopt Kayne’s (1981) proposal, then our two questions are directly answered.
First, English double object formation (and its analogues in Dutch and Danish) will not
in fact involve suppression of Oblique (Dative) Case, as in the ill-formed German and
Russian examples in (68b) and (70b). The Case assigned by to will be Objective; hence,
its absorption is fully parallel to absorption of the Objective Case assigned by verbs.
Likewise, the cross-linguistic differences in the availability of Dative Shift will follow.
Assuming that Oblique Case cannot be suppressed (perhaps for the reasons suggested
by Freidin and Babby (1984)), it will be possible to have Dative Shift only when to
(or its equivalent) is an Objective Case assigner. Dative Shift will thus be impossible in
French, Italian, and Spanish, where Oblique Case is assigned by P, but possible in the
preposition-stranding languages like English, Dutch, and Danish.46
7. THE PROJECTION OF RAISING STRUCTURES
In analyzing dative and double object constructions, D-Structure forms like the follow-
ing have played a central role:
(73) VP
SpecV′ V′
V VP
e XP V′
V YP
The obvious question arises: Where do such structures come from? How are they
projected? In this section I will briefly suggest an answer, involving a restricted version
of X-bar theory operating together with principles for mapping thematic relations onto
syntactic structure.
7.1. The Single Complement Hypothesis

Chomsky (1970) proposes that phrasal configurations in natural language subscribe to
a universal schematism known as X-bar structure. This schematism is now widely held
to be given by the two rules in (74) (from Stowell (1981)),
(74) a. XP SpecX′ X′
b. X′ X YP*
where X ranges over any category (N, V, and so on), where SpecX′ is the position of
specifiers for XP, including determiners and subjects, and where YP* is a finite string of
complements (possibly null) of the head X.
The X-bar scheme in (74) accepts a view of long standing within the grammatical
tradition, namely, that there is a fundamental, twofold asymmetry between subjects
and complements. First, there is a basic structural asymmetry. The head together with
its complements jointly constitutes a predicate phrase (X′) that excludes the subject. As
a result, complements have a subordinate hierarchical status vis-a-vis subjects. Second,
there is a basic numerical asymmetry. As the presence of the (Kleene) star “*” indicates,
although X can have arbitrarily many complements, XP can have at most one subject.
These two properties constitute an empirical hypothesis about how natural language
realizes the relations between a predicate and its arguments; neither is required from a
purely formal point of view. In artificial languages a three-place relation like GIVE(x,y,z)
can equally well be represented as taking all of its arguments jointly, with no hierarchi-
cal differences among them (75a); as taking the last two and predicating the result of the
first (75b); or as taking the last and predicating the result of the first two (75c):
(75) a. b. c.
GIVE x y z x x y
GIVE y z GIVE z
Choice among these is purely a matter of formal convenience.

Consider now eliminating one of the two subject-complement asymmetries just men-
tioned. In particular, consider amending the rules for X-bar structure as follows:47
(76) a. XP SpecX′ X′
b. X′ X YP
Like the rules in (74), (76a,b) impose a fundamental structural asymmetry between
subjects and complements. The latter remain subordinate to the former. Unlike the
rules in (74), however, (76a,b) eliminate the numerical asymmetry between subjects and
complements. According to (76b), just as there can be at most one subject per maxi-
mal projection, so there can be at most one complement. In intuitive terms, one might
understand this revision as making the following claim: natural language distinguishes
one kind of relation as fundamental, namely, the transitive one. This involves a relation
between two arguments, a subject and an object. We will say that (76a,b) embody a
Single Complement Hypothesis about X-bar structure.
7.2. Principles of Argument Realization and Projection

In addition to the Single Complement Hypothesis, I will assume two principles govern-
ing the projection of arguments vis-a-vis their predicates. The first concerns the domain
in which arguments of a predicate are realized:
P1
If α is a predicate and is an argument of α, then must be realized within a projec-

tion headed by α.
This principle imposes a very tight relation between thematic and categorial structure.
For example, it virtually forces the analysis of clauses argued for by Kitagawa (1986),
Kuroda (1988), Speas and Fukui (1986), and Sportiche (1988), where the subject of IP
is located underlyingly within VP. According to these authors, a sentence like John saw
Mary begins with a structure like that in (77),
(77) IP
NP I′
I VP
[Tense]
NP V′
John V NP
see Mary
where John is an underlying VP subject and hence realized within a projection of the
predicate from which it receives a θ-role (see). On these views, John raises to IP subject
position at S-Structure in order to receive Case (see above references for discussion and
supporting arguments).
The second principle governs the relative subordination of arguments in D-Structure.
Assume the following hierarchy of thematic relations due essentially to Carrier-Duncan
(1985):
Thematic Hierarchy
AGENT > THEME > GOAL > OBLIQUES (manner, location, time, . . .)
Then the roles assigned by a verb are linked to arguments according to P2:
P2
If a verb α determines θ-roles θ1, θ2, . . . , θn, then the lowest role on the thematic
hierarchy is assigned to the lowest argument in constituent structure, the next low-
est role to the next lowest argument, and so on.
Thus, P2 translates relative position on the thematic hierarchy into relative structural
subordination of complements, with arguments bearing the lowest-ranked role being
most subordinate.48
Finally, I will assume an interpretation of θ-theory according to which if a predicate
α determines n thematic roles, then it also determines n argument positions, whether its
roles are actually assigned to those positions or not. The point of this specification is
to ensure that an argument position for a given role is projected even when the role in
question is demoted and assigned in an adjunct configuration. What I am saying by this
interpretation is that a-theory is, in effect, “blind” to adjunct assignment—that in order
to satisfy θ-theory a structure must show as many A-positions as it has thematic roles.
7.3. Illustration: Raising Structures for Give

Let us examine how these principles apply in sentences containing a verb like give (that
is, a verb having more than two roles to project). The X-bar scheme in (76) admits the
structure (78a). Assuming now that all roles are projected into A-positions (in other
words, no roles are demoted), the result is structure (78b), in which (according to P2) α
is assigned the Goal role, and β is assigned the Theme role:
(78) a. VP b. VP
XP V′ NP V′
V YP β V PP
give give to α
The latter structure leaves one argument—the Agent—unprojected. At this point we

seem to encounter conflicting requirements imposed by θ-theory, X-bar theory, and
the realization principle P1. The first demands that the Agent role be expressed. The
second excludes the expression of this argument within the simple X-bar projection
of V in (78b). Finally, the third demands that the Agent argument be realized within a
projection headed by V.
I want to suggest that this situation is resolved by the grammar through the projec-
tion of structures like those assumed here:
(79) VP
NP V′
γ V VP
e NP V′
β V PP
give to α
Here VP becomes the complement of an X-bar “shell,” whose head is empty and hence
without independent thematic requirements, and whose specifier is γ Structure (79)
constitutes something like the “minimal, purely structural elaboration” of (78) that
supplies an A-position for the Agent argument γ of give (satisfying θ-theory), conforms
to X-bar theory, and allows for satisfaction of the principle P1. The latter is satisfied by
V-Raising, which places γ within a projection headed by give.49
The projection of a double object D-Structure form proceeds as in the oblique case;
however, instead of projecting the Theme into the VP subject position (78b), we demote
this role and realize it as an adjunct (78b′). Since (by assumptions) give must determine
as many A-positions as roles, a VP shell is again generated and the Agent role is assigned
(up to demotion) to its specifier, as in (80). The S-Structure derivation then proceeds as
discussed above.
(78) b′. VP
NP V′
e V′ NP
V PP β
give to α
(80) VP
NP V′
γ V VP
e NP V′
e V NP
V PP β
give to α
Since the projection of empty structure for a verb like give crucially depends on the
presence of a third, external argument, this entails that psych verbs of the sort discussed
earlier will not involve V-Raising. Recall that such verbs involve two internal arguments
but no external argument. Accordingly, the VP for such examples will contain all argu-
ments of the verb:
(81) VP
V′ NP*
V NP β
worry α
This structure, which is just the one argued for by Belletti and Rizzi (1988), satis-
fies θ-theory, the restricted X-bar theory in (76), and the realization principle P1.
Hence, no empty V projections are licensed, and no V-Raising occurs. Nonetheless,
as observed earlier, all requirements that would normally compel raising are met in
this structure.50
7.4. V′ Reanalysis Again

The version of X-bar theory adopted here provides a rather natural motivation for
V′ Reanalysis, which was introduced earlier in connection with “Heavy NP Shift”
and which played an important part in the analysis of Case assignment with dou-
ble objects. Consider two alternative characterizations of the notion “transitive
predicate”:
(82) a. α is a transitive predicate = df α has one unsaturated internal argument
b. α is a transitive predwcate = df α is an X0 category
The first definition is made available by θ-theory. In terms of thematic grids, a transi-
tive predicate is one taking a direct object, that is, one determining an internal thematic
role. The second is made available by our restricted version of X-bar theory. Recall that
according to (76), heads-X0 categories-determine a subject (SpecXP) and a complement;
hence, the notion of head and transitive predicate fall together.
Now of course definitions (82a) and (82b) do not coincide precisely. In fact, they
may fail to coincide in one of two ways: a head may not determine even one internal
θ-role, as with “unergative” verbs like run and sneeze. This results in a D-Structure
form with less than the full X-bar characterization of “transitive predicate.” Alterna-
tively, a head may determine more than one internal a-role, as with dative verbs like
give and worry. This results in some proper projection of V (for example, a V′ like give
to Mary) meeting the a-theoretic characterization of “transitive predicate” rather than
V itself.
In the first case general principles appear to be operating that freely allow unergative
verbs to be “thematically reconstrued” as transitives. The result is the appearance of
so-called cognate objects, as in run a race, jump a mighty jump, sneeze a little sneeze,
die a painful death:
(83) Unergatives: Cognate Object Formation

VP VP
NP V′ NP V′
John V YP John V NP
die die a painful

death
Thematic intransitive Thematic transitive
Categorial transitive Categorial transitive
I would like to suggest that V′ Reanalysis represents something like the counterpart
of cognate object formation for the second case. Whereas unergative verbs like run and
die undergo an “adjustment” in the thematic structure of V to match its status as a lexi-
cal category (V0), verbs like give and worry undergo an adjustment in category to match
the status of V as a thematic transitive:
(84) Ditransitives: V′ Reanalysis

V′ V
V PP V PP
give to him give to him
Thematic transitive Thematic transitive

Categorial nontransitive Categorial transitive
In both instances the outcome is the same: the θ-theoretic and X-bar-theoretic notions
of transitive predicate are “realigned.”
NOTES
I thank Joseph Aoun, Mark Baker, Greg Carlson, Hyon Sook Choe, Noam Chomsky, Yoshio
Endo, Alessandra Giorgi, Jacqueline Gueron, James Higginbotham, Kyle Johnson, Richard
Kayne, William Ladusaw, Marta Lujan, Alec Marantz, David Pesetsky, Tim Stowell, and two
anonymous Linguistic Inquiry (LI) reviewers for comments on earlier versions of this work, as
well as audiences at the University of Massachusetts, Amherst; MIT; University of Quebec at
Montreal; and SUNY at Stony Brook. Preparation of this manuscript was supported by the MIT
Center for Cognitive Science under grants from the A. P. Sloan Foundation’s particular program
in cognitive science.
1. The judgments here are Barss and Lasnik’s. Not all of the unacceptable sentences in (3) merit
a full star in my idiolect; however, in each case the contrast observed by Barss and Lasnik is
notable.
2. Kayne (1983a) and Czepluch (1982) argue that double objects involve empty PP structure:
(i) a. VP b. VP
V V PP NP2
PP NP2
P NP1
P NP1
e
e
(Kayne) (Czepluch)
Whatever their other virtues, these proposals are clearly no improvement on (4a,b) with respect
to the facts observed by Barss and Lasnik; (ia,b) predict that NP2 should asymmetrically
c-command NP1, which is incorrect.
The structure (ia) assumed by Kayne (1983a) does contain one very important element
that is incorporated in the account developed here: in double object structures the comple-
ment of V is clauselike, with NP1 having an “inner subject” status.
3. This analysis is developed more extensively in Fillmore (1965).
4. These observations also bear on other, nondative multiple complement constructions:
(i) a. Mary sent John packing.
b. Mary took John { for
in
marriage (archaic) }
granted .
c. Mary put John { }
through the wringer/his paces .
to work/sleep
The idiomatic status of the indicated elements argues that they too form an underlying
constituent.
A very few dative structures of the form V + NP + PP do seem to involve genuine V + NP
idioms. For example:
(ii) a. Our ignorance [gave way] to enlightenment.
b. Mary [gave birth] to a bouncing baby boy.
c. This event [gave rise] to a lot of trouble.
However, although these are unexpected thematically from the point of view of (8), they
are easily accommodated syntactically in a way that preserves the constituent status of the
relevant idiomatic portion. Under the analysis of Chomsky (1955/1975), idioms like those
in (iia–c) can be listed as basic Vs (give way, for example, would be comparable to the
simplex verb yield). On the other hand, idioms like those in (l0a–d) (send to the showers,
throw to the wolves, take into consideration, and so on can be listed as basic V’s. Under
(6a,b) only the former can be accommodated in a way that preserves the constituent status
of the idiom.
5. The basic analysis of Chomsky (1955/1975) and Fillmore (1965), in which dative comple-
ments are analyzed as more intimate arguments than direct objects, is adopted in the “Right
Wrap” Categorial Grammar analyses of Dowty (1978) and Bach (1979) (the latter explicitly
acknowledges the connection to Chomsky (1955/1975)). In these analyses the surface form
of give a book to Mary arises, not by extraposing the prepositional phrase rightward, but
by wrapping the phrase give to Mary around its object a book so that the verb ends up first.
Jacobson (1983, 1987) gives a Generalized Phrase Structure Grammar translation of the
Right Wrap accounts using a V-Raising structure similar to what is assumed here. (Jacobson
does not, however, assume the subject–predicate form for VP adopted here, nor any of the
proposals about phrase structure that underlie it.)
6. See note 49 for a somewhat different motivation for V-Raising.
7. Roberts (1985) articulates this notion under a theory of “V-visibility,” whereby tense and
agreement information have much the same status for V-projections that Case has for nomi-
nal projections. Both fall under an extended “inflectional filter” that requires them to be
marked in an appropriate way.
8. I assume a definition of government involving c-command in the sense of Reinhart (1979).
9. This derivation is also sanctioned under proposals in Chomsky (1986a). Once V raises to [v e],
the lower VP will be L-marked and no barrier to movement or Case assignment.
10. This general analogy is noted by Dowty (1982) and Jacobson (1987).
11. An LI reviewer points out that this account of (18), when extended to examples like (i),
entails that a letter to Mary in the morning and a note to Max during the afternoon must be
constituents:
(i) I wrote a letter to Mary in the morning and a note to Max during the afternoon.
This reasoning is correct. Under the semantic analysis of adverbs assumed here (essentially
that of McConnell-Ginet (1982)), adverbs are not the outermost adjuncts of V but rather
its innermost complements. The underlying structure of the first conjunct of (i) is thus (ii),
where write combines with the adverb before either the direct or the indirect object phrase,
and where the correct surface form arises by iterated V-Raising:
(ii) VP
SpecV′ V′
V VP
e NP V′
a letter V VP
e PP V′
to Mary V PP
write in the morning
12. Johnson (1987) and Giorgi (1987) argue independently for a definition of governing category
in terms of CFC.
13. The subject status of complements illustrates an interesting feature of the present analysis
regarding “anaphor orientation.” As is well known, English differs from a number of other
languages (for example, Danish and Icelandic) in permitting either an IP subject or an object
to serve as antecedent for an anaphor in oblique position (Danish example (ib) from Pica
(1987)):
(i) a. John showed Mary to himself/herself.
b. *Jeg fortæller Jorgeni om sigi.
I told Jorgen about himself
This fact is often described by saying that although other languages have “strict subject
orientation” in their anaphors, English does not.
On the account sketched above, the term “strict subject orientation” becomes somewhat
misleading since under a V-Raising analysis, certain objects—that is, NPs governed by V at
some stage in the derivation—are also subjects, specifiers of the maximal phrase VP. The
analysis is thus compatible with the view that “subject orientation” is not a property of ana-
phors in particular languages but rather a universal property, and that cross-linguistic varia-
tion is not located in the grammatical function of potential antecedents but follows from some
other difference. This result appears compatible with work by Pica (1987), who suggests that
“strict subject orientation” arises from the full versus defective phrasal status of the anaphoric
elements in question. Under Pica’s analysis, for example, the defective phrasal status of sig in
Danish (versus himself in English) forces it to move to Infl (at Logical Form), where it can take
only the IP subject as antecedent, resulting in reduced binding possibilities.
14. The analyses proposed by Chomsky (1955/1975) and Fillmore (1965) make available a
nonmovement analysis of Heavy NP Shift similar to that given here. In particular, Heavy NP
Shift can be viewed as arising just when Extraposition (or “Separation”) of the inner verbal
complement does not occur. Similarly, under Categorial Grammar analyses involving Right
Wrap (Dowty 1979, Bach 1979), Heavy NP Shift can be taken to arise when a transitive
verb phrase (TVP) is analyzed as a basic lexical unit; Right Wrap then applies vacuously to
yield a concatenation of the latter to its object. Jacobson’s (1983, 1987) Generalized Phrase
Structure Grammar analysis also makes use of complex predicate raising similar to that
assumed here (although she appeals to a syntactic category TVP, which has no status in
the present account). Finally, Haegeman and Van Riemsdijk (1986) argue for reanalysis of
verbal projections and verb projection raising in a number of Germanic languages.
15. Such an analysis of examples like (22a) is in fact proposed explicitly in Chomsky (1955/1975).
16. This analysis of “NP Shift” and its consequences for parasitic gap phenomena are explored
in Larson (1989).
17. See Herslund (1986) for remarks to this effect.
18. These remarks only imply, of course, that one of the two dative constructions-oblique or
double object-should be derived from the other. See Johns (1984) for an argument that in
certain Eskimoan languages the double object form is basic and the oblique structure derived.
Dryer (1987) attempts (unconvincingly in my view) to argue a similar position for English.
19. See sections 5.1 and 6 for more on the Case-marking status of to.
20. The proposal of Argument Demotion is closely analogous to Keenan’s (1975) formulation
of relational advancement. According to Keenan, rather than being ousted from its gram-
matical relation by an advancing phrase, an argument undergoes spontaneous demotion in
its grammatical relation and so permits the latter to advance.
21. Promise is treated here as a double object verb, with the infinitive corresponding to the
underlying direct object. The relevance of this analysis for control properties of promise is
explored in Larson (1991).
22. Further arguments that “Heavy NP Shift” does not involve A-movement of NP are given in
Larson (1989). In addition to its relevance for “Heavy NP Shift,” Norwegian also provides
a clue about why inner object extraction is unavailable in English. Note that just as Nor-
wegian differs from English in the A-moveability of the inner object, it also differs in the
A-moveability of the outer object (examples from Christensen (1982)):
(i) a. Barnai ble overrakt ti blomstene.
the childreni were handed ti the flowers
b. Blomstenei ble overrakt barna ti.
*the flowersi were handed the children ti
In section 4 English examples like *The flowers were handed the children are analyzed
as involving a Case theory violation. Essentially, when the outer object moves to subject
position, the structural Case assigned to the inner object is suppressed, leaving the chil-
dren Caseless. The well-formedness of (ib) in Norwegian indicates that the latter has some
stronger means for Case-assigning the inner object-for example, that both the inner and the
outer object receive an inherent Case in Norwegian double object structures. This in tum
suggests that the well-formedness of extraction of inner objects in Norwegian versus its ill-
formedness in English may be traced to the stronger Case identification of the inner object
in the former. I hope to develop this analysis in detail elsewhere.
23. These remarks do not imply, of course, that Dative Shift or its equivalents cannot be mor-
phologically marked. Such verbal inflection does indeed show up in languages with so-called
applicative constructions (see examples in (61)). This inflection is not parallel to the -en
marking of passive, however; rather than being associated with the demoted argument, it
records the thematic role that the promoted argument bears. I propose in section 5.2 that
this function is precisely what allows for the greater productivity of these constructions as
compared with English Dative Shift.
24. As an anonymous LI reviewer points out, the nonoptionality of the theme in double object
constructions must be understood modulo the lexical properties of specific verbs. For exam-
ple, the verb write has the specific property of allowing its theme to remain implicit:
(i) a. Bill wrote a long letter to his mother.
b. Bill wrote his mother a long letter.
c. Bill wrote to his mother.
Accordingly, we expect, and find, a corresponding double object form in which the theme is
absent:
(i) d. Bill wrote his mother.
As the reviewer also points out, write differs in this respect from verbs like pay and serve,
which allow the theme to be implicit only when the latter has been demoted to adjunct status:
(ii) a. Fred paid the ransom to the agent.
b. Fred paid the agent the ransom.
c. *Fred paid to the agent.
d. Fred paid the agent.
(iii) a. Patty served two desserts to the fat man.
b. Patty served the fat man two desserts.
c. *Patty served to the fat man.
d. Patty served the fat man.
For more on the interaction of Dative Shift and “Object Deletion,” see Dowty (1979).
25. An alternative view of θ-role assignment to -en versus to the by-phrase would be to take the
former as θ-role assignment in the lexicon (parallel to Rizzi’s (1986) account of small pro)
and the latter as θ-role assignment in the syntax. Our principle would then require demoted
θ-roles to be assigned to an adjoined element at whatever level assignment takes place.
26. Technical details aside, this view of Case assignment to the outer object is essentially identi-
cal to that suggested in Marantz (1984).
27. Jaeggli (1986) makes a related proposal about Objective Case assignment, appealing to two
structural Cases rather than a structural and an inherent Case.
28. This proposal suggests an appealing general view of Case systems: we might picture the
structural or grammatical Cases (nominative, accusative) as determined uniformly by Infl,
and the inherent or semantic Cases (objective, dative, benefactive, and so on) as determined
strictly by V:
(i) Nom Acc
NP I NP
V NP t NP . . .
Obj Dat
Under this idea, the direct object would be, in effect, a position where the two Case systems
intersect—a position where two Cases, accusative and objective, are “superimposed” on a
single argument.
This may clarify a number of empirical issues. Belletti and Levin (1985) have argued that
verbs taking an external argument always have a “direct Case” to assign, even when they
appear to determine an oblique complement as a matter of lexical properties. They argue
that with verbs like talk, which determine a dative complement (John talks to Mary), the
preposition obligatorily reanalyzes (or incorporates) as part of V to permit Objective Case
assignment:
(ii) John [Infl [vP talk-to Mary]]
Given the above remarks, we might understand the obligatory direct Case detected by Belletti
and Levin as the structural, accusative Case determined by Infl. The situation in (ii) is then
as follows: the verb talk determines an inherent oblique Case for its object; however, Infl
determines a structural objective Case that also must be assigned to the object. These two
competing demands are reconciled by reanalysis/incorporation of P: the inherent oblique
Case component determined by talk is expressed; however, the reanalyzed complex assigns
objective Case under government by Infl.
The scheme in (i) may also shed light on the phenomenon of “ergative splits,” where the
clitic (or agreement) morphemes associated with Infl exhibit a nominative-accusative organi-
zation, whereas the independent nominal arguments show an ergative-absolutive paradigm.
(The Australian aboriginal language Warlpiri is an example (Hale 1973; Jelinek 1984).) We
might view such systems as resulting from a strict separation of the structural and inherent
Case systems by means of clitics:
(iii) Nom Acc Erg Abs Dat
CI I CI NP V NP t NP . . .
The nominative-accusative organization of the clitic morphemes would then reflect the (universal)
demands of the structural Case, whereas the ergative-absolutive-dative organization of the overt
arguments would reflect the inherent, semantic Case assignments determined by V.
29. This view is compatible with the proposal of Rouveret and Vergnaud (1980) that a passive
participle loses its Case-assigning possibilities by virtue of its derived adjectival status. We
might say that because adjectives are Case-receiving categories, any Case transmitted to
them is absorbed and not passed along.
30. Stowell (1981) and Czepluch (1982) also propose analyses in which dative passives receive
a direct derivation. However, the relations among passives, double objects, and dative pas-
sives that result under their accounts differ sharply from those determined here.
31. Cliticization may also be responsible for another familiar fact about pronouns in double
object contexts, namely, that if the outer object is pronominal, then the inner object must be
pronominal (and unstressed) as well:
(i) a. i. *?I sent my father it yesterday.
ii. ?I sent’im it yesterday.
b. i. *?Give Felix it!
ii. ?Gimme it!
Suppose that English (non-wh) objective pronouns must occur adjacent to a lexical Case
assigner, regardless of whether the Case they bear is structural or inherent. This condition
is not met in (iai) and (ibi) since my father and Felix intervene. In (iaii) and (ibii), however,
if it and me are analyzed as having undergone cliticization, as their reduced form suggests,
then it will be adjacent to the V + Cl complex, satisfying the adjacency condition.
If the cliticization strategy discussed above does indeed exist, it evidently must be avail-
able only as a “last resort.” Cliticization cannot be allowed to save potential Case Filter
violations in examples like (ii) where an alternative, licit derivation exists:
(ii) *It was killed’im.
(compare He was killed.)
32. Other languages for which “3 → l” advancement has been proposed include Cebuano (Bell
1983) and Kinyarwanda (Gary and Keenan 1977). For discussion of this issue within the
framework of Relational Grammar, see Perlmutter and Postal (1983).
33. Belletti and Rizzi (1988) do not actually draw this conclusion, although it is compatible with
the structure they assign to psych verbs of the preoccupare class.
34. David Pesetsky has pointed out to me that Belletti and Rizzi’s analysis parallels certain
Relational Grammar accounts of psych verbs, where the latter have only 2 and 3 arcs, and
where “3 → I” advancement takes place. Given the preceding discussion, this parallelism is
not surprising.
35. See Belletti and Rizzi (1988) for discussion.
36. Belletti and Rizzi’s account of verbs like worry and frighten can be carried over almost
without modification into the analysis proposed here. As it turns out, psych verbs do not
(or, at any rate, need not) involve V-Raising (see section 7.3). Nonetheless, they satisfy
all principles that normally force V-Raising. Thus, in the structure in (53) worry heads
a VP governed by Infl, as required. Furthermore, inherent Objective Case assignment to
NP* proceeds in a way fully analogous to what occurs with the outer object in a double
object construction: NP* is sister to a V′ whose θ-grid contains a single unsaturated internal
argument. The latter is thus subject to V′ Reanalysis and able to Case-mark NP* (see the
discussion of (37)).
37. A related proposal is made by Givón (1984). See also the argument of Culicover (1982)
that the availability of Dative Shift depends on the particular semantic contribution made
by to.
38. Green (1974) points out that resistance to Dative Shift is not a fact about verb-particle
constructions per se. Other such combinations do permit Dative Shift:
(i) a. I will send off a letter to them in the morning.
I will send them off a letter in the morning.
b. Pick out a coat for me.
Pick me out a coat.
(Green 1974, p. 82)
The latter is a for-dative alternation; the former involves send and a directional adverbial.
(ia) is particularly interesting because of its contrast with (58a,b). Off does not appear to
affect the goal status of the indirect object, with the result that send off is largely synony-
mous with send. The same is not true, of course, with give versus give away.
39. In a similar vein, Pinker (1984) discusses data from the acquisition of dative constructions by
children and suggests that the dative alternation is internalized by children as a productive
process “whose domain of application is partially constrained” (p. 322). He further proposes
that “successful avoidance of ungrammatical forms is a consequence of eventually acquir-
ing appropriate constraints on these rules” (p. 322), where, for Pinker, these constraints are
both phonological and semantic-thematic in character.
40. See also Givón (1984). It is interesting to note that Bantuists commonly refer to applicative
forms as “voices” of the verb; thus, one speaks of the dative, instrumental, and locative
“voice,” and so on. This suggests an at least implicit recognition of the connection between
passives and double object constructions pursued explicitly here.
41. This view does not require applicative morphemes to be analyzed as independent θ-role
assigners—prepositions that have been “merged with” or “incorporated into” V—as in
Marantz (1984) and Baker (1985). Thus, there is no necessary expectation that applica-
tive affixes will show synchronic or diachronic relations with prepositions, nor even that
applicative affixes will be forbidden from appearing in oblique constructions. These results
appear to be empirically correct: applied affixes and their prepositional counterparts often
show no morphological relatedness (see (6la–d)); applied affixes often derive historically
from nonprepositional sources (such as reduced verbs; K. Hale (personal communication));
and applicative morphology can in fact occur redundantly in the oblique construction in
languages with applicatives. Chung (1976) points out examples like (ia,b), in which the
applied affix kan and the benefactive preposition kepada co-occur
(i) a. Laki2 itu meng-irim(-kan) surat kepada wanita itu.
man the trans-send-Ben letter to woman the
‘The man sent a letter to the woman.’
b. Anak laki2 itu mem-bajar(-kan) lima dolar kepada polisi itu.
child male the trans-pay-Ben five dollar to police the
‘The boy paid five dollars to the policeman.’
42. In the literature Burzio’s Generalization is generally stated as requiring that a verb assign
Case to its object iff it assigns a thematic role to its subject:
(i)
V assigns Case ↔ V assigns θ
However, this formulation is problematic if, as argued by Marantz (1984), verbs do not
assign thematic roles to subjects, but rather verb phrases do. Given Marantz’s claim, a more
accurate formulation would appear to be (ii):
(ii) V assigns Case ↔ Vi assigns θ
This revision allows for an interesting view of the intuitive content of Burzio’s Generalization.
As stated in (ii), Burzio’s Generalization can be seen as establishing a correlation between two
distinct notions of “predicate”—in particular, between the notion “syntactically well-formed
predicate” (one that integrates an object phrase in the grammatically licensed way) and the
notion “semantically well-formed predicate” (one that assigns a θ-role).
43. Nothing in this proposal hinges on whether the absence of the relevant preposition for
spare reflects historical accident or deeper facts about the “space” of human prepositional
concepts.
An anonymous LI reviewer suggests that the explanation for why spare does not permit
NP-PP complementation may be weakened by the existence of the verb deny. Deny takes a
dative argument that superficially does not seem to be a goal:
(i) This law denies to felons the right to vote.
In fact, however, the nongoal status of the to-object is not clear. Dictionaries standardly
define deny in the sense intended in (i) in terms of a dative-that is, to ‘deny’ is to ‘refuse to
grant or give’:
(ii) The law refuses to grant to felons the right to vote.
This close relation between give and deny also underlies the intended contrast in (iii):
(iii) You give her everything and deny her nothing.
It thus seems plausible to conjecture that the to-object of deny actually does bear the goal role
by virtue of the implicit dative relation “contained within” or entailed by the meaning of deny.
44. Affectedness appears to play a role in a variety of cases involving dative alternations. As
pointed out to me by K. Johnson, Oehrle (1976) notes that in pairs of examples like the fol-
lowing the second more strongly carries the implication that the students have actually learned
the subject matter:
(i) a. Max taught French to the students.
b. Max taught the students French.
We might understand this as reflecting the affected-object status of the students in (ib) versus
(ia). Canonically, one is affected by teaching insofar as one learns. Consider also for-dative
alternations of the following kind noted by Kayne (1975):
(ii) a. I knitted this sweater for our baby.
b. I knitted our baby this sweater.
Although (iia) is perfectly acceptable as an utterance by a pregnant wife to her husband,
the second is decidedly odd in this context because it appears to require the baby’s present
existence. Again, we can understand this judgment as resulting from the fact that our baby
occupies the position of affected arguments in (iib), and only extant individuals can be
affected.
45. I am grateful to both anonymous LI reviewers for directing me to these issues, and to one
in particular for discussion of the facts in (68).
46. Kayne (1981, 1983a) also takes the contrast in prepositional Case assignment in French and
English to explain the absence of double objects in the former. However, the use he makes
of this idea is quite different from what is assumed here.
47. The version of X-bar theory in (76) is closely related to Montague’s (1974) use of curried
functions and effectively embodies Kayne’s (1983b) binary branching requirement.
48. The thematic hierarchy is essentially the thematic relations counterpart of the Relational
Hierarchy of Perlmutter and Postal (1983).
49. This result suggests an alternative to our earlier assumption that V-Raising is forced by the
demands of Case assignment and tense agreement (see the discussion of (14)). Given the
above remarks, it becomes possible to motivate V-Raising through a requirement on the
mapping of categorial and thematic structure: each argument must be governed by its head
at some derivational stage.
The considerations adduced for give apply equally to any three-argument verb, including,
for example, put and talk. These too will involve VP complementation structures:
(i) a. John [V′ put [VP a fly [V′ t in the soup]]]
b. John [V′ talked [VP to Felix [V′ t about Mary]]]
Similarly for a two-argument verb that takes an adverbial modifier. According to the thematic
hierarchy, adverbials and obliques will be projected in the innermost complement position
(see also note 11); this will force the creation of a VP shell with subsequent V-Raising:
(i) c. John [V′ saw [VP Mary [V′ t recently]]]
These principles will also dictate the projection of multiple levels of VP complementation,
and iterated V-Raising, when a three-argument verb itself takes a modifier:
(i) d. John [v′ sent [ VP a note [V t [ VP to Max [V t on Tuesday]]]]]
′ ′
50. Recall that the right-peripheral position of the specifier of VP (NP*) permits V′ Reanalysis
and Case assignment identical to what occurs with double objects.
REFERENCES
Allerton, D. (1978) “Generating Indirect Objects in English,” Journal of Linguistics 14: 21–33.
Aoun, J. and D. Sportiche (1983) “The Formal Theory of Government,” The Linguistic Review
2: 211–236.
Baker, M. (1985) Incorporation: A Theory of Grammatical Function Changing. Doctoral disserta-
tion, MIT, Cambridge, MA.
17: 347–354.
Bell, S. (1983) “Advancements and Ascensions in Cebuano,” in D. Perlmutter, ed., Studies in Rela-
tional Grammar I. (pp. 143–218) Chicago: University of Chicago Press.
Belletti, A. and B. Levin (1985) “On VP-PP Constructions,” unpublished manuscript MIT,
Cambridge, MA.
Belletti, A. and L. Rizzi (1988) “Psych-verbs and Theta-theory,” Natural Language and Linguistic
Theory 6: 291–352.
Burzio, L. (1986) Italian Syntax. Dordrecht: Reidel.
Carrier-Duncan, J. (1985) “Linking of Thematic Roles in Derivational Word Formation,” Linguis-
tic Inquiry 16: 1–34.
Chomsky, N. (1955/1975) The Logical Structure of Linguistic Theory. Chicago: University of
Chicago Press.
———. (1970) “Remarks on Nominalization,” in R. Jacobs and P. Rosenbaum, eds., Readings in
English Transformational Grammar. (pp. 184–221) Waltham, MA: Ginn.
———. (1981) Lectures on Government and Binding. Dordrecht: Foris.
———. (1982) Some Concepts and Consequences of the Theory of Government and Binding.
———. (1986a) Barriers. Cambridge, MA: MIT Press.
———. (l986b) Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger.
Christensen, K. (1982) “On Multiple Filler-Gap Constructions in Norwegian,” in E. Engdahl
and E. Ejerhed, eds., Readings on Unbounded Dependencies in Scandinavian Languages.
(pp. 77–98) Umea Studies in the Humanities 43. Stockholm: Almqvist & Wiksell.
———. (1987) “Modem Norwegian Ingen and the Ghost of an Old Norse Particle,” in R.D.S.
Allen and M.P. Barnes, eds., Proceedings of the Seventh Biennial Conference of Teachers if
Scandinavian Studies in Great Britain and Northern Ireland, (pp. 1–17). London: University
College London.
Chung, S. (1976) “An Object-Creating Rule in Bahasa Indonesia,” Linguistic Inquiry 7: 41–87.
Culicover, P. (1982) Syntax. New York: Academic Press.
Czepluch, H. (1982) “Case Theory and the Dative Construction,” The Linguistic Review 2: 1–38.
Dowty, D. (1978) “ Rules in a Montague Grammar,” Linguistic Inquiry 9: 393–426.
———. (1979) “Dative ‘Movement’ and Thomason’s Extensions of Montague Grammar,” in S. Davis
and M. Mithun, eds., Linguistics, Philosophy and Montague Grammar. (pp. 153–222) Austin:
University of Texas Press.
———. (1982) “Grammatical Relations and Montague Grammar,” in P. Jacobson and G. K. Pullum,
eds., The Nature of Syntactic Representation. (pp. 79–130) Dordrecht: Reidel.
Dryer, M. (1987) “On Primary Objects, Secondary Objects and Antidative,” Language 62: 808–845.
Emonds, J. (1972) “Evidence That Indirect Object Movement Is a Structure-Preserving Rule,”
Foundations of Language 8: 546–561.
Engdahl, E. (1983) “Parasitic Gaps,” Linguistics and Philosophy 6: 5–34.
Feldman, H. (1978) “Passivizing on Datives in Greek,” Linguistic Inquiry 9: 499–502.
Fillmore, C. (1965) Indirect Objects and the Ordering of Transformations. Mouton Freidin, R. and
L. Babby (1984) “On the Interaction of Lexical and Syntactic Properties: Case Structure in
Russian,” in Cornell Working Papers in Linguistics 6. (pp. 71–104) Ithaca, NY: Cornell Uni-
versity, Ithaca, New York.
Gary, J. and E. Keenan (1977) “On Collapsing Grammatical Relations in Universal Grammar,”
in P. Cole and J. Sadock, eds., Syntax and Semantics 8: Grammatical Relations. New York:
(pp. 83–120), Academic Press.
Giorgi, A. (1984) “Toward a Theory of Long Distance Anaphors,” The Linguistic Review 3:
307–361.
———. (1987) “The Notion of Complete Functional Complex: Some Evidence from Italian,”
Givón, T. (1984) “Direct Object and Dative Shifting: Semantic and Pragmatic Case,” in F. Plank,
ed., Objects. (pp. 151–182), London: Academic Press.
Green, G. (1974) Semantics and Syntactic Regularity. Bloomington: Indiana University Press,
Bloomington.
Haegeman, L. and H. van Riemsdijk (1986) “Verb Projection Raising, Scope, and the Typology of
Rules Affecting Verbs,” Linguistic Inquiry 17: 417–466.
Hale, K. (1973) “Person Marking in Walbiri,” in S. Anderson and P. Kiparsky, eds., A Festschrift
for Morris Halle. (pp. 308–344) New York: Holt, Rinehart and Winston.
Hawkins, R. (1981) “On ‘Generating Indirect Objects in English’: A Reply to Allerton,” Journal
of Linguistics 17: 1–9.
Herslund, M. (1986) “The Double Object Construction in Danish,” in L. Hellan and K. Chris-
tensen, eds., Topics in Scandinavian Syntax. (pp. 125–147) Dordrecht: Reidel.
Jacobson, P. (1983) “Grammatical Relations,” paper presented at the 1983 Winter meeting of the
Linguistic Society of America, Minneapolis, MN.
———. (1987) “Phrase Structure, Grammatical Relations and Discontinuous Constituents,” in
G. Huck and A. Ojeda, eds., Syntax and Semantics 20: Discontinuous Constituency. (pp. 27–69)
New York: Academic Press, New York.
Jaeggli, O. (1986) “Passive,” Linguistic Inquiry 17: 587–622.
Jelinek, E. (1984) “Empty Categories, Case and Configurationality,” Natural Language and
Linguistic Theory 2: 39–76.
Johns, A. (1984) “Dative ‘Movement’ in Eskimo,” in D. Testen, V. Mishra, and J. Drogo, eds.,
Papers from the Parasession on Lexical Semantics. (pp. 162–172) Chicago Linguistic Society,
Johnson, K. (1987) “Against the Notion, ‘SUBJECT,’” Linguistic Inquiry 18: 354–361.
Kayne, R. (1975) French Syntax. Cambridge, MA: MIT Press.
———. (1981) “On Certain Differences between French and English,” Linguistic Inquiry 12: 349–371.
———. (l983a) “Datives in French and English,” in Connectedness and Binary Branching.
(pp. 193–202) Dordrecht: Foris.
———. (1983b) “Unambiguous Paths,” in Connectedness and Binary Branching. (pp. 129–164)
Dordrecht: Foris.
Keenan, E. (1975) “Some Universals of Passive in Relational Grammar,” in R. Grossman, L. J. San,
and T. Vance, eds., Proceedings of the Eleventh Regional Meeting of the Chicago Linguistic
Society. (pp. 340–352) Chicago: University of Chicago
Kimenyi, A. (1980) A Relational Grammar of Kinyarwanda. Berkeley: University of California Press.
Kitagawa, Y. (1986) Subjects in Japanese and English, doctoral thesis. University of Massachusetts-
Amherst.
Koopman, H. (1986) “On Deriving Deep and Surface Order,” in C. Jones and P. Sells, eds., Papers
from the Fourteenth Annual Meeting of NELS. (pp. 220–235) Amherst, MA: GLSA, University
of Massachusetts-Amherst.
Kuroda, Y. (1988) “Whether We Agree or Not: A Comparative Syntax of English and Japanese,”
Linquisticae Investigationes 12: 1–47.
Larson, R. K. (1991) “Promise and the Theory of Control,” Linguistic Inquiry 22: 103–139.
———. (1989) “Light Predicate Raising,” Lexicon Project Working Papers No, 27. Center for
Cognitive Science. Cambridge, MA: MITWPL.
McConnell-Ginet, S. (1982) “Adverbs and Logical Form,” Language 58: 144–184
Montague, R. (1974) “The Proper Treatment of Quantification in Ordinary English,” in
R. Thomason, ed., Formal Philosophy. (pp. 247–270) New Haven, CT: Yale University Press.
Oehrle, R. (1976) The Grammatical Status of the English Dative Alternation. Doctoral disserta-
tion, MIT, Cambridge, MA.
———. (1983) “The Inaccessibility of the Inner NP: Corrections and Speculations,” Linguistic
Analysis 12: 159–171.
Perlmutter, D. (1978) “Impersonal Passives and the Unaccusative Hypothesis,” in Jeri J. Jaeger, ed.,
Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society, (pp. 157–190).
Berkeley: University of California.
Perlmutter, D., ed. (1983) Studies in Relational Grammar 1. Chicago: University of Chicago Press.
Perlmutter, D. and P. Postal (1983) “Some Proposed Laws of Basic Clause Structure,” in D. Perlmut-
ter, ed., Studies in Relational Grammar 1. (pp. 81–128) Chicago: University of Chicago Press.
Perlmutter, D. and C. Rosen, eds. (1984) Studies in Relational Grammar 2. Chicago: University
of Chicago Press.
Pica, P. (1987) “On the Nature of the Reflexivization Cycle,” In J. McDonough and B. Plunkett, eds.,
Proceedings of the 17th Annual Meeting of the North East Linguistic Society, (pp. 483–499).
Amherst, MA: GLSA. University of Massachusetts-Amherst.
Pinker, S. (1984) Language Learnability and Language Development. Cambridge, MA: Harvard
University Press.
Reinhart, T. (1979) “Syntactic Domains for Semantic Rules,” in F. Guenthner and S. Schmidt, eds.,
Formal Semantics and Pragmatics for Natural Language. (pp.107–130) Dordrecht: Reidel.
Rizzi, L. (1986) “Null Objects in Italian and the Theory of pro,” Linguistic Inquiry 17: 501–557.
Roberts, I. (1985) “Agreement Parameters and the Development of English Modal Auxiliaries,”
Rouveret, A. and J.-R. Vergnaud (1980) “Specifying Reference to the Subject: French Causatives
and Conditions on Representations,” Linguistic Inquiry 11, 97–202.
Sag, 1., G. Gazdar, T. Wasow, and S. Weisler (1985) “Coordination and How to Distinguish
Categories,” Natural Language and Linguistic Theory 3: 117–171.
Shimizu, M. (1975) “Relational Grammar and Promotion Rules in Japanese,” in R. Grossman,
L. J. San, and T. Vance, eds., Proceedings of the Eleventh Regional Meeting of the Chicago
Linguistic Society . (pp. 529–535), University of Chicago.
Speas, M. and N. Fukui (1986) “Specifiers and Projection,” unpublished manuscript MIT,
Cambridge, MA.
Sportiche, D. (1988) “A Theory of Floating Quantifiers and Its Corollaries for Constituent
Structure,” Linguistic Inquiry 19: 425–449.
Sproat, R. (1985) “Welsh Syntax and VSO Structure,” Natural Language and Linguistic Theory
3: 173–216.
Stowell, T. (1981) Origins of Phrase Structure. Doctoral dissertation, MIT, Cambridge, MA.
Tenny, C. (1987) Grammaticalizing Aspect and Affectedness. Doctoral dissertation, MIT, Cam-
bridge, MA.
Travis, L. (1985) Parameters and Effects of Word Order Variation. Doctoral dissertation, MIT,
Cambridge, MA.
Whitney, R. (1983) “The Syntactic Unity of Wh-Movement and Complex NP Shift,” Linguistic
Analysis 10: 299–319.
Ziv, Y. and G. Scheintuch (1979) “Indirect Objects Reconsidered,” in P. Clyne, W. Hanks, and
C. Hofbauer, eds., Proceedings of the Fifteenth Regional Meeting of the Chicago Linguistic
Society. (pp. 390–403), University of Chicago.
Zubizarreta, M.-L. (1985) “The Relation between Morphophonology and Morphosyntax,”
2 Double Objects Revisited
Reply to Jackendoff
Richard K. Larson
Jackendoff (1990) discusses the analysis of double objects proposed in Larson (1988), tak-
ing issue with the general analysis and many of its specific claims. In this article I review
Jackendoff’s points concerning linear order and binding (section 1), structure projection
(section 2), complement alternations (section 3), Dative Shift (section 4), and modifiers
and nonconstituent conjunction (section 5). I argue that the prospects of the analysis in
Larson (1988) are not nearly so bleak as Jackendoff portrays them-that the questions he
raises can be answered in natural ways, and that the counterproposals he makes are not
superior to the views originally suggested.
1. LINEAR ORDER AND DOMAIN ASYMMETRY
Larson (1988) analyzes certain domain asymmetry facts observed by Barss and Lasnik
(1986) in double object examples and proposes that they can be explained directly
under a nonstandard theory of complement structure using a purely hierarchical defini-
tion of “syntactic domain.” Jackendoff argues that various additional facts necessitate
a “mixed” definition of syntactic domain, involving hierarchy and linear order. He also
suggests that linearity is natural in the account of intrasentential domains, since it is
independently needed in the account of intersentential anaphora. Jackendoff is wrong
on both counts, however. The data he cites do not establish the conclusion he draws.
Although the facts are compatible with his position, they are also compatible with the
configurational approach to domain asymmetry in Larson (1988). Moreover, Jackend-
off’s argument from intersentential anaphora involves circular reasoning.
1.1. Linearity versus Hierarchy

At the beginning of his article Jackendoff tries to establish the necessity of linear order
on the basis of simple surface inspection of various paradigms. He reviews the behav-
ior of domain-sensitive items in nonalternating double objects, and alternating NP-PP
complements, in double PP complements with free order, and in double PP complements
in nominals, and he observes that, quite generally, complements that appear on the right
can be in the domain of those on the left but not vice versa. He concludes, “The over-
whelming generalization [is] that linear order plays a role in these phenomena” (p. 430).
He presents the facts as “having shown that Larson has chosen the wrong way out of the
apparent difficulties presented by Barss and Lasnik’s observations” (p. 436).
Jackendoff’s conclusion is not justified, however, and indeed cannot be reached by
simple inspection. This is because there is another possible explanation for the facts.
Granting the surface generalization that domain effects show left-right asymmetry, two
major hypotheses are available: (a) domain effects reflect linear order, in which case
the surface generalization is the true generalization, and phrase markers are potentially
quite flat; or (b) domain effects reflect only structure, in which case the surface gener-
alization is an illusion, and phrase markers are downward branching to the right. To
establish his conclusion, Jackendoff must eliminate possibility (b), but this can only be
done by a consideration of structure.
A rightward downward branching analysis is not an abstract possibility but repre-
sents precisely the approach taken in Larson (1988). In that analysis, and in contrast
to more standard views, elements appearing on the right—including obliques—are
typically lower in the phrase marker than elements to their left. As a consequence,
dependencies that superficially appear to be linear will also be describable in simple
hierarchical terms. As an illustration, consider (1a,b), involving a negative polarity item
(any day this week) and a potential trigger (few friends):1
(1) a. John visited few friends any day this week.
b. *John visited anyone few days this week.
(compare John visited someone few days this week.)
On a standard structure, in which adverbs are adjoined to VP and hence higher in

the tree than the direct object (2a), this pair would support an ordering restriction on
negative polarity licensing. Licensing would require some version of m-command by
the trigger (that is, mutual containment in maximal projections), together with linear
precedence. By contrast, on the sort of structure proposed in Larson (1988), with its
uniform rightward downward branching (2b), these facts follow directly under a simple
first branching node definition of c-command; the direct object c-commands the tempo-
ral adjunct, but not conversely. The apparent precedence restriction is merely apparent:
(2) a. IP
NP I′
John I VP
VP NP
V NP any day
this week
visit few friends
b. VP
NP V′
John Vi VP
NP V′
visit
few friends Vi NP
e any day
this week
Double Objects Revisited 193
The same general outcome holds with Jackendoff’s own data, as we will see. Thus,
in section 3.3 I argue that the alternation blame X on Y/blame Y for X involves two
distinct D-Structure forms in which the PP is uniformly lower than the direct object.
As a consequence, what appears to be a linear asymmetry is accounted for under first
branching node c-command, even ignoring the structure introduced by PP:
(3) a. VP
NP V′
John V VP
e NP V′
the V PP
accident
blame on Max
b.
VP
NP V′
John V VP
e NP V′
Max V PP
blame for the

accident
Similar results obtain with other alternating NP-PP complements (for instance,
spray-load verbs, as discussed in section 3.1), with double PPs (as discussed in sections
3.2 and 3.3), and with nonalternating double objects (as discussed in Larson (1988)).
Despite Jackendoff’s claims, none of the data he cites actually chooses between the pure
structural and the mixed structural proposals. None of them forces the conclusion that
the correct statement of syntactic domain must take account of linear order as well as
hierarchy. To secure this claim, one must look at more than strings.
1.2. Restrictiveness
If the facts cited by Jackendoff do not decide between a pure structural and a “mixed”
approach to syntactic domains, the former nonetheless has an important methodologi-
cal advantage over the latter, one that Jackendoff himself alludes to.
A purely hierarchical analysis of domain asymmetry assuming first branching node
c-command rules out many initially plausible double object structures, as discussed by
Barss and Lasnik (1986); and it compels a more complex branching configuration. In the
analysis of Larson (1988), this more complex form has numerous further consequences
for the analysis of conjunction, “Heavy NP Shift,” and discontinuous idioms.
By contrast, a notion of domain involving both structure and order entails very
few structural consequences and is in fact compatible with all possible structurings
of V–NP1–NP2. For example, if the definition of syntactic domain involves both linear
precedence and first branching node c-command, then Barss and Lasnik’s results are
compatible with both (4a) (Oehrle 1976) and (4b) (Kayne 1983). Similarly, if the defini-
tion of syntactic domain involves linear precedence and m-command, then Barss and
Lasnik’s results are compatible with (4a) and (4c) (Chomsky 1981):
(4) a. b.
give Mary a present give Mary a present

c.
give Mary a present
Thus, although the two approaches are equal on the data Jackendoff cites, they are not
equal in more general terms. The pure hierarchical analysis appears to yield a more
restrictive theory overall.
This is not a surprising result, but is in fact a rather familiar one in modern syntactic
study, where the choice between elaborating the account of specific constructions and
elaborating the content of general principles arises routinely. Although choice in such
matters is ultimately an empirical issue, it has often proven useful as a research strategy
to prefer complex structures to complex principles, simply because the former tends to
yield a more restrictive theory overall, and hence one to be preferred under the usual
logic of the language acquisition problem.2
1.3. Linearity in Discourse?

Jackendoff suggests that linearity is natural in the description of intrasentential binding
since it is independently necessary for the description of intersentential anaphora in
discourse (5) and in conjunctions (6). He rejects any strict separation of the domains
of sentence and discourse grammar that would allow the apparent linearity effects in
(5) and (6) to be dismissed as “mere pragmatics.”
(5) a. Johni came in. Hei was tired.
b. *Hei came in. Johni was tired.
(6) a. Johni came in and hei was tired.
b. *Hei came in and Johni was tired.

Jackendoff’s reasoning misses the mark in two important ways, however.
First, the syntactic domain phenomena at issue here—those discussed by Barss and
Lasnik—are quite distinct from simple cross-sentential anaphora. Though the latter
extends across nonsubordinate clauses, the former—even those with very weak local-
ity requirements like quantifier binding, each . . . the other dependencies, and negative
polarity phenomena—do not:
(7) a. *No one came in. Anyone was tired.

(negative polarity)
b. *No boy came in. He was tired. (quantifier binding)
c. *Each man came in. The other nodded. (each . . . the other)
Because of this difference, the relevance of (5)–(6) for the definition of syntactic domains
is simply indeterminate. Even if the explanation of (5b) and (6b) must appeal to linear
precedence, nothing is entailed about the phenomena discussed by Barss and Lasnik.3
Second, even if we accept that constraints on intra- and intersentential pronoun
anaphora are not disparate phenomena, we are not obliged to accept what Jackendoff
concludes from this. Indeed, it is equally plausible to draw a conclusion from (5b) and
(6b) opposite to his. Rather than taking these data as evidence for linearity in sentence
grammar, we could instead view them as evidence for hierarchy in discourse grammar.
Such a view would not be absurd. Research in Discourse Representation Theory (Kamp
1981; Heim1982) and in the processing of discourse anaphors (Cohen 1987; Grosz
1978; 1981; Grosz and Sidner 1986; Grosz, Pollack, and Sidner 1989; Guindon and
Colleagues 1986) has argued precisely that discourses are structured objects whose
hierarchical constituent relations condition discourse anaphora.
To illustrate a simple theory of discourse structure that yields these results, consider
the following three proposals relating intra- and intersentential anaphoric processes:
(a) Intrasentential anaphora between elements α, β depends on the relative hierarchi-

cal relations of α, β themselves; intersentential anaphora between α, β depends on
the relative hierarchical relations of the Ss containing α, β.
(b) Coordination structures fall under X-bar theory and have conjunctions as their
heads.
(c) In their default form, discourses are extended coordinations.
Principle (a) says, in effect, that S-internal structure is inaccessible to discourse gram-
mar-that anaphoric relations between elements in separate sentences (including separate
sentences of a sentential conjunction) are determined by the relations of the sentences con-
taining them. Principle (b) takes the view of Ross (1967) and Collins (1988) that coordina-
tions are endocentric, are headed by their conjunctions, and fall under the familiar X-bar
structure [XP YP [X′ X ZP]]; (c) is self-explanatory. Under these proposals, (5b) and (6b)
would receive the structure in (8) (borrowing the category label & from Hale (1989)):
(8) &P
S &′
he came in & S
(and)
John was tired
The relevant coreference relations might then be blocked by the equivalent of Prin-
ciple C of the binding theory: in parallel to the intrasentential case, we could say that an
S containing an R-expression cannot be c-commanded by an S containing a coreferential
phrase.
The point here is not, of course, to present and defend a full-fledged theory of dis-
course anaphora, but rather to observe that Jackendoff’s argument rests on the premise
that linearity is the right explanation for the phenomena in (5) and (6). This premise must
be defended, not simply assumed. If discourses are in fact structured objects (as the
work cited above strongly argues), then the premise is far from self-evident. As it stands,
Jackendoff’s argument is basically circular: it argues for linearity over structure (in
intrasentential relations) under the assumption of linearity over structure (in intersen-
tential relations).
2. THE V-RAISING ANALYSIS
Jackendoff finds serious fault with the V-Raising analysis proposed in Larson
(1988) questioning not only its empirical correctness but its very coherence. He
judges it to involve a radical departure from current views of θ-assignment and
D-Structure and to be unfaithful to one of its chief motivating principles: the Uni-
formity of Theta Assignment Hypothesis (UTAH). In reality, Jackendoff’s claim of
a radical departure is mistaken. Furthermore, the departure from UTAH in Larson
(1988) is no greater than elsewhere in the Extended Standard Theory (EST), and
the version of UTAH that it retains still suffices to motivate a derivational approach
to double objects.
2.1. Structure Projection in Larson (1988)

The account of structure projection in Larson (1988) makes use of three basic elements
or principles: a principle of hierarchy (9), a principle of location (10), and a principle
of direction (11):4
(9) a. XP → YP Xʹ
b. Xʹ → X ZP
(10) If β is an argument of α, then β must be realized within a projection of α.
(11) Arguments of a predicate α are projected according to the hierarchy AGENT
> THEME > GOAL > OBLIQUE, such that if θ1 > θ2 on the thematic hierarchy, then
the recipient of θ1 c-commands the recipient of θ2.
To illustrate this with the verb put, these principles yield initial VP structures like
(12), where the relative structural prominence of John, some beer, and in the cooler
reflects the relative thematic prominence of agent, theme, and location:
(12) VP
NP V′
John V VP
e NP V′
some beer V PP
put in the cooler
The presence of the empty V head follows from the X-bar theory in (9). Since maxi-
mal phrases may contain at most one complement per projection, only two of put’s
arguments can be fitted into the minimal VP that it heads; this forces an upward branch-
ing of X-bar structure to secure an argument position for the agent. This in turn brings
along an empty head position as a pure consequence of X-bar theory.5,6
2.2. A “Radical Shift in Theory”?

Jackendoff labels this account “a radical shift in theory” (p. 451), claiming that it implic-
itly abandons the view that θ-roles are assigned in D-Structure. His reasoning is as follows:
since the subject of put lies outside the lower VP at D-Structure, it cannot be θ-marked by
the verb until the latter raises into the upper VP. Jackendoff deems this a serious departure
from the θ-Criterion of Chomsky (1981), whose intuitive content (according to him) is “to
prevent NPs from acquiring θ-roles in the course of a derivation” (p. 451).
We can assess Jackendoff’s claim of a major departure by simply comparing a rep-
resentation from Larson (1988) with a more conventional structure, taking equivalent
domains of θ-assignment. (13a) is the D-Structure realization of a dative VP under the
analysis in Larson (1988); (13b) is the D-Structure realization of a double object IP
under the analysis in Chomsky (1981):7
(13) a. VP
NP V′
V VP
John
e NP V′
V PP
a book
gave to Mary
b.
IP
NP I′
John I VP
e V′ NP
V NP a book
gave Mary
These structures differ in the domain of a-assignment (VP for Larson (1988), IP for
Chomsky (1981)) and in the linear order of the object NP and V′. However, they do
not differ in the respect Jackendoff draws attention to. Just as the agent lies outside the
(smallest) VP in (13a), so it lies outside the VP in (13b). Contrary to what Jackendoff
claims, then, we might expect accounts of θ-assignment for the two phrase markers to
be largely similar.
This expectation is correct. Chomsky (1981) distinguishes two basic cases in which an
element α θ-marks an element β: one involving structures of the general form [α′ . . . α . . .
β . . . ] and one involving structures where β is the subject of α. The former is referred to
as direct θ-marking, and the latter as indirect θ-marking (see Chomsky 1981, pp. 36–38).
Assuming that the goal phrase Mary is an argument in (13b), Mary is directly θ-marked,
and John is indirectly θ-marked. The status of the theme argument a book is left somewhat
indeterminate in Chomsky (1981); however, Chomsky later generalizes the definitions of
subject and object to all NPs in the configurations [NP, XP] and [NP, X′], respectively
(Chomsky 1986b, 161). Under these criteria, a book is a subject in (13b) and hence counts
as indirectly θ-marked in this structure.
Given the homology of structure between (13a) and (13b), and given the category-
neutral character of the definitions of θ-marking and the grammatical functions, it is
clear that the account of θ-marking for (13b) will extend without modification to (13a).
The object (to) Mary will be directly θ-marked by give, whereas the two subjects John
and a book will be indirectly θ-marked.8
We see then that Jackendoff’s claim that structures like (13a) mark a “radical shift in
theory” is based on a misunderstanding of current theory. It is based on the assumption
that all θ-marking by α is required to occur within the minimal maximal projection of
α. This principle is not part of current versions of the EST.
2.3. The Status of UTAH

Jackendoff observes that although the derivational account of double objects in Larson
(1988) is in part motivated by appeal to Baker’s (1988) UTAH the analysis does not
actually conform to UTAH strictly understood since, for example, dative and double
object sentences receive distinct underlying structures despite their identical thematic
relations. Jackendoff’s point is, of course, quite correct and indeed holds much more
generally. To my knowledge, no current EST analysis holds to the strict form of UTAH-
including that of Baker (1988), the author of the hypothesis. This is clear, for example,
from the fact that all recent analyses of Passive assume distinct structures for active–
passive pairs, despite the identical thematic relations involved.
Departures from strict UTAH in Larson (1988) arise in two ways. The first involves
the projection of optional oblique arguments. Compare the VP in (2b), where visit
appears with an optional temporal phrase, with that in (14), where the temporal modi-
fier is absent:
(14) VP
NP V′
V NP
John
visit few friends
In (2b) the object few friends is realized structurally as a specifier of V′; in (14) it is
realized as a complement of V. Such a departure from strict UTAH will occur whenever
an optional argument is realized: a complement in one structure will be realized as a
specifier in the other, despite bearing the same thematic relation in both.
A second departure from UTAH involves “passive” alternations in the general sense
discussed in Larson (1988). Compare (13a), the D-Structure realization of John gave a
book to Mary, with (15), the D-Structure realization of the counterpart double object
form John gave Mary a book:
(15) VP
NP V′
John V VP
e NP V′
V′ NP
e
V NP a book
gave Mary
In (13a) the object a book is realized structurally as a specifier of V′; in (15) it is real-
ized as an adjunct of V′. Again, this departure from strict UTAH will arise whenever an
argument is demoted: a specifier in one structure will be realized as an adjunct in the
other, despite bearing the same thematic relation in both.
It is natural to ask whether a modified form of UTAH is compatible with Larson
(1988) given these results, and bearing in mind that some weakening of strict UTAH is
required under any current version of the EST. The following is embodied implicitly in
Larson (1988):
Relativized UTAH
Identical thematic relationships are represented by identical relative hierarchical

relations between items at D-Structure.
Under Relativized UTAH a set of thematic relations may be realized in formally dif-
ferent D-Structure representations. However, the latter will all share an important
property: the relative structural prominence relations between role-bearing elements
will be the same.
It is easy to see informally that the D-Structure representations in Larson (1988)
conform to Relativized UTAH. In (2b) the addition of an optional oblique has shifted
the absolute position of the theme vis-a-vis the experiencer subject in (14); however, the
relative structural prominence of the two is preserved across this difference. John asym-
metrically c-commands few friends in both. Similarly, in (15) argument demotion has
shifted the absolute position of the theme vis-a-vis the agent and goal in (13a). However,
their relative structural prominence is preserved: John asymmetrically c-commands a
book, which in turn asymmetrically c-commands (to) Mary.
Relativized UTAH is not a formal principle of Larson (1988) but rather follows
as a consequence of its principle of structure projection (11), taken together with
argument demotion. The former requires the structural prominence of arguments
to reflect the thematic hierarchy AGENT > THEME > GOAL > OBLIQUE. The
latter allows a role assigned to a specifier of X; to be assigned (up to optionality)
to an adjunct of X; Principle (11) is sufficient to guarantee that optional argument
alternations like those in (2b) and (14) meet Relativized UTAH: the two argu-
ments found in both structures preserve their relative prominence in virtue of the
thematic hierarchy. Alternations involving argument demotion like (13a) and (15)
meet Relativized UTAH as a consequence of (11) taken together with the simple
structural fact that all constituents of XP in the configuration (16) are asymmetri-
cally c-commanded by both the specifier of V′ (a) and an adjunct of V′ (13), and any
constituent asymmetrically c-commanding such a VP asymmetrically c-commands
both α and β.
(16) [VP α [V’[V’ V XP] β]]
We see then that although the strict form of UTAH is not upheld in Larson (1988) (or in
any other recent version of the EST), a slightly weaker, more flexible version is available.
The latter is not a principle of Larson (1988) but rather follows from its assumptions.
Relativized UTAH preserves the basic motivation for a derivational account of double
objects. As discussed in Larson (1988), S-Structure realizations for oblique dative and
double object examples show inverse domain relations: the theme behaves as hierarchi-
cally superior to the goal in the former, but as hierarchically inferior in the latter. Given
that the same thematic relations are involved in each, Relativized UTAH requires the
underlying relative hierarchical relations in the two to be the same. It follows that one
of the two forms—dative or double object—must be derived.9
3. COMPLEMENT ALTERNATIONS
Jackendoff brings up a number of alternations beyond those involving double objects

and notes various questions they raise for the account in Larson (1988). I take up Jack-
endoff’s points in considering the three ways that alternations in complement order and
form may arise in that analysis.
3.1. NP Movement
One source of complement alternation is VP-internal NP Movement. Pairs related this
way involve identical underlying hierarchical relations but different surface grammatical
relations. One instance is the dative-double object alternation discussed in Larson (1988).
Another instance, I believe, is the well-known spray-load alternation noted by Jackendoff
and discussed by Partee (1965), Anderson (1971), and many subsequent authors:
(17) a. John sprayed paint on the wall.
b. John sprayed the wall with paint.
(18) a. John emptied water from the pail.
b. John emptied the pail of water.
I suggest that the first member of these pairs is analogous to an oblique dative, and the second
is analogous to a double object form. Thus, (17a) involves the D-Structure form in (19a), and
(17b), the D-Structure form in (19b). The wall raises to the empty VP specifier position by
NP Movement, and spray raises to the empty head position, as in the case of double objects.
Under this proposal, we predict the domain relations that Jackendoff observes: the outer PP
object will always be in the domain of the direct object, but not conversely.
(19) a. VP
NP V′
John V VP
e NP V′
paint V PP
spray on the wall

b. VP
NP V′
John V VP
e NP V′
e V′ PP
V NP with paint
spray the wall
There is suggestive evidence from quantifier scope for the general parallel between
double object forms and spray-load verbs proposed here. As noted (first, to my knowl-
edge) by D. Lebeaux (personal communication), oblique dative-double object pairs such
as (20a,b) show an asymmetry in scope interpretation when the two complements are
quantified. Specifically, although (20a) may be understood with either the theme or goal
taking wide scope (ONE-EVERY versus EVERY-ONE), (20b) is not similarly ambigu-
ous. The latter strongly selects the reading in which the scope of the quantifiers matches
their surface order (ONE-EVERY); hence, (20b) is understood as asserting that some
one particular student is assigned all the problems:
(20) a. The teacher assigned one problem to every student.
b. The teacher assigned one student every problem.
Schneider-Zioga (1988) has noted the same asymmetry in spray-load pairs, with the
locative variant patterning similarly to the oblique dative form and the with variant
patterning similarly to the double object form:
(21) a. The worker loaded one box on every truck.
b. The worker loaded one truck with every box.
(22) a. Max sprayed some slogan on every wall.
b. Max sprayed some wall with every slogan.
(23) a. I drained one chemical from every tank.
b. I drained one tank of every chemical.
Thus, (21a) is again ambiguous with either the theme or the locative taking wide scope.
(21b), by contrast, is not ambiguous, strongly selecting the reading in which the scope of
the quantifiers matches their surface order (ONE-EVERY). And similarly for (22)–(23)
in my judgment. Dative-double object pairs thus pattern analogously to spray-load pairs
in this respect.10
As noted by Jackendoff, spray-load alternations depart from double object structures
in one obvious way. In the latter the “demoted” argument appears as a bare NP, whereas
in the former it appears in a PP headed by with or of. In fact, this does not represent a
“sharp” difference between dative and spray-load pairs. Dative alternations with sup-
ply and provide also require a preposition and do not allow for a bare outer NP (24),
and for many speakers (including myself) dative alternations with award seem to allow
either structure (25):
(24) a. The teacher supplied one pencil to every student.
The teacher supplied one student *(with) every pencil.
b. The teacher provided one pencil to every student.
The teacher provided one student *(with) every pencil.
(25) a. We awarded one prize to every contestant.
b. We awarded one contestant (with) every prize.
Given these results, it is natural to ask what conditions the distribution in (17)–(25).
Specifically, what requires the presence of a preposition, and how is the particular choice
of preposition determined? I suggest that the relevant factor in both cases is recoverabil-
ity; however, I will postpone discussion of this point until section 4.3.
3.2. Light Predicate Raising

A second source of alternation is Light Predicate Raising (LPR), the operation respon-
sible for so-called Heavy NP Shift phenomena in Larson (1988, 1989). Pairs related by
LPR involve identical underlying structures and identical surface grammatical relations.
As an example, consider structure (26):
(26) VP
NP V′
John V VP
e
NP V′
some new V PP
imported
beer
put in the cooler
If put is raised to the empty V position directly, the result is the simple “nonshifted”
version of this sentence, John put some new imported beer in the cooler. As an alterna-
tive, however, the lower V′ may undergo a reanalysis operation open to any predicate
projection that is thematically monotransitive.11 When this occurs, the whole reanalyzed
phrase raises around the object:
(27) VP
NP V′
John V VP
e NP V
some new V PP
imported
beer
put in the cooler
This yields the “NP-shifted” variant of this sentence, John put in the cooler some new
imported beer, although the direct object does not actually move in producing it.
Since V′ Reanalysis mentions neither the categorial identity nor the role of the com-
plement moved over, we predict a more general phenomenon of “XP Shift.” In Larson
(1989) it is suggested that double PP complement pairs, like those discussed by Jackend-
off, may be an instance of this:
(28) a. John talked to Mary about Bill.
b. John talked about Bill to Mary.
Specifically, it is proposed that (28b) results by V′ Reanalysis and raising of [V talk about
NP], as shown in (29).
(29) VP
NP V′
John V VP
e NP V
to Mary V PP
talk about Bill

Contrary to what Jackendoff (1990) claims (see note 2), I myself do find a heaviness
effect present in such examples in the form of relatively greater stress borne by Mary in
(28b) than by Bill in (28a). This effect also shows up in the differential availability of
pronoun reduction in the outer PP, as pointed out to me by John Frampton:
(30) a. John talked to Mary about him\’m.
b. John talked about Mary to him\*’m.
A derivation of this form also accounts for an important difference in domain prop-
erties distinguishing (28a,b) in my judgment. For me, (31a–c) are considerably weaker
than their counterparts in which to and about are inverted:
(31) a. *John talked about the men to each other.
(compare John talked to the men about each other.)
b. ?*John talked about no mani to hisi son.
(compare John talked to no mani about hisi son.)
c. ?*I talked about each man to the other.
(compare I talked to each man about the other.)
d. ??I talked about nothing to anyone.
e. *Which mani did you talk about to hisi son?
(compare Which mani did you talk to about hisi son?)
f. *Who did you talk about to which boy?
*Which boy did you talk about to who?
(compare Who did you talk to about which boy?)
These results would follow under the structure in (29). Since the object of about fails
to c-command that of to in the resulting structure, we correctly predict domain relations
to fail between them, as in (31a–d). We also predict weak crossover and superiority
effects, as in (31e,f).
3.3. Alternative Projection

A last source of complement alternation is alternative projection. Pairs related in this
way involve different underlying structures and different surface grammatical relations.
Alternative projection may arise in two ways. The thematic roles borne by the comple-
ments may simply be different in the relevant cases, or else the roles in question might
be unordered with respect to the thematic hierarchy.
A plausible case of the former is the alternation in (32), cited by Jackendoff (see
his (54)):
(32) a. John blamed the accident on Max.
b. John blamed Max for the accident.

Jackendoff takes the thematic relations in these sentences to be identical; however, a
closer look reveals basic differences. Specifically, (32a) patterns as a locative construction
in which the object is a theme and the PP a locative argument, whereas (32b) patterns
as a dative construction in which the surface object is a goal and the PP an adjunct
benefactive/malefactive phrase.
The locative character of (32a) is reflected in the close paraphrase relation it bears to
the construction in (33a) with put and in their shared subcategorization properties. Note
that just as put requires both a theme and a location (33b), so does locative blame (33c):
(33) a. i. John blamed the accident on Max.
ii. John put the blame for the accident on Max.
b. i. *John put the blame for the accident.
ii. *John put on Max.
c. i. *John blamed the accident.
ii. *John blamed on Max.
Locative blame also shares selectional properties with put. Just as the latter allows either
animate or inanimate NPs in the locative, so does the former:
Max.
(34) a. John put the blame for the accident on
the weather.
Max.
b. John blamed the accident on
the weather.
By contrast, (32b) appears to be notionally parallel to constructions involving thank

and to the corresponding dative forms with give (35). Dative blame shares the subcat-
egorization properties of these predicates. Just as the latter allow the benefactive phrase
to be absent, so does the former (36):
(35) a. i. John blamed Max for the accident.
ii. John thanked Max for the gift.
b. i. ?John gave the blame for the accident to Max.
ii. John gave thanks for the gift to Max.
(36) a. John blamed Max (for the accident).
b. John thanked Max (for the gift).
Dative blame also shares a selectional property of thank that distinguishes the two from
locative blame. In the first case, unlike the second, it seems that the cause or object of
blame/thanks must be animate:12
(37) a. ??John blamed The weather for his bad luck.

The temperature
b. ??John thanked The weather for his good luck.
The temperature
This result is, of course, straightforward if blame is dative in this construction and its
object is a goal; goals typically show an animacy restriction.
Given the different thematic relations involved with locative and dative blame, it
follows that these verbs will involve different D-Structure representations. In particular,
under the thematic hierarchy in (11) we derive the two structures for (32a) and (32b)
given earlier in (3) (repeated in (38)):
(38) a. VP
NP V′
John V VP
e NP V′
the V PP
accident
blame on Max
b. VP
NP V′
John V VP
e NP V′
V PP
Max
blame for the

accident
These structures in turn predict the asymmetries of syntactic domain noted by Jack-
endoff.13
Locative and dative blame illustrate the case where alternative projection arises
from different underlying thematic relations. However, alternative projection might
also occur with no difference in underlying thematic relations. This is possible if the
thematic hierarchy is not totally ordered, so that distinct roles θ1 and θ2 are unordered
with respect to each other. In this circumstance the principle in (11) would be vacuously
satisfied both when the argument bearing the first is projected in a superior position to
the argument bearing the second, and in the converse case.
This is a possible alternative view of the double PP complements discussed above.
Suppose that the roles borne by the two PPs are simply unordered with respect to each
other in the thematic hierarchy. Then both of the structures in (39a,b) satisfy principle
(11). Assuming that subsequent reanalysis of P and V allows the object of the higher PP
to c-command out (as discussed in Chomsky (1981 pp. 225–226)), we then account for
Jackendoff’s domain judgments regarding double PPs-namely, that relations available
with to-about are available with the converse ordering. Note that such cases would
present the strongest superficial evidence for linear ordering, since the complements
would appear to be distinguished only by linear order but nonetheless show domain
asymmetry. We see again, however, that with rightward downward branching such
evidence is merely apparent and that a purely structural analysis of the asymmetry
remains available.
(39) a. VP
NP V′
John V VP
e PP V′
to Max V PP
talk about Felix
b. VP
NP V′
John V VP
e PP V′
V PP
about Felix
talk to Max
Jackendoff suggests that “free” projection of complements as in (39a,b) would rep-

resent a “lexical” analysis of the double PP alternation. On the view sketched above,
however, this claim is false. Nothing particular to the verb in question would be involved
in accounting for these facts. Rather, free projection would result from a universal
property of the thematic hierarchy—namely, from the fact that certain roles are left
unordered with respect to it.
In summary, then, this discussion confirms the judgment made in section 1.1 regard-
ing Jackendoffs double complement paradigms and their consequences for domain
asymmetry. We see that the alternations Jackendoff cites can be brought within the
analysis of Larson (1988) and that under that analysis, domain asymmetries can
be accounted for in purely structural terms-without appeal to the linear order of
complements.
4. CONSTRAINTS ON DATIVE SHIFT
Jackendoff discusses the analysis of Dative Shift in Larson (1988), faulting its account of
why verbs like donate fail to undergo Dative Shift and noting apparent problems raised
by for-dative and motional to-dative constructions. In fact, Jackendoff’s criticism of the
former appears to be based on a mistaken view of his own counterevidence. Further-
more, for-datives and motional to-datives tum out to be easily incorporated into Larson
(1988) under a proposal that Jackendoff himself suggests.
4.1. Beneficiaries and Affectedness

Larson (1988) proposes that the oblique-double object alternation is constrained by
two basic conditions:
(a) The object of the oblique must be an argument of V—that is, it must be specified
in V′s thematic grid.
(b) The set of θ-roles assigned by V must subsume the set assigned by the P in the
oblique phrase.
Under these two conditions, the oblique preposition reduces to Case marking and can
be “absorbed” under the equivalent of Passive; this then triggers NP Movement. Lar-
son (1988) suggests that the verb donate fails as a double object verb by failing to meet
the second requirement; the proposal is that although donate selects a third object, it
assigns the latter the role of beneficiary, and not the role assigned by to, which is goal.
Accordingly, any attempt to absorb or suppress to would violate the equivalent of recov-
erability of deletion.14
Jackendoff disputes the claim that donate marks a beneficiary but not a goal. He cites
pseudocleft paradigms like (40a,b) ( = his (62a,b)) as a test for beneficiary status and
concludes from them that it is not a beneficiary with donate:
(40) a. What Bill did for Harry was give him a book.
b. ??What Bill did for the library was donate a book to it.
In fact, I think Jackendoff is mistaken about what the contrasts in (40) reveal. I suggest
that acceptability in such paradigms is not a matter of what role the indicated pronomi-
nal element bears (beneficiary, goal, and so on) but rather its position. More precisely,
I suggest that acceptability in such pseudocleft frames requires the pronominal element to
be understood as “affected” (in the sense of Tenny (1987)) and that to obtain an affected
reading an element must occur (at D- or S-Structure) in direct object position—the posi-
tion of arguments governed by V when the latter is governed by I.15
This counterproposal predicts all the data Jackendoff cites. In (40a) him occurs as a
direct object, whereas in (40b) it is the object of P. Thus, under the alternative proposal,
we expect the first to be good as compared with the second. This view also explains the
otherwise curious reversal of judgments that Jackendoff notes when (40a) occurs in its
oblique form:
(41) ??What Bill did for Harry was give a book to him.
To account for this fact, Jackendoff is forced to split roles into “affected” and “unaf-
fected” beneficiaries. But notice that similar results obtain in dative and spray-load
alternations where the PP in the pseudocleft is headed by to instead of for (42):16
(42) a. i. What Bill did to the book was send it to me.
ii. ??What Bill did to the book was send me it.
b. i. What Bill did to the hay was load it on the truck.
ii. ??What Bill did to the hay was load the truck with it.
c. i. ??What Bill did to the truck was load the hay on it.
ii. What Bill did to the truck was load it with the hay.
Here the constructions involve goals, themes, and locations; hence, to maintain Jack-
endoff’s view we would have to split these roles into “affected” and “nonaffected”
subroles. This misses the fact that in each well-formed example the pronoun occurs in
direct object position. These results thus suggest that Jackendoff’s argument is based on
a false assumption and that his pseudocleft paradigms test not for the beneficiary role
but simply for affectedness. Contrasts like the one in (40) are thus not counterevidence
to the claim that donate assigns a beneficiary role, and they do not refute the proposed
analysis.
4.2. For-Datives and Motional To-Datives

Jackendoff discusses the for-dative alternation and observes two apparent problems
that the construction raises for the account in Larson (1988). First, benefactive for-
phrases behave as adjuncts by various tests, that is, as unselected elements. But verbal
selection is required on the analysis in Larson (1988). Second, only a semantically
restricted class of oblique for-dative constructions can have corresponding double
object forms. Roughly put, the verbs in question must denote events of preparation or
creation, and the created or prepared objects must be intended for the benefit of the
beneficiary. This lexical conditioning appears problematic for Larson (1988), which
attempts to account for the alternation purely on the basis of the respective roles
assigned by V and P.
Similar results apply with a class of verbs expressing causation of motion, among
which Jackendoff lists hit, throw, kick, shin, and send. These predicates allow the dative
alternation but do not intuitively require a recipient. Moreover, the verbs of this class
allowing a double object form are semantically restricted: intuitively, the agent must
set the theme in motion along a trajectory. Here again there is evidence of lexical con-
ditioning.
Jackendoff considers how benefactive double object constructions might be analyzed
under alternative approaches. He mentions approvingly the following proposal (citing
Oehrle (1976), Grimshaw (1989), and Pinker (1989)): “a lexical rule optionally adds
a beneficiary argument to transitive verbs of creation and preparation, so the indirect
object comes to be θ-marked by the verb” (p. 448). Presumably Jackendoff would advo-
cate a similar solution for the class of motional to-datives.
4.2.1. Argument Augmentation

The observations that Jackendoff makes about the semantic conditioning of for-datives
and motional to-datives are convincing, and the questions they raise for the account in
Larson (1988) are important ones. I would like to suggest, however, that these ques-
tions can be answered straightforwardly within Larson (1988) if we adopt the kind of
solution Jackendoff recommends but simply “uncouple” it from the dative alternation.
Suppose we assume, following Jackendoff and the authors he cites, that there are two
lexical rules in English that augment the argument structure of transitive verbs. We may
state them as in (43) and (44) for concreteness:
(43) Benefactive Augmentation (Optional): Add θBENEF to the θ-grid of α.
Condition: α denotes an event of creation or preparation.
Result: The theme is for the benefit of the beneficiary.
(44) Goal Augmentation (Optional): Add θGOAL to the θ-grid of α.
Condition: α denotes an event of motion in which the agent imparts a trajectory to the theme.
These rules apply to predicates that meet their conditioning clause. Thus, Benefactive
Augmentation can apply to transitive bake (45a) to yield ditransitive bake (45b). And
Goal Augmentation can apply to transitive hit (45c) to yield ditransitive hit (45d). Sup-
pose also that for has the simple thematic grid in (45e):
(45) a. bake: {θAGENT, θTHEME}
bake: {θAGENT, θTHEME, θBENEF}

b.
hit: {θAGENT, θTHEME}

c.
hit: {θAGENT, θTHEME, θGOAL}

d.
for: {θBENEF}
e.
We can now project the derived ditransitives similarly to the way give is projected
in Larson (1988). Thus, ditransitive bake can be projected into the oblique structure
(46a), with for Case-marking the benefactive argument and redundantly assigning it the
benefactive role. This yields the oblique form John baked a cake for Mary. Alternatively,
given its thematic redundancy, for can be “absorbed” as Case marking and the theme
argument projected into an adjoined position (46b). After NP Movement this yields the
double object form John baked Mary a cake:
(46) a. VP
NP V′
John V VP
e NP V′
a cake V PP
bake for Mary

b. VP
NP V′
John V VP
e NP V′
V′ NP
e
V NP a cake
bake Mary
Analogously for derived ditransitive motion verbs such as hit, throw, and send.
This analysis directly answers the questions raised by Jackendoff. Thus, we see that
although benefactive for-phrases are indeed generally adjuncts, with certain verbs they
are able to become arguments—namely, with those undergoing Benefactive Augmenta-
tion. Since the latter is an optional rule, these verbs are also able to appear without a
benefactive phrase or to appear with a benefactive functioning as a pure adjunct. Simi-
larly for directional to-phrases when they occur with motion verbs.
Furthermore, we capture the lexical conditioning observed in these construc-
tions, and we explain why it presents itself superficially as a constraint on double
object formation. Since the class of verbs undergoing Benefactive or Goal Augmen-
tation is semantically constrained, and since only augmented verbs can undergo
Dative Shift, the appearance is one of semantic constraints on the double object
alternation. This is an illusion, however. On the proposed view, lexical condition-
ing is equally present in oblique dative/benefactive examples like John baked a cake
for Mary and John hit the ball to Mary when the verb has been augmented. We
simply aren’t able to “see” the conditioning directly in such examples given their
homophony with sentences involving a transitive verb + adjunct, unlike what we
find in the double object case.
This proposal comports naturally with an attractive and highly restrictive view of the
interaction between syntax and semantics—namely, that semantic information enters
only into the projection of initial structures, and there only in the form of thematic
roles. On this proposal, more complex semantic properties and more specific lexical
information affect structure projection only insofar as they affect the number of roles
a predicate bears and/or their identity. The analysis suggested above is faithful to this
view: the fact that certain verbs denote events of creation, or denote events in which an
object follows a ballistic trajectory, does not enter into the projection of oblique versus
double object structures directly. Rather, these facts are relevant to argument augmenta-
tion, which adds roles of specific kinds. Structure projection then occurs according to
the universal principles in (9)-(11). These lexical properties also do not constrain the
operation of NP Movement. Grammatical rules like Passive, Dative Shift, and the like
(that is, Affect a) are left free of semantic constraints, up to completely general principles
like recoverability.
4.2.2. Donate Again

This view of for-datives and motional to-datives suggests an approach to the failure of
donate to Dative-Shift different from that taken in Larson (1988). Instead of analyzing
donate as a three-argument verb that violates prepositional recoverability with Dative
Shift, we might instead take it as a two-argument verb that fails to undergo Goal Aug-
mentation. Dative Shift would then be blocked by the nonargument status of goal NPs
with donate.
Such a proposal has in fact been advanced by Randall (1987) and (in a somewhat
different form) by Grimshaw (1989), and gains plausibility from the fact that unlike
give, but like other verbs expressing (very roughly) events of dispersal, donate allows
the dative to be optional without being implicitly presupposed:
(47) a. John gave that money *(to Mary).

donated
b. John gave away that money (to charities).
distributed
dispersed
Since only arguments can undergo Dative Shift, the failure of donate (give away, distrib-
ute, disperse, and the like) to shift would then be straightforward.
This proposal has been challenged by Gropen et al. (1989) on the grounds that verbs
that take an optional to- or for-phrase nonetheless can undergo the alternation. The
challenge is not a decisive one, however. As we see, argument versus adjunct status can-
not be judged simply by whether the goal and benefactive phrases are optional versus
obligatory. Rather, the crucial issue is whether the predicates in question can undergo
argument augmentation, which in turn depends on the conditions for application of the
latter and the precise semantics of the former.17
The general argument augmentation analysis suggested for for-datives and motional
to- datives can, I believe, be extended to cover spray-load alternations as well. Thus, just
as transitive feed undergoes Goal Augmentation yielding a three-argument dative that
projects with an oblique (48a) or shows NP Movement (48b), so feed might undergo
an augmentation rule yielding a three-argument locative that projects with an oblique
(48c) or shows NP Movement (48d):
(48) a. John fed data to the machine.
b. John fed the machine data.
c. John fed data into the machine.

d. John fed the machine with data.
Nonetheless, spray-load forms exhibit an interesting additional complexity.

Spray-load verbs differ from simple datives in two important (and, I believe,
related) ways. As observed earlier, spray-load verbs show a with or ofPP in their
derived forms. They also differ significantly in the range of prepositions appearing in
the oblique form. Oblique datives permit to but do not allow directional Ps with an
explicit locative component (in, on, and so forth) (49). By contrast, oblique spray-
load verbs require directional Ps with additional locative material and do not permit
a bare to (50)–(51):
to
* onto
(49) John gave the book * into
Bill.
* on
* to
on(to)
(50) John loaded the hay the truck.
in(to)
on
from
(51) John cleared the dishes the table.
off of
One way of summarizing this situation intuitively is as follows: spray-load alternations

are analogous to dative alternations, except that they show an additional prepositional
component “at both ends.” In their oblique forms, spray-load verbs show the direc-
tional component of datives, plus additional locative material. In their derived forms,
spray-load verbs show additional prepositional material in the form of with or of.
I would like to suggest tentatively that the with and offound in the derived form of
spray-load paradigms are actually (very general) locative prepositions in this usage and
that they are in fact the counterparts of the locative component found in the oblique
form of the spray-load paradigm. That is, intuitively, I suggest that spray-load alterna-
tions are in fact basically dative alternations, with a locative element “added on as a
constant.” This locative element is present in the oblique form as on, in, off (“not on”),
and so forth, and in the derived form as with or of. In terms of the proposals in Larson
(1988), a simple way to view this is via recoverability: suppose that spray-load verbs are
basically datives and that although they select a directional-locative as a third argument,
their thematic grid subsumes only the goal role. Then “absorption” of onto, into, off of,
and the like would involve unrecoverable loss of the locative component. As a result, a
locative must be present in the derived form as well.18
The particular choice between with and of appears to me to follow naturally under
proposals by Hale (1986), who suggests that all locative notions rest on a basic distinc-
tion between what he terms central and noncentral coincidence. Roughly, the former
relation holds when the center of one object coincides (up to pragmatic limitations) with
the center of another; and the latter relation holds when one object stands to another
as its terminus (either initial or final) (see Hale (1986) for details). Hale suggests of as
a preposition expressing noncentral coincidence, and I would like to propose with as
its “central” counterpart. Thus, with spray-load verbs denoting processes that result in
their object participants coming to spatially overlap (such as spray and load), the rel-
evant locative preposition is with, expressing central coincidence. And with spray-load
verbs denoting processes whose result is that their objects come to not-coincide (such
as empty, clear, and drain), the relevant locative preposition is of, expressing noncen-
tral coincidence. These remarks are sketchy, of course, but they do suggest an intui-
tively plausible way of bringing the locative alternation within a derivational analysis
involving NP-Movement, and a way of approaching the particular preposition choice
involved.
5. RESIDUAL ISSUES
5.1. Modifiers
Jackendoff criticizes Larson (1988) for “neutraliz[ing] the structural distinction between
arguments and modifiers” (p. 452). In support of this distinction, Jackendoff lists a
number of analogous sentence pairs involving modifiers and arguments, in which the
two diverge in grammatical behavior. But Jackendoff does not indicate how (or why)
these differences could only follow from a structural distinction between arguments and
modifiers in which the former is a complement and the latter is adjoined.19
It is dubious that a general argument of this kind could be given. On a theory in
which modifiers and arguments project differently, such projection is presumably based
on prior thematic differences—for example, on the fact that modifiers are not θ-selected
whereas arguments are. Conceptually, this allows the possibility of explaining the rel-
evant facts directly in terms of the thematic property (here θ-selection), without appeal
to structure. Such a line has indeed been widely pursued. Huang (1982) argues for a
Condition on Extraction Domains (CED) in which differences in extraction behav-
ior from arguments versus adjuncts follow from the fact that the former are selected
whereas the latter are not. Similarly, Chomsky (1986a) proposes a definition of barriers
for movement in terms of the notion of L-marking, which in turn crucially involves
the notion of θ-government. The latter bounds movement out of adjuncts, not by the
fact that they occur in adjoined position, but by the fact that they are not θ-governed
and hence not L-marked. The projection of modifiers into “complement position” (so-
called) thus does not relinquish our grasp on modifier-argument asymmetries a priori.20
There is some empirical evidence supporting the account of modifiers in Larson
(1988)—in particular, for the idea that modifiers can be structurally subordinate to
direct objects. One argument comes from examples like (1a) (repeated below as (52)).
Negative polarity items typically require S-Structure c-command by their triggers. The
fact that an affective object (jew friends) can license a temporal negative polarity item
(any day this week) thus implies that the former c-commands the latter:
(52) John visited few friends any day this week.

A second argument is due to Contreras (1984), who notes obviation effects between the
indicated NPs in examples like (53a–d):
(53) a. *John filed them [without reading Mary’s articles].
b. ??Felix dismissed her [before consulting Jane’s mother].
c. *Criticize her [because Eunice arrived late]!

d. *Max burned it [although Bill had given him the present].
As Contreras observes, obviation here is plausibly analyzed as a Principle C effect; how-

ever, such an account requires us to view the adjunct clause as within the c-command
domain of the direct object.21
A final argument involves sloppy identity. Reinhart (1983) has argued that a sloppy
reading is available in examples like (54a,b) but not in (54c) because the antecedent
c-commands the pronoun in the first conjunct of (54a,b) but not in the first conjunct of
(54c). The generalization is that sloppy identity requires c-command in the antecedent–
pronoun relation.
(54) a. Zelda brought Siegfried a picture of his wedding day, and Felix too.
Rosa wore a pink dress to her bridal party, but not Zelda.
b.
c. Felix persuaded [a friend of Rosa] to visit her, but not Zelda.
(≠ Felix persuaded a friend of Zelda to visit Zelda.)
Similar-looking contrasts in sloppy versus strict readings arise in examples like (55a–c),
which involve Gapping rather than VP-Ellipsis:
(55) a. I interviewed Max after Mary brought him, and you Felix.
(“I interviewed Max after Mary brought Max, and you interviewed Felix after Mary
brought Max” (STRICT) or “I interviewed Max after Mary brought Max, and you
interviewed Felix after Mary brought Felix” (SLOPPY))
b. I always visit Max without calling him first, and you Felix.
(“I always visit Max without calling Max first, and you always visit Felix without
calling Max first” (STRICT) or “I always visit Max without calling Max first, and
you always visit Felix without calling Felix first” (SLOPPY))
c. I interviewed [a friend of Max] after Mary brought him, and you Felix.
(≠ “I interviewed [a friend of Max] after Mary brought Max, and you interviewed
Felix after Mary brought Felix” (SLOPPY))
If objects c-command adjuncts, the account of VP-Ellipsis and Gapping contrasts can
be assimilated. We can say that a sloppy reading is available in (55a,b) but not in (55c)
because only in the former pair does a c-command relation hold between the antecedent
and the pronoun in the initial conjuncts.
The analysis of adjuncts in Larson (1988) also receives some conceptual motiva-
tion from work by Parsons (1985, 1991) and Davidson (1967). Parsons argues that
verbs should be analyzed semantically as unary predicates of events, with thematic roles
interpreted as relations that connect events to their participants, their time and place of
occurrence, their manner of execution, and so on. On this analysis, (56a) has the logical
form in (56b), according to which the sentence is true iff there is some event of giving e,
of which John is the agent, of which Fido is the theme, of which Mary is the goal, which
was on Boston Common, and which was at 3:00p.m.:
(56) a. John gave Fido to Mary on Boston Common at 3:00 p.m.
∃e[giving(e) & AGENT)(e,j) & THEME(e,f ) & GOAL(e,m) & ON(e,BC)

b.
& AT(e,3pm)]
Observe that argument and adjunct phrases are not distinguished semantically here;
indeed, the only “true argument” of V is the event argument e, all other participants
being linked to e by binary thematic relations.
The event analysis suggests a natural rationale for the view of complement structure
in Larson (1988): arguments and adjuncts are projected analogously in syntax because
they are fundamentally analogous in semantics. It also affords a rather natural view of
the order of complements fixed by the thematic hierarchy AGENT > THEME > GOAL
> OBLIQUE. We can take the latter to reflect a hierarchy of individuation criteria for
events.22 Recall that the lower an expression is on the thematic hierarchy, the “closer” it
is to the verb, so that in an example like (57) the causal adverb because he didn’t want
to awaken his wife is actually the closest complement of V and the agent subject, the
most distant:
(57) John buttered the toast carefully, in the bathroom, at midnight, because he didn’t want
to awaken his wife.
Interestingly, this order seems to track rather closely the criteria that philosophers have
suggested for individuating events, where the causes and effects of events are preeminent,
followed, in descending order, by their time of occurrence, their place of occurrence,
their manner of execution, and their participants (see Davidson (1967) for discussion).
Under this view, complements would thus be positioned with respect to V at D-Structure
in a way that mirrors their relative prominence in event individuation; complements
whose relations are most closely involved in distinguishing e are positioned closest to V.
Although I cannot defend this proposal further here, these points are enough to suggest
that under recent attractive semantic analyses of the clause, the syntax of complementa-
tion proposed in Larson (1988) is not only prima facie plausible but even natural.
5.2. Nonconstituent Coordination and Gapping

Larson (1988) suggests a V-Raising account of examples like (58a–c), which are often
said to exhibit “nonconstituent coordination.” These are assimilated to simple VP con-
junction plus across-the-board V-Raising (58d):23
(58) a. John sent a letter to Mary and a book so Sue.

b. Max put the cat in the closet and the dog in the basement.
c. We consider Alice intelligent and Bill hardworking.
d. [VP NP [V′ V [VP[VP NP [V′ t XP]] and [VP NP [V′ t XP]]]]]
Jackendoff disagrees, arguing that such examples result from Gapping.

Sentences like (58a–c) have occasioned lively debate in the literature, with a variety
of analyses being proposed—including Gapping—and with authors regularly changing
their minds on the subject.24 Hence, it is safe to take the issue as unsettled. Nonetheless,
a number of points can be made regarding the arguments Jackendoff provides and the
general plausibility of the V-Raising account.25
First, an across-the-board account of examples like (58a–c) is at least plausible in
principle given similar facts in other languages that do appear to involve V-Raising.
Neijt (1979) proposes that Dutch examples like (59a) derive by across-the-board Verb-
Second with geven ‘give’; the latter moves from an underlying final position. (59b–d)
(due to Riny Huybregts) would presumably be analyzed similarly:
(59) a. Jan gaf [Marie een appel t] én [Piet een pear t]
‘John gave Mary an apple and Pete a pear.’
b. Jan legde [de worteltjes in de gootsteen t] én [het brood op tafel t]
‘John put the carrots in the sink and the bread on the table.’
c. ‘Jan vindt [Felix intelligent t] én [Max vlijtig t]
‘John considers/finds Felix intelligent and Max hardworking.’
d. Jan zag [Willem studeren t] én [Marie uitflippen t]
‘John saw Bill study and Mary goof off.’
Such a Verb-Second derivation is directly analogous to what is urged for the parallel
English cases in Larson (1988).26
Second, Jackendoffs claim to the contrary, there is in fact additional evidence for the
constituency of phrases like a letter to Mary, the cat in the closet, and Alice intelligent in
examples like (58a–c). Early in the history of discussions of Right Node Raising (RNR),
it was observed that although RNR is in general a strong test for constituency, there is
a class of cases for which it seems to fail. Thus, Grosu (1976) and Abbott (1976) cite
(60a–c) and (61a–e) (respectively) as exhibiting Right Node Raised elements that are
not phrases (examples from Erteschik-Shir (1987)):
(60) a. John has sliced, and Mary also seems to have sliced, [a large piece of cake with a
shining new knife].

b. Bill may present, and Mary certainly will present, [a series of papers at tomorrow’s
linguistic meetings].
c. Mary may have conducted, and Bob certainly will conduct, [a large number of tests
in the large oval laboratory].
(61) a. Smith loaned, and his widow later donated, [a valuable collection of manuscripts to
the library].
b. I borrowed, and my sisters stole, [large sums of money from the Chase
Manhattan Bank].
c. Leslie played, and Mary sang, [some C&W songs at George’s party].
d. Mary baked, and George frosted, [twenty cakes in less than an hour].
Under the analysis suggested here, the bracketed expressions are all constituents—
indeed, they are all VPs. The right-peripheral expression in (61b), for example, has the
structure in (62):
(62) VP
NP V′
large sums V PP
of money
e from the Chase
Manhattan Bank
These data, which are problematic under other accounts and are not readily assimilated
to Gapping, provide additional evidence for the V-Raising constituency.27,28
5.3. On Gapping
Structures of the kind in Larson (1988) suggest an interesting approach to many cases
of ellipsis usually attributed to Gapping. Under standard views of constituency, Gapping
examples like (63) involve removing a discontinuous sequence, consisting of the verb
together with an oblique complement, a modifier, or a secondary predicate. The result
is a stranded direct object:
(63) a. John gave a record to Mary, and Bill a tape. (give to Mary)
b. Alec put a dollar in the machine and Max fifty cents. (put in the machine)
c. John worded the letter carefully, and Mary the memo. (word carefully)
d. Alice saw you in the park yesterday, and I Doris. (see in the park yesterday)
e. Max painted the barn red, and Bill the house. (paint red)
f. Eunice hammered the metal flat, and Gertrude the tin. (hammer flat)
g. Hector ate the beef raw, and Alonzo the fish. (eat raw)
As noted by Hoeksema (1987), such examples contrast in acceptability with cases like
(64), where Gapping attempts to take out the verb and direct object, stranding the
oblique complement, modifier, or secondary predicate (in each case the initial NP in the
second conjunct is to be understood as its subject):
(64) a. *John gave a record to Mary, and Bill to Alice.
b. *Alec put a dollar in the machine and Max in the collection plate.
c. *John worded the letter carefully, and Mary tactlessly.
d. *Alice saw you in the park yesterday, and I in the museum.
e. *Max painted the barn red, and Bill pink.
f. *Eunice hammered the metal flat, and Gertrude smooth.
g. *Hector ate the beef raw, and Alonzo marinated.
Hoeksema observes a similar result with Pseudogapping, which replaces the verb by
the auxiliary do in contexts involving (primarily) comparatives of equality and inequal-
ity (65). Once again, attempts to affect the verb and direct object yield less acceptable
results (66):
(65) a. Max painted more barns red than he did houses.
b. Eunice hammered as much zinc flat as she did tin.
c. Hector ate less beef raw than he did fish.
d. Felix painted the barn red the same way that he did the house.
(66) a. ??Max painted more barns red than he did pink.
b. ??Eunice hammered as much zinc flat as she did smooth.

c. ??Hector ate less beef raw than he did marinated.
d. *Felix painted the barn red the same way that he did blue.
If branching VP structures of the kind suggested in Larson (1988) are correct, then
we can provide a relatively tidy configurational account of all of these facts by saying
that in the gapped and pseudogapped examples, it is precisely a (thematically monotran-
sitive) V′ constituent that is elided or replaced with do:29
(67) VP
NP V′
α V XP
give to Mary
put in the machine
word carefully
see in the park t yesterday
paint red
hammer flat
cat raw
On more conventional views of phrase structure, no simple configurational account of

these facts is forthcoming.
NOTES
I am grateful to Lisa Cheng and Michael Hegarty for helpful comments on an earlier version of this
article.
1. I assume that any day this week is a bare NP adverb as discussed in Larson (1985). See
Emonds (1987) and McCawley (1988) for alternative analyses.
2. This situation is also illustrated in formulations of the Empty Category Principle (ECP),
where early versions contained a disjunctive condition mentioning both lexical and ante-
cedent government. More recent versions (Chomsky 1986a) have attempted to simplify the
general principle by eliminating one of the disjuncts (lexical government). The result has
been more complex structures involving traces, but a more restrictive theory overall.
3. Jackendoff’s discussion of his (34)–(38) mixes together a number of phenomena and also fails
to distinguish between command, locality, and levels. Simple anaphora (Principle B) requires
c-command by an antecedent and the presence of the antecedent in the anaphor’s governing
category. As Belletti and Rizzi (1988) have argued, however, anaphora does not appear to
require surface (S-Structure) c-command. The locality condition explains the ill-formedness
of Jackendoff’s (34b) and (37b). The possibility for D-Structure c-command (under Bel-
letti and Rizzi’s theory of psych verbs) explains the well-formedness of (35b). By contrast,
each . . . the other and quantifier binding appear to require (at least) LF c-command, but no
strict locality for the bound item. This accounts for (34a), (36a), and (38b). The ill-formed
(35a) and (36b) fall together as weak crossover violations under the assumption that the
other is bound by the each phrase as his is bound by every boy. Finally, negative polarity
items appear to demand S-Structure c-command and to resist factive environments. This
accounts for (37a) and (38a), respectively.
The important point to note here is that these facts are all fully compatible with a defini-
tion of domain that uniformly requires what Jackendoff refers to as a “dominance-based
condition.” In particular, although these various items differ in how close a binder must be
or at what level binding occurs, all depend on command.
4. The term direction is borrowed from Russell (1903); the X-bar principles in (9) embody
a formal symmetry between subjects and complements, namely, that there can be only a
single instance of either in a single maximal phrase. Contrary to what Jackendoff implies
in the discussion of his (52), formal symmetry is nowhere advanced in Larson (1988) as a
reason for accepting this version of X-bar theory. The reasons for accepting (or rejecting)
this theory are the same as for any other scientific proposal: the understanding (or lack
thereof) that it brings to the phenomena-in this case the range of phenomena discussed in
Larson (1988).
5. Given the account of thematic assignments in Larson (1988), the fact that some beer is struc-
turally an “inner subject” (VP specifier) of put in no way entails that it must be understood
as an agent. This error is made by Aoun and Li (1989). The thematic role of some beer
is determined by the relation between structural prominence and the thematic hierarchy,
which in this case requires the NP to be the theme.
6. The projection of empty head positions in Larson (1988) is analogous to the projection
of empty argument positions under standard proposals. On the usual view, the presence
of a functional head (such as Infl) compels a specifier position to be projected, even in the
absence of a θ-role to assign the latter. The compulsion is X-bar theory: heads require speci-
fier positions. Similarly, on the view in Larson (1988), the presence of a specifier compels a
head position to be projected, even when it receives no role from the latter.
7. Structures containing an I′ like (13b) do not actually occur in Chomsky (1981); thus, (13b)
represents a slight “updating” of that work, incorporating the general X-bar theory for
minor categories adopted in Chomsky (1986a, 1986b). However, this point is irrelevant
to the issue of θ-assignment locality discussed in the text. The same issue of θ-assignment
outside the maximal projection headed by V arises under structures where NP, I, and VP
directly depend from S, as in Chomsky (1981).
8. The addition of modifiers to dative structures like (13a) will change the assessment of direct
and indirect θ-marking. As discussed in section 2.3, modifiers are added below arguments;
hence, what was previously a directly θ-marked complement (here (to) Mary) will become
an indirectly θ-marked VP subject.
9. As noted in Larson (1988), UTAH does not itself settle the direction of derivation between
oblique and double object structures. Several investigators, including Bowers (1981), Johns
(1984), Dryer (1987), and Aoun and Li (1989), have advocated deriving the oblique from
the double object form.
10. I do not currently have an account of the “scope freezing” effect of double object alterna-
tions within the analysis of Larson (1988). Thus, at present these facts argue only for a
general correlation between double object structures and the with/of variant of spray-load
paradigms, and not for the specific analysis in Larson (1988). Schneider-Zioga (1988) and
Aoun and Li (1989) propose an analysis of these scope facts based on the small clause analy-
sis of Kayne (1983). Simplifying somewhat, they suggest that in double object structures, the
two objects occur in a small clause (SC), and they propose that a quantified outer object is
absolutely confined to the SC domain:
(i) The teacher assigned [SC [some student] [every problem]].
There is evidence against the view that the scope limitations with double objects (and related
constructions) involve absolute confinement of the outer quantifier. First, double object struc-
tures show Antecedent-Contained Deletion in an outer quantified object, as in (ii):
(ii) a. John gave someone [everything that Bill did [ VP e]].
b. Max wants to give someone [everything that you do [ VP e]].
On the analysis proposed in Sag (1976), May (1985), and Larson and May (1990), correct
reconstruction of the empty VP requires the quantified NP to receive scope at least as wide as
the VP serving as reconstruction source. In (iia) this entails that everything that Bill did e must
get scope at least as wide as the VP headed by give, that is, outside the putative small clause. In
(iib) everything that you do e must receive scope at least as wide as the matrix VP headed by
want in order to get the reading where the sentence is understood as “Max wants to give some-
one everything that you want to give them.” Again, this is outside the putative small clause.
Second, outer objects show de dicto/de re ambiguities with respect to higher predicates.
Thus, (iii) is ambiguous between a reading where every apartment in the building is read
opaquely and a reading where it is read transparently with respect to promise. (This sentence
also shows the scope effect for quantified double objects, so that if every apartment in the
building is de re, someone must be de re as well.)
(iii) I promised to rent someone every apartment in the building.
On standard analyses of such ambiguities, a de re reading of every apartment in the building
will involve assigning this NP scope beyond promise.
These results suggest that an outer quantified object is not absolutely confined in double
object structures, but only confined relative to the inner object. James Higginbotham (per-
sonal communication) notes further evidence in support of this view from the following
contrast:
(iv) a. No one gave Bill anything.
b. ??No one gave someone anything.
Higginbotham points out that if polarity items must occur at Logical Form (LF) within the
immediate scope of their triggers (as argued by Linebarger (1987)), and if an outer object is
confined to take relative scope narrower than an inner object, then the contrast is explained.
In (iva), where the inner object is unquantified, anything moves at LF into the immediate
scope of its trigger no one. In (ivb), however, anything cannot move into the immediate scope
of no one since it must stay within the scope of someone. Hence, the sentence is excluded. I
hope to take up at a later date the issue of how the relative scope confinement of outer and
inner quantifiers is obtained.
11. This thematic restriction on reanalysis is discussed in Larson (1988, section 7.4)
12. I am indebted to Mark Aronoff for pointing out these animacy contrasts, and to Mark
Aronoff and Dan Finer for general discussion of dative versus locative blame. Aronoff points
out that the anomaly in (37a,b) weakens precisely to the extent that one personifies or
anthropomorphizes the weather or temperature.
13. Note further that the constraint in relative quantifier scope observed with double objects
and spray-load verbs does not arise in the dative/locative blame alternation. In my judgment
both forms are equally ambiguous with quantified objects:
(i) a. John blamed some mistake on every subordinate.
b. John blamed some subordinate for every mistake.
The contrast in scope ambiguities between this alternation and those analyzed as involving NP
Movement supports the general view that the alternations should be analyzed differently.
14. Jackendoff finds the invocation of recoverability “curious” (p. 446), observing that recipients
are notionally predictable with donate and that donate permits only the preposition to to mark
this recipient. His view is thus that since the preposition and its role are as predictable with
donate as they are with give, they should be equally recoverable. A similar concern is voiced
in Pinker (1989). The response to this is that recoverability and predictability are simply not
equivalent. Recoverability is a grammatical notion requiring a subset relation between sets of
assigned roles. Predictability is an independent pragmatic or semantic notion. It seems to me
that it is proper to distinguish such notions, and that their independence is familiar from other
contexts. It is predictable on semantic grounds that arrivals involve arriving somewhere. It is
also predictable that when arrive occurs with a locative PP specifying terminus of motion, the
preposition will be at—no other P is allowed (compare arrive at, *arrive to, *arrive on, and
so on). However, from this nothing follows regarding the roles arrive assigns. In particular,
it does not follow that arrive assigns a locative role to the P-object in cases like John arrived
at the party. Accordingly, it does not follow that at could be suppressed under the notion of
recoverability discussed above. Thus, although Jackendoff’s point about the general predict-
ability of to with donate may be correct, it does not jeopardize the proposal in Larson (1988).
Recoverability and predictability are simply not equivalent under that account.
15. Here again I basically follow Tenny (1987). Note that under this view it is not necessary for
an affected NP to occupy the direct object position at S-Structure; the latter may instead bind
a trace there. This will accommodate familiar examples like (ia), under the analysis in (ib):
(i) a. The garden swarmed with bees.
b. [npi The garden] swarmed ti with bees.
Note that the claim here is not that an argument is affected if it is a direct object; the falsity
of this is easily seen in cases like ??What I did to John was hear him. Rather, an argument
is affected only if it is a direct object.
16. It has been widely noted that double object structures do not freely tolerate outer pronomi-
nal objects; hence, the assessment of (42ai) versus (42aii) must control for this factor. We
can do so by observing the relative acceptability of (i) versus (42aii):
(i) John will send me it.
In general, the “outer pronoun effect” is ameliorated if the inner object is a pronoun as well.
We see then that the anomaly of (42aii) does genuinely seem to issue from the pseudocleft
structure.
17. Assuming that donate fails to Dative-Shift by virtue of failing to undergo Goal Augmenta-
tion, the natural question arises as to why donate falls outside the scope of this rule. I suggest
the reason relates to an important constraint on recipients of donate versus give, namely,
that recipients of donate must be organizations or groups and not persons. This proposal
seems to be compromised by examples like (i):
(i) Felix just donated two million dollars to Jesse Jackson.
In fact, however, it seems that in cases like (i) we understand the person designated in the
goal phrase (Jesse Jackson) qua representative of an organization (the Jackson political
campaign).
This restriction on the goal argument of donate is codified to some extent in legal defini-
tion and practice. Corpus Jurus Secundum distinguishes gifts and donations as follows: “The
term ‘donation’ . . . is often used as equivalent in meaning to gift; but a donation, it has been
held, need not have all of the essentials of a gift . . . The term ‘donation’ is more aptly used to
describe that which is given to a public cause or charity than to indicate a bounty to an indi-
vidual” (Kiser 1943, vol. 38, 783–784). This distinction between charitable contributions-
that is, donations-and gifts is also embodied in federal tax law: “The contributions or gifts
of any taxpayer must be made to charitable organizations [author’s emphasis] in order to be
deductible as charitable contributions. Except where a donation is made to an individual as
an agent for charitable organizations, no deduction may be taken for amounts donated to
individuals” (Research Institute of America 1978, vol. 15, 32,086). Similarly: “Donations are
deductible only if made to the organizations described in the statute as eligible for deduct-
ibility . . . If the donor overly limits the class of beneficiaries for whom the charity may use
his gift, it will be construed as a gift (or compensation or other payment) directly to those
beneficiaries, not to the charitable organization” (McNulty 1978, 187).
This discussion appears largely commensurate with the suggestion in Larson (1988) that
predicates like donate, distribute, disperse, give away, and give out fail to undergo Dative Shift
in virtue of being “verbs of dispersion,” wherein the goal of the event is not a “point target” (an
individual) but rather a “region” (a group). In present terms, the proposal would be that Goal
Augmentation requires the third argument to be at least a potential individual recipient, but the
inherent semantics of donate, distribute, disperse, give away, and give out forbids this.
For recent discussion of the double object alternation adopting a different view of donate
than that proposed here, see Hegarty (1989).
18. Jackendoff (1983) notes that morphologically complex directionals like into and onto should
be analyzed semantically as “to a point in” and “to a point on” (respectively), where the prepo-
sition that appears outermost corresponds to the inner complement. Given the discussion of
with as the core locative, it is interesting to note the presence of forms like within, where once
again what is morphologically outermost would correspond to an inner complement.
19. This omission is nontrivial since examples similar to the ones Jackendoff cites show different
grammaticality results. (For example, compare It was the park that John wrote a letter to
Mary in with It was the mailbox that John put a letter to Mary in.)
20. Jackendoffs facts about each other can be analyzed directly in these terms. Suppose that
extraction is governed by the CED, as in Huang (1982); suppose further that the interpreta-
tion of each other involves movement of each to the local domain of its antecedent at LF
(as argued by Lebeaux (1983), Chomsky (1986b) and Heim, Lasnik, and May (1989)).
Then we predict the unacceptability of Jackendoff’s (72b) versus the acceptability of (72a).
The former will involve LF extraction out of an unselected complement, contra the CED.
whereas the latter will not.
21. This argument is not entirely straightforward, encountering two important classes of coun-
terevidence. First, Solan (1983) notes that the obviation effect in question appears to be
substantially weaker with adverbial clauses than with complement clauses ((ia,b) record my
judgments; Solan (1983, 94) actually finds (ia) fully acceptable):
(i) a. ?They booed him before the candidate finished his speech.
b. *They told him that the candidate would not finish the speech.
Second, and relatedly, Hornstein and Weinberg (1990) observe that Principle C effects mani-
fested by bound epithets in complement clauses are entirely absent with bound epithets in
modifying clauses:
(ii) a. *Every mani thinks [cp that Mary likes the idioti].
b. Every contestanti was given a prize [pp before the idioti could protest].
Although I have no secure explanation to offer for these facts, the contrasts in question may
be explainable by reference to the level at which obviation must occur. As discussed in Larson
(1987), a variety of clausal PP adverbials (including temporal before- and after- clauses) license
Antecedent-Contained Deletion and show de dicto/de re ambiguities:
(iii) a. I talked to everyone [pp before Bill did [VP e]].
b. John thinks Mary arrived [pp before she did arrive].
These facts suggest that such adverbials are quantificational, undergoing scope assignment
at LF:
(iv) a. [np everyone], [pp before Bill did [ VP e]]i [ip I talked to ti, tj]
b. [pp before she did arrive]i [ip John thinks Mary arrived ti]
Suppose then that binding of epithets does not take place until LF, when scope assignment
occurs. Then (iia) will involve a Principle C violation; because the clausal complement is not
scopal, the idioti will end up bound by ti:
(v) [np every man]i [ip ti thinks [pp that Mary likes the idioti]]
On the other hand, scopal movement of the before-clause in (iib) will bring the epithet
outside the domain of ti at LF (under a first branching node definition of c-command) and
remove the potential Principle C violation:
(vi) [np every contestant]i [pp before the idioti could protest]j [ip ti was given a prize tj]
The distinction between Solan’s and Contreras’s examples might be explained in a similar
way. If the former involve quantificational adverbials (before and after) whereas the latter
(without) do not, the Principle C violation could be rescued in the former, as opposed to the
latter, by the intervention of Quantifier Raising (QR), which breaks up the illicit c-command
relation. See Belletti and Rizzi (1988) for recent arguments that principles of the binding
theory must be allowed to apply at different levels.
22. See also Carlson (1984) for the proposal that events are individuated by the participants
linked to them through thematic relations.
23. An analysis of these data similar to the one proposed in Larson (1988), but involving
closely related proposals by Jacobson (1987), is independently suggested in Dowty (1988).
Dowty also independently notes the facts concerning Right Node Raising discussed below
in section 5.3.
24. Thus, Jackendoff (1971) analyzes such sentences as nongapped and derived by Conjunction
Reduction, whereas Jackendoff (1990) now advocates Gapping. Sag (1976) analyzes them
as involving Left Peripheral Deletion, whereas Sag et al. (1985) draw them under a general
conjunction rule that includes Gapping. Hudson (1982) argues specifically that such sen-
tences do not derive by Gapping and continues to separate them from gapped examples in
Hudson (1989). Neijt (1979) analyzes them as gapped, and Stillings (1975) analyzes them
as nongapped.
25. Jackendoff presents one positive argument for a Gapping analysis of “nonconstituent coor-
dination” based on the claim that in gapped sentences only two constituents may appear in
the second conjunct, one before and one following the gap (ia) (Jackendoff’s (48a)). He then
argues that the “nonconstituent coordinations” discussed in Larson (1988) show this same
distribution, judging (ib) (Jackendoff’s (50)) to be similarly bad:
(i) a. ??Harry bought a book at 6:00 in Harvard Square, and Fred at 9:15 in Watertown.
b. ?I wrote nothing to Mary in the morning and hardly anything to Max during the afternoon.
This argument is compromised in two ways, however. First, the data judgments Jackend-
off records are highly dubious, and not supported in the literature. I myself find (ib) fully
acceptable. Similarly for the following example, cited by Hudson (1982) against the view
that Jackendoff is defending here:
(ii) John gave the books to Mary at Christmas and the records to Sue for her birthday.
As Hudson notes, one can extend sentences like (ii) to include additional modifiers, without
serious degradation:
(iii) John gave the books to Mary at Christmas and the records to Sue for her birthday after
the wedding.
Second, as pointed out by Sag et al. (1985), it is not clear how valid the “two-constituent”
test is in any event. Sag (1976) cites acceptable Gapping examples like (iv), which involve
more than two constituents:
(iv) a. Peter talked to his boss on Tuesday, and Betsy to her supervisor on Wednesday.
b. John talked to his supervisor about his thesis, and Erich to the dean about depart-
mental politics.
Thus, neither the data in this argument nor the test they are invoked in support of appear
very secure.
26. Neijt (1979) herself rejects an across-the-board extraction analysis of the English cases on
the basis of data involving English both and Dutch en ‘both’. Neijt claims the acceptability
contrast between the Dutch and English examples recorded in (i) and (ii). She argues that
this contrast can be explained by appeal to the independent fact that both and en are licensed
with phrasal conjunction, but not with S conjunction. If the Dutch cases involve VP conjunc-
tion plus across-the-board raising, then en will join phrases—VPs—as desired. On the other
hand, if the English cases involve S conjunction and Gapping, then we predict that both will
be unable to appear—a correct prediction, according to Neijt’s judgments:
(i) a. *John gave both Mary an apple and Peter a pear.
b. Jan gaf en Marie een appel en Piet een peer.
(ii) a. *John gave either an apple to Mary or a pear.
b. Jan gaf of een appel aan Marie of een peer aan Piet.
As a nonspeaker of Dutch, I am unable to evaluate the Dutch/English contrast fully. Nonethe-
less, I observe that if Neijt’s argument is correct, we predict a strong contrast in acceptability
between (ia)/(iia) and (iiia,b), where overt VP conjunction is involved and the question of S
conjunction does not arise. The latter pair should be much better:
(iii) a. John both gave Mary an apple and gave Peter a pear.
b. John either gave an apple to Mary or a pear.
I myself find no such contrast. In my speech, (ia), (iia), and (iiia,b) are equally acceptable.
And in certain cases, such as (iva, b), I judge the nonconstituent coordination with both to
be better than its counterpart with full VPs:
(iv) a. John considers both Mary intelligent and Peter hardworking.
b. John both considers Mary intelligent and considers Peter hardworking.
Hence, Neijt’s argument for a fundamental difference between English and Dutch appears
unconvincing. The behavior of both does not seem to distinguish overt VP coordination and
“nonconstituent coordination” in any significant way.
27. A number of authors have argued plausibly that RNR involves a form of discontinuous con-
stituency in which the “raised” expression is a simultaneous daughter of two mother nodes.
Thus, McCawley (1982) analyzes RNR as involving trees with “crossing branches,” and
Erteschik-Shir (1987) develops a related view using the analysis of across-the-board extrac-
tions in Williams (1978) (see also Levine 1985 and McCloskey 1986). These proposals, if
correct, do not prejudice the point made here—namely, that the right-peripheral element is
a constituent.
28. In unpublished work, Bowers and Williams have suggested that examples like (i), showing
both “nonconstituent conjunction” and Light Predicate Raising, create problems for Larson
(1988):
(i) a. ?You left on the shelf all the shirts and in the suitcase all the socks.
b. I consider intelligent anyone who can add and ambitious anyone who can subtract.
I believe such data can be accommodated under the recent proposal by Pollock (1989) and
Chomsky (1989) that English has verb movement to functional head positions. Suppose that
(ia) involves an underlying structure like (ii), with a V′ conjunction and a functional head
position [α e] above VP:
(ii)
[α e [VP you [ V′ [V′ e [VP all the shirts [V′ leave on the shelf]]] and
[V′ e [VP all the socks [V′ leave in the suitcase]]]]]]
First, the two instances of leave raise separately to the empty verb positions in the V′ con-
juncts, and then they extract across-the-board to [α e]:
(iii) [α leave [VP you [V′ [V′ t [VP all the shirts [V′ t on the shelf]]] and
[V′ t [VP all the socks [V′ t in the suitcase]]]]]]
Next, there is V′ Reanalysis in the separate conjuncts and raising around the object NPs:
(iv) [α leave [VP you [V′ [V′ [V t on the shelf] [VP all the shirts t]] and
[V′ [V t in the suitcase] [VP all the socks t]]]]]
Finally, you raises around leave to subject position (not shown), deriving the surface order
of (ib). Interesting technical questions arise in this analysis regarding the identity of α and
its precise position. But such examples seem to pose no insuperable difficulty for LPR and
the V-Raising account of nonconstituent conjunction taken together.
29. Hoeksema (1987) also uses these facts to argue for the general constituency in secondary
predication adopted here. Note that this account suggests the intriguing possibility of uni-
fying standard VP-Deletion and a large number of Gapping cases as alternative forms of
V′-Ellipsis. The former would correspond to Intransitive V′-Ellipsis, where the largest V′ not
containing the subject is elided; the latter would correspond to Transitive V′-Ellipsis, where
the largest V′ not containing the direct object is elided.
REFERENCES
Abbott, B. (1976) “Right Node Raising as a Test for Constituenthood,” Linguistic Inquiry 7:
639–642.
Anderson, S. (1971) “On the Role of Deep Structure in Semantic Interpretation,” Foundations of
Aoun, J. and Y.-H. A. Li (1989) “Scope and Constituency,” Linguistic Inquiry 20: 141–172.
Baker, M. (1988) Incorporation. Chicago: .University of Chicago Press.
17: 347–354.
Belletti, A. and L. Rizzi (1988) “Psych-Verbs and θ-Theory,” Natural Language and Linguistic
Theory 6: 291–352.
Bowers, J. (1981) The Theory of Grammatical Relations. Ithaca, NY: Cornell University Press.
Carlson, G. (1984) “Thematic Roles and Their Role in Semantic Interpretation,” Linguistics 22:
259–279.
Chomsky, N, (1981) Lectures on Government and Binding. Dordrecht: Foris
———. (1986a) Barriers. Cambridge, MA: MIT Press
———. (1986b) Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger.
———. (1989) “Some Notes on Economy of Derivation and Representation,” in I. Laka and A. Maha-
jan, eds., MIT Working Papers in Linguistics 10. (pp. 43–74) Cambridge, MA: Department of
Linguistics and Philosophy, MIT.
Cohen, R. (1987) “Analyzing the Structure of Argumentative Discourse,” Computational Linguistics
13: 11–24.
Collins, C. (1988) “Conjunction Adverbs,” unpublished manuscript, MIT, Cambridge, MA.
Contreras, H. (1984) “A Note on Parasitic Gaps,” Linguistic Inquiry 15: 698–701.
Davidson, D. (1967) “The Logical Form of Action Sentences,” in N. Rescher, ed., The Logic of
Decision and Action. (pp. 81–120) Pittsburgh, PA: University of Pittsburgh Press.
Dowty, D. (1988) “Type Raising, Functional Composition, and Non-Constituent Conjunction,”
in R. Oehrle, E. Bach, and D. Wheeler, eds., Categorial Grammars and Natural Language
Structures. (pp. 153–197) Dordrecht: Reidel.
Dryer, M. (1987) “On Primary Objects, Secondary Objects and Antidative,” Language 62: 808–845.
Emonds, J. (1987) “The Invisible Category Principle,” Linguistic Inquiry 18: 613–632.
Erteschik-Shir, N. (1987) “Right Node Raising,” in M. Browning, E. Czaykowski-Higgins, and E. Rit-
ter, eds., MIT Working Papers in Linguistics 9. (pp. pp. 105–117) Cambridge, MA: Department
of Linguistics and Philosophy, MIT.
Grimshaw, J. (1989) “Getting the Dative Alternation,” in I. Laka and A. Mahajan, eds., MIT
Working Papers in Linguistics 10. (pp. 113–122) Cambridge, MA: Department of Linguistics
and Philosophy, MIT.
Gropen, J., S. Pinker, M. Hollander, R. Goldberg, and R. Wilson (1989) “The Learnability and
Acquisition of the English Dative Alternation,” Language 65: 203–257.
Grosu, A. (1976) “A Note on Subject Raising to Object and Right Node Raising,” Linguistic
Inquiry 7: 642–645.
Grosz, B. (1978) “Discourse Knowledge,” in D. Walker, ed., Understanding Spoken Language
(pp. 228–345) New York: North-Holland.
———. (1981) “Focusing and Description in Natural Language Dialogues,” in A. Joshi, B. L.
Webber, and I. A. Sag, eds., Elements of Discourse Understanding, (pp. 85–105) Cambridge:
Cambridge University Press.
Grosz, B., M. Pollack, and C. Sidner (1989) “Discourse,” in M. Posner, ed., Foundations of Cogni-
tive Science. (pp. 437–468) Cambridge, MA: MIT Press.
Grosz, B. and C. Sidner (1986) “Attention, Intensions, and the Structure of Discourse,” Computa-
tional Linguistics 12: 175–204.
Guindon, R., P. Sladsky, H. Brunner, and J. Conner (1986) “The Structure of User-Adviser Dialogues:
Is There Method in Their Madness. Proceedings of the 24th Conference of the Association for
Computational Linguistics, (pp. 224–230). New York: Columbia University.
Hale, K. (1986) “Notes on World View and Semantic Categories: Some Warlpiri Examples,” in
P. Muysken and H. van Riemsdijk, eds., Features and Projection. (pp. 233–254) Dordrecht:
Foris.
———. (1989) “Subject Obviation, Switch Reference, and Control,” paper presented at the MIT
Workshop on Control, MIT, March 1989.
Hegarty, M. (1989) “Verb Raising, Adverbs, and Double Objects,” unpublished manuscript MIT,
Cambridge, MA.
Heim, I. (1982) The Semantics of Definite and Indefinite Noun Phrases. Doctoral dissertation,
University of Massachusetts, Amherst.
Heim, I., H. Lasnik, and R. May (1989) “Reciprocity and Plurality,” unpublished manuscript
University of California, Irvine.
Hoeksema, J. (1987) “The Status of Small Clauses in Dutch and English,” unpublished manuscript
University of Pennsylvania, Philadelphia.
Hornstein, N. and A. Weinberg (1990) “The Necessity of LF,” The Linguistic Review 7: 129–168.
Huang, C.-T. J. (1982) Logical Relations in Chinese and the Theory of Grammar. Doctoral disserta-
tion, MIT, Cambridge MA.
Hudson, R. (1982) “Incomplete Conjuncts,” Linguistic Inquiry 13: 547–550.
———. (1989) “Gapping and Grammatical Relations,” Journal of Linguistics 25: 57–94.
Jackendoff, R. (1971) “Gapping and Related Rules,” Linguistic Inquiry 2: 21–35.
———. (1983) Semantics and Cognition. Cambridge, MA: MIT Press.
———. (1990) “On Larson’s Analysis of the Double Object Construction,” Linguistic Inquiry
21: 427–456.
Jacobson, P. (1987) “Phrase Structure, Grammatical Relations and Discontinuous Constituents,”
in G. Huck and A. Ojeda, eds., Syntax and Semantics 20: Discontinuous Constituency. (pp. 27–69)
Johns, A. (1984) “Dative ‘Movement’ in Eskimo,” in D. Testen, V. Mishra, and J. Drogo, eds., Papers
from the Parasession on Lexical Semantics. (pp. 162–172) Chicago: Chicago Linguistic Society,
Kamp, H. (1981) “A Theory of Truth and Semantic Representation,” in J. Groenendijk, T. M. V.
Janssen, and M. Stokhof, eds., Truth, Interpretation and Information. (pp. 277–322) Dordrecht:
Foris.
Kayne, R. (1983) “Datives in English and French,” in Connectedness and Binary Branching. (pp.
193–202) Dordrecht: Foris.
Kiser, D. (1943) Corpus Jurus Secundum. Vol. 38, Brooklyn, NY: American Law Book Co.
Larson, R. (1985) “Bare-NP Adverbs,” Linguistic Inquiry 16: 595–621.
———. (1987) ‘“Missing Prepositions’ and the Analysis of English Free Relative Clauses,” Linguistic
Inquiry 18: 239–266.
———. (1988) “On the Double Object Construction,” Linguistic Inquiry 19: 335–391.
———. (1989) “Light Predicate Raising,” MIT Lexicon Project working Papers 27, Cambridge,
MA: MIT.
Larson, R. and R. May (1990) “Antecedent Containment or Vacuous Movement: Reply to Baltin,”
Lebeaux, D. (1983) “A Distributional Difference between Reciprocals and Reflexives,” Linguistic
Inquiry 14: 423–440.
Levine, R. (1985) “Right Node (Non-)Raising,” Linguistic Inquiry 16: 492–497.
Linebarger, M. (1987) “Negative Polarity and Grammatical Representation,” Linguistics and Phi-
losophy 10: 325–387.
May, R. (1985) Logical Form. Cambridge, MA: MIT Press.
McCawley, J. (1982) “Parentheticals and Discontinuous Constituent Structure,” Linguistic Inquiry
13: 91–106.
McCawley, J. (1988) “Adverbial NPs,” Language 64: 583–590.
McCloskey, J. (1986) “Right Node Raising and Preposition Stranding,” Linguistic Inquiry 17:
183–186.
McNulty, J. (1978) Federal Taxation of Individuals in a Nutshell. St. Paul, MN: West Publishing Co.
Neijt, A. (1979) Gapping. Dordrecht: Foris
Oehrle, R. (1976) The Grammatical Status of the English Dative Alternation. Doctoral dissertation,
MIT, Cambridge, MA.
Parsons, T. (1985) “Underlying Events in the Logical Analysis of English,” in E. LePore and B. Mc-
Laughlin, eds., Actions and Events. (pp. 235–267) Oxford: Basil Blackwell.
———. (1991) Events in the Semantics of English. (pp.235–267) Cambridge, MA: MIT Press.
Partee, B. (1965) Subject and Object in Modern English. Doctoral dissertation, MIT, Cambridge,
MA (Published as Subject and Object in Modern English, New York: Garland 1979.)
Pinker, S. (1989) The Acquisition of Argument Structure. Cambridge, MA: MIT Press.
Pollock, J.-Y. (1989) “Verb Movement, Universal Grammar, and the Structure of IP,” Linguistic
Inquiry 20: 365–424.
Randall, J. (1987) Indirect Positive Evidence: Overturning Generalizations in Language Acquisi-
tion, unpublished manuscript, Indiana University Linguistics Club, Bloomington, IN.
Reinhart, T. (1983) Anaphora and Semantic Interpretation. London: Croom Helm.
Research Institute of America, Inc. (1978) Federal Tax Coordinator 2d. New York: Research
Institute of America, Inc.
Ross, J. R. (1967) Constraints on Variables in Syntax. Doctoral dissertation, MIT, Cambridge,
MA. (Published as Infinite Syntax, Norwood, NJ: Ablex, 1986.)
Russell, B. (1903) The Principles of Mathematics. New York: Norton
Sag, I. (1976) Deletion and Logical Form. Doctoral dissertation, MIT, Cambridge, MA.
Sag, I., G. Gazdar, T. Wasow, and S. Weisler (1985) “Coordination and How to Distinguish Catego-
ries,” Natural Language and Linguistic Theory 3: 117–171.
Schneider-Zioga, P. (1988) “Double Objects and Small Clauses,” unpublished manuscript, University
of Southern California Los Angeles, California.
Solan, L. (1983) Pronominal Reference. Dordrecht: Reidel.
Stillings, J. (1975) “The Formulation of Gapping in English as Evidence for Variable Types in Syntactic
Transformations,” Linguistic Analysis 1: 247–273.
Tenny, C. (1987) Grammaticalizing Aspect and Affectedness. Doctoral dissertation, MIT, Cambridge,
MA.
Williams, E. (1978) “Across-the-Board Rule Application,” Linguistic Inquiry 9: 31–43.
3 Promise and the Theory of Control
Richard K. Larson
Control in promise constructions like (1a) raises a number of interesting questions for
grammatical theory. As is well known, promise is one of a small number of verbs in
English that select an object and an infinitive and show subject control. This behavior
departs from the far more common pattern of verbs like persuade, and force, which take
an object and an infinitive and show object control (1b):
(1) a. John promised Mary to return home by 5:00 p.m.
{
persuaded
b. John forced }
Mary to return home by 5:00 p.m.
One of the most intriguing facts about the “special” control behavior of promise is
that it is not an isolated difference; promise actually diverges from verbs like persuade
and force in a number of important respects. The suspicion naturally arises then that
some reduction can be made—that it might be possible to derive the control behavior
of promise from other, independent properties of the verb.
In this article I argue for such a reduction. Pursuing an observation by Bowers (1973),
I propose that the control behavior of promise derives from its status as a dative verb, and
from a resulting formal parallel between examples like (1a) and double object constructions
like John promised Mary a sports car. A key element in this analysis is the account of double
object structures developed in Larson (1988b), in which these constructions are derived syn-
tactically from more basic. oblique forms. I show that the D-Structure representations avail-
able in this account make it possible to predict control structurally using a Minimal Distance
Principle analogous to that of Rosenbaum (1970). The control and selectional properties of
promise are thus linked together in a simple and intuitively satisfying way.
In section 1 I briefly review the syntactic behavior of promise, suggesting its status as a
double object verb. In section 2 I introduce the account of complementation and double
object structures from Larson (1988b), applying it to persuade, force, and promise. In sec-
tion 3 I relate these structures to control via the Minimal Distance Principle and contrast
the analysis with the structurally based theories of Stowell (1981), Thomason (1976),
Bach (1979), and Bach and Partee (1980). In section 4 I briefly consider how the struc-
tures assumed in this account are projected and licensed. Finally, in section 5 I discuss two
classes of problems for a configurational account of control with promise.
1. PROMISE AS A DOUBLE OBJECT VERB
The underlying syntactic character of infinitival promise constructions is directly sug-

gested by the paradigm in (2), first noted (to my knowledge) by Bowers (1973), and
subsequently by Stowell (1981):
Promise and the Theory of Control 231
(2) a. i. John promised [Mary] [a sports car].
ii. John promised [a sports car] [to Mary].
b. i. John promised [Mary] [to leave].
ii. ?John promised [to leave] [to Mary].
Along with its NP-Infinitive subcategorization, promise also permits two NP objects,
which may appear in either a “double object” configuration (2ai) or an oblique dative
configuration (2aii). Furthermore, the usual V-NP-Infinitive construction with prom-
ise (2bi) also has a (somewhat marginal) oblique variant (2bii). These facts suggest
that the pairs in (2a,b) are formally parallel; more precisely, they imply that promise-
NP-Infinitive constructions are analogous to double object structures, with the NP
object and infinitive of the former parallel to the inner and outer objects (respectively)
of the latter.
There is evidence beyond the parallels in (2) supporting this view. As noted
by Stowell (1981), infinitival promise constructions resist extraction of their NP
objects (3a–c) just as double object structures resist extraction of their inner objects
(4a–c):1
(3) a. ??Who do you think John promised to leave?
b. *John was tough to promise to leave.
c. ??John promised to leave [all the people who didn't want him there].
{
(4) a. ??Who do you think John promised
gave }
a sports car?
{
b. *John was tough to promised
gave }
a sports car.
{
c. ??John promised
gave }
a sports car [everyone that helped him].
Correlatively, promise permits the extraction of its infinitival complement, just as dou-
ble object verbs permit the extraction of their outer objects:
(5) a. What did John promise Mary?
(Ans.: To leave by five o’clock.)
b. What John promised Mary was [to leave by five o’clock].
{
(6) a. What did John promised
gave }
Mary?
(Ans.: A sports car.)
{
b. What John promised
gave }
Mary was [a sports car].
Promise with an infinitive also patterns analogously to a double object structure in the
realization of its complements:
(7) a. John promised ϕ to leave.
b. ??John promised Mary ϕ.
(8) a. {
John promised
gave }
ϕ a donation.
{
b. ??John promised
gave }
the charity ϕ.
The distribution of other predicate’s in English superficially similar to promise shows

clearly that such behavior is correlated with double object taking status. Consider the
verbs force and persuade. The latter resemble promise in selecting the V-NP-Infinitive
complement frame:
{ }
(9) a. John persuaded Mary to leave.
forced
b. John promised Mary to leave.
But they differ in having neither double object nor oblique dative subcategorizations
(compare (2)):
(10) a. *John persuaded Mary a conclusion.
*John forced Mary an action.
b. *John persuaded a conclusion to Mary.
*John forced an action to Mary.
c. John persuaded Mary to leave.
John forced Mary to leave.
d. *John persuaded to leave to Mary.
*John forced to leave to Mary.
Force and persuade also sharply diverge from promise with respect to the data in (3)–(8).
In fact, the complements of the former show behavior essentially the inverse of the latter.
Thus, force and persuade allow extraction of their NP objects but resist extraction of
their infinitival complement (compare (3a–c), (5a,b)):
{persuaded
(11) a. Who do you think John forced }
to leave?
{ }
b. John was tough to persuade to leave.
force
{
c. John forced }
persuaded to leave [all the people who had no business being there].
{ }
(12) a. *What did John persuade Mary?.
force
(Ans.: To leave by five o’clock.)
{persuaded
b. *What John forced }
Mary was [to leave by five o’clock].
And force and persuade require the presence of an object while permitting absence of
the infinitive (compare (7a,b)):
{persuaded
(13) a. *John forced }
ϕ to leave.
b. John finally persuaded Mary ϕ.
John forced Mary ϕ against her will.
These contrasts are straightforward under the premise that infinitival promise con-
structions are double object structures whereas infinitival force and persuade construc-
tions are not. Recall that a salient property of double object structures is that the under-
lying grammatical relations of their complements appear “inverted” in surface form;
the underlying indirect object surfaces as a direct object, and the underlying direct
object surfaces as an oblique of some kind. This means that the associations between
grammatical relations ofthe complements of promise versus persuade and force are as
follows:
promise NP INFINITlVE
persuade
force NP INFINITIVE
Given this situation, it comes as no surprise that complements of promise should behave
“inversely” to those of force and persuade, with the NP of the former patterning with
the infinitive of the latter two, and conversely.2
2. THE SYNTAX OF DOUBLE OBJECT CONSTRUCTIONS
Assuming that NP-Infinitive complements to promise involve a double object struc-

ture, let us turn to the analysis of the latter. The account I will adopt here is one
developed in Larson (1988b), whose central idea is that double object structures are
syntactically related to oblique forms by a modern version of “Dative Shift.” The
relevant D-Structure realizations derive from early proposals by Chomsky (1955)
and their more recent elaboration in the work of Dowty (1978), Bach (1979), and
Jacobson (1983, 1987). The basic idea is that in an example like (14a), all verbal
arguments are initially structured within VP in the subject-predicate form, as shown
in (14b):
(14) a. John [VP promised a reward to Mary].
b. VP
NP V′
John V VP
e NP V′
a reward V PP
promise to Mary
On this view, the VP underlying promise a reward to Mary is a binary branching structure;
it consists of an empty V taking a VP complement whose specifier is a reward, whose head
is promise, and whose complement is the PP to Mary. The intuitive content of this struc-
ture is that promise takes the complement to Mary, forming a small predicate promise-
to-Mary. This is predicated of an “inner subject” a reward, forming a VP with clauselike
structure: a reward promise to Mary. The latter is then predicated of the subject John.
The correct surface ordering of sentence constituents, shown in (15), derives by
movement of promise to the empty V position and by raising of the VP-internal subject
to SpecIP position:
(15) IP
NPi I′
John I VP
NPi V′
t Vj VP
promise NP V′
a reward Vj PP
t to Mary
This movement is suggested to follow from general principles governing the assign-
ment of Case and Agreement: promise must head a projection governed by Infl to
receive Tense and Agreement information, and the object a reward must be governed
(and hence c-commanded) by V in order to receive Case. The verb raises to meet
these joint requirements. Likewise, the subject NP must receive Case and raises to
SpecIʹ for nominative Case assignment.3 Since we will be concerned here only with
relations holding within VP, we will henceforth ignore IP structure and subject raising
for simplicity.
Double object examples like John promised Mary a reward derive transformation-
ally from to-dative forms similar to (14b). In particular, they arise when the familiar
set of operations responsible for passive sentences applies within VP. Under standard
proposals, the derivation of passives involves two central effects: withdrawal of Case
from an object and dethematization of a subject position. This triggers NP Movement
of the object to subject position. The suppressed subject θ-role is (optionally) realized
by an adjunct phrase attached to a V projection (V′):
(16) VP
NPi V′
Mary V′ PP
V NPi by John
seen t
Assume now that the dative preposition to has the status of Case marking. And con-
sider imposing the general suite of effects involved with passives on the lower clauselike
VP in (14b), as shown in (17). Case is withdrawn from Mary, which, in this instance,
amounts to saying that the preposition to is absorbed. Furthermore, the VP-subject posi-
tion is dethematized, and hence empty. This constellation of effects triggers NP-Movement
of the Caseless indirect object to the empty VP subject position. The suppressed VP subject
θ-role is again realized by an adjunct phrase attached to a V projection (V′):
(17) VP
NP V′
John V VP
e NPi V′
Mary V′ NP
V NPi a reward
promise t
Finally, as before, the verb raises, achieving the correct surface order of constituents
shown in (18):
(18) VP
NP V′
John Vj VP
promise NPi V′
Mary V′ NP
Vj NPi a reward
t e
This is (modulo Subject Raising) the S-Structure configuration for a double object verb
phrase. Evidently, the VP in (16) and the lower VP in (18) are closely analogous.4
2.1. Persuade and Force versus Promise

This analysis of complementation and double object formation can be applied to yield
very different structures for the two classes of verbs discussed above taking a surface
object and an infinitival complement. Consider first examples with persuade and force
like (1b) (repeated here):
(1) b. John { persuaded

forced } Mary to return home by 5:00 p.m.
Under the present account we can assign such examples the D-Structure form in (19),
where Mary is an underlying subject of a small predicate persuade to return home by
5:00 p.m. (or force to return home by 5:00 p.m.):5
(19) VP
NP V′
John V VP
e NP V′
Mary V α
} persuade
force } to return
home by 5:00 p.m.
This structure subsequently undergoes V-Raising at S-Structure, as in (20):
(20) VP
NP V′
John V VP
} persuade
force } NP V′
Mary V α
e to return
home by 5:00 p.m.
No further movements apply in the derivation of (1b); the relative positions of the
verbal complements (NP and α) remain constant. This accords with our earlier observa-
tion that constructions with persuade and force preserve their underlying grammatical
relations in surface form: the surface direct object (Mary) is also a deep direct object,
and so on.
Consider now examples with promise like (1a) (repeated here):
(1) a. John promised [Mary] [to return home by 5:00 p.m.].
As noted earlier, such sentences are formally analogous to double object construc-
tions. Under the analysis of double objects presented above, this entails an underlying
D-Structure representation like that in (21):
(21) VP
NP V′
John V VP
e NP V′
e V′ α
V NP to return
home by 5:00 p.m.
promise Mary
As in (17), the VP-subject position is dethematized and hence empty. The notional direct
object (to return home by 5:00 p.m.) appears as a V′ adjunct. Furthermore, the Case of
the indirect object (the preposition to) has been absorbed. These circumstances trigger
NP-Movement of the indirect object to direct object (VP-subject) position. And there
is also the usual raising of V, yielding the derived VP structure of (1a) shown in (22):
(22) VP
NP V′
John
Vi VP
promise
NPj V′
Mary
V′ α
Vi NPj to return
home by 5:00 p.m.
e e
Here, unlike the case of persuade and force, there is a change of deep and surface
grammatical relations. Promise involves an inversion of direct and indirect object gram-
matical relations, just as standard passives involve an inversion of subject and object
grammatical relations. Despite the surface similarities between (1a) and (1b), then, their
structures and derivational histories are radically different. The latter is an “active”
form, whereas the former is fundamentally a “passive.”
It is natural to inquire why the oblique counterparts of infinitival promise construc-
tions are somewhat marginal—why examples like (23a), with the structure in (23b), are
less acceptable than (15):
(23) a. Jon promised [to return home by 5:00 p.m.] [to Mary].
b. VP
NP V′
John Vi VP
promise
α V′
to return Vi PP
home by 5:00 p.m.
e to Mary
I suggest the relevant factor is “Case resistance” in the sense of Stowell (1981). Briefly,
in the structure in (23b), the infinitival to return home by 5:00 p.m. is in the position
of direct objects, and hence in the position of structural accusative Case assignment.
Assume (essentially following Stowell (1981)) that infinitival complements are analo-
gous to PPs in being “intrinsically Case-marked.” Furthermore, assume (again follow-
ing Stowell) that this status is incompatible with receiving structural accusative Case.
These assumptions entail that an infinitive cannot be realized as a surface direct object
even when it has this notional status underlyingly. (23b) therefore has the status of a
Case theory violation.
The double object structure provides a resolution for the conflicting demands of Case
theory and θ-theory involved with infinitival promise constructions. For even though
the infinitive is notionally a direct object, the double object derivation with promise
allows it to be realized as an adjunct, and hence in a position not targeted by structural
accusative Case.6
3. CONTROL
The structures proposed above for persuade, force, and promise have been introduced
independently of the issue of control. Nonetheless, they can be related very directly to
the facts of controller choice. We can make the connection through the following ver-
sion of Rosenbaum’s (1970) “Minimal Distance Principle”:
Minimal Distance Principle (MDP)
An infinitive complement of a predicate P selects as its controller the minimal

c-commanding noun phrase in the functional complex of P.
The notion of a “functional complex for predicate P” derives from Chomsky (1986),
where it is used to define the local domain of anaphors under the binding theory. In brief,
the “functional complex of P” is the structural domain in which thematic roles determined
by P are discharged. Anaphors are required to obtain an antecedent within the functional
complex of their governor. Here we are, in effect, treating control as a form of anaphora
with the extra proviso that infinitives select the closest available antecedent.7
Suppose we assume now that the MDP applies at D-Structure, the level where the-
matic relations are transparently represented. Then we predict controller choice cor-
rectly for our two verb classes. In constructions involving persuade and force such
as (19), the surface object (Mary) is the closest NP in the functional complex of the
verb that c-commands the infinitive at D-Structure. Hence, we predict object control.
Contrastingly, in constructions involving promise such as (21), the eventual surface
direct object (Mary) fails to c-command the infinitive at D-Structure and hence is not
a possible controller. The closest available NP in the functional complex of promise is
the higher VP subject John. Hence, we predict subject control.
Under this view, the “special” control behavior of promise dissolves as such. Con-
troller choice follows directly from the status of promise as a double object verb under
the MDP.
3.1. Two Generalizations

This analysis can accommodate two well-known generalizations concerning argument
manipulations with control verbs.
3.1.1. Visser’s Generalization
Visser (1963–73) observes that passivization is possible with object control verbs (24a)
but not with subject control verbs (24b,c):8
{
persuaded
(24) a. John was forced }
to leave.
b. *It was tried to leave.
c. *John was promised to leave.
This generalization may be taken to follow from our principles of control. Note first
that (24a–c) all involve D-Structure representations whose highest VP-subject position
is empty. In (24a) this situation has no consequences for control; at D-Structure the
infinitive will have a c-commanding NP to serve as controller (namely, the underlying
object John). The latter will thus be selected as controller just as in the active case, as
shown in (24aʹ). The surface form is generated by raising John to SpecVP position and
then to subject position, and by raising V to [V e).
With (24b,c), however, the absence of an underlying VP subject does have conse-
quences. In these examples there simply is no c-commanding NP at D-Structure to
serve as controller for to leave. All commanding argument positions are nonthematic
and hence empty, as in (24bʹ) and (24cʹ). Since some controller is required, but none is
available, these sentences are ruled out.9,10
(24) a′. IP
NP I′
e I VP
was NP V′
e V VP
e NP V′
John V α
persuade to leave
(24) b′. IP
NP I′
e I VP
was NP V′
e V α
try to leave
(24) c′. IP
NP I′
e I VP
was NP V′
e V VP
e NP V′
e V′ α
V NP to leave
promise John
3.1.2. Bach’s Generalization

Bach (1979) points out that detransitivization is available with subject control verbs but
proscribed with object control verbs:
(25) a. John promised to leave.
{ }
b. *John persuaded to leave
forced
This observation cannot be explained along the same lines as Visser’s Generalization,
where ungrammaticality results from the absence of a controller. If we take examples
like (25b) to arise by simply not projecting the direct object at D-Structure, then the
VP subjects of persuade and force will constitute potential controllers for the infinitive
under the MDP, as shown in (25bʹ):
(25) b′. VP
NP V′
John V α
persuade to leave
force
To account for Bach’s Generalization, we will take a different tack: we appeal to the
general difference in deletability of complements noted earlier (recall (7)–(8) and (13)):
(26) a. John promised the charity a donation.
John promised a donation.
*John promised the charity.
b. John persuaded Mary of a certain conclusion.
*John persuaded of a certain conclusion.
John persuaded Mary.
(26a,b) show that the “inverse” behavior of promise and persuade with respect to which
complements can “delete” is independent of control. It is observed not only with infini-
tival complements as in (7), (13), and (25) but also with NP objects, where anaphoric
dependence is not involved.
On the present account, the contrast in (25)–Bach’s Generalization–may be assimi-
lated to this latter, independent difference. Essentially, it seems that only internal argu-
ments other than themes are optionally projected at D-Structure in English. Structures
like (25bʹ) are thus ruled out, not on grounds of illicit control relations, but because they
violate the Projection Principle (Chomsky (1981)). They fail to syntactically represent
an argument that in fact must be structurally represented.
3.2. An Alternative Account

It is revealing to contrast the account of control in promise constructions developed here
with the general family of proposals by Stowell (1981), Thomason (1976), Bach (1979),
and Bach and Partee (1980).
Stowell’s (1981) analysis involves the structure shown in (27) for promise-NP-Infinitive:
(27) IP
NP I′
John I VP
will V S′
V NP PRO to leave
promise Mary
In (27) the verb and NP object form a constituent that excludes the infinitive. Stowell
suggests that the category of this small constituent is V—that promise and Mary in effect
make up a complex lexical verb taking the infinitive as an object.
Essentially the same structure is proposed within the framework of Categorial Gram-
mar by Thomason (1976), Bach (1979), and Bach and Partee (1980). They assign the
following derivation tree:
(28) John will promise Mary to leave
John promise Mary to leave
promise Mary to leave
promise Mary
Here constituency is indicated in the relative order of combination of the verb and its
complements. Promise first combines with its NP argument to yield a complex predi-
cate promise Mary, as in Stowell’s analysis; the latter then combines with the infinitive
to leave.
These structural analyses can be related to control under a Minimal Distance Prin-
ciple just as in the account proposed here. Thus, in Stowell’s (27) John is the minimal
c-commander of the infinitive; hence, this structure predicts subject control. The same
result can be obtained for analysis trees, under the following, trivially modified form
of the MDP:
MDP‫׳‬
An infinitive α selects as its controller the first NP that combines with a constituent
containing α.
In (28) the first NP composed with a phrase containing to leave is the subject John;
hence, MDPʹ predicts subject control.
These proposals are similar in spirit to the one advanced here in attempting to
predict controller choice structurally under a Minimal Distance Principle. However,
they diverge sharply from the present account on the issue of where the MDP
applies and the structure it applies to. Stowell, Thomason, Bach, and Partee assume
that the level where the MDP applies, and where the surface NP object of promise
fails to c-command the infinitive, is S-Structure (or the equivalent). In contrast, the
analysis proposed here assumes that the level where the MDP applies, and where
NP fails to c-command the infinitive, is D-Structure. At S-Structure, these relations
are inverted by (Dative Shift,) and c-command between NP and the infinitive is
established.
As it turns out, the assumption that the MDP applies at S-Structure to structures
like (27) and (28) entails a serious drawback: it requires us to abandon the idea
that control properties of promise follow from its status as a double object verb.
The reason for this is straightforward. It can be shown directly that the structures
in (27) and (28) are not double object structures. This means that Stowell, Thoma-
son, Bach, and Partee must assume quite distinct analyses for promise-NP-Infinitive
and promise-NP-NP, and so must lose the possibility of an explanatory connection
between them.
That (27) and (28) are not double object structures is demonstrated by the results
of Barss and Lasnik (1986). The authors cite a variety of data showing that in a double
object structure, the inner object asymmetrically c-commands the outer object. For
example, anaphors must be c-commanded by their antecedents. In a double object
structure the inner object may be the antecedent for an outer object anaphor, but not
conversely:
{
Mary herself
(29) a. I showed *herself Mary . }
In general, a quantifier must c-command a pronoun at S-Structure if it is to bind it. In
double object structures a quantified inner object may bind a pronominal outer object,
but not conversely:
(29) b. I gave { *itseveryowner

i
contestant his prize
i i
every prize i
}.
Constructions of the form each . . . the other such as Each man saw the other or Each
man saw the other’s friend may have a reciprocal reading when and only when the each-
phrase c-commands the other-phrase. Double object structures show an asymmetry
with respect to the each . . . the other construction on its reciprocal reading:
{ *theeachother’s
(29) c. I sent
friend each man}.
man the other’s socks
Finally, negative polarity items must be c-commanded by an S-Structure “trigger,”

such as a negation or negative quantifier, in order to be licensed. In double object struc-
tures an inner object trigger may license an outer object negative polarity item, but not
conversely:
(29) d. I promised { *anyone nothing }.

no one anything
The derivational analysis of double object structures proposed in section 2 is com-
patible with these data, since, as noted, in post–“Dative Shift” structures the inner
object (asymmetrically) c-commands the outer complement (consider (18) and (22)).
On the other hand, the structures assumed by Stowell, Thomason, Bach, and Partee are
not compatible with Barss and Lasnik’s data. In (27) and (28) the inner object fails to
c-command the outer object.11
These results show that the basic points of our analysis of the control behavior of
promise are intimately connected. Essentially, if we wish to maintain (a), then we can
make use of (b) only if we also hold (c) and (d):
(a) Controller choice is determined by the MDP.

(b) Control constructions involving promise are double object structures.
(c) Double object structures are syntactically derived.
(d) Controller choice is fixed at D-Structure.
Suppose we adopt the MDP. Then since the inner object of a double object structure
c-commands the outer object at S-Structure, assuming that promise-NP-Infinitive is a
double object form entails assuming that NP c-commands the infinitive at S-Structure.
This not only means that the MDP cannot apply at S-Structure, on pain of yielding the
wrong results for promise, but also entails that double object structures must have a
level in which the c-command relations of their complements are inverted and at which
the MDP can apply.
4. PROJECTION OF CONTROL STRUCTURES
Under the view that control is determined at the level of initial syntactic configurations
through the MDP, controller choice largely reduces to the issue of how such initial struc-
tures are projected. In Larson (l988) V-Raising structures are proposed to follow from
the interaction of three simple components: the restricted X-bar theory in (30) and the
principles of argument realization in (31a,b):
(30) a. XP → YP X′
b. X′ → X ZP
(31) a. If α is an argument of β, then α must be realized within a projection of β.
b. Roles determined by a predicate α are projected according to the hierarchy

Agent > Theme > Goal > Oblique, such that if θ1 > θ2, then the argument
to which θ1 is assigned c-commands the argument to which θ2 is assigned.
Consider first VPs headed by promise, which determines an agent, a theme, and a
goal. These roles must be projected in conformity with X-bar theory. However, the X-bar
theory in (30) permits at most two arguments of promise to be realized within a single
projection of V. Assuming that infinitives may bear the role of theme like ordinary NPs
(for instance, a reward, a sports car), (31a,b) determine the initial tree shown in (32a):
(32) a. VP
α V′
to leave V PP
promise to Mary
This structure leaves the agent role unprojected, and no site to project it in. To accommo-
date the remaining argument, we must make some elaboration of (32a). (32b) is, in effect,
the minimal structural elaboration compatible with the three principles given above:
(32) b. VP
NP V′
John V VP
e α V′
to leave V PP
promise to Mary
Here an X-bar “shell” with an empty head has been projected to accommodate the agent
phrase. In this structure, all arguments of promise are realized. The structure satisfies
X-bar theory (30). Prominence of roles on the thematic hierarchy is properly reflected
in the c-command relations of the arguments bearing those roles (31b). Finally, all argu-
ments either do (or, after V-Raising, will) fall within a projection headed by V (31a).
D-Structure realizations for “Dative Shift” sentences are determined in a similar way.
We proceed as before, but rather than projecting the theme infinitive to leave in a specifier
position, we instead realize it as an adjunct attached to V′, as shown in (33a):
(33) a. VP
NP V′
e V′ α
V NP to leave
promise Mary
On the assumption that θ-theory is “blind” to whether roles are realized as arguments
or adjuncts, the VP projection in (33a) will count as “filled” (see Larson (1988b) for
discussion); we therefore require another argument position for the agent. This require-
ment is satisfied by projecting the X-bar shell shown in (33b), as above:
(33) b. VP
NP V′
John V VP
e NP V′
e
V′ α
V NP to leave
promise Mary
The derivation then proceeds as in (22), subject control having been fixed from this
structure according to the MDP.
I assume that configurations involving object control verbs like persuade and force are
projected analogously to the oblique promise construction in (32b), but with the infini-
tival complement corresponding to the goal phrase of the latter (recall (19)). This analysis
is independently supported by the data in (34) (brought to my attention by H. Nakajima):
{
into the corner
(34) a. John forced Mary into leaving }
.
b. John persuaded Mary into filling out the complaint.
As (34a,b) illustrate, infinitival complements to persuade and force can be directly

replaced by simple directional PPs expressing goal or result of action. It seems plausible
to conclude, then, that infinitival clauses bear the goal role when realized as comple-
ments to these verbs.12
5. OBJECTIONS
Configurational theories of the kind pursued here encounter two well-known lines of
criticism, one involving the specific verb promise and one involving other superficially
similar “double object” verbs like ask, teach, and allow.
5.1. “Shifting Control”

It has been widely observed that passivization in the infinitival complement of promise
seems to affect control relations in certain cases. Thus, in well-known examples like
(35a) the embedded clause is most naturally construed with the matrix object Mary and
not the matrix subject John. Such behavior appears problematic for theories of control
like the one proposed here. The structural relations among the complements of promise
in (35a) do not appear different from those in the nonpassivized example (35b); hence,
a shift in control is unexpected:
(35) a. John promised Mary to be allowed to leave.

A plausible view of (35a) can be motivated by appeal to certain general seman-

tic properties of dative verbs discussed by Oehrle (1975). Consider typical dative and
double object constructions like (36a,b):
(36) a. John gave an apple to Mary.
b. John sent Mary a letter.
Oehrle points out that such constructions all share an important class of entailments
involving “transfer of possession,” which we might express as follows:
(37) X–Vs–Y–to Z ⇒ Z gets Y for dative Verbs V
Thus, if (36a,b) are true, then as a matter of the semantics of give and send, (38a,b) will
be true as well:
(38) a. Mary got an apple.
b. Mary got a letter.
If John gives an apple to Mary, then possession of the apple is transferred to Mary. Like-
wise, if John sends Mary a letter, then in a course of events where things go as planned,
the letter comes into Mary’s possession. This entailment holding of dative give and send
holds of dative promise as well. Thus, if (39a) is true, then some appropriately modal-
ized version of (39b) is also true:
(39) a. John promised Mary a sports car.
b. Mary gets a sports car.
If John promises Mary a sports car, then in some course of events where things go as
promised, a sports car is transferred to Mary.
Notice that transfer of possession entailments go through not only with concrete
theme nominals like an apple, a letter, or a sports car, but with abstract nominals as well.
Thus, (40a,b) entail (41a,b), respectively (in appropriately modalized forms):
(40) a. John promised permission to leave to Mary.
b. John promised Mary authorization to visit Albania.

(41) a. Mary gets permission to leave.
b. Mary gets authorization to visit Albania.
Interestingly, in the latter examples a further entailment holds. If Mary gets what is
denoted by the second object in (41a,b), then appropriately modalized versions of
(42a,b) are also true:
(42) a. Mary leaves.
b. Mary visits Albania.
If Mary has permission to leave, then in some permissible course of events she leaves.
And if Mary has authorization to visit Albania, then in some permissible course of
events she visits Albania. Mary thus comes to be understood as the subject of the
infinitives in (40a,b) through a chain of entailments: (40a) implies (41a), which in
turn implies (42a), and similarly with (40b). Control is not involved here, as is clear
from the fact that Mary does not c-command the infinitive in (40a) although it is the
notional subject.
I want to propose that what is at work in “shifting control” examples like (35a) is
a species of what is observed in (40)–(42). Specifically, I suggest that the interpretation
of the infinitival in (43a) is determined not by control, as in (43b), but rather through
entailments, as in (43c):13
c. John promised Mary permission to leave.
On this view, the infinitive in (43a) is without a surface controller and hence analogous
to the occurrences of to be allowed to leave in (44):14
{a. unusual
}
(44) a. [To be allowed to leave] is b. always pleasant .
Its construal with Mary in (43a) reflects the fact that, as a dative construction, this sen-
tence entails (a modalized version of) (45a), which in turn entails (a modalized version
of) (45b):
(45) a. Mary gets to be allowed to leave.
b. Mary leaves.
This proposal ties the occurrence of shifting control crucially to the presence of a
dative verb. We thus predict that passivization in the complements of nondative verbs
like persuade and force will not induce similar effects. This prediction appears correct:
(46) a. #I persuaded John to be allowed to leave.

b. #I forced John to be allowed to leave.
The possibility of subject control in such examples is quite marginal in my judgment,
much weaker than in the comparable cases with promise. Given object control, and the
semantics of the complement, the sentences are anomalous.
Further evidence linking shifting control to datives can be seen in the class of infini-
tival complements where shifting control occurs. As has been frequently noted, not
all passive complements of promise allow object construal (47a,b). Moreover, not all
complements that show shifting control are passives. Farkas (1988) observes examples
like (47c), which some English speakers can understand as allowing subject or object
construal of to stay up:
(47) a. #John promised Mary to be kissed by Felix.
b. #Max promised Mary to be rumored to be leaving.
c. The mother promised the children to stay up.
This result can again be understood via the semantics of dative verbs. As noted ear-
lier, datives entail a “transfer” of the theme out of the keeping of the agent, and into
the keeping of the goal. This of course requires the theme to have been in the keeping of
the agent at the point oftransfer.15 Now infinitival complements do not denote objects
that can be possessed like apples or dollar bills. Nonetheless, there is a natural sense in
which the kind of object picked out by an infinitive—an event, or course of action e—
can be “in the keeping” of a person x, namely, if x can bring e about. Following Farkas
(1988), we might say that when x can bring e about, x is potentially responsible for e.
Given then that responsibility for e is the equivalent of possession of e by the source,
we expect shifting control to be possible only when the infinitive denotes an event for
which the source is responsible.16
This expectation is borne out for the good cases of object construal (35a) and (47c)
(repeated here):
(47) c. The mother promised the children to stay up.
We understand the possibility of leaving to be in the control of John in (35a). And we

understand bedtime schedules to be in the control of the mother in (47c). Likewise, the
expectation is borne out for the unacceptable cases (47a,b) (repeated here):
(47) a. #John promised Mary to be kissed by Felix.
b. #Max promised Mary to be rumored to be leaving.
We understand responsibility for kissing to lie with Felix and not John in (47a); hence,
object construal fails. Similarly, we understand the responsibility for the rumor to lie
with no one in (47b), so again object construal fails. These results lend plausibility
to the view that shifting control with promise does not involve control but instead
involves certain entailments associated with the verb. The notion ·’transfer of posses-
sion” appears to illuminate both how the infinitive can be construed with the object of
promise when subject control fails, and also what kind of infinitives will permit such
construal.17
One important question that I must leave unresolved here, but which must be
addressed ultimately if this suggestion is to amount to an actual solution, is when
precisely an infinitival complement is subject to control versus construal by entail-
ments. The issue is evidently a tricky one; in view of (47c), for example, it seems
that application of Passive in the complement of promise is not necessary to trig-
ger entailment construal. Nonetheless, the general division of labor suggested here
between control and entailment seems plausible and seems moreover to answer to
the general intuition that examples of shifting control differ in status from their
controlled counterparts—that although they can be made acceptable to a degree,
they are never as natural as the latter, and indeed are highly variable in their accept-
ability by native speakers. On the view explored here, this would directly reflect the
grammatical status of the control relation versus the semantic/pragmatic status of
the construal relation.
5.2. Apparent Object-Control Double Object Verbs

Under the proposals made above, the subject control properties of promise follow
directly from its status as a double object verb. This leads us to expect subject control
quite generally with double object verbs taking an infinitival complement. Consider the
behavior of verbs like ask, teach, and allow, however. These verbs show a double object
form with NP arguments:
(48) a. Max asked Felix a question.
b. John taught Oscar a song.
c. The judge allowed Lizzie a last request.
But when they occur with infinitival complements, parallel to promise, the result is
object control:
(49) a. Max asked Felix to leave.

b. John taught Oscar to sing.
c. The judge allowed Lizzie to escape.
Superficially, ask, teach, allow, and similar predicates in English (order, tell, permit, and
so on) appear to contradict our generalization linking complementation and control.
On closer examination it becomes clear that verbs like those at issue do not in fact
jeopardize the correlation between control and selection. For we observe that when
these predicates occur with infinitival complements, they diverge from promise in more
than control behavior. For example, note that although (48a–c) have oblique variants,
(49a–c) do not:
(50) a. i. Max asked a question of Felix.
ii. *Max asked to leave of Felix.
b. i. John taught a new song to Oscar.
ii. *John taught to sing to Oscar.

c. i. The judge allowed a last request to Lizzie.
ii. *The judge allowed to escape to Lizzie.
This contrasts with the behavior of promise when it takes an infinitive:
(2) b. ii. ?John promised to leave to Mary.
Furthermore, note that the constraints on extraction of the inner object holding for
infinitival promise (recall (3a–c)) are not observed with ask, teach, or allow:
{asked
(51) a. Who do you think John taught }
allowed to sing?
{ }
ask
teach
b. John was tough to allow to sing.
{
asked
}
c. John taught to sing [all the kids from school].
allowed
Realization of complements, which we saw above to be independent of control, also

distinguishes teach and allow from promise (compare (7)):
(52) a. John taught/allowed Felix to sing.
b. John taught/allowed Felix.
c. *John taught/allowed to sing.
Finally, observe an interesting difference in the extractability of the outer comple-

ment with ask, teach, and allow versus promise. In my judgment, there is a contrast in
the acceptability of NP versus infinitival answers in (53)–(55):
(53) a. What did John ask Felix?
A question about tarantulas./*To sing.
b. What did John ask Felix to do?
To sing.
(54) a. What did John teach Oscar?
A new song./*To sing.
b. What did John teach Oscar (how) to do?
To sing.
(55) a. What did the judge allow Lizzie?
A last meal./*To escape.
b. What did the judge allow Lizzie to do?

To escape.
Unlike promise (recall (5)), ask, teach, and allow seem to permit extraction of an outer
object NP but not an outer infinitive.
The upshot of these facts is clear: when they take infinitival complements, ask, teach,
allow, and similar verbs not only diverge from promise with respect to control but fail to
show any of the latter’s double object behavior. Instead, the three pattern like persuade,
force, and other object control verbs. The correlation between control behavior and
double object complement structure observed with promise thus is maintained, even in
the case of ask, teach, and allow. What is different about these verbs is that their double
object complementation with NP objects is not carried over to infinitival complementa-
tion. Instead, a new pattern seems to assert itself, one like the pattern observed with
force and persuade.
This last result raises a natural question: why does a new pattern of complement
structure emerge with ask, teach, and allow but not with promise? To answer this, we
must examine the verbs in question more closely.
5.2.1. Ask
The problematic status of ask is directly clarified by certain additional facts. Observe first
that ask departs sharply from promise in its NP complementation. Specifically, whereas
the latter shows the familiar oblique dative/double object alternation (56a), the former
does not (56b). In place of to, ask must employ the nondative preposition of (56c):
(56) a. Mary promised John an apple.
Mary promised an apple to John.
b. Mary asked John a question.
*Mary asked a question to John.
c. Mary asked a question of John.
This fact is significant since it is not clear that V-NP-of-NP and V-NP-NP frames are
transformationally related; nor is it clear, even if they are related, that it is the oblique
form that represents the underlying c-command relations (as with promise). Conse-
quently, although promise and ask share a surface V-NP-NP form, it is not evident that
this corresponds to the same syntactic structure in the two cases.
This point is strengthened by a second important difference between ask and prom-
ise. Whereas ask shows V-NP-NP and V-NP-Infinitive forms like promise, it also permits
a V-NP-Interrogative complementation, which promise does not:
(57) a. Mary asked John to{

a question/the time
sing
when to sing . }
b. Mary promised John to {
a song
sing
*whether Bill would sing . }
Notice furthermore that, intuitively, the interrogative semantics of ask overtly expressed
by the wh-clause in (57a) is also covertly present with the NP and infinitival comple-
ments. Infinitives with ask appear always to be understood as implicit polarity ques-
tions; for example, (58a) seems closely related in meaning to (58b):
{
a. to sing
(58) a. Mary asked John b. if he would sing . }
Likewise, outer NPs with ask are largely restricted to two kinds: NPs that refer to ques-
tions (explicitly or implicitly) (59a), or NPs constituting “concealed questions” in the
sense of Grimshaw (1979)(59b):18
{
that same question
(59) a. Mary asked John something . }
{
the time
b. Mary asked John the height of the building . }
{
what the time is
(Compare Mary asked John what the height of the building is .) }
Taken together, these data suggest the following view. Despite surface appearances,
ask-NP-NP and ask-NP-Infinitive are not analogous to promise-NP-NP and promise
NP-Infinitive. Whereas the latter are dative constructions, the former are essentially
“disguised” variants of ask’s V-NP-Interrogative pattern. This conclusion is supported
by the nondative behavior of ask with NPs (56) and by the semantics of infinitival and
outer NP complements (57)–(59), which are always interpreted as “concealed” ques-
tions, or as referring to questions.
Given this result, the issue of control with ask now reformulates itself in an interesting
way. Notice that object control with ask will be explained if outer complements to ask
are uniformly interrogatives, and if interrogatives project “lower” than the inner, theme
object: the latter will be the minimal c-commander for an outer infinitive, yielding object
control under the MDP, as in (60). As it turns out, there are independent grounds for
expecting interrogatives to project in this way. Interrogative complements to ask (and
similar verbs such as inquire and wonder) evidently represent a form of indirect speech,
and as discussed by Munro (1982), indirect (and direct) speech complements are treated
quite generally across natural languages as adjuncts or oblique phrases, failing to trig-
ger transitivity marking and other syntactic processes expected with genuine objectlike
arguments (see Munro (1982) for detailed discussion). If interrogative complements to
ask are indeed thematically adjuncts or obliques, then their inferior structural position
vis-á-vis the object is directly accounted for by the principle in (31 b). The latter ranks
themes higher than obliques on the thematic hierarchy, and furthermore requires this
thematic prominence to be reflected in structural prominence at D-Structure.19
(60) VP
NP V′
Mary V VP
e NP V′
John V XP
ask a question
to leave
if he would leave
5.2.2 Teach
The verb teach is similar to ask in allowing interrogative as well as nominal and infini-
tival complements:
{
a. that song
(61) a. Mary never taught John b. to sing
c. why he should care about tumblebugs
.}
Furthermore, again as with ask, an intuitive interrogative semantics appears to assert
itself with infinitives and certain nominals. Infinitives like (62a) are very naturally
understood as concealed “how to” questions, as in (62b). And nominal examples like
(62c) are naturally construed along the lines of (62d):
(62) a. John taught Felix to sing.
b. John taught Felix how to sing.
c. John taught Felix a song.
d. John taught Felix how to sing a (certain) song.
These data suggest that at least some examples with teach like (6lb) and (62a) might be
analyzed parallel to ask, with infinitives projected similarly to an interrogative clause.
However, there is an important additional fact about teach that bears on control. Unlike
ask, teach is a true dative-shifting verb:
(63) a. Mary taught a new song to John.
b. Mary taught John a new song.
Since teach has not only an interrogative pattern like ask but also a dative pattern like
promise, it follows that we cannot rest with saying that teach shares the former’s control
behavior. Clearly, we must also say something about why it doesn’t seem to show the
control behavior of promise. Since teach exhibits both of the basic complementation
patterns, why doesn’t it exhibit both control paradigms as well?
The answer appears to lie in the interpretation of infinitives when they are inserted
into promise-type structures with teach. Recall that infinitival complements of teach and
promise distribute differently in that the latter allows an oblique dative form whereas
the former generally does not:
?John promised to leave to Mary.
b. John taught Mary to sing.
*John taught to sing to Mary.
Note also that simple infinitival complements are fully well-formed with promise, but
generally ill-formed with teach:
(65) a. John promised to sing.
b. *John taught to sing.
(Compare John taught how to sing.)

Interestingly, there is a context in which both sets of examples improve-namely, when
they can be understood as conveying an injunction of some kind, typically moral:
{ [to honor thy father and mother]

(66) a. Jesus taught [to love thy neighbor as thyself ] .}
{
“Honor thy father and mother!”
b. Compare jesus taught “Love thy neighbor!” . }
Under this type of construal, infinitival complements of teach also behave analogously
to those of promise in another respect. In this situation such complements denote some-
thing that can be referred to by a nominalized form of the verb:
(67) a. John promised [to sing].
Q: What was his promise?
A: To sing.
b. Jesus taught [to love thy neighbor as thyself].
Q: What was his teaching?
A: To love thy neighbor as thyself.
These data point to the following view. Contrary to initial impressions, teach does in
fact realize both its ask-type complementation and its promise-type complementation
with infinitives. The former is represented by infinitives on their “how to” reading; the
latter is represented by infinitives on their “injunctive” reading. This then suggests why
the expected subject control behavior of teach is missing. If the promise—type infinitives
with teach always have an injunctive reading-in other words, are always understood
essentially as “concealed imperatives”—then their understood subjects are always in
a sense antecedently fixed.20 As a matter of semantics, they must always be construed
as referring analogously to generic you or one in examples like You should always
look both ways before crossing the street or One should always look both ways before
crossing the street. The expected infinitival complementation pattern with teach is thus
present, but promise-type subject control is excluded on independent grounds.
It is natural to wonder why infinitives with teach take on injunctive force in promise-
type constructions. Although I don’t have a secure answer, the data in (67) suggest a
clue. Recall that in promise-type constructions, infinitives are inserted in the position
of themes and thus behave essentially as NPs-that is, as “thing”–denoting expressions-
and not as clauses. Suppose the way this occurs is that the infinitive is understood as
nominalizing the action of the verb: as denoting “the promise” or “the teaching.”21
Interestingly, in the latter case there appears to be a strong tendency for the notion of
“teachings” to be understood with some kind of implicit proclamatory force. One might
speculate, then, that the injunctive reading of infinitives with teach arises through the
requirement that they be understood as nominalizations of the verbal action, together
with the way such nominalizations are understood with this particular verb.
5.2.3. Allow
The verb allow is similar to teach in showing a true double object alternation (68)
and hence raises a similar question regarding why this alternation is apparently not
expressed with infinitives:
(68) a. The judge allowed a last request to Lizzie.
b. The judge allowed Lizzie a last request.
As in the previous cases, certain additional properties of allow not manifested by prom-
ise suggest an answer. Recall first that allow differs significantly from promise in per-
mitting Exceptional Case Marking (ECM) complements; this difference is illustrated by
the availability of expletives and idiom chunks in the complement of allow versus their
unavailability with promise:
{
allow
}
(69) a. The government will never *promise there to be a demonstration.
{
allowed
}
b. John *promised the cat to get out of the bag.
Thus, whereas V-NP-Infinitive with promise involves a double object structure, with
allow the situation is evidently more along the lines of (70), where what follows allow
is a clausal constituent:22
(70) VP
NP V′
the judge V IP
allow NP I′
Lizzie I VP
to escape
Given this result, our original question with allow assumes a different and somewhat
sharper form. Instead of inquiring why the dative complementation of allow fails to
manifest itself when the latter occurs with infinitives, we are now interested in how
dative and ECM complementations are related with allow.
A plausible answer can be adapted from Mittwoch (1976), who proposes, in effect,
that V–NP–Infinitive frames for allow and permit license an ECM structure together
with an implicit dative argument.23 On this view, examples like (71a) receive the form
in (71b), where (to Lizzie) is the implicit oblique:
(71) a. The judge allowed Lizzie to escape.
b. The judge allowed [Lizzie to escape] (to Lizzie).
This analysis can capture the familiar deontic versus epistemic ambiguity observed in
(72), where, on the former reading, the government grants permission to John, and
where, on the second reading, the government simply allows the situation to occur:
(72) The government allowed John to leave.

Specifically, we can understand the former as the case where the implicit dative is con-
strued with the ECM subject (73a), and the second as the case where the implicit dative
is construed as a generic or arbitrary pronoun equivalent to one (73b):
(73) . . . allow/permit [α to VP] (to β)
a. Deontic: β = α
b. Epistemic: β = proarb
If this proposal is on the right track, then the differential control behavior of promise
and allow is explained straightforwardly. We see that despite superficial similarity, the
pair in (74) involves control only in the first member; the second is an ECM structure:
b. John allowed Mary to leave.
Furthermore, we observe that the dative complementation of allow manifested with NP

objects is in fact preserved in its infinitival complementation. Under our adaptation of
Mittwoch (1976), ECM structures with allow are actually oblique dative structures with
the ECM complement corresponding to the theme object and with an implicit oblique dative
argument whose identity determines the understanding of the complement as deontic versus
epistemic. As in the previous cases, then, the status of allow as a counterexample to the MDP
appears to dissolve on a more careful examination of its specific grammatical properties.
NOTES
I am grateful to members of the 1986 Linguistic Society of America Summer Institute workshop
“‘Syntax and Semantics: Logical Form and Its Semantic Interpretation” for the stimulating discus-
sion that led to this work, and to Mark Baker, Noam Chomsky, Yoshio Endo, Donka Farkas, Dan
Finer, Riny Huybregts, Anita Mittwoch, Sunseek Oh, David Pesetsky, and Tom Roeper for their
comments. I also thank participants in the 1989 University of California at Irvine Syntax Work-
shop, where some of this material was presented.
1. In Larson (l988b; 1989) it is proposed that “Heavy NP Shift” (as in (3c) and (4c)) does not
actually involve rightward movement of the object NP (ia) but instead involves leftward
movement of a complex predicate phrase (ib):
(i) a. John saw t at the concert [everyone he knew].
b. John [saw at the concert] everyone he knew t.

Under this view, (3a-c) are not in fact all instances of extraction from the inner object
position-only (3a,b) are. (Similarly for (4a-c).) The existence of this alternative analysis of
Heavy NP Shift does not affect the central point at issue, which is the parallel behavior of
promise-NP-NP and promise-NP-Infinitive with respect to this phenomenon.
2. Y. Endo points out to me that surface Case marking of accusative and dative complements in
Japanese appears to exhibit the dative character of promise and the “inverse” complement
relations of promise and persuade in a direct way. Thus, Japanese yakusokusi ‘promise’ exhib-
its the same Case pattern as a dative verb ageru ‘give’ (ia,b). And just as the double object
(double accusative) form is forbidden with the datives, so it is forbidden with ‘promise’ (iia,b):
(i) a. Taroo-ga
Hanako-ni hon-o age-ta.
Taro-nom Hanako-dat book-acc give-past
‘Taro gave a book to Hanako.’
b. Taroo-ga Hanako-ni daigaku-e ikukoto-o yakusokusi-ta.
Taro-nom Hanako-dat college-to go-acc promise-past
‘Taro promised to go to college to Hanako.’
(ii) a. *Taroo-ga Hanako-o hon-o age-ta.
Taro-nom Hanako-acc book-acc give-past
‘Taro gave Hanako a book.’
b. *Taroo-ga Hanako-o daigaku-e ikukoto-o yakusokusi-ta.
Taro-nom Hanako-acc college-to go-acc promise-past
‘Taro promised Hanako to go to college.’
Finally, the equivalent of persuade in Japanese departs from ‘promise’ in showing the NP
object as accusative and the sentential complement in the unmarked position of the oblique
(iii) (compare (ib)):
(iii) Taroo-ga daigaku-e ikuyoo-ni Hanako-o settokusi-ta.
Taro-nom college-to go-so-as-to Hanako-acc persuade-past
‘Taro persuaded Hanako to go to college.’
Endo observes that in certain cases the object of persuade surfaces as a dative:
(iv) Taroo-ga Hanako-ni daigaku-e ikuyoo-ni settokusi-ta.
Taro-nom Hanako-dat college-to go-so-as-to persuade-past
‘Taro persuaded Hanako to go to college.’
However, he points out that in such examples the meaning of the verb shifts so that the sen-
tence no longer carries the implication that the object was “totally affected” by the subject-
specifically, there is no implication that Hanako was persuaded, only that she was urged.
Such examples appear to illustrate what is referred to as “2-to-3 demotion” in Relational
Grammar, and to involve a derived sense of ‘persuade’.
3. Directionality of Case assignment as a motivation for verb raising is proposed by Koopman
(1983) and Travis (1985); inflectional marking by Infl as a motivation for verb raising is
proposed by Roberts (1985).
4. In Larson (1988b), Case assignment to the outer object in a double object structure is
licensed through a rule of V′ Reanalysis that optionally recategorizes V′ as V whenever the
former is thematically equivalent to a monotransitive verb (that is, whenever V′ has a the-
matic structure with one unsaturated internal argument). Applied to (18), this yields (i):
(i) VP
NP V′
John V VP
promise NPi V′
Mary
V NP
V NPi a reward
e t
In this structure the outer NP a reward is in the canonical configuration of objective Case assign-
ment [ vʹ V NP]—and is assumed to receive an inherent objective Case from the reanalyzed V.
5. In what follows I will take no position on the categorial status of infinitives, or on the related
issue of whether they contain an empty subject (PRO). For various points of view on these
questions, see Chomsky (1981), Koster and May (1982), Chierchia (1984), and Borer (1986).
6. Recall from note 4 that Case assignment to the outer object in a double object structure
takes place through optional reanalysis of the lower V′ as V. We may assume that in Dative
Shift derivations involving infinitives, such as (22), V′—Reanalysis simply does not apply,
suspending Case assignment to the adjunct infinitive phrase.
The distribution of infinitives in nominals supports the view that unacceptability in (23a)
arises for Case reasons. Note that (ia,b) are both fully well-formed:
(i) a. John made a promise to Mary to leave.
b. John made a promise to leave to Mary.
On the theory adopted here, this behavior follows directly from the fact that although verbs
assign Case, and hence forbid an adjacent infinitival complement, nominals do not assign
Case, and so permit one.
The proposal raises a further question as to why the verb and its infinitive may appear
adjacent in simple, so-called intransitive uses of promise:
(ii) John promised to leave.
One possible answer might be that promise assigns Case only when it realizes a nominal
argument, analogously to what occurs with other verbs (e.g., eat), which, when they occur
without a nominal object, allow an adjacent Case-resisting category (e.g., PP):
(iii) a. John ate [NP an apple].
b. John ate [PP on the veranda].
A second possibility, following Stowell (1981), is that in examples like (ii) the infinitive has
actually been vacuously extraposed and thus occupies a Case-free position.
7. The notion that control is fundamentally a form of anaphora has been discussed by Koster
(1978, 1984), Bach and Partee (1980), Chomsky (1981), Manzini (1983), and Williams
(1980), among others.
8. “Visser’s Generalization actually makes the wider point that structures of subject-oriented
predication resist Passive, as shown by the following examples from Koster (1984):
(i) a. He strikes his friends as pompous.
*His friends are struck (by him) as pompous.
b. The boys made Aunt Mary good little housekeepers.
*Aunt Mary was made good little housekeepers (by the boys).
c. Max failed her as a husband.
*She was failed by him as a husband.
These data lend support to the view of Williams (1980) that both anaphora and (obligatory)
control are at bottom a form of predication. They also suggest that, properly understood,
the MDP might be derivable from principles governing how predicates are associated with
their subjects. I will not pursue this point further here.
9. Hoekstra (1984) and Koster (1984) also suggest that examples like (24b,c) are ruled out by
lack of a suitable controller.
10. As discussed by Koster (1984), the close relation between control and anaphora may
also shed light on why the object of a by-phrase cannot control an infinitive in promise
constructions. Note that (ia) seems as bad as (24c). This fact appears to be related to
the general difficulty in anaphoric relations where the target antecedent occurs in a by-
phrase (ib):
(i) a. *John was promised to leave by Mary.
b. *John was introduced to herself by Mary.
If control is fundamentally a form of anaphora, then the ill-formedness in (ia,b) can be
assimilated.
11. Bach (1979), Bach and Partee (1980), and Thomason (1976) do not in fact assume that
promise-NP-Infinitive is a double object structure. Rather, they adopt an alternative deriva-
tion for the latter in which the proper c-command relations do hold:
(i) John promised Marya reward
John promise Marya reward
Mary promise a reward
promise a reward
This derivation is compatible with Barss and Lasnik’s facts but abandons the idea of a con-
nection between the differing behaviors of promise in its various complementations.
D. Pesetsky has observed that standard c-command tests appear to show that the
inner object c-commands the infinitive in promise-NP-Infinitive structures, and hence
refute the Stowellffhomason/Bach and Partee theories directly. He points to the follow-
ing data:
(ii) a. I promised each child to visit the other’s parent.
b. I promised no one to talk to any of the committee members.
c. I promised every man to have Mary visit his mother.
12. Proposals appealing to hierarchies of thematic roles or grammatical relations, like this one,
evidently represent at best descriptive approaches to the question of argument projection.
Dowty (1988) has proposed that such general hierarchies can be dispensed with in favor of
a binary template involving two “Proto-roles”: Proto-Agent and Proto-Patient. He further
suggests that the basic prominence of Proto-Agent over Proto Patient may be an iconic
reflection of the logical/temporal precedence of causes over effects, as organized by our
conceptual apparatus.
13. The approach to shifting control adopted here is similar to that taken in Oh (1988).
14. The parallel between (43a) and (44) is further supported by the following contrasts:
(i) a. John promised Mary to leave.
b. *John promised Mary for him to leave.
(ii) a. To be allowed to leave at 5:00 p.m. is unusual.
b. For him to be allowed to leave at 5:00 p.m. is unusual.
(iii) a. John never promised Mary to be allowed to visit Paris.
b. John never promised Mary for her to be allowed to visit Paris.
In obligatory control structures like (ia) the infinitive is not replaceable by one containing
a for complementizer and an overt subject even when the latter is coreferential with the
intended controller. However, in nonobligatory control structures like (iia) such replacement
is possible. Note now that (43a) appears to pattern like the nonobligatory examples in that
the “controlled” infinitive is replaceable with an infinitive with an overt subject, where the
same level of acceptability is maintained in my judgment.
15. For example, if John loses a dollar and a gust of wind by chance brings it into my hand, I
cannot report this (except facetiously) using the sentence John gave me a dollar. The point
is that although a dollar has come into my possession, John has not transferred it into my
keeping from his.
16. See Farkas (1988) for more on the notion “responsibility.”
17. It is worth emphasizing that although we appeal to entailments here in accounting for
sentences like (36a) and (48), the result is not a general entailment theory of control as in
Chierchia (1984) or Dowty (1985). On this analysis, control and construal by entailment
would be distinct phenomena, the former syntactic and the latter essentially semantic.
Furthermore, the entailments appealed to here would not be control entailments in the
sense of Chierchia or Dowty. On the suggested view, what governs object construal with
promise is simply the standard set of entailments that this verb carries by virtue of its
dative status. In general, then, construal of the understood subject of an infinitive has no
specific principles here. Construal by control follows from the (independently motivated)
shape of D-Structure. And construal by entailment follows from the general entailments
of the verb.
18. Examples involving ask . . . a favor represent a rather complex case. Note that a favor in (i)
is not understood as a concealed question, nor does the NP itself refer to a question. Note
also that with a favor, ask permits not only an oblique form with of, but also one with for
(iib):
(i) John asked Mary a favor.
(≠ John asked Mary what a favor is.)
(ii) a. John asked a favor of Mary.
b. John asked Mary for a favor.
It seems that a favor may be understood both like an infinitival-that is, as a disguised polar-
ity question (iii) and like the theme object of for; however, the source of this “ambiguity” is
unclear to me at the moment:
(iii) a. John asked Mary to do a favor.
b. John asked Mary if she would do a favor.
(See note 19 for further discussion.)
19. T. Stowell has pointed out to me that the account of shifting control given in section 5.1 for
promise will accommodate the corresponding phenomenon with ask. Observe that although
ask is normally object controlled, Passive in its complement appears to license subject con-
strual, despite the fact that no double object derivation is involved:
(i) a. John asked Mary to leave.
b. John asked Mary to be allowed to leave.
As Stowell observes, in certain usages the intuitive semantics of ask seems to involve a
transfer of possession, as with promise. However, the transfer of possession is understood
as going in different directions in the two cases: to the subject from the object in the former.
and from the subject to the object in the latter. Compare:
(ii) a. John asked Mary a favor (and he got it).
b. John promised Mary a favor (and she got it).
Suppose then that examples like (ib), like their counterparts with promise, do not in fact
involve control but instead involve construal by “transfer of possession” entailments. Then
we expect the subject to be associated with the infinitive under (an appropriately modalized
form of) the entailment:
(iii) X–asks–Y–for Z ⇒ X gets Z
20. The notion of “concealed imperatives” appears to apply in other cases beyond teach. Con-
sider the verbs order and tell in examples like the following:
{
a. ordered
(i) a. John b. told }
Mary to leave immediately.
Intuitively, ordering and telling (in the sense of (ib)) involve the issuing of a command. Canonically, the
semantic notion of “command” is realized by an imperative, as in the direct quotational variant of (ib):
(ii) John told Mary. “Leave immediately!”
Note that if we assume imperatives to be projected like other direct and indirect speech
complements. we derive that the “concealed imperative” infinitives in (i) are projected lower
than the object NP and hence must show object control under the MDP.
21. In this connection it is interesting to recall the familiar paraphrase relation between (ia) and (ib):
(i) a. I promise you that X.
b. I give you [my word that X].
This once again illustrates the basic dative character of promise and underscores the notion that
in promising, the clause—understood as the promise—is conceptualized as a “thing” passing
from the agent to the goal.
22. Mittwoch (1976) observes that allow and permit have the apparent notional status of caus-
atives-of-modals. Thus, (ia) is conceptually quite close to what is represented in (ib):
(i) a. John allowed Mary to leave.
b. John CAUSE [Mary can/may leave].
Given the canonical character of causatives as ECM verbs, this observation perhaps offers
the beginnings of an explanation of why allow and permit take an ECM complement.
23. For discussion of implicit dative arguments, see Brody and Manzini (1988) and Larson ( 1988a).
It should be stressed that the account of permit and allow given in Mittwoch (1976) actually
involves both ECM structures with an implicit dative argument and a “deontic” reading, and a
control structure involving a PRO subject. Hence, the analysis proposed here, which eliminates
the second and retains only the first, is a simplification of Mittwoch’s own views.
REFERENCES

Bach, E. and B. Partee (1980) “Anaphora and Semantic Structure,” in J. Kreiman and A. Ojeda, eds.,
Papers from the Parasession on Pronouns and Anaphora. (pp. 1–28) Chicago Linguistic Society.
17: 347–354.
Borer, H. (1986) “1-Subjects,” Linguistic Inquiry 17: 375–416.
Bowers, J. (1973) Grammatical Relations. Doctoral dissertation, MIT, Cambridge, MA.
Brody, M. and M. R. Manzini (1988) “On Implicit Arguments,” in R. Kempson, ed., Mental Repre-
sentations: The Interface between Language and Reality. (pp. 105–130) Cambridge: Cambridge
University Press.
Chierchia, G. (1984) Topics in the Syntax and Semantics of Infinitives and Gerunds. Doctoral dis-
sertation, University of Massachusetts, Amherst.
Chomsky, N. (1955) The Logical Structure of Linguistic Theory, manuscript. (Published as The
Logical Structure of Linguistic Theory, New York: Plenum, 1975.)
———. (1986) Knowledge of Language: Its Nature, Origin, and Use, New York: Praeger.
Dowty, D. (1978) “Governed Transformations as Lexical Rules in a Montague Grammar,” Lin-
guistic Inquiry 9: 393–426.
———. (1985) “On Recent Analyses of the Semantics of Control,” Linguistics and Philosophy
8: 291–331.
———. (1988) “Thematic Proto-Roles, Subject Selection, and Lexical Defaults,” colloquium paper
presented at the 1987 Winter Meetings of the Linguistic Society of America, San Francisco, CA.
Farkas, D. (1988) “On Obligatory Control,” Linguistics and Philosophy 11: 27–58.
Grimshaw, J. (1979) “Complement Selection and the Lexicon,” Linguistic Inquiry 10: 279–326.
Hoekstra, T. (1984) Transitivity: Grammatical Relations in Government-Binding Theory. Dor-
drecht: Foris.
Jacobson, P. (1983) “Grammatical Relations,” talk presented at the 1983 Winter Meetings of the
———. (1987) “Phrase Structure, Grammatical Relations, and Discontinuous Constituents,” in G.
Huck and A. Ojeda, eds., Syntax and Semantics 20: Discontinuous Constituents. (pp. 27–69)
Koopman, H. (1983) The Syntax of Verbs. Dordrecht: Foris.
Koster, J. (1978) Locality Principles in Syntax. Dordrecht: Foris.
———. (1984) “On Binding and Control,” Linguistic Inquiry 15: 417–459.
Koster, J. and R. May (1982) “The Constituency of Infinitives,” Language 58: 117–143.
Larson, R. (l988a) “Implicit Arguments in Situation Semantics,” Linguistics and Philosophy 11:
169–201.
Larson, R. (1988b) “On the Double Object Construction,” Linguistic Inquiry 19: 335–391.
———. (1989) “Light Predicate Raising,” Lexicon Project Working Papers 27. Center for Cogni-
tive Science. Cambridge, MA: MITWPL.
Manzini, M. R. (1983) “On Control and Control Theory,” Linguistic Inquiry 14: 421–446.
Mittwoch, A. (1976) “Raising or Equi or Both: Another Look at Root Modals and at Permissive
Allow,” Papers in Linguistics 10: 55–75.
Munro, P. (1982) “On the Transitivity of ‘Say’ Verbs,” in P. Hopper and S. A. Thompson, eds.,
Syntax and Semantics 15: Studies in Transitivity. (pp. 301–318) New York: Academic Press
Oehrle, R. (1975) The Grammatical Status of the English Dative Alternation. Doctoral dissertation,
MIT, Cambridge, MA.
Oh, S. (1988) “A Promising Control Theory,” in L. MacLeod, G. Larson, and D. Brentani, eds.,
Papers from the 24th Meeting of the Chicago Linguistic Society, (pp. 290–303). Chicago:
Chicago Linguistic Society.
Rosenbaum, P. (1970) “A Principle Governing Deletion in English Sentential Complementa-
tion,” in R. Jacobs and P. Rosenbaum, eds., Readings in English Transformational Grammar.
(pp. 20–29) Waltham, MA: Ginn.
Thomason, R. (1976) “Some Extensions of Montague Grammar,” in B. Partee, ed., Montague
Grammar. (pp. 77–117) New York: Academic Press.
Cambridge, MA.
Visser, F. T. (1963–73) An Historical Syntax of the English Language. Leiden: Brill.
Williams, E. (1980) “Predication,” Linguistic Inquiry 11: 203–238.
Complex Predicates: Background
Although originally applied to dative argument alternations and related control struc-
tures, VP shells had implications for phrases not typically identified as verbal argu-
ments. This point has already been noted in connection with circumstantial adverbs.
In order to extend the Dative Shift analysis of Larson (1988) or its modern A-Shift
counterpart from to-datives to the wider class of applicative structures, it is necessary to
regard such “adverbials” as oblique complements, following proposals from Relational
Grammar. However, the most radical consequences for complementation came through
an alternative analysis of “Heavy NP Shift” phenomena that shell theory offered, that
is, Light Predicate Raising. In the fall of 1988 I gave a seminar at MIT pursuing the
empirical and theoretical implications of Light Predicate Raising, and ultimately wrote
up the conclusions as Larson (1989).
Because the consequences of Light Predicate Raising are so extensive, I cannot
attempt a full consideration of the issues here equivalent to that offered for datives.1
Below I provide some historical background to the analysis, a brief review of its major
semantic and structural implications, and comments on where those implications have
proven fruitful in the years since the analysis was first proposed. I also consider some
basic questions that have arisen for the analysis, particularly in light of subsequent theo-
retical developments, and sketch a possible updated version of Light Predicate Raising
in terms of the account of structure projection offered here.
1. HEAVY NP SHIFT AS LIGHT PREDICATE RAISING
As noted in the General Introduction, Montague Grammarians routinely appealed to

derivations like (1a,b), involving underlying complex predicative constituents like per-
suade to leave and hammer flat. The deep constituency of these phrases is masked by
Right Wrap, which interpolates the direct object.
(1) a. John persuade Mary to leave Left Concatenation

/ \
John persuade Mary to leave Right Wrap
/ \
Mary persuade to leave Right Concatenation
/ \
persuade to leave
b. John hammer the metal flat Left Concatenation
/ \
John hammer the metal flat Right Wrap
/ \
the metal hammer flat Right Concatenation
/ \
hammer flat
Jacobson (1983, 1984, 1987) observes that this view opens an intriguing possibility for
examples like (2a,b), standardly assumed to derive from (1a,b) by an operation of “NP
Shift” that moves the direct object rightward from postverbal position, adjoining it on
the edge of VP (3a,b):
(2) a. John persuaded to leave a mysterious woman in black.
b. John hammered flat all the metal that he been given to work with.
(3) a. John [VP persuaded ___ to leave] [a mysterious woman in black]
b. John [VP hammered ___ flat] [all the metal that he been given to work with]
In place of this picture, (2a,b) could be seen as arising through derivations where Left Con-
catenation replaces Right Wrap, allowing the complex predicative constituent to preserve
its underlying integrity (4a). In effect, this would be to treat persuade to leave like a simplex
transitive verb (kiss), letting it undergo the same sequence of derivational steps (cf. 4b):
(4) a. John persuade to leave a mysterious woman in black Left Concatenation

/ \
John persuade to leave a mysterious woman in black Left Concatenation
/ \
a mysterious woman in black persuade to leave Right Concatenation
/ \
persuade to leave
b. John kissed a mysterious woman in black Left Concatenation

/ \
John kissed a mysterious woman in black Left Concatenation
/ \
a mysterious woman in black kissed
Although novel in the early 1980s when Jacobson made it, this proposal was not in
fact a new one. In considering examples like (5a–d), Chomsky (1955/1975) had already
argued at length for their derivative status, taking them to arise from sentences like (6a–d),
respectively, by a transformation that broke up a complex verbal constituent.2 Chomsky
referred to expressions like bring in, consider a fool, and so on as “V-complement con-
structions,” analyzing them, in effect, as complex transitive verbs and labeling them VT.
(5) a. The detective brought the suspect in.
b. They consider John a fool.
c. The teacher gave several books to him.
d. I found the boy studying in the library.
(6) a. The detective brought in the suspect.
b. They consider a fool John.
c. The teacher gave to him several books.
d. I found studying in the library the boy.

Chomsky (1955/1975) remarks that while the examples in (5a) and (6a) are both gram-
matical, “in general the separability of the preposition is determined by the complexity
of the NP object,3 thus we could hardly have . . . The detective brought the man who
was accused of having the stolen automobile in . . . As the object becomes more com-
plex . . . the naturalness of the transformation decreases” (1955/1975, p. 477). Thus,
for Chomsky too, Heavy NP Shift examples offer a window into structure, revealing the
underlying constituency of the boldfaced elements in (6).
As discussed in the General Introduction, Jacobson (1983, 1984, 1987) was the first
to attempt to translate derivations with Right Wrap into more conventional phrase
markers, proposing a verb-raising analysis. She extended this view to verb-particle alter-
nations like (5a)/(6a). Like Chomsky (1955/1975), Jacobson took the verb + particle
to form an underlying complex equivalent to a transitive verb (TVP). Examples like
(5a), with the verb and particle discontinuous in the surface string, derive by raising the
verbal head alone, as in (7a). Examples like (6a), with the verb and particle contiguous,
derive by raising the entire verbal complex, as in (7b) (cf. Jacobson 1987, p. 45).
(7) a. VP b. VP
V NP TVP V NP TVP
bring the suspect V V Part the suspect V
V Part bring in V Part
bring in bring in
Jacobson (1987) proposes no general extension of this analysis to Heavy NP Shift

examples like (2a,b)/(6b–d), and indeed no direct extension is available in her account
as it stands.4 Observe that (7) analyzes bring in as an underlying complex lexical head
with structure [V[V bring][Part in]]. This analysis is crucial in providing two distinct
instances of V for raising, and accords well with standard dictionary practice, which
lists composites like bring in, bring off, bring out and look up as phrasal verbs under
their lexical head (bring, look, etc.). Such a view is not plausible, however, for cases
like persuade to leave, hammer flat, consider a fool, and so on. While these phrases do
contain lexical Vs (persuade, hammer, and consider, respectively), they are not lexical
Vs themselves. No one would expect to find persuade to leave, hammer flat, and so on
listed in a dictionary.
Nonetheless, an extension of Jacobson’s approach does suggest itself. Observe that
if complex TVPs were allowed to reanalyze categorially as Vs, or if TVP were itself
allowed to raise, counterpart to a simplex V, then Heavy NP Shift could be brought
under the same general view that Jacobson urges for verb-particle combinations:5
(8) a. VP b. VP
V NP TVP TVP/V NP TVP/V
consider John John

V NP V NP V NP
consider a fool consider a fool consider a fool

On this picture, the examples in (2a,b) and all of those in (6) would arise, not by Heavy
NP Shift, but rather by Light Predicate Raising.
Larson (1988, 1989) makes the extension just noted. As discussed in the General
Introduction, shell theory analyzes verbal projections as binary, right-descending struc-
tures in which oblique phrases are projected lowermost. Accordingly, Mary gave every-
thing that he demanded to John receives the structure in (9a), where to John is the first/
lowest complement of the verb. The usual derivation raises give to the empty V posi-
tion, yielding the unmarked V–NP–PP word order. However, Larson (1989) proposes
an alternative derivation in which the constituent give to John raises around the object,
yielding the V–PP–NP word order of NP Shift. This derivation is made possible by an
optional rule of “V′ Reanalysis,” which permits certain V′s to be reanalyzed categorially
as Vs, and hence to undergo raising (9b):
(9) a. VP
NP V′
Mary V VP Unmarked V-O Word Order
e NP V′
V PP
everything that
he demanded give to John
b. VP
NP V′
Mary V VP
V′ Reanalysis +
give to John NP V′ ⇒V Light Predicate Raising
everything that e
he demanded
V′ Reanalysis is formulated in Larson (1989) as follows:
V′ Reanalysis:
Let α be a phrase [Vʹ . . .] whose θ-grid contains one undischarged internal θ-role.
Then α may be reanalyzed as [V . . .].
This rule instantiates one version of the TVP analysis suggested above: V′s that are
thematically transitive (hence equivalent to TVPs in the Montagovian framework) are
permitted to relabel as simplex Vs and so to undergo raising to a V-head position in a
structure-preserving way.
2. THE INTERFACE LOGIC OF LIGHT PREDICATE RAISING
Classic Heavy NP Shift assumes rightward movement of a noun phrase with adjunction
to VP and a trace or copy left in a Case-marked position (i.e., a variable) (10).
(10) VP
VP NP
. . . NP . . . α
α
This view assimilates NP Shift to the general class of movements including Wh-
movement, Topicalization, Quantifier Raising, and so on, all of which create operator-
variable structures.
The Light Predicate Raising analysis can be seen as instantiating a different
“interface logic.”6 As discussed in the General Introduction, Fregean approaches to
sentence semantics assign (11a) an analysis as in (11b), where the verb (kiss) cor-
responds to a binary relation (kiss) on individuals (m, j) of the sort denoted by the
argument phrases. By contrast, neo-Davidsonian approaches offer the very different
view in (11c), where the verb contributes a unary event predicate, and where the
arguments are linked to it by means of conjoined binary thematic relations.7 This
account permits the analysis of many kinds of adverbs as additional conjuncts to the
basic event predication (12):
(11) a. Mary kissed John.
Kiss(m,j)
b.
c.
∃e[kissing(e) & Agent(e,m) & Theme(e,j)]
(“There was a kissing, and it was by Mary, and it was of John.”)
(12) a. Mary kissed John quickly.
∃e[kissing(e) & Agent(e,m) & Theme(e,j) & quick(e,C)]

b.
(“There was a kissing, and it was by Mary, and it was of John, and it was quick.”)
Formulae like (11c) and (12b) represent event quantification as unrestricted or

unstructured; the existential quantifier attaches to a “flat” sequence of co-equal con-
juncts. Herburger (2000) proposes an attractive theory of focus that turns on this point.
Compare sentence (11a) with the focal variant of it in (13a). Whereas (11a) simply
asserts the existence of a kissing of John by Mary, (13a) is taken to divide this infor-
mation up into a presupposition or background entailment of Mary’s kissing someone
and an assertion that the individual in question was John. Herburger (2000) corre-
lates this difference in “information packaging” with a difference in event representa-
tion. Whereas (11a) gets the flat, unstructured event quantification in (11c), (13a) gets
the structured event quantification in (13b), where the nonfocused material forms the
restriction on the event quantifier and the focused material constitutes its scope.8
(13) a. Mary kissed JOHN.

i. Presupposes/Background-entails: Mary kissed someone.
ii. Asserts: The individual was John.
b. ∃ e [kissing(e) & Agent(e,m)] (Theme(e,j))
Q Restriction Scope
Presupposed/B-Entailed Asserted
“For some kissing by Mary, its theme was John.”
Assuming the restriction in a structured quantifier is presupposed/background-entailed,

and the scope asserted, the difference in information packaging follows directly from
these representations.9 Alternative foci can be accommodated along the lines shown in
(14)–(16):
(14) a. MARY kissed John.
b.
∃e [kissing(e) & Patient(e,j)] (Agent(e,m))
“For some kissing of John, its agent was Mary.”
(15) a. Mary KISSED John.
b.
∃e [Agent(e,m) & Patient(e,j)] (kissing(e))
“For some event with Mary as agent and John as patient, it was a kissing.”
(16) a. Mary kissed John QUICKLY.
b.
∃e [kissing(e) & Agent(e,m) & Patient(e j)] (quick(e,C ))
“For some kissing with Mary as agent and John as patient, it was quick.”
Herburger’s analysis of focus as structured event quantification can be integrated with

Light Predicate Raising by analyzing the latter as fundamentally a focusing structure,
and by appealing to the influential “Mapping Hypothesis” of Diesing (1992), which
correlates hierarchical, syntactic structure with quantificational semantics. Generalizing
slightly on Diesing’s original formulation, her essential proposal is that high material
maps to the restriction and low material to the scope:
Mapping Hypothesis:
Lowest material from VP is mapped to the nuclear scope. The residue is mapped
into a restrictive clause.
Consider now the Light Predicate Raising structure in (17) and the interpretation given
beneath. Here VP material initially projected below the direct object has been amalgam-
ated by V′ Reanalysis (V bar) and raised above it, stranding the latter:10
(17) VP
NP V′
Mary V V′
NP V
give to John
everything that he demanded e
∃e [giving(e) & Ag(e,m) & Gl(e,j)] (Theme(e,ethd))
Q Restriction Scope
Presupposed Asserted
“For some giving by Mary to John its theme was everything that John demanded.”
Under the Mapping Hypothesis, a direct correspondence becomes available between the
meaning of the Light Predicate Raising structure and its form: the lower material maps
to the scope—what is asserted in focal quantification—whereas the higher material
maps to the restriction—what is presupposed or background-entailed.
On this view, Light Predicate Raising can be understood as deriving an appropriate
configuration for focus interpretation. The direct object is made lowermost, and hence a
suitable interpretive target for the scope under Diesing (1992), whereas the remaining VP
material is swept upward with the verb and made part of the quantificational restriction.
3. STRUCTURAL IMPLICATIONS OF LIGHT PREDICATE RAISING
The Light Predicate Raising analysis carries three key implications for structure. First,
in examples showing NP Shift order, the verb and verb phrase elements occurring to the
left of the direct object NP must form a single underlying constituent, specifically one
of category V′.
Constituent (V′)
Subject [V′ V X Y X ] NP
N ___.
Second, the moved constituent should have the thematic status of a complex transitive
verb, essentially following Chomsky (1955/1975) and Jacobson (1983, 1984, 1987).
Thematic Transitive
N ___.
Third, since Light Predicate Raising involves leftward raising of a V projection around
the direct object, then the NP must itself be in situ, where it was base-merged.
In Situ Object
⇓
Subject [V′ V X Y X ] NP ___.
As noted in Larson (1989), all three implications clash directly with structural assump-
tions and analyses commonly held at the time, and indeed still widely held today.11
3.1. Modifiers
The idea that verb phrase elements to the left of an NP-Shifted object form a sin-
gle underlying constituent with V conflicts with standard views of modifiers. Many
researchers have analyzed the boldfaced items in examples like (18a) as VP-adjuncts,
attaching recursively to the verb phrase in a right-ascending structure. This view is
expressed not only in standard phrase structure representations like (18b) but also in
classical Montague Grammar derivations like (18c), where verbal modifiers are ana-
lyzed semantically as VP-functors, right-concatenating to intransitive verb phrase (IVP)
expressions and yielding expressions of the same category:
(18) a. John saw Mary [briefly] [in Rome] [last week].
b. VP
VP NP
VP PP last week
VP AdvP in Rome
saw Mary briefly
c. saw Mary briefly in Rome last weekIVP Right Concatenation

/ \
saw Mary briefly in RomeIVP last week Right Concatenation
/ \
saw Mary brieflyIVP in Rome Right Concatenation
/ \
saw MaryIVP briefly
/ \
saw Mary
Under (18b)/(18c), the verb + modifiers do not form a constituent that excludes the
direct object. This analysis virtually mandates a classical NP Shift account of examples
like (19a), in which the direct object raises rightward to the edge of VP (19b):
(19) a. John saw briefly in Rome last week a woman he had known since childhood.
b. VP
VP NP
VP NP a woman . ..
VP PP last week
VP AdvP in Rome
saw NP briefly
... NP Shift
In contrast, Light Predicate Raising requires V (see) and its modifiers (briefly, in
Rome, last week) to form a constituent that excludes the direct object. On this view,
(20a) and (20b) both derive from the right-descending structure in (20c), wherein the
modifiers constitute the innermost complements of see. The unmarked verb–direct
object–modifier order of (20a) derives by raising see alone to the empty V position in
the highest VP. The verb–modifier–direct object (NP Shift) order (20b) results by reana-
lyzing the circled V′ as V, with subsequent raising of the entire complex:
(20) a. John saw a woman he had known since childhood briefly in Rome last week.
b. John saw briefly in Rome last week a woman he had known since childhood.
c. VP
NP V′
John V VP
e NP V′
a woman . . . V VP
V-O Order see AdvP V′
briefly V VP
see PP V′
in Rome V NP
see last week
‘NP Shift’ Order
An analysis like (20c) obliges us to revisit the range of arguments that have been offered
in favor of the adjunct modifier account. Larson (2004, reprinted in this volume) examines
the key domain of scope. Data of the sort in (21) and (22) provide simple scopal evidence
for a right-descending analysis of modifiers. As (21a) shows, an inner, downward-entailing
adverb (rarely) can license a negative polarity item (any) contained within an outer modifier.
On the usual view that negative polarity items must be within the c-command domain of
their trigger at Spell-Out, this result is predicted by the right-descending structure (21b) but
not by the right-ascending one (21c). (22a–c) demonstrate the same point for a downward-
entailing direct object (few people) and a verb phrase modifier. Again the former behaves as
if it c-commands the latter from its base position.12
(21) a. John spoke [rarely] [during any of our meetings]
b. John [VP spoke [VP rarely [V′ t during any of our meetings]]]
c. John [VP [VP spoke rarely] during any of our meetings]
(22) a. John met [few people] [during any of his visits]
b. John [VP spoke [VP rarely [V′ t during any of our meetings]]]
c. John [VP[VP spoke rarely] during any of our meetings]

Some researchers, notably Ernst (1994, 2001), have construed this problem for right-
ascending syntax as a narrowly technical one, to be addressed simply by replacing
c-command in the licensing condition for domain-dependent items like negative polar-
ity items with an alternative formal construct—for instance, m-command plus linear
precedence. In my view, this response seriously underestimates the stakes involved. As
Reinhart (1983) observes, in a theory of the syntax-semantics interface embracing local
compositionality—a highly desirable property—the structural notion of c-command
maps transparently to the semantic notion of scope. The elements that a phrase c-com-
mands syntactically are exactly the ones that it scopes over semantically, and hence the
ones that can depend on it. This is crucial in the analysis of binding on any composi-
tional semantic account that I am familiar with (see Larson and Segal 1995 and Büring
2005 for discussion). The same holds true of negative polarity items like any and the
downward-entailing items in whose semantic scope they must be interpreted. Whereas
a right-descending structure like (20c) correctly maps syntactic domain to composi-
tional semantic scope, a right-ascending one does not, and no tinkering with syntactic
licensing conditions alone will correct this disparity. C-command and m-command plus
precedence are thus not on equal footing, and theories appealing to one versus the other
are not to be equally valued. C-command has “interface validity” that its competitors
simply lack.
Interestingly, the right-ascending view of modifiers appears to have supporting
scopal evidence of its own. Consider the semantic contrast in (23), noted by Andrews
(1983):
(23) a. John knocked on the door intentionally twice.
b. John knocked on the door twice intentionally.
(23a,b) appear to assert two different things. The former asserts that John did some-
thing twice, namely, knock on the door intentionally. The latter asserts that John did
something intentionally, namely, knock on the door twice. Andrews’s diagnoses this
meaning difference as one of scope, with twice taking scope over intentionally on the
first reading, and intentionally taking scope over twice on the second. Andrews’s diag-
nosis can be accommodated neatly with a right-ascending syntax and a standard Mon-
tague Grammar–style semantics. Assuming the two adverbs adjoin recursively to VP,
(23a) gets the structure in (24a), where twice is highest, and (23b) gets the structure in
(24b), where intentionally is highest.
(24) a. John [VP [VP [VP knocked on the door ] intentionally ] twice ]
b. John [VP [VP [VP knocked on the door ] twice ] intentionally ]
Suppose further a semantics wherein VP adverbs are functor expressions, taking (inten-
sions of) VP-denotations as their arguments and returning VP-denotations as their val-
ues (Thomason and Stalnaker 1973; Montague 1974). Applying this analysis to (24a,b)
yields the results in (25), where twice’ (the interpretation of twice) has the widest scope
in (25a) and where intentionally’ has the widest scope in (25b).
(25) a. twice’(întentionally’(^knocked on the door’))
intentionally’(^twice’ (^knocked on the door’))

b.
Syntax and semantics thus come together neatly.
Although these results are compelling when taken on their own, it’s important to
note the serious tension between them and the previous results. As we saw, c-command
and right-ascending syntax yield incorrect predictions with respect to (21) and (22),
whereas m-command together with linear precedence allows the correct ones (26a).
But now, with (23), c-command seems to give the right results, whereas m-command
plus linear precedence yields exactly the wrong ones. On a right-ascending adjunction
structure, inner modifiers (here intentionally) asymmetrically m-command and precede
outer ones (twice) (26b):
(26) a. [VP [VP spoke rarely] during any of our meetings]

C-command: X
M-command + Precedence:
b. [VP [VP [VP knocked on the door ] intentionally ] twice ]

C-command:
M-command + Precedence: X
Note further that there is no clear way to separate the two phenomena, at least under
Andrews’s assessment of what’s going on in (23). In both cases we are dealing with
scope: scope with respect to a downward-entailing item in (26a) and relative adverb
scope in (26b).
Apparent “scope antinomies” like these have led a number of researchers (Pesetsky
1995; Phillips 2003) to conclude that c-command is the correct licensing condition
in all cases and hence that both a right-descending and a right-ascending analysis of
modifiers are required. In other words, the grammar must be allowed to generate
both (19b) and something like (20c) in order to accommodate the full range of facts.
However, Larson (2004) argues that this conclusion is hasty, at least as far as the data
in (23) are concerned, and that a strictly right-descending analysis is in fact sufficient.
A key point is the interpretive parallel between (23a,b) and the examples in (27a,b),
respectively:
(27) a. John’s intentional knockings on the door were two (in number).
b. John’s double-knock on the door was intentional.
Like (23a), (27a) asserts that John did something twice—that is, knock on the door
intentionally. And like (23b), (27b) asserts that John did something intentionally— that
is, knock on the door twice. Crucially, however, the semantics in (27) does not involve
scope, as Andrews’s original diagnosis would have it, but rather predication. The right-
most adjectives (two and intentional) constitute the basic predicates in their respective
sentences.
Larson (2004) suggests that the semantics of (23) and (27) be assimilated using struc-
tured event quantification of the sort discussed above. Briefly, (23a), repeated below as
(28a), has the structure in (28b) and the interpretation in (28c), where the adverb twice
constitutes the scope of event quantification and the adverb intentionally is part of the
restriction:
b. VP
NP V′
John V VP
knock PP V′
on the door V VP
knock AdvP V′
intentionally V AdvP
knock twice
c. ∃ E [ e[Ee → knocking(j,d,e) & intentional(e)]] (two(E))
A
Q Restriction Scope
(28c) can be read as saying that there was an event E consisting of subevents e, each of
which was an intentional knocking on the door by John, and that larger event E was 2
in number. This matches the interpretation of the nominalization (27a).
By contrast, (23b), repeated below as (29a), has the structure in (29b) and the inter-
pretation in (29c), where the adverb intentionally constitutes the scope of event quanti-
fication and the adverb twice is part of the restriction:
(29) a. John knocked on the door twice intentionally.
b. VP
NP V′
John V VP
knock PP V′
on the door V VP
knock AdvP V′
twice V AdvP
knock intentionally
c. ∃E [ e[Ee → knocking(j,d,e) & two(E)] (intentional(E))

A
Q Restriction Scope
(29c) can be read as saying that there was an event E consisting of two subevents, each
of which was a knocking on the door by John, and that larger event E was intentional/
intended. This matches the interpretation of the nominalization (27b).
Thus the scopal analysis of (23a,b), based on right-ascending syntax and a standard
Montague Grammar–style semantics involving VP-functors, is replaced whole cloth
with a predicational analysis based on right-descending syntax and a semantics involv-
ing Davidsonian event quantification. The latter improves on the former in not only
capturing the relevant semantic difference but also relating it to a parallel difference
in sentences involving nominalization, a connection that the functor analysis sheds no
light on.
Scope is not the only line of supporting evidence and argument that has been offered
for a right-ascending analysis of modifier attachment. A full examination of these issues
is beyond what can be attempted here, however.13 The discussion above is simply offered
as a prima facie evidence that data long taken to yield decisive support for right ascent
of modifiers not only can be reanalyzed under a right-descending view but yield an
account more attractive than its competitor.
3.2. Reduced Clauses

Constituent status for verb phrase elements to the left of an NP-Shifted object also
conflicts with widely held views about reduced clauses. (30a) is an Exceptional Case
Marking (ECM) or “Raising to Object” structure; (30b) is a so-called small clause
construction; (30c) is a resultative. Many researchers have taken the direct object and
postverbal boldfaced material in such examples to form a constituent—specifically, a
clausal projection counterpart to the overt finite clauses in (31a–c) (respectively).
(30) a. Mary believes John to be intelligent.
b. Mary judges John competent (to stand trial).
c. Mary sanded the hull smooth.
(31) a. Mary believes [John is intelligent].
b. Mary judges [that John is competent (to stand trial)].
c. Mary sanded until [the hull was smooth].
Familiar structural analyses like (32a–c) instantiate this clausal view:
(32) a. VP b. VP
V IP V SC
believe NP I′ judge NP AP
John to be intelligent John competent
c. VP
V(P) SC
sand NP AP
the hull smooth
Under (32a–c), the verb and boldfaced material do not form a constituent that excludes
the direct object. Again, this exerts strong pressure toward a classical account of NP
Shift examples like (33a), wherein the direct object raises rightward, again presumably
to the VP edge (33b):
(33) a. Mary sanded smooth all of the hull aft of the centerboard case.
b. VP
VP NP
V(P) SC all of the hull . . .

sand NP AP
... smooth
NP Shift
In contrast, Light Predicate Raising requires V (sand) and the postverbal AP (smooth)
to form a constituent that excludes the direct object. (30c) and (33a) both derive from
the right-descending structure in (34). The basic V–DO–AP order in (30c) derives by
raising of V alone. The V–AP–DO (NP Shift) order in (33a) results by reanalysis of the
circled V′ with subsequent raising of the entire complex:
(34) VP
NP V′
Mary V VP
e NP V′
all of the hull . . . V AP

V-O Order sand smooth
‘NP Shift’ Order
Similarly for the examples like (35a) and (36a), with ECM and small clause structures
(respectively):
(35) a. Mary believes anyone who can tie his own shoes to be intelligent.
b. Mary [believes to be intelligent] anyone who can tie his own shoes. ___
(36) a. Mary judges anyone who passes standard pretrial psychiatric exams competent.
b. ___
Mary [judges competent] anyone who passes standard pretrial psychiatric exams ___.
The situation with reduced clauses resembles that of modifiers insofar as the syntax
in (32a–c) can be paired with a standard semantics that appears to fit it snugly. Since
Chomsky (1981) and Stowell (1981), ECM and small clause–taking verbs have been
widely analyzed as proposition-taking, that is, as showing the same selection that they
do with finite clausal complements. On this view, (30a)/(31a) and (30b)/(31b) receive
the same semantics, that is, (37a) and (37b) (respectively).14 Both pairs involve a relation
between an individual (Mary) and a proposition (denoted by întelligent’(John) and
^competent’(John) in 36a,b, respectively).
(37) a. believe’(Mary, întelligent’(John))
b. judge’(Mary, ^competent’(John))
Resultatives like (32c) are widely analyzed under the “result state” semantics given
roughly in (38).15 According to the latter, (32c) means that Mary brings into existence
(by hammering) a certain state, that is, the hull being smooth.
(38) cause(Mary, ^become(^smooth’(the hull))
Note the presence of a propositional nucleus in (38), denoted by ^smooth’(the hull).

Once again the structure in (32c) matches propositional content with a clause.
Given our results with modifiers and event quantification, it’s natural to ask whether
right-descending structures for reduced clause examples can be given their own natural
semantic analyses—ones not equivalent to those just noted.16 I am not aware of such an
alternative for ECM and small clause constructions at present. In the case of resultatives
like (30c), however, a new semantics has arisen that fits the right-descending syntax
in a very natural way, one that is not equivalent to the widely assumed “result state”
account. The core of this new analysis is the familiar observation that whereas the verb
in a resultative is typically atelic, hence compatible only with durative for-adverbials
(39a), the resultative construction itself is uniformly telic, hence compatible only with
endpoint in-adverbials (39b):17
(39) a. Mary sanded the hull (for ten hours/??in ten hours).
b. Mary sanded the hull smooth (in ten hours/??for ten hours).
A crucial property of resultatives is thus the contribution of telicity.
3.2.1. The Nature of Telicity

In a series of influential studies of verbal aspect, Krifka (1989, 1992, 1998) proposes
that telic verbs are composed of two separable components: (i) a predicate of events
denoting an activity and (ii) a bounded scale that maps the theme argument onto the
event. To take a concrete example, consider (40), with the telic predicate eat:18
(40) Mary ate an apple in an hour.
This sentence describes an event of apple-eating lasting one hour. As the event progresses,
the apple is eaten, with successively smaller portions of the apple’s consumable volume
remaining, until, finally, that volume reaches zero. At the zero point, the apple is con-
sumed, and apple-eating is over. In effect, boundedness in the apple projects boundedness
in the apple-eating. Krifka expresses this formally by saying that telic predicates like eat,
in addition to describing an event with an affected theme, encode a homomorphism that
maps some scalar structure associated with their theme onto the part-whole structure
of the event. Specifically, eat projects a bounded scale of decreasing consumable volume
upon the eating event, ranging from 100% down to zero (41):
(41) Apple
Eating
This mapping permits the theme to bound the event whenever the theme itself denotes a
bounded quantity—whenever it is quantized. Of course, when the theme argument is not
quantized, for example, when it is a plural or mass noun, no boundedness can be induced
from it, and hence the event described by the predicate will be unbounded itself (42a,b):
(42) a. Mary ate apples (for an hour/?*in an hour).
a. Mary ate apple (for an hour/?*in an hour).
The precise scale according to which a telic predicate maps its theme argument varies
with the predicate, as shown by the list in (43) (partially adapted from Wechsler 2005).
Verbs of consumption (43a–c) map their theme via a scale of decreasing consumable
volume;19 verbs of creation (43d,e) map their theme by how much of it has been brought
into existence or how far along one is in a standard process for producing it; “phase-
state transitives” (43f–h) map their theme by temperature (upward or downward) to the
point of phase change. And so on.
(43) Predicate Scale (X = affected theme argument.)
drink a glass of wine consumable volume of X remaining

a.
eat a sandwich
b. consumable volume of X remaining
read a letter
c. consumable volume of X remaining (to be read)
write a letter
d. amount of X in existence
bake a cake
e. stage of X in production process
freeze the soup

f. temperature of X measured downward to phase change
liquify the nitrogen

g. temperature of X measured downward to phase change
boil the water
h. temperature of X measured upward to phase change
An important requirement of telicity is that the predicate invoke not merely a scale by
which to map the theme but a bounded scale. This requirement is met in all of (43a–h).
The scale of consumable-volume-of-X-remaining is bounded by the zero point. The scale
for cake production is bounded by the timed end of a process of heating, which is itself
situated in a larger preparation script. The scale for phase-state transitives is bounded
by a phase-state change. Not all predicates encoding a scale have this property, however.
Consider the deadjectival transitive verbs cool and dim (44a,b). These predicates clearly
involve a scale that maps their theme to the event they describe: an event of cooling or
dimming progresses as the theme decreases in temperature or brightness (respectively):
(44) Predicate Scale (X = affected theme argument.)
cool the soup temperature of X

dim the lights brightness of X
Nonetheless, neither cool nor dim must be understood as telic, even with a quantized
theme; (45a) and (45b) are both acceptable with both kinds of temporal modifiers:
(45) a. Mary cooled the soup for ten minutes/in ten minutes.
b. Mary dimmed the lights for ten minutes/in ten minutes.
The reason is straightforward: neither temperature nor brightness is an inherently

bounded scales; conceptually, objects can be made hotter or colder without limit.20
Cooled and dimmed are thus understood as telic only when some bound is contextually
given, and indeed the situations where we interpret (45a,b) as telic have exactly this
property. (45a), with in ten minutes, must be understood with respect to some contextu-
ally understood bound like “cool enough for serving,” “cool enough for eating,” and so
on. Similarly for (45b). Summarizing, then, the basic picture of telicity is that in (46):
(46) Telic Predicate = Activity Predicate + Bounded Scale for mapping Theme
A telic predicate is a semantic complex made up of a transitive activity involving an

affected theme together with some bounded scale for mapping a property of the theme
to the event denoted by the predicate.
3.2.2. The Resultative Construction as Complex Telic Predicate Formation

With these elements in place, consider now resultative constructions like sand the hull
smooth. As we saw, the latter is telic but is formed from a nontelic activity predicate
(sand). Under the equation in (46), this suggests that the AP must supply the bounded
scale required for telicity. This is exactly the proposal of Wechsler (2005), who argues
that resultatives are, in effect, instances of complex telic predicate formation, wherein
an activity verb combines with an expression denoting a bounded scale to create a
derived telic verb. Thus sand describes an action performed on an affected theme: ‘to
polish or scrape something with sand or sand paper’.21 In and of itself, however, sand
doesn’t imply any specific scale by which the sanded object is mapped to the sanding
event.22 Lacking a scale, sand is thus a transitive activity verb, as shown by its co-
occurrence with for-phrases ((38a), repeated below):
(38a) Mary sanded the hull for ten hours/?*in ten hours.
Consider now predicates denoting a bounded scale that may apply to the sanded object.
A ship’s hull may become progressively smoother by sanding, until it is finally smooth; it
may become flatter, until it is finally flat; cleaner, until it is completely clean. Wechsler’s
proposal suggests that any of these expressions should combine with sand to yield a
resultative, since all would yield a complex telic predicate. This prediction is correct, as
shown by (47a–c):
(47) a. Mary sanded the hull smooth in ten hours/?*for ten hours.
b. Mary sanded the hull flat in ten hours/?*for ten hours.
c. Mary sanded the hull clean in ten hours/?*for ten hours.
Wechsler’s proposal also illuminates the puzzling issue of why certain APs do not function
readily as resultative predicates (48a–c), a co-occurrence restriction that many have assim-
ilated to selection and that is difficult to accommodate under the reduced-clause view:
(48) a. Mary sanded the hull smooth/*rough.
(cf. Mary sanded the surface [until it was smooth/rough].)
b. Mary wiped the table dry/*wet.
(cf. Mary wiped the table [until it was dry/wet].)
c. Mary bent the rod straight/*curved.
Recall that telic predicates require, not simply a scale, but a bounded scale: one with an
inherent endpoint. Interestingly, smooth/rough, dry/wet, and straight/curved appear to
differ exactly in this way: the former member of each pair is a bounded or closed-scale
adjective—one that implies its own endpoint—whereas the latter member is an open-
scale adjective—one that does not. An independent test for closed-scale status is the
applicability of modifiers like completely, totally, or perfectly, which select an endpoint
on an adjectival scale and hence apply only to adjectives having such an endpoint. As
(49a–c) show, these modifiers sort adjectives in the same way as the resultative construc-
tion itself.23
(49) a. The table was perfectly dry/?*wet.
b. The hull was completely smooth/*?rough.
c. The rod was perfectly straight/*curved.
Wechsler’s highly attractive semantics for resultatives fits neatly with the right-descending
syntax discussed above. (50a) and its NP shift variant (50b) both derive from the struc-
ture in (50c). The composition of the lowermost constituent (sand smooth) can be
understood precisely as complex telic predicate formation, in which sand provides the
activity involving an affected theme and smooth provides the bounded scale mapping
the theme to the activity event:
(50) a. Mary sanded all of the hull aft of the centerboard case smooth.
b. Mary sanded smooth all of the hull aft of the centerboard case.
complex telic predicate
c. Mary e [VP all of the hull aft of the centerboard case [V' sand
d smooth
th ]]
activity + scale
The standard syntactic picture of resultatives (reduced clause) and the standard seman-
tics accompanying it (result state) are thus replaced with the right-descending syntax
required under Light Predicate Raising and a very new account of semantic composi-
tion involving events, one that provides an analysis of data like (48), which escape the
reduced-clause picture.
The semantic results described above do not apply to ECM and small clause struc-
tures of the sort in (30a,b), nor, arguably, to so-called nonthematic resultatives like
(51a,b), where the object nominal is not an affected theme argument of the main verb
because it is not an argument of V at all:24
(51) a. The audience laughed John off the stage.
(cf. *The audience laughed John.)
b. John shouted himself hoarse.
(cf. *John shouted himself.)
Once again, the discussion here is simply offered as prima facie evidence that a right-
descending not only is defensible for a major domain of data involving reduced clauses
but can in fact be associated with a better understanding of semantic composition than
that assumed for the more familiar structures.
4. TWO FURTHER IMPLICATIONS OF LIGHT PREDICATE RAISING
The discussions of modifiers and reduced clauses above both involve the first key struc-
tural implication of Light Predicate Raising, that is, that the verb and postobject mate-
rial form a constituent V′ that excludes the direct object.
Constituent (V′)
N ___.
The other two implications are that this V′ has a special thematic status and that the
direct object NP is in situ and unmoved. The latter implications are no less controversial
than the first, but here I will simply frame the issues involved, revisiting the thematic
status of the verbal complex in section 5.
4.1. The Thematic Status of Expletives

The Light Predicate Raising account adopts the view, following Chomsky (1955/1975)
and Jacobson (1983, 1984, 1987), that it is uniquely transitive predicates that are subject
to V′ Reanalysis and raising. NP Shift examples are thus predicted always to involve
transitive thematic structure. With this point in mind, consider examples (52a) and (53a):
(52) a. There arose off the coast a powerful storm.
(cf. There arose a storm off the coast.)
___
b. There [arose off the coast] a powerful storm ___.
(53) a. There walked into the room a tall, dark stranger.
(cf. ??There walked a stranger into the room.)
b. There [walked into the room] a tall, dark stranger ___.
Light Predicate Raising not only implies that arise off the coast and walk into the room
are underlying constituents, as shown in (52b) and (53b), respectively, but also that
these phrases have transitive thematic structure. The latter would in turn seem to imply
not only that the sentence-final nominal has the status of a thematic object but also
that the expletive there has the status of a thematic subject. This result clashes with
standard analyses of there as a “dummy” or pleonastic subject in constructions like (52)
and (53) —in effect, an overt placeholder for a nonthematic position. Light Predicate
Raising thus commits one to a decidedly nonstandard view of expletives. Larson (1989)
explores this view in a preliminary way.
4.2. The Nature of Parasitic Gaps

Light Predicate Raising also takes the view that the sentence-final direct object in an NP
Shift example does not achieve its position by moving rightward, but rather by having
everything else in VP raise leftward around it. The NP itself remains in situ, in its base
site. This conclusion brings Light Predicate Raising into direct conflict with standard
views about parasitic gaps.
Consider (54a), which appears to contain two gap positions: one after offend, asso-
ciated with the wh-word (who), and another after recognize. Wh-movement in the
main clause does not require the gap in the adjunct, as shown by (54b); however, the
adjunct gap is illicit without wh-movement in the main clause (54c). The second gap
thus appears to be “parasitic” upon the first. The usual view of such “parasitic gap con-
structions” is that they involve two A′-movement chains, one in the main clause, due to
an element like wh-, and another in the adjunct, due to a covert operator (OP). Chom-
sky (1986) proposes that the two A′-chains may compose, indicated by ⊕ in (54d), and
that chain composition is what licenses the second gap.
(54) a. Who did you offend __ by not recognizing __?
b. Who did you offend __ by not recognizing Mary?
c. *You offended John by not recognizing __.
d. Who did you offend __ [OP by not recognizing __]?
Consider now (55a), due to Engdahl (1983), which appears to involve a gap in the
adjunct clause (by not recognizing) that is parasitic upon the NP Shift structure
(cf. 55b):
(55) a. I offended [by not recognizing __] my favorite uncle from Cleveland.
b. *I offended my favorite uncle from Cleveland [by not recognizing __ ].

On the analysis rehearsed above, (55a) should involve A′-movement in the main clause
in order to license the parasitic gap in the adjunct. The standard account of NP Shift
examples secures this result.25 The object moves rightward to the VP edge, yielding two
composable A′-chains (56):
(56) I offended __ [OP by not recognizing __] [my favorite uncle from Cleveland]
In sharp contrast, the Light Predicate Raising analysis of (55a) is not compatible with
the standard account of parasitic gaps since it involves only V/V′-Raising, with no
matrix A′-movement chain generated (57):
(57) I [offended by not recognizing __] my favorite uncle from Cleveland __.
We seem compelled to view offend by not recognizing as a complex transitive predicate

in its own right, with no dependence on matrix A-bar movement in its licensing, contra
the evidence suggested in (54a–c), and (55a,b). Larson (1989) proposes that parasitic
gap structures are indeed complex predicate formations and shows how many of their
standard properties can be made to follow on this view.26
5. LIGHT PREDICATE RAISING REVISITED
Implications like those reviewed above are dramatic, interesting, and plainly worth
pursuing. For, if correct, they require significant revisions in our thinking, holding out
the possibility of genuine new insights into syntactic phenomena and their interface with
semantics. At the same time, Light Predicate Raising embodies certain basic assump-
tions, some of which were mysterious even at the time of its formulation and others
that have been rendered problematic by subsequent developments in theory. In this sec-
tion I review these assumptions and briefly consider a revised picture of Light Predicate
Raising based on the updated approach to syntactic composition developed in these
introductions and in Larson (in preparation).
5.1. Light Predicate Raising and Other VP-Internal Alternations

Classical NP Shift takes the view that what’s going on in (58a) is something specific
to nominals. This is embodied not only in the name of the operation (“NP Shift”)
but also in the assumption that what it creates is an operator-variable structure (58b).
Overwhelmingly, operators are nominals, and variables are the Case-marked/checked
traces of nominals.
(58) a. Mary gave to John [everything that he demanded].
b. Mary [vp gave ti to John] OPi
As we noted, Light Predicate Raising embodies a different interface logic. It doesn’t

assume creation of an operator-variable structure and so, in principle, seems extensible
to a broader class of VP alternations. Double PP pairs like (59a,b) are a potential case
in point:
(59) a. Max talked [to all the other witnesses] [about Bill].
b. Max talked [about Bill] [to all the other witnesses].
Larson (1989, 1990) proposes that these two examples are derived from the same under-
lying structure, the second by Light Predicate Raising via an extension of the notion
“transitive” to any predicate containing an undischarged internal argument, regardless
of whether the latter is an object NP or some other category of selected complement.
Assuming that both PPs in (59) are arguments of talk (i.e., talk is triadic in this context),
talk about Bill will involve one undischarged internal θ-role (Goal) and as such will be
subject to V′ Reanalysis and raising. Raising V alone yields the assumed base order in
(59a); raising the reanalyzed V′ yields the alternative order in (59b) (see 60):
(60) VP
NP V′
Max V VP
e PP V′ ⇒V
to all the ... V PP

Base Order talk about Bill
Light Predicate Raising Order
Light Predicate Raising thus extends naturally from NP Shifts to a broader class of word
order alternations in VP, including double PP structures like (59a,b). The latter can be
accommodated without stepping beyond the basic account.
With this result in mind, consider now the four examples in (61a–d), which also
involve two PPs, but a direct object in addition. Assume that bought carries four
θ-roles in this structure (Agent, Theme, Source, Remuneration) so that both PPs are
complements:27
(61) a. John bought [a very cute little puppy] [from Mary] [for $200].
b. John bought [from Mary] [for $200] [a very cute little puppy].
c. John bought [a very cute little puppy] [for $200] [from Mary].
d. John bought [for $200] [from Mary] [a very cute little puppy].
The first pair, (61a,b), can be analyzed in the by now familiar way. Assuming (61a)
exhibits the base order of arguments, the sentence can be derived by successive raising
of the simplex V (see 62). By contrast, (61b) can be derived by raising V to the inter-
mediate empty V position, followed by reanalysis of the middle V′, whose Source (from
Mary) and Remuneration (for $200) arguments have already been combined, and which
therefore involves exactly two remaining undischarged θ-roles (see 62).
(62) VP
NP V′
John V VP
e NP V′ ⇒ V
a very cute. . . V VP
buy PP V′
(61a) Order from Mary V PP
buy for $200
(61b) Order
But now what about (61c,d)? (61c) exhibits inversion of the two PPs only, with the
direct object remaining in place (cf. 61a). Furthermore, (61d) seems to be the NP Shift
counterpart of (61c) (cf. 61b). If the latter impression is correct, then however PP inver-
sion is achieved in (61c), it must yield a V′ that can be reanalyzed and raised around the
direct object to yield (61d) by Light Predicate Raising.
One possibility, discussed in Larson (1990), is that (61a) and (61c) simply represent
alternative base projection possibilities for the two PPs. Perhaps the two θ-roles are
unordered with respect to each other on the thematic hierarchy, making available the
variant of (61a) in (61c), with the two PPs exchanged. From this structure (61c,d) can
be derived in parallel to (61a,b), respectively (see 63).
(63) VP
NP V′
John V VP
e NP V′ ⇒ V
buy PP V′
(61c) Order for $200 V PP

buy from Mary
(61d) Order
However, to my ear, (61c) exhibits a focusing effect vis-à-vis (61a) similar to that
observed in Light Predicate Raising cases. The sentence-final source PP (from Mary)
seems to receive some stress or accent and to represent new or asserted information. If
this is correct, then it seems (61c) should represent a derived form, with focus resulting
by rearrangement from a more basic order.
Consider therefore another possibility. Suppose we retain (61a) as representing the
base source for all four examples but change our view of V′ Reanalysis. Suppose we take
the latter to be thematically unconstrained and freely available.28 Then (61c) could be
derived by reanalysis of the lowest V′, raising it to the intermediate empty V position,
with subsequent raising of the lexical V (buy) alone.
(64) VP
NP V′
John V VP
e NP V′
V PP PP V′ ⇒ V
buy for $200 from Mary V PP

buy for $200
(61c) Order
And (61d) could be derived by the same first step, followed by reanalysis of the entire
middle V′ with raising to the highest empty V, a “snowballing derivation” (65):
(65) VP
NP V′
John V VP
e NP V′
V PP PP V
buy for $200 from Mary V PP

buy for $200
(61d) Order
As in (59a,b), we would be capturing an alternation in PP order by V′ Reanalysis and

raising, but here we would in fact be stepping beyond the account developed so far. In
particular, we would be abandoning the transitivity thematic constraint on V′ Reanaly-
sis and its tie to the earlier Chomsky (1955/1975) and Jacobson (1987) proposals.
These points call attention to the operation of V′ Reanalysis and its status in the
Light Predicate Raising account. We may sensibly ask: what sense does a V′ Reanalysis
operation make? Does it constitute more than a stipulation? And why, even granting
such an operation, should it be tied to transitivity? What is the logic behind a reanalysis
rule with a thematic constraint?
5.2. V′ Reanalysis Examined

As discussed at length in the General Introduction, shell projection in Larson (1988) was
tied to templatic X-bar theory, specifically to a particular “transitive” format consisting
of a single specifier (YP), a single head (X), and a single complement (ZP) (66):
(66) XP
YP X
X ZP
Syntactic structure was largely restricted to recursive, right-descending instances of this

single pattern, with many classes of circumstantial modifiers analyzed as specifiers and
complements, and with surface word obtained by raising the verbal head (67):
(67) VP
NP V′
John V VP
buy NP V′
Fido V VP
buy PP V′
from Mary V PP
buy for $200
Within this approach, the core device of V′ Reanalysis invoked in Light Predicate Raising
can be related to the assumed primacy of (66). Reanalysis of the middle (circled) V′ node
in (67) allows it to raise as a single element, producing the equivalent of (68), in which
the predicate material forms, in effect, a single, syntactically complex, transitive head.
(68) VP
NP V′
John V NP
... ...
buy from Mary for$200 Fido
V′ Reanalysis can thus be seen as “aligning” the structural notion of transitivity embod-
ied in (66) (i.e., α is transitive = α is a head X0) with a semantic notion of transitivity
deriving from θ-theory (i.e., α is transitive = α has exactly one undischarged internal
θ-role, or α has exactly two undischarged θ-roles, etc.). This is one way of implement-
ing the view of Chomsky (1955/1975) and Jacobsen (1987) noted at the outset—that
transitivity is somehow at the root of NP shift phenomena.
5.2.1. Problems with the Account

One problem this account faced at the time was the correct formulation of thematic
transitivity. The rule of V′ Reanalysis offered in Larson (1988) and reproduced in (69a)
implies the notion of thematic transitivity in (69b):
Vʹ Reanalysis: Let α be a phrase [vʹ . . .] whose θ-grid contains one undischarged
(69) a.
internal θ-role. Then α may be reanalyzed as [v . . .].
b. α is transitive = α has exactly one undischarged internal θ-role.

The latter crucially makes reference to the notion “internal θ-role,” which derives from
the work of Williams (1981). Williams was working in the context of structures like
(70) and proposed that a predicate’s θ-roles be segregated into ones assigned externally
versus. internally to VP.
(70) S
NP VP
Mary V NP PP
gave Fido to John

<Agent, Theme, Goal>
Ditransitive give is assigned the lexical “θ-grid” shown below it. The agent role is identi-
fied by underlining as an external θ-role: one that must be assigned externally to VP (i.e.,
to the subject). The goal and theme roles are internal θ-roles and assigned within VP.
Despite its invocation in (69), the notion of an internal θ-role is in fact problematic in
Larson (1988). The basic principles of shell structure projection require all roles associ-
ated with a predicate θ to be assigned within a projection of θ. Thus, in contrast to (70),
a shell structure for Mary give Fido to John requires all arguments of give to fall within
VP (71); none can be external in Williams’s sense.29
(71) VP
NP V′
to Mary V VP
give NP V′
Fido V PP
give to John
Absent a coherent notion of internal θ-role, (69b) is not well defined. An alternative
definition of transitivity is required.30
The motivation rehearsed above for V′ Reanalysis also became problematic in the
light of later theory. We noted in the General Introduction that the dissolution of tem-
platic X-bar theory under Chomsky (1994) undercut the original account of shell struc-
ture projection offered in Larson (1988). In the same way, it undercut any claims of
primacy for structure (66). While instances of [XP YP [X′ X ZP]] might derive by syn-
tactic composition, the latter could not be seen as some kind of preexisting structural
“mold” into which sentence forms had to be cast and/or toward which structures should
“align.” The transitive pattern simply lost its privileged status.
Finally, the whole idea of bar-level reanalysis rules became suspect. Chomsky (1994)
argues convincingly that bar level should not be understood as a primitive of linguistic
theory, but instead derivatively, depending on whether an item is projected or project-
able. In brief, items that are unprojected but projectable are heads (X0). Items that are
projected but not further projectable are maximal phrases (XP). Items that are both pro-
jected and further projectable are intermediate (X′). With bar level defined relationally
in this way, reanalysis of V′ as a head becomes impossible. The V′ in (67), for example,
is a projected element; as such it is simply not a head, and there is no way to regard
it as such under the new definitions. Chomsky (1994) goes on to question whether
intermediate-level phrases—being neither heads nor maximal projections—should be
legitimate targets of linguistic rules at all, making the status of V′ Reanalysis all the more
questionable. Chomsky (1994) thus essentially collapses the justification for transitive
V′ Reanalysis offered in Larson (1988, 1989, 1990) and preempts any appeal to bar-
level reanalysis as a mechanism for Light Predicate Raising.
As with shell structure projection generally, it is fruitful to consider whether Light
Predicate Raising can be reconstructed in more modern terms, wherein the prob-
lematic aspects of the earlier account are eliminated. We may begin by examining
its basic assunptions of transitivity and lexicality. Is a thematic constraint on Light
Predicate Raising really necessary? Are only certain thematically determined kinds of
predicates allowed to raise? Furthermore, is there compelling evidence for the lexical
status of the phrase moved by Light Predicate Raising? Do we really need to analyze
it as an X0?
5.2.2. Transitivity Reexamined

For concreteness, we will construe the claim that Light Predicate Raising is limited to
transitive predicates as the claim that only V′s of a binary addicity are allowed to raise.
Evidence against such a constraint would be facts showing that V′s of other addicities
must be allowed to raise as well.
We have already considered potential evidence that ditransitive/triadic V′s should be
permitted to raise. Alongside the example in (72a) and its NP Shift variant in (72b), we
noted the PP-inversion variant in (72c). The latter can be brought within the current
account by permitting V′-Raising followed by V-Raising, as in (72d):
(72) a. John bought [a very cute little puppy] [from Mary] [for $200].
b. John bought [for $200] [from Mary] [a very cute little puppy].
c. John bought [a very cute little puppy] [for $200] [from Mary].
d. John bought [a very cute little puppy] [V′ bought for $200] [from Mary] t.
.
However, the V′ [V′ bought for $200] is not a thematic transitive. At the point of raising,
bought for $200 has three roles to assign (agent, theme and source), not two, and hence
the phrase constitutes a complex ditransitive. Thus (72a,b) might be seen as evidence
that V′-Raising should extend to both transitive and ditransive V′s.
Likewise, there is evidence that V′-Raising should apply to intransitive/monadic V′s.
Larson (1988, 1989) notes the well-known asymmetry in NP Shift possibilities in PP
dative constructions (73) versus double object constructions (74), attributing this con-
trast to the transitivity constraint on V′ Reanalysis.
(73) a. Mary gave everything he demanded of her to John.
b. Mary gave to John everything he demanded of her.
(74) a. Mary gave everyone she met on Thursday a present.
b. ??Mary gave a present everyone she met on Thursday.
In (73b) the target V′ consists of triadic give with one of its arguments (to John). Since
the latter discharges one role (θgoal) in the former, V′ has the status of a transitive predi-
cate and can reanalyze as V and raise (75).
(75) VP
NP V′
Mary V VP
e NP V′ V′ ⇒ V
everything he V PP
demanded . . .
give to John
< θagent, θtheme, goal >
By contrast, the target V′ in (74b) consists of triadic give with two of its arguments:
the V′-adjoined theme a present and the trace of the goal phrase everyone she met on
Thursday, which has itself raised. Since the latter discharge two roles of give, V′ has the
status of an intransitive predicate and cannot reanalyze as V and raise. Larson (1988)
suggests this as the reason why (74b) is ill-formed.
(76) VP
NP V′
Mary V VP
e NPi V′ V′ ⇒ V
everyone she V NP
met
V NP a present
give ti
< agent, theme, goal >
This account of the asymmetry in (73)/(74) does not seem correct, however. As noted
in Larson (1990) and in “Datives: Background” in this volume, spray-load alternations
also appear to be instances of the dative alternation, with the to-variant counterpart to
the PP dative and the with-variant counterpart to the double object form. Following the
reasoning given above, we would predict a parallel asymmetry in NP Shift possibilities
in (77) and (78):
(77) a. John sprayed day-glow orange latex paint on the wall.
b. John sprayed on the wall day-glow orange latex paint.
(78) a. John sprayed all the walls in the Administration Building with green paint.
b. John sprayed with green paint all the walls in the Administration building.
In my judgment, however, there simply is no such difference. Both pairs in (77) and (78)
seem fully acceptable.31 By hypothesis, (78b) involves the same structure as (74b) and
should block V′ Reanalysis and raising in the same way if transitivity is required (cf. 79).
(79) VP
NP V′
Mary V VP
e NPi V′ V′ ⇒ V??
all the walls… V PP

V NP with green paint
spray ti
< agent, theme, loc >
But raising is not blocked. Thus, either spray-load pairs are not instances of the double
object/applicative alternation after all, or else what’s wrong with (74b) is unconnected
to the transitivity of the V′ predicate. The former conclusion seems unlikely given the
extensive parallels between the two constructions found in English and their counter-
parts in applicative languages.32 Hence we must conclude the latter.33 This implies that
raising of intransitive V′ predicates, including spray-load variants like (78b), is indeed
available.
The data in (72) and (78)/(79) thus appear to speak directly to the first question
raised above: Is a thematic constraint on Light Predicate Raising justified? Is it confined
to predicates of a certain valence? The answer appears to be “no.” Given the data, rais-
ing appears to be available with V′ predicates of greater than transitive addicity (as in
(72)) and with V′ predicates of less than transitive addicity (as in (78)/(79)). V′-Raising
thus appears to be thematically unconstrained, contra Larson (1988, 1989) and indeed
contra the original assumptions of Chomsky (1955/1975) and Jacobson (1987).
5.2.3. V0 Status Reexamined

What then about the second question raised above: Is there compelling evidence for the
lexical status of the phrase moved by Light Predicate Raising? Do we really need to (re)
analyze it as an X0?
In Larson (1988) the need for V′ to V0 reanalysis is largely a theory-internal require-
ment. Since raising is conceived as substitution into an empty head position ([V0 e]), and
since movement to head positions is restricted to heads themselves (i.e., X0 categories)
under some version of structure preservation, reanalysis of V′ to V0 is required before
movement can take place. The only empirical evidence offered in Larson (1989) for the
lexical status of the moved phrase concerns the “freezing effects” of Wexler and Culi-
cover (1980). Wexler and Culicover note that NP Shift as in (80a) and (81a) appears to
block further extraction from VP, as in (80b,c) and (81b)
(80) a. John gave to Bill the picture that was hanging on the wall.
b. *Who did John gave to __ the picture that was hanging on the wall?
(cf. Who did John give the picture that was hanging on the wall to?)
c. *Bill would be easy for John to give to __ the picture that was hanging on the wall.
(cf. Bill would be easy for John to give the picture that was hanging on the wall to.)
(81) a. They elected President of Mauritania the colonel who had engineered the recent
coup.
b. *Which country did they elect President of __ the colonel who had engineered the
recent coup?
(cf. Which country did they elect the colonel who had engineered the recent coup
President of __ ?)
If the boldfaced predicates in (80a) and (81a) are analyzed as V0s, as V′ Reanalysis pro-
poses, then extraction as in (80b,c)/(81b) can be blocked by appeal to “lexical integrity”
constraints that, roughly speaking, forbid syntactic operations from affecting material
inside a lexical category (V0) (82).
(82) a. Who did John [V0 give to ?

X
b. *Which country did they [V0 elect President of ?
X
But in fact a freezing account of (80b,c) and (81b) was not unproblematic even at
the time of Larson (1988, 1989), given the widely held assumption that verbs raise to
T to receive their tense specification. As noted explicitly in Larson (1989), raising give
to T in (82) would also seem to violate lexical integrity just as much as wh-movement;
hence some special stipulation was required for this case.
(83) [TP John [T′ T [VP John [V′ [V0 give to Bill] the picture that was hanging on the wall.
X
Furthermore, attractive alternative accounts of the ill-formedness in (80b,c) and (81b)

are conceivable that don’t appeal to lexical integrity. For example, we noted above
that Light Predicate Raising structures can be associated with event quantificational
structures wherein the sentence-final XP represents focused/new information and the
remaining material represents presupposed/old information (84):
(84) [John gave to Bill] the picture that was hanging on the wall.
∃e [giving(e) & Agent(e,j) & Goal(e,b)] (Theme(e,the-picture))
Presupposed/Old Focused/New
Interestingly, it is widely assumed that wh-extraction structures also involve a division

into a presupposition and a focus, with the answer corresponding to focused informa-
tion and the remainder corresponding to the presupposition (85) (see Kawamura 2007
and references therein):
(85) Who did John give the picture to __?
WH:x ∃e [giving(e) & Agent(e,j) & Theme(e,the-picture)] (Goal(e,x))
Presupposed/Old Focused/New
This suggests that, rather than lexical integrity, the problem with (80b,c) and (81b)
might be one of appropriate focus/presupposition representation. Wh-movement in
(80b), for example, might be seen as imposing a mapping that conflicts with the one
imposed by Light Predicate Raising, as shown in (86), where the partition required
by wh- is shown above and the one required by Light Predicate Raising is shown
below:
(86) *Who did John give to __ the picture that was hanging on the wall?
WH: Presupposed/Old Focused/New
WH:x ∃e [giving(e) & Agent(e,j) Goal(e,x) Theme(e,the-picture)
LPR: Presupposed/Old Focused/New
Potential evidence for this view is the contrast in (87) observed by Rochemont and
Culicover (1991):
(87) a. For whom did Bill purchase last week an all-expense-paid ticket to Europe?
b. *Who did Bill purchase for __ last week an all-expense-paid ticket to Europe?
Rochemont and Culicover (1991) assume identical source structures for (87a) and
(87b), with the difference in acceptability lying in whether PP versus NP is extracted.
This assumption seems incorrect, however, given the form of an answer to (87a), for
example, (88). The latter demonstrates that wh-extraction in (87a) is not from within
the boldfaced material (equivalent to 87b) but from a sentence-final site:
(88) Bill purchased last week an all expense-paid ticket to Europe [pp for Mary].
Notice now that under a Light Predicate Raising analysis in which ditransitive V′s are
permitted to raise, (88) can be derived as in (89), where purchase last week raises to the
middle V′ and the verb purchase then extracts to the highest empty V site:34
(89) VP
NP V′
Bill V VP
e NP V′
an all-expense V′ VP
V NP PP V′
purchase last week for Mary V NP

purchase last week
With the PP for Mary stranded in lowest position, the division of information structure
imposed by Wh-movement and Light Predicate Raising will now align, as shown in (90).
Both impose the same partition of presupposed/old and focused/new information:
(90) For whom did Bill purchase last week an all-expense-paid ticket to Europe ___ ?
WH: Presupposed/Old Focused/New
WH:x ∃e [purchasing(e) & Agent(e,b) & Theme(e,ticket) & Temp(e,lw)] (Goal(e,x))
LPR: Presupposed/Old Focused/New
Thus information structure partitioning appears to offer a potential alternative approach

to the “freezing effects” in (80)/(81), one involving no appeal to the lexical (X0) status
of the raised material.
6. LIGHT PREDICATE RAISING AS X′ INDETERMINACY?
The above remarks undermine both core elements of the V′ Reanalysis mechanism in
Larson (1988, 1989). Neither thematic transivity nor bar-level change appears empiri-
cally well grounded. Furthermore, the theoretical assumptions underlying V′ Reanalysis
have either lapsed or simply become untenable in the light of subsequent developments.
As in the case of shell theory itself, we may again ask whether the general approach is
forfeit or whether reconstruction is available in more contemporary terms.
I would like to briefly suggest an approach to Light Predicate Raising based on the
updated version of shell theory developed here; it turns on the observation, noted above,
that under Chomsky (1994) intermediate-level projections have an indeterminate status
as neither unprojected (and hence not heads, X0) nor unprojectable (and hence not
maximal phrases, XP).
To illustrate what I have in mind, consider (91), which represents the derivation of
a sentence like John gave Fido to Mary at the point where we have merged the little v
[AG] voice head (co-selected with give) and VP:
(91) v′
v VP
[uAGval[ ]]
Fido V′
[iTH[2]]
give PP
[uAG]]
[uTHval[2] to Mary
[uGL[1] [uGLval[1]] [iGL[1]]
By assumptions, v bears a strong V feature and attracts a verbal head to it. What can
v identify as an appropriate verbal target for raising? There are three potential tar-
gets in its scope: [VP Fido give to Mary], [V′ give to Mary], and give. Since the relevant
nodes stand in dominance relations to each other, they do not stand in any c-command
relations. Hence each counts as equally close to v from the standpoint of economy/
minimality.
The candidate give is present in the Numeration, the basic selection of X0 items from
the lexicon. Hence I assume it can be identified as a verbal head. This makes give a target
for raising by little v, yielding the by now familiar structure in (92):
(92) v′
v VP
v give Fido V′
[uAGval[3]] [uAG[ 3]] [iTH[2]]
give PP
[uTHval[2]
[uAG[ ]]
[uGL[1] to Mary
[uTHval[2]
[uGL[1]] [uGLval[1]] [iGL[1]]
The candidate [VP Fido give to Mary] has merged as the complement of little v in (91),
giving it the relational status of a maximal phrase, that is, a construction not further
projected. Assuming that v must attract a head, this eliminates [VP Fido give to Mary] as
a target for raising by little v.
Consider then the phrase [V′ give to Mary]. As noted earlier the latter is both pro-
jected and further projectable: it is intermediate. Suppose we interpret intermediacy as
indeterminacy. In other words, since give to Mary doesn’t fully match the specifications
of either heads or maximal phrases, suppose the grammar allows either analysis. If we
analyze [V′ give to Mary] as a maximal phrase, this case will fall together with the previ-
ous one. Assuming v must attract a head, [V′ give to Mary] will be eliminated as a target
for raising to little v. By contrast, if we analyze [V′ give to Mary] as a head, then this case
falls together with the first one: [V′ give to Mary] becomes a target for raising to little v,
yielding the Light Predicate Raising structure in (93):
(93) v′
v VP
v V′ Fido V′
[uAGval[3]]
give PP give PP
[uAG[ ]] [uAG[ ]]
[uTHval[2] to Mary [uTHval[2] to Mary
[uGL[1] [uGLval[1]] [iGL[1]] [uGL[1] [uGLval[1]] [iGL[1]]
Hence what was identified as V′ Reanalysis in Larson (1988, 1989) and associated with
a special privileged status for transitive structures can be reanalyzed as “V′ Indetermi-
nacy” and associated with a special privileged status for intermediate (X′) projections
generally, under the bare phrase structure theory of Chomsky (1994).
The reformulation of Light Predicate Raising offered above preserves two of the
core consequences discussed earlier. The V and VP elements occurring to the left of
a “shifted” direct object NP must still form a single constituent of category V′. The
implications of NP Shift data for modifier attachment and reduced-clause structures
thus remain unchanged.
Constituent (V′)
N ___.
Likewise, the direct object in an NP Shift structure continues to be analyzed as in situ,

in its base-merge position. The implications of NP Shift data for the analysis of parasitic
gap phenomena therefore also remain unchanged.
In Situ Object
⇓
Subject [V′ V X Y X ] NP ___.
What is different is that the raised V′ need no longer have the thematic status of a transi-
tive verb. With the connection to transitivity eliminated, derivations like those in (64)
and (65) noted above become possible.35
The reasoning about V′ Indeterminacy offered above is perfectly general, and hence
carries the expectation of X′ Indeterminacy and the equivalent of Light Predicate Rais-
ing in every category. However, I must leave exploration of these consequences for
another occasion.36
NOTES
1. See Larson (in preparation) for a fuller discussion of the issues.

2. Chomsky (1955/1975, p. 510) extends this analysis to virtually all of the complex predica-
tive constructions considered later by categorial Montague Grammar.
3. Interestingly, Chomsky was already aware by this early point that the relevant variable
was not heaviness, despite subsequent terminology. Thus he observes, “It is apparently not
the length in words of the object that determines the naturalness of the transformation,
but rather, in some sense, its complexity. Thus ‘they brought all of the leaders of the riot
in’ seems more natural than ‘they brought the man I saw in.’ The latter, though shorter, is
more complex on the transformational level . . . A good deal of further study is needed to
determine the nature of this process and to define properly the relevant sense of complexity
of the object” (Chomsky 1955/1975, p.477).
4. Jacobson (1987) makes preliminary remarks on Heavy NP Shift, stating on p. 32: “We
return to this in Section 4.3.” However, section 4 of the paper ends at subsection 4.2, and
the final section 5 doesn’t discuss the issue.
5. “TVP/V” in (8b) is to be read “TVP or V,” with the choice made in parallel in the two positions.
6. The discussion below repeats some material from Larson (2004, reprint in this volume),
which provides additional details.
7. I ignore tense for simplicity.
8. This proposal represents a development and extension of ground-breaking work by Partee
(1991).
9. See Diesing (1992), Heim and Kratzer (1997), and Herburger (2000) for discussion of the
presuppositional contribution of quantificational structure.
10. For simplicity I ignore the quantificational structure of everything that he demanded of John
and treat this expression as if it denoted a constant. A fuller exposition might represent the
interpretation as approximately:
∃e [giving(e) & Ag(e,m) & Gl(e,j)] (∀x[∃e′[demanding(e) & Ag(e′,j) & Theme(e′,x)](Theme(e,x)));
that is, “There was a giving by Mary to John, and its theme was every x such that there had been
a demanding of x by John.”
11. For some critical reactions to the Light Predicate Raising account, I refer readers to Roche-
mont and Culicover (1991) (however, see Culicover and Rochemont (1997) for a rethinking
of their position), Kuno and Takami (1993), and Williams (1994).
12. Within a right-adjunction analysis, it is tempting to appeal to Quantifier Raising and Logical
Form scope in explaining (22a) (Ernst 1994), but under the widely held view that frequency
adverbs do not undergo Quantifier Raising this proposal will not be extensible to (21a).
13. Other crucial arguments involving VP ellipsis and binding are analyzed in Larson (in prepa-
ration), where it is shown that Davidsonian event quantification sheds new light on these
domains as well.
14. For simplicity I ignore the contribution of tense.
15. See, for example, Dowty (1979), Hoekstra (1988), Carrier and Randall (1992) and Kratzer
(2004), among many others.
16. By “not equivalent to those just noted” I mean to exclude “function-composition” analyses
wherein, for example, proposition-taking believe combines with the predicate intelligent to
yield a result as in (ia), where P ranges over NP intensions. Composing (ia) with John then
yields (ib), what one would have obtained by directly composing John with to be intelligent,
and the result with believe.
(i) a. believe intelligent ⇒ λP[believe’(întelligent’(P))]
b. believe John intelligent ⇒ believe’(întelligent’(John))]
Function composition does allow one to graft the propositional semantics onto the complex
predicate syntax. But plainly this is not the same as providing a semantics that makes a
complex predicate syntax the expected one.
17. Note that (39a) improves with an in-adverbial if some kind of predicate equivalent to
smooth is understood in context (flat, clean, fully, etc.); this simply reinforces the basic
observation in the text.
18. See Tenny (1994) for related ideas and proposals.
19. Reading is often conceptualized as a form of (media) consumption. Cf. John devoured the
latest Stephen King thriller.
20. It was a scientific discovery, not something given by human intuition, that temperature is in
fact a bounded scale—that there exists an absolute zero (0° K).
21. According to the American Heritage Dictionary of the English Language.
22. Possibly this is because sanding can affect an object equally along a variety of scalar dimensions,
for example, making it smoother, flatter, cleaner, fairer (in the technical shipbuilding sense), and
so on. Since all are equal as effects, none is privileged by being encoded by sand itself.
23. Some care must be exercised in applying this test. The table was completely wet has a perfectly
sensible construal according to which it asserts, not that the table is wet to the maximal degree,
but rather that all relevant parts of the table are wet. Only the availability of the former read-
ing is relevant to the evaluation of whether an adjective is closed versus open scale.
24. Larson (in preparation) notes the existence of nonthematic resultatives like (ia–d), where V +
resultative XP form an apparent idiom. The existence of such idioms is compatible with the
right-descending syntactic analysis in (iia), where V + AP/PP form a constituent (knock dead).
Resultative V–AP/PP idioms are not compatible with the reduced-clause syntax (iib).
(i) a. Mary knocked the audience dead. (‘Mary impressed the audience greatly.’)
≠ Mary knocked the audience.
b. Mary drank John under the table. (‘Mary out-drank John.’)
≠ Mary drank John.
c. Max swept Alice off her feet. (‘Max romanced Alice to great effect.’)
≠ Max swept Alice.
d. Mary ran the business into the ground. (‘Mary destroyed the business.’)
≠ Mary ran the business. (on the relevant sense of run)
(ii) a. Mary knock the audience [V′ knock dead].
b. Mary knock [SC the audience dead].
25. More precisely, it does so under the view that displacement to the right VP edge is A′-movement.
26. See Nissenbaum (2000) for an independent elaboration and discussion of the complex predi-
cate view.
27. Cf. the General Introduction, specifically, the discussion of example (18).
28. This suggestion is due to Noam Chomsky (personal communication).
29. This point holds for any theory of projection embodying the VP-Internal Subject Hypoth-
esis; on all such theories Williams’s internal/external theta-role distinction is void.
30. Larson (1989) explores a definition from logic, wherein transitives correspond to binary
relations; that is, (i)
(i) α is transitive = α has exactly two undischarged θ-roles.
31. For me the same judgments hold for analogous to-with alternations observed with award
and present. Both variants allow NP Shift of the object:
(i) a. i. John awarded/presented distinguished flying cross medals to the airmen.
ii. John awarded/presented to the airmen distinguished flying cross medals.
b. i. John awarded/presented all the soldiers assembled on the parade grounds with medals.
ii. John awarded/presented with medals all the soldiers assembled on the parade grounds.
32. See Bruening (2001) for insightful discussion of the parallels in “scope freezing.”
33. It has been suggested by Oehrle (1983) that the ill-formedness of (74b) and counterpart
constructions where the object is wh-extracted is a processing fact having to do specifically
with the double-NP status of the dative alternation. This is plausible insofar as with the
spray-load and award-present alternations, where we are dealing with an NP and a PP, the
effects are absent.
34. This derivation is parallel to (65) above.
35. This change also allows for an interesting reanalysis of the expletive data, in which the
latter continue to be viewed as arguments, but projected lower in the structure (following
Tortora 1997) and moved to their final higher position by the equivalent of Applicative Shift
(A-Shift). See Larson (in preparation) for development.
36. See Larson (1991, reprinted in this volume) for potential instances of Light Predicate Raising
within the DP projection.
REFERENCES
Andrews, A. (1983) “A Note on the Constituent Structure of Modifiers,” Linguistic Inquiry 14:
695–697.
Bruening, B. (2001) “QR Obeys Superiority: Frozen Scope and ACD” Linguistic Inquiry 32:
233–273.
Büring, D. (2005) Binding Theory. Cambridge: Cambridge University Press.
Carrier, J. and J. Randall (1992) “The Argument Structure and Syntactic Structure of Resulta-
tives.” Linguistic Inquiry 23: 173–234.
Chomsky, N. (1955/1975) The Logical Structure of Linguistic Theory. New York: Plenum Press.
——— (1981) Lectures on Government and Binding. Dordrecht: Foris.
——— (1986) Barriers. Cambridge, MA: MIT Press.
——— (1994) “Bare Phrase Structure,” MIT Occasional Papers in Linguistics. Cambridge, MA:
Culicover, P. W. and M. Rochemont. (1997) “Deriving Dependent Right Adjuncts in English.” In
H. van Riemsdijk, D. LeBlanc, and D. Beermann, eds., Rightward Movement (pp. 279–300).
Amsterdam: John Benjamins.
Diesing, M. (1992) Indefinites. Cambridge, MA: MIT Press.
Dowty, D. (1979) Word Meaning and Montague Grammar. Dordrecht: Kluwer.
Ernst, T. (1994) “M-Command and Precedence,” Linguistic Inquiry 25: 327–335.
——— (2001) The Syntax of Adjuncts. Cambridge: Cambridge University Press.
Heim, I. and A. Kratzer (1997) Semantics in Generative Grammar. Malden, MA: Blackwell.
Herburger, E. (2000) On What Counts. Cambridge, MA: MIT Press.
Hoekstra, T. (1988) “Small Clause Results,” Lingua 74: 101–139.
Jacobson, P. (1983) “Phrase Structure and Grammatical Relations,” LSA Forum Lecture, Linguis-
tic Society of America Annual Meeting, Minneapolis, MN.
——— (1984) “Verb-Particle Constructions in Phrase Structure Grammar,” invited colloquium
presentation, University of Pennsylvania, Philadelphia, December.
——— (1987) “Phrase Structure, Grammatical Relations and Discontinuous Constituents,” in
G. Huck and A. Ojeda, eds., Syntax and Semantics 20: Discontinuous Constituency (pp. 27–69).
Kawamura, T. (2007) Some Interactions of Focus and Focus Sensitive Elements. Doctoral disserta-
tion, Stony Brook University, Stony Brook, NY.
Kratzer, A. (2004) “Building Resultatives,” unpublished manuscript, University of Massachusetts,
Amherst, MA.
Krifka, M. (1989) “Nominal Reference, Temporal Constitution and Quantification in Event
Semantics,” in R. Bartsch, J. van Benthem, and P. van Emde Boas, eds., Semantics and Contex-
tual Expression (pp. 75–115). Dordrecht: Foris.
——— (1992) “Thematic Relations as Links between Nominal Reference and Temporal Consti-
tution,” in I. A. Sag and A. Szabolcsi, eds., Lexical Matters (pp. 29–54). Stanford, CA: CSLI
Publications.
——— (1998) “The Origins of Telicity,” in S. Rothstein, ed., Events and Grammar, (pp. 197–236).
Dordrecht: Kluwer.
Kuno, S. and K. Takami (1993) Grammar and Discourse Principles: Functional Syntax and GB
Theory. Chicago: University of Chicago Press.
Larson, R. (1988) “On the Double Object Construction, Linguistic Inquiry 19: 335–391.
——— (1989) “Light Predicate Raising,” Lexicon Project Working Papers No. 27. Center for
Cognitive Science, MIT. Republished in this volume.
——— (1991) “The Projection of DP and DegP,” unpublished manuscript, Stony Brook University,
Stony Brook, NY
——— (2004) “Sentence-Final Adverbs and ‘scope’,” in M. Wolf and K. Moulton, eds., Proceed-
ings of NELS 34 (pp. 23–43). Amherst, MA: GLSA, University of Massachusetts-Amherst.
———. (in preparation) VP and DP.
Larson, R. and G. Segal (1995) Knowledge of Meaning. Cambridge, MA: MIT Press.
Montague, R. (1974) Formal Philosophy. New Haven, CT: Yale University Press.
Nissenbaum, J. (2000) “Covert Movement and Parasitic Gaps,” in M. Hirotani, A. Coetzee,
N. Hall, and Y.-J. Kim, eds., Proceedings of NELS 30 (pp. 541–555). Amherst MA: GLSA,
University of Massachusetts-Amherst.
Oehrle, R. (1983) “The Inaccessibility of the Inner NP: Corrections and Speculations,” Linguistic
Analysis 12.2: 159–171.
Partee, B. (1991) “Tropic, Focus and Quantification,” in S. Moore and A. Z. Wyner, eds., Proceed-
ings from SALT 1, 159–187. Ithaca, NY: CLC Publications, Cornell University.
Phillips, C. (2003) “Linear Order and Constituency,” Linguistic Inquiry 34: 37–90.
Reinhart, T. (1983) Anaphora and Semantic Interpretation. Chicago: Chicago University Press.
Rochemont, M. and P. Culicover (1991) “In Defense of Rightward Movement,” Toronto Working
Papers in Linguistics 11. Toronto, ON: University of Toronto.
Stowell, T. (1981) Origins of Phrase Structure. Doctoral dissertation, MIT, Cambridge MA.
Tenny, C. (1994) Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer.
Thomason, R. and R. Stalnaker (1973) “A Semantic Theory of Adverbs,” Linguistic Inquiry 4:
195–220.
Tortora, C. (1997) The Syntax and Semantics of the Weak Locative. Doctoral dissertation, Uni-
versity of Delaware, Newark.
Wechsler, S. (2005) “Resultatives under the Event-Argument Homomorphism Model of Telicity.”
in N Erteschik-Shir and T. Rapoport, eds., The Syntax of Aspect—Deriving Thematic and
Aspectual Interpretation (pp. 255–273). Oxford: Oxford University Press.
Wexler, K. and P. Culicover (1980) A Formal Theory of Language Learnability. Cambridge, MA:
MIT Press.
Williams, E. (1981) “Argument Structure and Morphology,” The Linguistic Review 1: 81–114.
——— (1994) Thematic Structure in syntax. Cambridge, MA: MIT Press.
4 Light Predicate Raising*
Following Ross (1967/1986) many linguists have assumed that sentence pairs like
(1)–(4) are related by a rule of “Heavy NP Shift”:
(1) a. Mary gave everything that he demanded to John.
b. Mary gave to John everything that he demanded.
(2) a. Max put all the boxes of home furnishings in his car.
b. Max put in his car all the boxes of home furnishings.
(3) a. I would consider anyone who leaves his doors unlocked foolish.
b. I would consider foolish anyone who leaves his doors unlocked.
(4) a. You see large numbers of Dr. Who fans at such conventions.
b. You see at such conventions large numbers of Dr. Who fans.
In essence, this rule derives the second member of each pair from the first by rightward
movement of a phonologically “heavy” noun phrase:
(5) a. Mary gave t to John [everything that he demanded].
b. Max put t in his car [all the boxes of home furnishings].
In this article I explore an alternative to the Ross analysis. On the proposed account,
the “b” examples in (1)–(4) arise, not by rightward movement of an NP, but rather by
leftward movement of a verbal category:
(6) a. Mary [gave to John] everything that he demanded t.
b. Max [put in his car] all the boxes of home furnishings t.
What relates such pairs is thus not “NP Shift,” but a form of predicate raising.
In section 1, I motivate the analysis by considering some data that are mysterious
under an NP Shift view of (1)–(4). In section 2, I introduce the predicate raising account
and show that it illuminates these puzzles as well as a number of other simple facts
about the construction. Sections 3 and 4 explore two more extended consequences; sec-
tion 3 examines NP Shift in expletive constructions with existential and presentational
interpretations, where a postverbal NP appears to undergo rightward movement. Sec-
tion 4 discusses the radical consequences of predicate raising for the analysis of para-
sitic gaps. Finally, in section 5, I briefly consider the general question of why a rule of
rightward NP Movement should be absent.
1. ASYMMETRIES BETWEEN “HEAVY NP SHIFT”

AND A′-MOVEMENT
Under standard views, Heavy NP Shift involves rightward movement of NP with

adjunction to VP:1
(7) VP
VP NPi
... NPi ... α
This assimilates Heavy NP Shift to the general class of A′-movements, and hence leads
us to expect that it will pattern similarly to Wh-movement, Topicalization, Quantifier
Raising, and so on. Interestingly, there are a number of cases in which the two pattern
quite differently.
1.1. Absence of P-Stranding

One well-known divergence between Heavy NP Shift and A′-movement involves extrac-
tion from PP. As noted by Ross (1967/1986), Bresnan (1976), and Stowell (1981),
among others, Wh-movement freely allows preposition stranding in English, whereas
Heavy NP Shift strongly resists it and demands pied-piping of PP (8)–(10):
(8) a. Who did you talk to about Jonnie’s problems?
b. *I talked to about Jonnie’s problems all of the teachers.
c. I talked about Jonnie’s problems to all of the teachers.
(cf. I talked to all of the teachers about Jonnie’s problems.)
(9) a. Who did he bargain with about wages?
b. *He bargained with about wages three senior officials.

Light Predicate Raising 305
c. He bargained about wages with three senior officials.
(cf. He bargained with three senior officials about wages.)
(10) a. Which city did she fly off to after the semester?
b. *She flew off to after the semester the oldest city in Mongolia.
c. She flew off after the semester to the oldest city in Mongolia.
(cf. She flew off to the oldest city in Mongolia after the semester.)
This asymmetry is not limited to English. Christensen (1987) notes analogous facts for
Norwegian (11)–(12), and parallel data exist in Swedish (13)–(14) (Swedish examples
due to Elisabet Engdahl (personal communication):
(11) a. Vi har lant den interessante boken du nevnte til Petter.
we have lent the interesting book you mentioned to Peter
b. Vi har lant til Petter den interessante boken du nevnte.
(12) a. Vi skal lese om den interessante boken du nevnte i morgen.
we shall read about the interesting book you mentioned tomorrow
b. *Vi skal lese om i morgen den interessante boken du nevnte.
(13) a. Max gav allt han hade med sig till John.
Max gave everything he had with him to John
b. Max gav till John allt han hade med sig.
(14) a. Jag talade med alla mina lärare om mina problem.
I talked with all my teachers about my problems
b Vem ska jag tala med om mina problem?
who shall I talk with about my problems
c. *Jag talade med om mina problem alla mina lärare.
I talked with about my problems all my teachers
Standardly, this divergence between A′-movement and Heavy NP Shift in extraction

from PP is simply stipulated. Ross (1967/1986) imposes a nonstrandability condition
on the Heavy NP Shift operation:2
(15) [PP P ___ ] NP

X
Similarly, Bresnan (1976) assumes distinct transformations of Heavy NP Shift and Wh-
movement and states structural conditions for application that demand pied-piping
in the latter case. Stowell (1981) elaborates Kayne’s (1981) proposal that preposition
stranding is licensed by a reanalysis operation incorporating V and P. Stowell sug-
gests an “antecedent condition” that requires a moved element to be to the left of a
V + P complex governing its trace. This forbids reanalysis in the case of rightward
movements like Heavy NP Shift, and so blocks preposition stranding. Here too a rule-
specific condition is assumed, although one governing reanalysis and not Heavy NP
Shift per se.
Under all of these proposals, the reason why constraints should hold of NP Shift (or
rightward movement) but not other, putatively similar movements goes unexplained.
We are thus left with an asymmetry between the two.
1.2. Bounding Effects with Adjuncts

A second, less familiar asymmetry between Heavy NP Shift and A′-movement concerns
their interaction with certain adjuncts. As is well known, English allows for a variety of
unselected depictive phrases that may be understood as predicated of either the subject
(16) or the object (17):3
(16) a. John left the party [angry].
b. I played the game [blindfolded].
c. Alice drove the car [drunk].
(17) a. Felix found the door [open].
b. Jude ate the fish [raw].
c. Edith drinks tea [cooled with ice].
In all such examples, the object is fully extractable under leftward A′-movement:
(18) a. Which party did John leave angry?
b. That game I never play blindfolded.
c. The car that Alice drove drunk is parked over there.
(19) a. Which door did Felix find open?
b. This fish you should never eat raw.
c. A tea that Edith drinks cooled with ice is easy to find.
However, Heavy NP Shift seems to be licensed smoothly only when the adjunct has
object orientation (20)–(22); Heavy NP Shift with subject-predicated adjuncts is con-
siderably weaker (23)–(25):4
(20) a. Felix found the Midville music library door open.
b. Felix found open the Midville music library door.
(21) a. Jude never eats fish over two days old raw.
b. Jude never eats raw fish over two days old.
(22) a. Edith always drinks herbal tea that is made from chamomile and lemongrass
cooled with ice.
b. Edith always drinks cooled with ice herbal tea that is made from chamomile and
lemongrass.
(23) a. John left the party for the ambassador from Ulan Bator angry.
b. *John left angry the party for the ambassador from Ulan Bator.
(24) a. Max discussed the recent Broadway musical by Sondheim nude.
b. ?*Max discussed nude the recent Broadway musical by Sondheim.
(25) a. Alice never drives her red, fuel-injected ’68 Chevy drunk.
b. *Alice never drives drunk her red, fuel-injected ’68 Chevy.
It is tempting to try to view this result in terms of different attachment sites for
subject- versus object-oriented adjuncts (Williams 1980) and an associated bounding
condition on Heavy NP Shift. The facts in (20)–(25) might seem to follow if subject-
oriented adjuncts are always attached to S and if Heavy NP Shift is required to take a
VP-internal adjunction. However, Andrews (1982) argues convincingly on the basis of
data from VP-Preposing (26a), Though-Movement (26b), and Wh-clefting (26c) that
both subject- and object-oriented adjuncts must have a VP-internal attachment:
(26) a. John said he would eat the meat nude/raw, and eat the meat nude/raw he did.
b. Eat the meat nude/raw though John did, nobody thought he was crazy.
c. What John did was eat the meat nude/raw.
In view of this, no simple bounding condition seems possible for the divergence
between Heavy NP Shift and other A′-movements. We are thus left with a second
apparent asymmetry.
1.3. The Distribution of meng- in Modern Indonesian

Chung (1976) describes the distribution of the transitivity marker meng- in the gram-
mar of modern Bahasa Indonesian, which exhibits a third distinction between Heavy
NP Shift and other forms of movement. Examples (27a–c) (from Chung 1976) illustrate
the occurrence of meng- in simple clauses:
(27) a. Saja (me-)lihat diri saja dalam air.

1SG TRANS-see self my in water
‘I saw myself in the water.’
b. Mereka (me-)masak ikan untuk saja.

3PL TRANS-cook fish for 1SG
‘They cooked a fish for me.’
c. Mereka ber- lajar ke Amerika.
*me
3PL INTRANS sail to America
TRANS
‘They sailed to America.’
Thus (27a,b) show meng- licensed by reflexive and nonreflexive object nominals, and
(27c) shows that when the verb is intransitive, the intransitive marker ber- must occur
and not meng-.
The basic rule for the distribution of meng- appears to be that this element attaches
to transitive verbs that are adjacent to their direct object in surface form. Accordingly,
although meng- appears smoothly in examples like (27a–c), it cannot occur in sentences
where the direct object has undergone movement, as in passives (28a), object-preposing
constructions (28b), relatives (28c), clefts (28d), or questions (28e):
(28) a. Dia (*men-)di-pukul oleh mereka.
3sg trans-pass-hit by 3pl
‘He was hit by them.’
b. Buku itu saja (*mem-)batja.
book the 1sg. trans-read
‘That book I read.’
c. Surat jang anak itu sedang (*mem-)tulis pandajang-nja tiga halaman.
letter comp child the prog trans-write length-its three page
‘The letter that the child was writing is three pages long.’
d. Kamu-lah jang saja (*men-)tunggu.
2sg-emph comp 1sg trans-wait
‘It’s you that I’m waiting for.’

e. Apa jang mereka (*me-)masak untuk pesta?
What comp 3pl trans-cook for party
‘What are they cooking for the party?’
Interestingly, there is one apparent exception to the generalization just stated: Heavy
NP Shift constructions. Although the direct object does not occur adjacent to its verb
in (29a–c), having been shifted rightward on the usual view, meng- is nonetheless able
to appear:
(29) a. Iwan me-masukkan ke dalam truk lima2 andjing jang ribut.
Iwan trans-force to inside truck five dog comp noisy
‘Iwan forced into the truck five dogs that were barking loudly.’
b. Yati me-masak untuk Ali ikan jang di-tangkap oleh kakak-nja kemarni.
Yati trans-cook for Ali fish comp pass-catch by sibling-her yesterday
‘Yati cooked for Ali the fish caught by her brother yesterday.’
c. Saja mem-beri kepada Ali semua uang jang kamu kasih.
1sg trans-give to Ali all money comp 2sg give
‘I gave to Ali all the money you gave to me.’
Chung herself gives the distribution in terms of linear order, stating that meng- is licensed
when the verb precedes its object in surface form. But since Heavy NP Shift is the only
rightward movement rule discussed in Chung (1976), and hence the only rule that results
in an object preceded by (but not adjacent) to its verb at S-Structure, in effect the pre-
cedence condition stipulates Heavy NP Shift as exceptional. We thus observe a third
asymmetry.
2. HEAVY NP SHIFT RECONSIDERED
The facts noted above can be illuminated under an alternative view of the basic “NP
Shift” phenomenon advanced in Larson (1988). The latter appeals to D-Structures deriv-
ing from early proposals by Chomsky (1955/1975) and Fillmore (1965), and their more
recent elaboration in work by Dowty (1978), Bach (1979) and Jacobson (1983,1987).
The basic idea is that in an example like (30a) involving multiple arguments, the latter
are initially structured within VP in subject-predicate form, with the outermost elements
being hierarchically most subordinate. In particular, the VP underlying put the box in
his car is a binary branching structure consisting of an empty V taking a specifier Max
and a VP complement. The latter is in turn headed by put and takes a specifier the box
and a complement in his car:
(30) a. Max will [VP put the box in his car].
b. IP
. . . VP
NP V′
Max V VP
e NP V′
the box V PP
put in his car
The intuitive content of this structure is that put takes the locative in his car, forming a
small predicate put-in-his-car. This is predicated of an “inner subject” the box, forming
a VP with clause like structure the box put in his car. The latter is then predicated of
the subject Max. The correct surface ordering of IP constituents derives by raising of
put to the empty V position and by raising of Max to IP specifier position.5 This move-
ment is taken to follow from general principles governing the assignment of Case and
agreement:6
(31) IP
NP I′
Max I VP
will NP V′
e V VP
put NP V′
the box V PP
e in his car
Heavy NP Shift examples like Max put in his car all the boxes of home furnishings
derive by a variant of the V-Raising operation in (31). Specifically, they arise when the
latter interacts with the following (optional) rule of V′ Reanalysis:
V′ Reanalysis
If α is a V′ and α is thematically monotransitive, then α may be reanalyzed

as V.
Understanding “thematically monotransitive” to mean that α determines exactly two
thematic roles, V′ Reanalysis says that any V′ that corresponds semantically to a binary
relation may be categorially reconstrued as a head.7 To illustrate, consider the structure
in (32):
(32) VP
NP V′
Max V VP
e NP V′
all the boxes of V PP

home furnishings
put in his car
Put selects three arguments—an agent, a theme, and a goal—and in the lowest V′ in
(32), the goal role has been discharged by the PP in his car. It follows that V′ is a predi-
cate with exactly two undischarged arguments—agent and theme—and hence is subject
to optional V′ Reanalysis.
If V′ Reanalysis does not apply, head-to-head movement of V proceeds as in (31)
above; put raises to the [V e] position, yielding the “nonshifted” version of VP: put all
the boxes of home furnishings in his car. On the other hand, if V′ Reanalysis does apply,
then the result is (33a). Raising now applies to the entire complex constituent put in his
car, yielding (33b):
(33) a. VP
NP V′
Max V VP
e NP V
all the boxes of V PP

home furnishings
put in his car
b. VP
NP V′
Max V VP
put in his car NP V
all the boxes of e

home furnishings
This movement positions the complex predicate headed by put where it may receive
tense and agreement information, and where it may Case-mark the object, as required.
2.1. Some Consequences

A “Light Predicate Raising” analysis sheds light on the NP Shift/A′-movement asym-
metries observed in section 1.
2.1.1. Preposition Stranding

We noted that prepositions are not strandable under NP Shift (34a). On the present
account, to generate (34a) by predicate raising we require a pre-raising structure in
which V (talk) and P (to) form a constituent that excludes the direct object:
(34) a. *We talked to about Jonnie’s problems all the teachers.
b. *We [talk to about J’s problems] all the teachers t.
There are two possible ways for such a V-P unit to arise: the verb and preposition
might form a D-Structure group—a lexical “merger” in the sense of Marantz (1984)
(35a). Alternatively, V and P might be amalgamated at some point prior to raising
by (downgrading) head-to-head movement—“P-incorporation” in the sense of Baker
(1988) (35b):
(35) a. VP
NP V′
We V VP
e NP V′
all the V PP
teachers
V P about J’s
problems
talk to
b. VP
NP V′
We V VP
e PP V′
Pi NP Vi PP
e all the V Pi about J’s

teachers problems
talk to
Both of these possibilities can be ruled out, however. (35a) can be excluded on the
grounds that English simply does not permit the free V + P lexical merger required
for D-Structure units like talk to (see Marantz (1984) for discussion). Since there is
no such lexical form, (35a) is simply unavailable. (35b), on the other hand, can be
ruled out as an Empty Category Principle (ECP) violation. As it stands, this structure
is equivalent to an incorporation from subject position, hence [P e] fails to be properly
governed (Baker 1988). Moreover, V′ Reanalysis and Light Predicate Raising will not
save the violation:
(36) VP
NP V′
We V VP
Vi PP PP V
V Pi about J’s Pi NP e
problems
talk to e all the
teachers
Assuming that heads share indices with their projections as a matter of X-bar theory,
but that adjoining elements share indices only with the node they adjoin to, it follows
that the index of P will label the V under which P adjoins in (35b), but not the higher
projection V′. This means that after V′ Reanalysis, the reanalyzed V′ will fail to bear
the index of P. The ECP is thus again violated in (36) since [P e] fails to be properly
governed.
This analysis predicts that examples like (34a) should be possible exactly when the
verb and preposition form a lexically specified constituent. This situation is exemplified
by verb-particle constructions like (37a–c):
(37) a. Max looked up the word in the dictionary.
b. Felix threw out the old papers.
c. They sealed off the last remaining exit.
Assuming that look up, throw out, and seal off form constituents at D-Structure, we
expect well-formed examples parallel to (34a) with a “stranded” particle. This expecta-
tion is met:8
(38) a. Max looked up in the dictionary the word that Bill had asked him about.
b. Felix threw out with the trash a manuscript that Oscar had been working on for
years.
c. They sealed off without incident the Blob’s only remaining exit from the cave.
2.1.2. Adjunct Orientation
V-Raising structures allow a simple account of the adjunct orientation facts noted
above. Recall that NP Shift was possible over object-oriented adjuncts, but not over
subject-oriented adjuncts:
(39) a. Jude never eats fish over two days old raw.
b. Jude never eats raw fish over two days old.
(40) a. John left the party for the ambassador angry.
b. *John left angry the party for the ambassador from Ulan Bator.
Suppose now (largely following proposals by Dowty (1979) and Schein (1995))
that differences in subject versus object orientation are represented structurally as
shown in (41) below. These structures represent object-oriented adjuncts as attach-
ing at a point that makes the direct object their closest c-commanding argument,
while subject-oriented adjuncts attach at a point that makes the subject their closest
c-commander.9
(41) a. VP b. VP
NP V′ NP V′
Jude V VP John V′ AP
e NP V′ V NP angry
the fish V AP leave the party

eat raw
We may suppose such configurations to follow from a simple principle governing the-
matic assignment with secondary predicate structures:
Principle
XP assigns a θ-role to NP iff XP is sister to a V projection that assigns a θ-role

to NP.
In (41a), V assigns a θ-role to the object; hence raw is sister to V when predicated of
the object. In (41b), V′ assigns a θ-role to the subject; hence angry is sister to V′ when
predicated of the subject.
These assumptions entail that object-oriented adjuncts will form a constituent with
V that excludes the direct object, and hence that they will involve verb raising. This in
turn yields the possibility of raising V + ADJUNCT as a unit, producing NP Shift with
object-oriented predicates:
(42) VP
NP V′
Jude V VP
e NP V
fish over two V AP

days old
eat raw
The sisterhood of V and AP involves no discharge of thematic roles; hence the lower
V′ preserves the valence of the original V. In the example at hand, this means that
eat raw, like eat, will correspond to a binary relation between agents and the entities
that they eat in a raw state. This permits V′ Reanalysis and raising of the complex
predicate.10
By contrast, subject-oriented adjuncts simply form no constituent with V that
excludes the direct object. This flatly excludes predicate raising from deriving a surface
form in which the direct object appears right-peripheral to a subject-oriented adjunct.
2.1.3. The Distribution of meng-

The facts concerning the “transitivity marker” meng- are also directly predicted under
this account. As we have seen, Light Predicate Raising leaves a direct object in situ. It
follows then that in both (43a) and (43b) below ikan jang ditangkap loeh kakaknja
kemarni, ‘the fish that I caught yesterday’, is the surface object of, and adjacent to, a
transitive verb. The only difference is that in the former it is the object of the simplex
transitive memasak, ‘cook’, while in the latter it is the object of the complex transitive
memasak untuk Ali, ‘cook for Ali’:
(43) a. Yati me-masak ikan jang di-tangkap oleh kakak-nja kemarni untuk Ali.
Yati trans-cook fish comp pass-catch by sibling-her yesterday for Ali
‘Yati cooked the fish caught by her brother yesterday for Ali.’
b. Yati me-masak untuk Ali ikan jang di-tangkap oleh kakak-nja kemarni.
Yati trans-cook for Ali fish comp pass-catch by sibling-her yesterday
‘Yati cooked for Ali the fish caught by her brother yesterday.’
In these circumstances we expect transitive marking with meng- as usual, and this is
just what we observe.
2.1.4. XP Shift
Under the Light Predicate Raising analysis, the derivation of NP Shift examples makes
no crucial appeal to the category NP. This predicts that categories other than NP should
participate in the same basic phenomenon. Consider (44)–(46) below:
(44) a. Max talked [to Mary] [about Bill].
b. Max talked [about Bill] [to Mary].
c. Max talked [about Bill] [to all of the other witnesses].
(45) a. Jack made a promise [to leave] [to Mary].
b. Jack made a promise [to Mary] [to leave].
c. Jack made a promise [to Mary] [to leave by 5:00 p.m. sharp].
(46) a. Marcia behaved [rudely] [toward Lisa].
b. Marcia behaved [toward Lisa] [rudely].
c. Marcia behaved [toward Lisa] [more rudely than I would have expected].
In each of these sets there is an intuition of unmarked order for the complement arrange-
ment in the “a” examples. And although the judgments are subtle, the inverse order
seems to become most natural only when the outer complements are stressed or “heavy”
(cf. the “b” and “c” examples).
Predicate raising permits such data to be assimilated to the general NP Shift phe-
nomenon, even though they do not involve NP. Consider, for example, the variants
in (44). Assuming that the two PPs represent internal arguments of talk, and that the
underlying VP is as in (47a), the lower V′ phrase talk about Bill will have a θ-grid with
two undischarged θ-roles, and hence will be subject to reanalysis. As with put, we can
choose not to apply V′ Reanalysis, and so raise only V. This yields talk to all the other
witnesses about Bill. Alternatively, we can apply V′ Reanalysis and raise the entire com-
plex predicate to the empty V position (47b):
(47) a. VP
NP V′
Max V VP
e PP V′
to all the V PP
other witnesses
talk about Bill
b. VP
NP V′
Max V VP
talk about Bill PP V
to all the e
other witness
This yields (44c) above, with its appearance of “Heavy PP Shift.” The remaining exam-
ples are analogous.
This analysis appears compatible with familiar facts showing “domain asymmetry”
between the two PPs in to-about constructions. The following examples, parallel to ones
discussed by Barss and Lasnik (1986) for double object structures, imply that the NP
contained in the to-PP asymmetrically c-commands the one contained in the about-PP:
(48) a. I talked to the men about each other.

*to each other about the men. (anaphora)
b. I talked to no mani about hisi son.
*to hisi son about no mani. (quantifier binding)
c. Which mani did you talk to about hisi son?
*Which boyi did you talk to hisi father about? (weak Crossover)
d. Who did you talk to about which boy?
*Which boy did you talk to who about? (superiority)
e. I talked to each man about the other’s son.
*to the other’s son about each man. (each . . . other)

f. I talked to no one about anything.
*to anyone about nothing. (negative polarity)
This result is problematic under more traditional views of phrase structure, even assum-
ing that the domain NP may extend out of PP. Standard representations of “double PP”
examples either predict no asymmetries between NP1 and NP2 based on hierarchical
relations (49a) or else predict that NP1 should be in the domain of NP2, but not con-
versely (49b):
(49) a. VP b. VP
V PP PP V′ PP
talk P NP1 P NP2 V PP P NP2
to α about β talk P NP1 about β
to α
On the other hand, under (47a) these facts are straightforward. Given the presence of
V′, the to-PP asymmetrically c-commands the about-PP and hence the NP nodes can be
related in the appropriate way.11
The distribution in (48) contrasts with the comparative uniform unacceptability of
parallel examples in which the about-PP precedes the to-PP:
(50) a. *I talked about the men to each other.

about each other to the men.
b. ?*I talked about no mani to hisi son.
about hisi son to no mani.
c. *Which mani did you talk about to hisi son?
*Which boyi did you talk about hisi father to?
d. ?*Who did you talk about to which boy?
*Which boy did you talk about who to?
e. I talked ?*about each man to the other.
*about the other to each man.

f. I talked ??about nothing to anyone.
*about anything to no one.
Again these facts are straightforward under (47b). After raising of the complex V, nei-
ther PP c-commands the other. This correctly predicts ill-formedness of any dependency
relations between complements requiring c-command.12
2.1.5. “Freezing”
Wexler and Culicover (1980) point out that when NP appears in right-peripheral posi-
tion after NP Shift, the associated VP becomes “frozen” for extraction:13
(51) a. John gave to Bill the picture that was hanging on the wall.
b. *Who did John give to the picture that was hanging on the wall?
(cf. Who did John give the picture that was hanging on the wall to?)
c. *Bill would be easy for John to give to the picture that was hanging on the wall.
(cf. Bill would be easy for John to give the picture that was hanging on the
wall to.)
(52) a. They elected President of Mauritania the colonel who had engineered the recent coup.
b. *Which country did they elect President of the colonel who had engineered the
recent coup?
(cf. Which country did they elect the colonel who had engineered the recent coup
President of?)
This result follows directly under the present analysis. Consider the VP structure
underlying (51a) after Light Predicate Raising:
(53) VP
NP V′
John Vi VP
give to Bill NP Vi
the picture that was e

hanging on the wall
The string give to Bill has the following status: it is a lexical category V; however, it is not
a basic lexical verb. In the terminology of Di Sciullo and Williams (1987), give to Bill is
“syntactic atom” although it is not a “morphological object.” As a lexical category, give
to Bill is expected to have the usual properties of V; for example, it may assign Case.
On the other hand, as a syntactic atom, give to Bill is not analyzable by syntactic rules.
We therefore derive the islandhood property of NP Shift constructions: since the raised
predicate is a syntactic atom, it is “opaque” to move α; hence extraction as in (51) and
(52) is forbidden.14
3. PREDICATE RAISING IN PLEONASTIC CONSTRUCTIONS
A central prediction of the Light Predicate Raising analysis is that only transitive
constructions should undergo NP Shift. This is because only transitive predicates
undergo reanalysis and subsequent raising around an internal argument. Pleonastic
constructions involving there appear to challenge this prediction in an interesting
way. Note that in such examples NP Shift occurs with intransitive verbs like be,
arise, fly, and dance (54)–(56) and with apparent intransitive predicates like enter
the room (57):
(54) a. There was an odd assortment of little green men in the attic.
b. There was in the attic an odd assortment of little green men.
(55) a. There arose a fierce storm off the coast.
b. There arose off the coast the fiercest storm in living memory.
(56) a. ??There walked/danced/flew a man into the hall.
b. There walked/danced/flew into the hall the tallest man any of us had ever seen.
(57) a. *There entered a tall, dark stranger the room.
b. There entered the room a tall, dark stranger.
To motivate an approach to these examples, we consider the nature of expletive

there.
3.1. Pleonasm as “Transitivity Alternation”

Within the Government-Binding theory, expletive there has standardly been analyzed as
a “dummy element,” inserted into unselected, nonthematic positions in order to satisfy
principles of grammar. The pair in (58) below represents a typical case:
(58) a. A fierce storm arose.
b. There arose a fierce storm.
Following proposals by Perlmutter (1978) and Burzio (1986), arise and other verbs
permitting “there insertion” have been widely analyzed as “unaccusatives”; the sole
argument of arise is an underlying object, and its deep subject position is athematic:
(59) [IP e [VP arise a storm]]

Simple intransitive surface forms like (58a) arise from (59) by movement of the object
NP to subject position. The “there-insertion” variant (58b) results when the NP remains
VP-internal, leaving the IP subject position nonthematic. This triggers insertion of a
purely formal NP—the expletive there.
Chomsky (1981) suggests that insertion of the expletive is required to satisfy cer-
tain general principles governing the distribution of empty categories. In brief, given
that the sole argument of arise remains in situ in (58b), the subject position will be
unfilled by movement. The result is an empty category in subject position, which must
be licensed like other empty elements. The typology of empty categories available in a
non–pro-drop language like English forbids a governed subject empty category of this
kind. Structures like (59) are thus ungrammatical as they stand. Insertion of the phono-
logically overt expletive element provides a means for rescuing such examples, essentially
by removing the problematic empty category.15
This view of expletives is plausible and widely assumed. Interestingly, however, a
number of questions arise in connection with its central premise: the view that there
is an unselected formal element. First, if there is indeed unselected, then we expect no
lexical restrictions on the predicates with which it may co-occur. As is well known,
however, such restrictions do in fact exist. Expletive there appears only with a spe-
cific semantic class of intransitive predicates—those connoting “existence” or “coming
into existence” and “availability” or “coming into availability.”16 Thus, in pairs like
appear/disappear, arise/fall, enter/exit, and so on, which otherwise involve the same
underlying syntax, only the former admits expletive there:
(60) a. There appeared a tall, dark stranger.
b. *There disappeared a tall, dark, stranger.
(61) a. There arose a great civilization.
b. *There fell a great civilization.
(62) a. There entered a small, shy child.
b. *There exited a small, shy child.

Other languages are analogous to English in this respect; sentences involving verbs of
the existence or availability class often show a variant involving a special pleonastic ele-
ment not available in nonthematic positions of other predicates with similar underlying
syntax. This behavior thus presents a puzzle: if there is unselected, as on the standard
view, then why (and how) is its occurrence constrained to verbs of a particular semantic
class?
A related question arises with regard to Case assignment. Belletti (1988) and Lasnik
(1992) have observed data suggesting that the postverbal NP is assigned Case directly by
V in expletive examples. As (63)–(65) show, expletive constructions appear to require
adjacency of V and the postverbal NP just as transitive constructions require adjacency
of V and a direct object. Such behavior is a widely accepted diagnostic for a Case-
assignment relation between V and NP:17
(63) a. There often arose fierce storms.
b. *There arose often fierce storms.
(64) a. There will usually be a question of legality.
b. *There will be usually a question of legality.
(65) a. John often saw Mary.
b. *John saw often Mary.
Interestingly, however, if the usual view of expletives is assumed, this result

appears to run afoul of the familiar observation by Burzio (1986) that, with great
regularity, verbs assign Case to an object if and only if they assign a thematic role
to—that is, select—a subject. The standard view of there as unselected in exple-
tive examples forces us to admit an instance of Case assignment without subject
selection.
Finally, the standard analysis of expletives appears to conflict with proposals by
Chomsky (1986b) that natural language subscribes to a “Full Interpretation Principle,”
according to which every element present in the syntactic representation must be inter-
preted at Logical Form. This general proposal has considerable intuitive appeal and
significant empirical consequences (see Chomsky (1986b) for discussion). However,
note that if the usual analysis is correct, expletives constitute a clear anomaly for Full
Interpretation. If expletives are in fact syntactically overt, unselected elements, they
precisely represent an element not interpreted at LF, and hence must be accommodated
in some special way.
Given these simple points, consider an alternative view of pleonastic there. Con-
sider the idea that the expletive is in fact selected by the verbs with which it co-occurs
and that pairs like (58a,b) do not arise from the same D-Structure source. Specifically,
suppose that the relevant predicates actually admit two distinct underlying represen-
tations. One is a basic unaccusative frame, with no selection of a subject, no Case
assignment to an object, and movement to subject position in the familiar way (66a).
The second, however, is a pseudo-transitive frame, with selection of a special subject
element (the expletive there), Case assignment to the object position, and no movement
of NP (66b):
(66) a. IP b. IP
. . . VP . . . VP
XP V′ NP V′
V NP there V NP
arise a fierce storm arise a fierce storm
On this view, the appearance of expletive there would represent a form of “transitivity
alternation” open to the specific semantic class of “existence” and “availability” verbs.
As a matter of their lexical properties, such verbs would have the option of projecting
an extra argument—an expletive subject.
This proposal resolves the questions of selection and Case assignment directly. Lexi-
cal restrictions on the occurrence of there insertion follow from the lexically deter-
mined nature of the alternation. Furthermore, conformity with Burzio’s Generalization
is restored, since precisely in the frame where V assigns a Case to its object we now
assume it to project a subject.18 Finally, the general conceptual problem raised by there
vis-à-vis Full Interpretation is eased. If pleonastics like there are indeed selected ele-
ments, they cease to present an anomaly for the principle.
Given that there does not refer like a typical argument NP, and makes no apparent
semantic contribution, the transitivity involved here is evidently a purely formal one.
The relevant Vs are transitive strictly in the sense of licensing two NPs in A-position as
a matter of their lexical properties. Under this proposal, then, the occurrence of pleo-
nastic there with licensing, unaccusative verbs is rather analogous to the occurrence of
cognate objects with unergative verbs. The latter are verbs that may occur intransitively,
selecting an underlying subject but no object (67a), and also transitively, projecting a
special cognate object (67b):
(67) a. John died.
Felix sneezed.
b. John died a terrible death.
Felix sneezed a wall-rattling sneeze.
Here too the special argument projected in this alternation appears to be semantically
empty to a large extent, and the “transitivity” largely formal.19
3.2. Transitivity Alternation as Reanalysis

This analysis of expletives can be brought together with certain other features of the
present account. In Larson (1988) it is suggested that V-Raising structures of the kind
assumed here are licensed by the X-bar theory in (68) operating together with the prin-
ciples of argument realization in (69):
(68) a. XP → YP Xʹ
b. Xʹ → X ZP
(69) a. If α is an argument of β, then α must be realized within a projection of β.
b. Roles determined by a predicate α are projected according to the thematic
hierarchy

agent > theme > goal > oblique, such that if θ1 > θ2, then the argument to which θ1
is assigned c-commands that to which θ2 is assigned.
To briefly illustrate their effect, consider once again a VP headed by the verb put. Put
determines three thematic roles—agent, theme, and location—all of which must be pro-
jected in conformity with X-bar theory. Note, however, that the X-bar theory in (68)
permits at most two arguments to be realized within a single projection of V. This means
that a single X-bar projection of put can include at most the theme and oblique argu-
ments (70a). This structure leaves the agent role unprojected and no site to project it in.
To accommodate the remaining argument we must therefore make some elaboration of
(70a). (70b) is, in effect, the minimal structural elaboration of (70a) compatible with
the three principles given above:
(70) a. VP
NP V′
the boxes V PP
put in the car
b. VP
NP V′
Max V VP
e NP V′
the boxes V PP
put in the car
Here an X-bar “shell” with an empty head has been projected to accommodate the agent
phrase. In this structure all arguments of put are realized. The structure satisfies X-bar
theory (68). The prominence of roles on the thematic hierarchy is properly reflected in
the c-command relations of the arguments bearing those roles (69b). And all arguments
either do or, after V-Raising, will fall within a projection headed by V (69a). The final
surface form of this example derives as in (31) by movement of the verb and highest VP
specifier.
Now, to say that phrases are limited to at most a single specifier and a single comple-
ment, as in (68), is to say, in effect, that the basic configuration admitted under X-bar
theory is the transitive one, in which a predicate relates two terms. That is, it is to
establish a natural correspondence between the syntactic notion of being lexical head—
an item heading an X-bar projection—and the semantic/thematic notion of assigning
(exactly) two thematic roles:20
(71) α is an X0 α determines two thematic roles.
Of course, these categorial and thematic notions do not actually coincide. In fact, coin-
cidence fails in either of the two possible ways. There are thematically transitive expres-
sions that are not lexical categories. This is the case with phrases like put in his car,
which determine an internal and an external argument but are of category V′. Likewise,
there are expressions that are heads categorically but are thematically intransitive. This
is the case with unergative verbs like die or jump, which project an external argument
but no internal argument and it is the case with unaccusatives like arise and exist, which
project an internal argument but no external argument.
In Larson (1988) it is proposed that V′ Reanalysis and Cognate Object Formation
are a reflection in grammar of the correspondence in (71). Thus the former represents
the case where a phrasal transitive predicate is reconstrued as a categorial head (72a).
And the latter represents the case where a lexical, intransitive predicate is reconstrued
as a transitive by projecting a special internal argument (72b):
(72) a. Ditransitives: V′ Reanalysis
V′ V
V PP ===> V PP
put in his car put in his car

Thematic transitive Thematic transitive
Categorial nonhead Categorial head
b. Unergatives: Cognate Object Formation
VP VP
===>
NP V′ NP V′
John V (ZP) John V NP

sneeze sneeze a powerful sneeze
Thematic nontransitive Thematic transitive
Categorial head Categorial head
Notice now that the projection of pleonastic subjects for unaccusatives may be
viewed in the same light. We can take this to represent the case where a lexical, intransi-
tive predicate is reconstrued as a transitive by projecting a special external argument:
(73) Unaccusatives: Pleonastic Subject Projection
VP VP
===>
XP V NP V
V NP there V NP
arose a storm arose a storm
Thematic nontransitive Thematic transitive

Categorial head Categorial head
On this view, all three operations fall together as instances of the general scheme
Affect α applying in the service of the correspondence in (71). The result is in each case
a form of pseudo-transitivity—either categorial or thematic.
3.3. Predicate Raising and Pleonastic Projection

Let us return to examples (54)–(57) involving pleonastic there, examining unaccusative
and presentational uses in turn.
3.3.1. Unaccusatives
Examples (54) and (55) (repeated below) exhibit NP Shift with unaccusative verbs:
(54) a. There was an odd assortment of little green men in the attic.
b. There was in the attic an odd assortment of little green men.
(55) a. There arose a fierce storm off the coast.
b. There arose off the coast the fiercest storm in living memory.
Consider the derivations underlying the latter pair. As an unaccusative, arise selects a
single internal argument. Furthermore, as an inchoative predicate, a verb of “coming
into existence,” arise may project a pleonastic subject. Suppose now that reanalyzed,
pseudo-transitive arise occurs with an oblique locative phrase, as in (55). Recalling the
earlier discussion of put, the result will be a “filled-out” X-bar projection headed by
arise with the theme argument a fierce storm in specifier position and the locative off
the coast in complement position (74a).21 This structure leaves one argument of arise
unprojected—that corresponding to the external argument there. Hence, as in the case
of put, we project an X-bar shell to accommodate it (74b):
(74) a. VP
NP V′
a fierce V PP
storm
arise off the coast
b. VP
NP V′
there V VP
e NP V′
a fierce V PP
storm
arise off the coast
Now, by assumptions, arise has the status of a transitive verb. Furthermore, the modi-
fier off the coast discharges none of its argument roles. It follows that the V′ arise off
the coast has the thematic status of a transitive verb and is subject to V′ Reanalysis. If
reanalysis does not apply, then V alone raises, resulting in the VP underlying There arose
a fierce storm off the coast (75a). On the other hand, if V′ Reanalysis does apply, then
arise off the coast is reconstrued as V and raises as a unit. The result is the VP underlying
There arose off the coast a fierce storm (75b).
(75) a. VP
NP V′
there V VP
arise NP V′
a fierce storm V PP
e off the coast
b. VP
NP V′
there V VP
arise off the coast NP V
a fierce storm e
Example (54), involving be, is analyzed similarly. Pleonastic Subject Projection and V′
Reanalysis thus combine to yield the range of examples in a simple way.22
3.3.2. Presentational Constructions

Consider now cases like (56) and (57) (repeated below) involving “presentational there”:
(56) a. ??There walked/danced/flew a man into the hall.
b. There walked/danced/flew into the hall the tallest man any of us had ever seen.
(57) a. *There entered a tall, dark stranger the room.
b. There entered the room a tall, dark stranger.
These examples diverge from ones with unaccusatives in a number of ways. First, NP
Shift is strongly preferred with presentationals. Such constructions are in general poor
with nonheavy NPs, as pointed out by Safir (1985):
(76) a. *There walked into the room John.

b. ?There walked into the room the Princess of Cleves.
c. There walked into the room the man that everyone thought would one day rule the
world.
Second, the verbs in question appear semantically or thematically anomalous for Pleo-
nastic Subject Projection. Unergatives like walk, dance, and fly are not verbs of exis-
tence or availability. Furthermore, while the verb enter is unaccusative in other uses,
it is clearly transitive in (57) and so should neither require nor permit its valence to be
“boosted.” Finally, expletive presentational constructions have a generally “marked”
status. Modern English speakers find them stylistically archaic, and languages that
allow other pleonastic constructions with the equivalent of there often forbid presenta-
tionals altogether (e.g., French).
Under the present account we can shed light on the special properties of presenta-
tionals. Consider D-Structures for (56) and (57), which are parallel to (74a) above:
(77) a. VP
NP V′
the tallest man that V PP

any of us had ever seen
walk into the room
b. VP
NP V′
a tall, dark stranger V NP
enter the room
As noted, the individual verbs in these examples are inappropriate for Pleonastic Subject
Projection, being either of the wrong notional semantic class or of the wrong valence.
Observe, however, that the V′ predicates in (77) do have the right thematic/semantic
character. Although walk is a simple intransitive motion verb, walk into the room is
plausibly an intransitive predicate of “coming into availability.” Similarly, although
enter is a transitive predicate of “coming into availability,” enter the room is an intransi-
tive availability predicate. Indeed, walk into the room, fly into the room and dance into
the room are essentially synonymous with enter the room up to the manner of motion
specified.
Suppose now that although Pleonastic Subject Projection (like Cognate Object For-
mation) is an operation on lexical Vs, as a marked option its domain can be extended
to intransitive predicates of existence and availability in general. In the case of walk into
the room in (77a), this means that an expletive there can be projected (78):
(78) VP
NP V′
there V VP
e NP V′

walk into the room
The salient properties of this structure now follow directly. The marked status of the con-
struction derives simply from the marked status of the rule application used to produce
it. Likewise, the apparent semantic anomaly of an unergative motion verb projecting a
pleonastic falls away. We see that it is not walk that licenses there but rather walk into the
room. The latter is a semantically appropriate predicate. Finally, the strong preference for
NP Shift is clarified. Note that after Pleonastic Subject Projection, V′s like walk into the
room, enter the room, and so on. will have transitive thematic status. Thus, if the marked
aspect of presentationals lies in the application of a lexical rule to a phrasal predicate,
then V′ Reanalysis offers a natural means for “erasing” the problematic structure (79):
(79) VP
NP V′
there V VP
e NP V

walk into the room
After reanalysis, the predicate in question now is a lexical category, as originally

required, and undergoes raising in the familiar way. An analogous account can be given
for constructions involving enter.
3.3.3. A Constraint on Presentationals

This account of presentational constructions appears to illuminate certain cross-linguistic
data from French. Safir (1985) observes that while French permits pleonastic subjects
equivalent to there with unaccusatives (80a), presentational uses parallel to (56) and (57)
are not possible (cf. (80,b,c) respectively):
(80) a. Il est arrive trois hommes.
there is arrived three men
There arrived three men

b. *Il se dirigeait vers la gare une grande foule qui etait composee de mutiles de guerre.
‘There walked toward the railroad station a large crowd composed of disabled war
veterans.’
c. *Il marchait vers la gare trois types qui voulaient nous tuer.
‘There marched toward the station three guys who wanted to kill us.’
This result appears to correlate with another systematic difference between English and
French discussed by Green (1974), Talmy (1985), and Rapoport (1986). These authors
note that French departs quite generally from English in forbidding resultative construc-
tions equivalent to (81), in largely forbidding verb-particle constructions equivalent to
(82) (from Green 1974), and in the lack of ambiguity in locative PPs like that in (83)
(from Rapoport 1986):
(81) a. Max shot him dead.
b. Felix drank himself silly.
(82) a. He filed the serial number off.
b. He swept the petitions aside.
c. John ate the dessert up.
d. She told him off.
(83) a. The bottle floated under the bridge.
‘The bottle floated around under the bridge.’
‘The bottle floated to a point under the bridge.’
b. Le bateau flottait sous le pont.
‘The bottle floated around under the bridge.’
#‘The bottle floated to a point under the bridge.’
As they point out, in each case what appears to be at issue is the unavailability in French
of a process of semantic “regrouping” (Green 1974) or “conflation” (Talmy 1985) that
brings inchoative elements into the verbal complex without a change of morphological
form. Thus the absence of resultatives and verb-particle constructions reflects a con-
straint on incorporating CAUSE-BECOME into the semantics of means, manner, or
instrument verbs ((84)–(86), adapted from Rapoport (1986)):
(84) a. hammer1: Betsy hammered the nail.
b. hammer2: Betsy hammered the nail flat.
c. hammer2 =def x CAUSE (y BECOME AP by hammer1-ing)
(85) a. file1: Max filed the serial number.
b. file2: Max filed the serial number off.
c. file2 =def x CAUSE (y BECOME Adv by file1-ing)
Similarly, the absence in French of a path reading in (83) reflects a constraint on incor-
porating BECOME into the semantics of float:
(86) a. float1: The bottle floated under the bridge.
b. float2: The bottle floated under the bridge.
c. float2 =def x BECOME PP by float1-ing

These points can be brought together with the data in (80), I believe. Recall that on
our account presentationals involve treating certain V′s as lexical Vs. Phrases like walk
into the room, enter the room, and so on are analyzed as complex inchoative verbs and
project pleonastic subjects. It seems plausible that the extension of lexical rules to the
syntax is constrained by what the lexicon can independently provide. So if applying a
lexical rule in the syntax involves treating its target as a complex inchoative word, we
expect complex inchoative words to be available in the lexicon. This appears to be what
is behind the divergence between English and French in presentational constructions.
As evidenced by (81)–(83), English very freely permits the lexical formation of incho-
ative predicates. Accordingly, Pleonastic Subject Projection in the syntax involves no
processes not independently available in the lexicon. On the other hand, French permits
very little latitude in the formation of lexical inchoatives, as evidenced by the lack of
examples parallel to (81)–(83). We thus do not have the latitude in forming syntactic
inchoatives that is required for presentationals.23
3.4. Definiteness Restrictions

It has been widely noted in the literature that pleonastic constructions with there show a
certain constraint on NPs occurring in postverbal position (Milsark (1974); Kayne (1979);
Guéron (1980); Safir (1985); Burzio (1986). Only indefinite noun phrases may appear:
(87) a. There was/were a/two/many/*the/*ever/*most mirror(s) on the desk.
b. There developed a/*the fierce storm off the coast.
This restriction disappears, however, when the NPs in question are in right-peripheral
position:
(88) a. There was/were on the desk

a mirror that I had never seen before.
the mirror that you asked me about.

most mirrors that had been collected during that period.
b. There developed off the coast

a fierce storm.
the fiercest storm they had experienced.
The question arises as to how these facts are accommodated in the present analysis.
Why does predicate raising (also known as NP Shift) void the definiteness restriction
on postverbal NPs?
The answer that I suggest derives from work by Belletti (1988) on Case assignment
in unaccusative constructions. Following Burzio (1986), Belletti assumes that unaccusa-
tives, like passives, do not assign structural objective Case to their underlying object NP.
This normally forces the latter to undergo movement to IP specifier position (89a). In
pleonastic constructions, however, it is proposed that unaccusatives assign an inherent
partitive case (89b):
(89) a. [IP a storm will [VP arise t ]]
b. [IP there will [VP arise a storm ]]

Partitive Case
This allows the NP to remain in situ and satisfy its case requirements. It also accounts
for the definiteness restriction, on the assumption that partitive case is incompatible
with the semantics of definite NPs (see Belletti (1988) for discussion).
I will adopt the substance of Belletti’s account, modifying it slightly to include the
analysis of structural case assignment in Larson (1988). In the latter, the structural case
assigned by Vs to their objects actually originates with INFL and is transmitted under
government by I:
(90) [I′ I ... [V′ V NP ]...]

Objective Case
On this picture, failure to assign objective case amounts to blocking objective case
transmission from INFL. We may take this to be a lexically specified property of the
passive morpheme, and of the general class of intransitives, including unaccusatives,
unergatives, and other predicates that do not intrinsically select both an internal and
an external argument.
Taking these points together, then, the general situation is as follows: unaccusatives
block case transmission to their objects and assign no thematic role to their subjects;
this triggers NP movement as in (89a) in the usual case. After Pleonastic Subject Projec-
tion, however, unaccusatives acquire a derived transitive status. They assign a special
θ-role to a subject (there) and, in conformity with Burzio’s Generalization, a special
pseudo–objective case (Partitive) to their objects. Their status as non–case transmitters
continues to block the structural case from INFL; hence assignment of Partitive enforces
the definiteness restriction.
Consider now the situation presented by NP Shift examples like (88b):
(91) IP
NP I′
there I VP
NP V′
t V VP
develop off NP V
the coast
the fiercest storm e
they had experienced
Develop off the coast has undergone V′ Reanalysis and raised around the “heavy” theme
NP. (Pleonastic there has also moved to subject position.) Not being a “morphological
object,” the complex predicate is not listed in the lexicon. So, in particular, it is not (and in
fact cannot be) specified as blocking Objective Case transmission from I. Since Objective
Case is assigned to the heavy NP, Partitive Case need not be. As a result, the definiteness
restriction enforced by Partitive Case is lifted, accounting for the divergence between
(87) and (88).24
3.5. “Heavy NP Shift” from Subject Position?

A number of languages have been described in the literature as exhibiting NP Shift
from subject position. Examples are given in (92)–(96). In each case, the phonologically
heavy NP appearing sentence-finally is the notional subject of IP:
(92) a. tiosc nach riabh t ar fáil an comhoibriú ó na feirmeorí
because comp(neg) be(pst) available the cooperation from the farmers
a bhí riachtanach.

comp be(pst) necessary
‘Because the necessary cooperation from the farmers was not forthcoming’
b. Tháinig t amach sa chaint an leithséal fill
come(pst) out in-the talk the excuse treachery(gen)
a tugadh don athair.

comp was-given to-the father
‘The treacherous excuse that had been given to the father emerged in the talk’
(Irish; Chung and McCloskey 1987, 225)
(93) Pa∂ munu kaupa essa bók margir stúdentar.
will buy this book many students
‘Many students will buy this book.’
(Icelandic; Platzack 1987)
(94) a. Ont telephones aujord’hui, tous les gens a qui vous aviez donne rendez-vous.
‘Have called today all the people with whom you had an appointment.’
b. Je crois qu’ont defile dans ce bureau, les gens les plus etranges que j’ai jamais
recontres.
‘I believe that have come to this office the strangest people I have ever met.’
(French; Deprez 1987)
(95) a. lo yadanu ’ im omnan tesapek et ha-hanhala ha toxnit

neg we knew if in-fact will satisfy acc the management the plan
le-hakamat megurei-keva.
for-the-construction dwellings-permanent
‘We didn’t know if, in fact, the plan to construct permanent dwellings would satisfy
the management.’
b. elu ha-kartisim Se-otam natan le-iSt-o dod-i ha-Samen
these the-tickets that-them gave to-wife-his uncle-mine the-fat
mi-herzeliya.
from-Herzeliya
‘These are the tickets that my fat uncle from Herzeliya gave to his wife.’
(Hebrew; Shlonsky 1987, 180)
(96) a. Mattu gi petta adyu ya:hu na taotao.

infl-arrive loc door that like-agr l man
‘That man that I like appeared at the door.’
b. Man-istaba gi tenda adyu i tres na lalahi ni un-tungu.

infl-were loc store that the three l men comp infl-know
‘The three men who we know were in the store.’
(Chamorro; Chung, personal communication)
These data raise a question for our analysis similar to that posed by intransitive con-
structions. Under the Light Predicate Raising view, only heavy internal arguments—
direct objects—should appear to undergo rightward movement; thus, how are we to
analyze apparent heavy external arguments at the right periphery?
Our results with predicate raising in unaccusative constructions suggest a plausible
view. Consider the English paradigm in (97) below. (97a) is a copular sentence with a
“heavy NP” subject. (97b) is a variant of it with the pleonastic there. (97c) is a variant
of the latter with shift of the postverbal NP. Suppose now (contrary to fact) that English
permitted not only (97a–c) but also (97d), equivalent to (92)–(96) above:
(97) a. An odd assortment of little green men was in the attic.

b. There was an odd assortment of little green men in the attic.
c. There was in the attic an odd assortment of little green men.
d. Was in the attic an odd assortment of little green men.
One view of this sentence might be that it is related to (97a) by rightward movement
of the NP (98a). There is another possibility, however. Rather than arising by NP Shift,
(97d) might instead be a variant of (97c)—one in which the element corresponding to
there is simply null (∅) (98b):
(98) a. t was in the attic [an odd assortment of little green men].
b. was in the attic an odd assortment of little green men t.
On the latter view, the apparent NP-Shifted subject would in fact be no subject at
all, but rather a postverbal NP. And the analysis of the right-peripheral position in
(97d) would essentially reduce to the analysis of the right-peripheral position of NP
in (97c).
A detailed defense of the view that (92)–(96) are actually null pleonastic construc-
tions whose “shifted” subjects arise by predicate raising is clearly beyond the scope of
this paper. However, two general points are notable in this context. First, it seems sig-
nificant that languages cited as showing subject Heavy NP Shift all appear to be either
pro-drop languages (which permit zero pleonastics and other null subjects quite freely)
or else languages for which null pleonastics have been independently argued in the lit-
erature.25 Thus the “null pleonastic analysis” always seems to be available for subject
Heavy NP Shift examples in principle.
Second, examples of subject Heavy NP Shift often manifest properties of expletive
constructions. In a number of cases the verbs involved are clearly unaccusative (consider
(92) and (96)). In others, even when the verb is not unaccusative, the right-peripheral
NPs are reported to show definiteness effects like those observed in impersonal con-
structions (see, e.g., Platzack (1987) on Icelandic). The only examples that do not show
earmarks of unaccusativity are those from French and Hebrew. Interestingly, in these
cases the subject position has been independently argued to contain an expletive (by
Kayne (1979) and Shlonsky (1987), respectively).
These points do not, of course, constitute an argument for the proposed view. But
they do suggest that the null pleonastic analysis is not a priori implausible, and hence
the extent of the problem posed by subject Heavy NP Shift for our account is not yet
clear.26
4. PARASITIC GAPS
The account of NP Shift presented here has strong consequences for the analysis of
parasitic gap constructions like (99a–d):
(99) a. Which article did John file without reading?
b. This is the kind of food that you must cook before you eat.
c. That’s the movie star that John flatters constantly in order to impress.
d. She’s a person that anyone who meets likes.
Following proposals by Taraldsen (1981), Engdahl (1983), and Chomsky (1982), such
sentences have been widely analyzed as involving two empty categories–one derived by
A′-movement (t) and one produced by some other means (e):
(100) Which article did you file t without reading e?
The latter is taken to be “parasitic” on the former on the basis of pairs like the following:
(101) a. *John filed that article without reading.
b. Which article did John file t without reading my directive?
These examples appear to show that e is licensed only in the presence of a matrix syn-
tactic variable t, although the converse is not true.
Chomsky (1982, 40) characterizes the abstract conditions relating a parasitic gap (e)
to an independent gap (t) and its binder (XP) as follows:
(102) XP . . . t . . . e, where XP is in A′-position and t does not c-command e.
These capture the fact that parasitic gaps are not licensed by A-movement (103a,b), nor
by A′-movement from subject position (103c):
(103) a. *These articles were filed without my reading e.
b. *John seemed to be happy despite Mary criticizing e.
c. *Who sneezed without Bill hearing e?
A further apparent property of the construction is that the independent gap must be
present at S-Structure. The LF trace of a quantifier (every article) or an in situ wh-phrase
(which article) does not license e:
(104) a. *John filed every article without reading e.
b. *Who filed which article without reading e?
The relevance of this phenomenon for Light Predicate Raising is straightforward. As

pointed out originally by Engdahl (1983), Heavy NP Shift appears to license parasitic
gaps ((105b) comes from Engdahl (1983, 12)):
(105) a. John filed without reading all the books on the third shelf.
b. I offended by not recognizing immediately my favorite uncle from Cleveland.
c. You should cook before you eat any food left out overnight.
d. John flattered constantly in order to impress the famous movie star from Laguna.
On a standard analysis of NP Shift, these examples are subsumed directly under the
generalization in (102). The structure of (105a), for example, is equivalent in relevant
respects to (100):
(106) John filed t without reading e [all the books on the third shelf].
On the present analysis, however, the generalization in (102) cannot apply. Examples
like (105a–d) will involve neither A′-movement nor production of a syntactic variable,
but rather movement of complex predicates like file without reading and cook before
you eat:
(107) a. John [V filed without reading e] all of the books on the third shelf t.
b. You should [V cook before you eat e] any food left out overnight t.
In the resulting structure there simply is no independent gap and hence no t for e to be
parasitic upon.
Light Predicate Raising therefore commits us to the following claims: first, the
descriptive generalization embodied in (102), with its view of parasitic gaps as cru-
cially involving A′-movement, is incorrect. Appearances to the contrary, the presence
of a syntactic variable must play no essential role in the licensing of parasitic gap
constructions like (105a–d). Second, and more positively, the parasitic gap phenom-
enon is rooted in the nature of certain complex predicates. Its basic properties derive
from how expressions like file without reading or cook before you eat are formed and
understood.
4.1. Parasitic Adjuncts as Secondary Predicates

The nature of the complex predicates involved with parasitic gaps is suggested by fur-
ther reflection on (105c) (repeated below):
(105) c. You should [cook before you eat] any food left out overnight.
Assuming this form to derive by Light Predicate Raising, the underlying structure of VP
is as in (108), where V–PP forms a V′ and where this V′ is thematically transitive and
subject to reanalysis:
(108) VP
NP V′
You V VP
e NP V′
any food left V PP

out overnight
cook P CP
before you eat e
Now, since the verb cook is itself already transitive, the addition of before you eat e
must involve thematic composition that preserves the transitivity of the head. Intui-
tively, this composition must link the theme role of cook and the theme role of eat.
We understand (105c) as stating that you should cook food x that has been left out
overnight before you eat x; any food left out overnight is, in effect, the joint object of
cook and eat:
(109) [NP any food ... ] [V′ cook before you eat ]
Theme
Interestingly, this situation is not unfamiliar. The circumstances with cook before you
eat e appear closely analogous to those involved with the object-oriented secondary
predicate structures discussed earlier in section 2.1.2. Recall that the latter show the
same underlying [V′ V XP ] configuration as (108):
(110) a. VP b. VP
NP V′ NP V′
the meat V AP the wash V PP
eat raw bring in
Furthermore, the thematic structure of these V′s is quite similar. Eat raw and bring in are
composite transitive predicates (recall that they undergo V′ Reanalysis (42)). And just as
the theme roles provided by V and PP are understood as linked and assigned to the same
NP in (109), so the theme roles of the AP and PP adjuncts in (110) are understood as linked
and jointly predicated of the verbal object. If John eats the meat raw, the meat is both eaten
and raw, and if John brings the wash in, the wash is brought and comes to be in as a result:
(111) a. [NP the meat] [v′ eat raw ]

Th eme
b. [NP the wash] [v′ bring in ]

Theme
These points thus suggest that complex parasitic gap predicates are a form of second-
ary predication structure in which a surface modifying clause (before you eat, without
reading, etc.) has an underlying status equivalent to an object-oriented resultative or
depictive adjunct.
4.1.1. Role Identification

We may make this view more precise by appeal to certain technical proposals in
Higginbotham (1985), which sets out a simple theory of thematic roles and their manip-
ulation.27 Specifically, we can appeal to the notion of θ-role identification, in which the
roles assigned by two predicates are merged. (112a–c) illustrate three basic cases where
this notion applies; here “<. . .>” represents a set of roles determined by a predicate, and
linking is indicated by a solid line:
(112) a. V′ b. V
V AP V V
cook unsalted cook CONJ V
<1, 2> <1> <1, 2>
and eat
<1, 2>
c. N′
N CP
man SpecC′ C′
<1>
Oi I saw t i
(112a) is a secondary predicate structure with identification of roles assigned by a verb

and adjectival predicate. (112b) is a coordination with role- identification between two
verbs across the conjunction and. Finally, (112c) is a restrictive relative clause with iden-
tification of roles via an operator. The role borne by man is linked to Oi in the specifier
position of CP, which in turn binds a trace receiving the role assigned by see. The result,
in all three cases, is that the relevant θ-roles are jointly assigned. In John cooked the
meat unsalted, the meat is understood as both cooked and unsalted. In John cooked and
ate the meat, the meat is understood as both cooked and eaten. Finally, in John is a man
I saw, John is understood as both a man and an individual that I saw.
These three cases of role identification fall under the following general constraint,
slightly reformulated from Higginbotham (1985, 564):28
Identification
For role bearers α and β, a role of α may be identified with a role of β iff α governs β.
Here “role bearer” denotes either a predicate that assigns a role or an operator to whose
trace a role is assigned. In (112a), cook and unsalted are mutually governing role bear-
ers. In (112c), man governs Oi on the usual assumption that a lexical head (N) governs
the specifier of its structural complement (CP). Finally, in (112b), cook and eat mutually
govern assuming as a basic property of conjunctions (perhaps their core property) that
the categorial structure they introduce forms no barrier to government.
With these points in mind, consider now the predicate cook before you eat. We wish
to identify the theme roles assigned by cook and eat. However, these roles cannot be
directly linked as in (110a,b) since the bearers do not stand in the necessary government
relation:
(113) V′
V PP
cook P CP
before you eat e
In (113), cook governs the PP before you eat, but it does not govern the role assigner
eat, which is embedded within CP, IP, and VP. It follows that if identification is to occur,
it must appeal to the operator strategy employed with relative clauses.
Modifying proposals by Contreras (1984) and Chomsky (1986b), suppose that an
empty operator Oi is generated in the object position of eat. The latter may move to
the specifier position of CP, as in (114a). In this configuration Oi and the verb cook are
in close proximity; however, government is still blocked by the intervening preposi-
tion before.29 Suppose then that along with their status as subordinating prepositions,
before, without, by, and so on also have the capacity to function “insubordinately”
(to adopt both a term and a suggestion by Huybregts and van Riemsdijk (1985)) and
to behave as coordinators. Given earlier remarks, this amounts to saying that these
elements may shed their status as barriers to government. Identification thus becomes
possible much as in the restrictive relative case (114b):
(114) a. V′
V PP
cook P CP
before SpecC′ C′
Oi you eat ti
b. V′
V PP
cook P CP
<1, 2>
before SpecC′ C′
Oi you eat ti
The result is a V–PP complex with the semantics of a transitive verb. Cook-before-you-eat
is a predicate true of some pair of individuals <x,y> just in case x cooks y before you eat y,
and so on.
Under this approach, the licensing of parasitic gaps in adverbial adjunct construc-
tions is completely independent of matrix A′-movement, as noted above. Thus (99a)
(repeated as (115a)) receives the underlying VP in (115b):
(115) a. Which article did John file without reading?

b. VP
NP V′
John V VP
e NP V′
which V PP
article CP
file P
without SpecC′ C′
Oi PRO reading ti
Identification and reanalysis produce the complex V file without reading t. Predicate
raising and Wh-movement then apply to yield the observed surface form.
4.1.2. Obligatory V′ Reanalysis

Since this account claims a fundamental homology between parasitic gap and second-
ary predication structures, it must address certain apparent differences between the two
construction types.
First, whereas there are subject-oriented adjuncts (116a,b), there are no subject-ori-
ented parasitic gap structures (116c,d) (where (116c) is starred on the reading ‘without
anyone seeing Sally’):
(116) a. John left the party [angry].
b. Felix painted the wall [nude].
c *Sally left the party [without anyone seeing].
d. *Oscar arrived [before I saw].
Second, whereas the verb may raise away from a simple object-oriented adjunct (117a)
or may raise with it in NP Shift constructions (117b), only the latter is possible with
parasitic gap adjuncts (117c,d):
(117) a. John [painted] the barn t red.
b. John [painted red] the barn that he inherited t.
c. *John [offended] his uncle t by not recognizing.
d. John [offended by not recognizing] his favorite uncle from Cleveland t.
Third, whereas other adjuncts can be dislocated from their verbs by movement (118a),
parasitic gap adjuncts must stay in place (118b,c):30
(118) a. Raw is no way to eat fish e.
b. After visiting Bill, who did you hire?
c. *After visiting e, who did you hire?
(cf. Who did you hire after visiting e?)
Fourth, and finally, whereas other secondary predication structures are compatible with
Passive, parasitic gap constructions, as we have noted, are not:
(119) a. The meat was eaten raw.
b. The wash was brought in.
c. *The article was filed without John reading e.

As it turns out, these data can be drawn together in an interesting way. All four surface
differences can be jointly explained if we assume that, unlike other secondary predi-
cates, parasitic gap structures involve obligatory V′ Reanalysis.
To see this, consider first the lack of subject orientation in parasitic gap adjuncts. As
discussed insection 2.1.2, subject-oriented adjuncts adjoin to V′:
(120) VP
NP V′
Sally V′ PP
V NP P CP
left the party without Oi anyone seeing ti
The V′ leave the party has only one unassigned role, the agent, and hence is intransi-
tive. When this role is identified with that borne by Oi, the result is a V′ (leave the party
without anyone seeing) that is still intransitive. Since V′ has the wrong thematic status,
reanalysis cannot apply. But by assumptions V′ Reanalysis must apply. Hence (120) is
ruled out. This account of ill-formedness extends directly to earlier examples involving
raising and movement from subject position:
(103) b. *John seemed t to be happy despite Mary criticizing e.
c. *Who sneezed without Bill hearing e?
Like (116c,d), these sentences require subject-oriented parasitic adjunct structures (be)
happy despite Mary criticizing and sneeze without Bill hearing. Hence they are excluded
on the same grounds.
Consider next the observation that verbs cannot be raised away from a parasitic
adjunct (117c), nor can the adjunct be moved away from the verb (118c). Here again, if
V′ Reanalysis is obligatory, the facts can be explained. Suppose that in creating a lexical
category, V′ Reanalysis “freezes” the resulting complex predicate for extraction. Then,
if reanalysis must apply, neither of the two movements in question will be permitted.
Only raising of the entire unit will be allowed. This accounts very simply for data taken
elsewhere to show the “S-Structure character” of the parasitic gap phenomenon. Recall
that sentences involving quantifiers and in situ wh-phrases do not license parasitic gaps
(104a,b), a result that is somewhat problematic on the view that syntactic variables
license parasitic gaps and that quantifiers and in situ wh create such a variable at LF:
(104) a. *John filed every article without reading e.
b. *Who filed which article without reading e?
Under the proposal made here, the ill-formedness of (104) just assimilates to that of
(117c). The former, like the latter, involves V movement out of a lexical constituent:
(121) a. John filed every article [ V t without reading e].
b. Who filed which article [V t without reading e]?

Finally, consider the observation that parasitic gaps are not licit with Passive while
other secondary predicates are:
(122) a. *These articles were filed without my reading e.
b. *My uncle was offended by my not recognizing e.
This result may again be derived from our assumption when it is taken together with
one additional proposal about how passive applies to complex predicates. Observe the
following examples:
(123) a. John was punched and kicked by Bill.
b. *John was punched and fell by Bill.
(123a,b) are cases of V conjunction (as we infer from the single, right-peripheral by-
phrase). They show that although V conjunction of two passives is licit, conjunction of
a passive and an unaccusative (fall) is not. What rules (123b) out? One plausible candi-
date is the familiar A-over-A Principle of Chomsky (1965). Suppose applying passive to
[V V and V] involves applying it to the whole, and hence uniformly to both conjuncts.
Then passive in (123b) requires Case absorption and external θ-role suppression with
both punch and fall. But this is impossible since the latter assigns neither Case nor an
external θ-role. Hence the sentence is out.
If passive must apply to a whole complex predicate, then the unacceptability in
(122) can be viewed analogously to that of (123b). Since file without reading e must
be analyzed as V, filed without reading e must be analyzed, in effect, as [V file without
reading e]-en. As above, Passive will be unable to apply uniformly to this predicate
for although it may absorb Case and suppress the θ-role of file, it will be unable to do
so with without Oi reading. Hence ill-formedness results. This account allows for the
acceptable passive secondary predicate examples like the wash was brought in. Since the
verb and the adjunct need not form a complex V in the latter case, passive can apply to
the verb alone ([V′ bring-en in]).
These results raise the natural question as to why V′ Reanalysis should be obligatory
with parasitic gaps but not with other secondary predicates. A plausible suggestion is
that it follows from the government requirement on role identification. Recall that in
order for the empty operator to be identified with a role assigned by the transitive verb,
it is necessary for the prepositions before, without, by, and so on to behave “insubor-
dinately.” That is, in order to obtain the transparency necessary for government, these
connectives must lose their status as independent governors, that is, as full-fledged lexi-
cal heads. This implies that in parasitic gap constructions, the PPs headed by before,
without, and so on should lose their status as independent complements, and hence that
the constituent made up of V and PP should lose the full head-complement status of a
V′ projection.
Suppose, as a principle of grammar, that in order to count as a proper projection of
X, X′ must contain a complement headed by an independent governor. Then since V–PP
must lose its status as a proper projection of V as the price of obtaining transparency
of government, parasitic gap structures will always necessarily contain a “defective”
V′. This defect will be eliminable just in case V′ can be reconstrued as V, and hence just
when the parasitic structure is object-oriented and the transitivity requirement for V′
Reanalysis is met. On any other orientation for the parasitic gap structure, reanalysis
will be blocked, the defective V′ will not be eliminable, and hence the structure will be
ill-formed.31,32,33
4.1.3. Pronoun/Trace Alternation

Our results illuminate a number of important distributional contrasts between examples
containing a parasitic gap and those with an overt pronoun:
(124) a. Which article did you file without reading e ?

b. it
The former examples involve an operator that must be θ-identified with the matrix V,
and hence an adjunct that must undergo V′ Reanalysis. The latter involve no such opera-
tor, hence no such identification, hence no obligatory reanalysis.
This difference correctly predicts that all the effects attributable to obligatory V′
Reanalysis should vanish when an overt pronoun appears. Thus V may raise away from
its adjunct, and the adjunct may raise away from V when a pronoun is present (125a,b);
likewise, Passive is possible in the matrix clause (125c):
(125) a. John filed every article without reading *e/it.
b. Without reading *e/it, which article did you file?
c. Every article was filed without John’s reading *e/it.
There are also interesting consequences with regard to basic syntactic position. It has
been observed in the literature (Mohanon 1982) that certain adjuncts with PRO typi-
cally show strict subject control when they contain no other empty category. Thus in
(126a,b), for example, PRO must be construed with the subject John and not the object
Bill despite the greater pragmatic naturalness of the latter:
(126) a. Johni strangled Billj while PROi/*j unconscious.
b. Johni spotted Billj without PROi/*j seeing himj/*i.
This fact may be simply explained if we suppose that controlled adjuncts like those in
(126) must take the high attachment of subject-oriented adjuncts (127) and if we grant
the usual assumption that control of PRO requires c-command by the antecedent:
(127) VP
NP V′
John V′ PP
V NP while PRO unconscious
strangle Bill
In this structure, only the subject John c-commands PRO, and hence only the subject is
a possible controller.
Note now that given these assumptions, the position of adjuncts like without PRO
reading α or by PRO not recognizing α will differ according to whether α is a parasitic
gap versus a pronoun. If the first, then the adjunct will take the by now familiar low
attachment as sister to V (128a). If the second, however, then the adjunct must take a
high attachment as sister to V′ (128b):
(128) a. VP
NP V′
I V VP
e NP V′
my uncle V PP
offend Oi by PRO not recognizing ei
b. VP
NP V′
V′ V′
V NPi by PRO not recognizing himi
offend my uncle
This in turn predicts an important contrast between the two with regard to NP Shift.
Observe that while (128a) will permit V′ Reanalysis and Light Predicate Raising, (128b)
will not. It follows that whereas a parasitic gap adjunct can appear with an object to its
right, the equivalent with a pronoun will be ill-formed:34
(129) a. I offended by not recognizing ∅ immediately

b. *himi
[my favorite uncle from Cleveland]i

[no/every uncle from Cleveland]i
We see then that under this account, adjuncts with a parasitic gap and adjuncts
with a pronoun have a fundamentally different syntax. The former involve complex
predicate formation with its attendant syntactic effects, and a close relation to the
verb. The latter involve no such formation and are in essence simple V′ predicate
modifiers.
4.2. Two Consequences

This analysis suggests certain connections between parasitic gaps and other phe-
nomena. In this section I briefly consider two such cases: a special class of parasitic
gap structures with promise and a set of facts involving extraction from coordinate
structures.
4.2.1. Parasitic Gaps with Promise
English contains a class of object-oriented secondary predicate structures that differ in
an important respect from those discussed so far:
(130) a. Max laughed John out of the door.
b. Alice drank Max under the table.

c. The wolves howled Igor into a frenzy.
d. Oscar ate himself sick.
As in more familiar cases, the possibility of Light Predicate Raising shows that verb and
adjunct together form a complex transitive predicate:
(131) a. Alice drank under the table the barkeep and his entire staff.
b. The wolves howl into a frenzy anyone who has to listen to them for more than an hour.
Note, however, that the mode of thematic composition in this case is quite different
from examples like eat raw or bring in. The role contributed by the adjunct is not iden-
tified with any role borne by V, and hence the direct object bears no thematic relation
to the simplex verb. In (130a), John bears no thematic relation to laugh—John is not
“laughed” in any sense. Similarly for Max and drink in (130b), and so on. The adjuncts
make an “absolute” contribution to the verbal complex in such cases, increasing the
valence of a basically intransitive predicate.
Given the analogy drawn here between secondary predication and parasitic gaps,
we are led to ask whether anything comparable to (130a–d) occurs in the domain of
parasitic gap phenomena. Are there cases where a parasitic gap operator makes an
absolute contribution to a basically intransitive predicate, boosting it to transitiv-
ity? Consider constructions with infinitival complements to promise such as (132a).
As has been noted in the literature, (Stowell 1981; Larson 1991), the surface direct
object in such constructions typically resists NP Shift (132b). Interestingly, how-
ever, when the infinitival complement contains a parasitic gap the result is much
improved (132c):
(132) a. John promised a little child with red hair to take Mary to the movies.
b. *John promised to take Mary to the movies a little child with red hair.
c. John promised to take e to the movies a little child with red hair.
In Larson (1991) (following a suggestion by Bowers (1973)) it is argued that infinitival

control structures involving promise are underlyingly double object constructions, and
hence that (133a,b) are formally parallel:
(133) a. John promised [the child] [to take Mary to the movies].
b. John promised [the child] [a toy bear].

It is furthermore proposed that double object examples derive from underlying oblique
forms and involve structures in which the goal phrase (the child) has been moved to
direct object position and the theme (to take Mary to the movies, a toy bear) is realized
as a V′ adjunct:
(134) IP
. . . VP
NP V′
John V VP
e NPi V′
the child V′ XP
V NPi to take Mary to the movies
a toy bear
promise t
This structure accounts for the unavailability of NP Shift in (132b). Observe that in
order to derive the latter by Light Predicate Raising, the indicated V′ in (134) would
have to undergo reanalysis. But this is not possible since V′ lacks the correct thematic
structure. In [V′ promise ti [to take Mary to the movies]], for example, ti discharges
the goal argument of promise and the infinitive discharges the theme. This means that
[V′ promise ti [to take Mary to the movies]] has only one argument, the agent, left to
assign—that is, it is intransitive. Since V′ cannot undergo reanalysis, it cannot raise as
a unit to yield (132b).
With these points in mind, consider now the following structure for (132c):
(135) IP
. . . VP
NP V′
John V VP
e NPi V′
the child V′ XP
V NPi Oj PRO to take tj to the movies
promise t
(135) differs minimally from (134) in containing a preposed empty operator in place of
Mary. Suppose we allow this operator in Spec of XP to contribute “absolutely” to the
thematic structure of the intransitive V′. That is, suppose we allow Oi to contribute a
role to V′ without being identified with any role assigned by the latter. As with (130a–d),
this would, in effect, create a transitive predicate from a former intransitive. Intuitively,
[V′ promise ti [Oj to take tj to the movies]] would denote the relation holding between x
and y just in case x promises the child to take y to the movies.35,36
With its thematic status thus altered, we correctly predict that (135) will allow V′
Reanalysis and subsequent Light Predicate Raising around the derived direct object:
(136) John [promised to take e to the movies] a little child with red hair t .
We also provide for another class of facts. Stowell (1981) notes that infinitival construc-
tions with promise, like double object structures generally, resist wh-extraction of their
“inner object” (137a,b). This too improves when the infinitive contains a parasitic gap
(137c):
(137) a. ?*Which child did you promise to take Mary to the movies?
b. ?*Which child did you promise those pictures of Oilcan Boyd?
c. Which child did you promise to take e to the movies?
Note now that if we take (137c) to derive from a structure like (134), with which child
the object of the derived complex predicate promise to take to the movies, then the for-
mer need no longer be viewed as involving extraction of an inner object:
(138) You [promise to take to the movies]j which child tj.
After predicate raising, which child is the object of the complex transitive verb. There
is, in effect, no second object involved here.37
These remarks do not, of course, constitute a formal analysis of promise-type para-
sitic gap phenomena; this waits upon an account of precisely when empty operators or
adjuncts can contribute absolutely to an intransitive V or V′. Nonetheless, these remarks
do suggest further potential for the connection made here between secondary predica-
tion and parasitic gaps. It appears that properties of former may also be found reflected
in properties of the latter.
4.2.2. Complex Coordinations

Ross (1967/1986), and more recently Lakoff (1986) have drawn attention to some interest-
ing facts concerning extraction and coordinate structures (examples from Lakoff (1986)):
(139) a. What did Harry go to the store and buy e?
b. How much can you drink e and still stay sober?
c. That’s the stuff that guys in the Caucasus drink e and live to be a hundred.
d. That’s the kind of firecracker that I set off e and scared the neighbors.
(140) a. Sam is not the sort of guy that you can just sit there and listen to e.
b. Sam is not the sort of guy that you can just sit there, listen to e and stay calm.
c. Sam is not the sort of guy that you can just sit there, listen to e, stay calm and not
want to punch e in the nose.

As is well- known, examples like these appear to violate the Coordinate Structures Constraint,
which requires extraction to remove material equally from the conjuncts of a coordination:38
(141) VP
what VP and VP
go to the store buy t
And, as is also familiar, acceptability in such cases appears to hinge on whether the
sequence of actions given by the conjuncts forms a “natural course of events.” (142a,b),
in which the events described by the conjuncts are intuitively unconnected, are notice-
ably weaker than (139) and (140):
(142) a. *Who did you hit a home run and recognize e?
b. *What did Harry send a telegram to Felix and eat e?
Notice now that “coordinate structure violations” similar to those in (139)–(140) also
appear in the context of NP Shift ((143) from Lakoff (1986, 153)):
(143) Max went to the store, bought e, came home, wrapped up e and put under the
Christmas tree [one of the nicest little laser death-ray kits I’ve ever seen].
Here again, the notion of a “natural course of events” seems to be active:
(144) a. You can eat and not get cancer [any of the herbs on this shelf].
b. *You can eat and leave for home [any of the herbs on this shelf].
Accepting these as genuine instances of the NP Shift phenomenon, and recalling previ-
ous discussion, a view of this phenomenon emerges very different from that entertained
by Ross and Lakoff. We analyze the NP Shift examples, not as involving rightward
extraction from a coordinate structure (145a), but rather as raising of a complex predi-
cate around an object (145b):
(145) a. VP NP
VP and VP any of the herbs on this shelf
eat t not get cancer
b. VP
NP V′
you V VP
eat and not get cancer NP V
any of the herbs on this shelf t
And we are led to analyze Wh-movement examples as involving, not extraction from
a coordinate structure (141), but extraction from the object position of a complex
verb (146):
(146) VP
NP V′
Harry Vi VP
go to the store and buy NP Vi
what t
Under this view, the violation of the Coordinate Structures Constraint in (139)–(140)
is illusory.
This result raises the question as to how complex coordinate Vs like those in (139)–
(140) are formed and understood. I suggest that they be analyzed along much the same
lines as the parasitic gap constructions discussed above—that is, that they have essen-
tially the status of secondary predications.39 To illustrate, I propose that the complex V
in (143) has the structure of a V′ conjunction, where conjuncts exhibiting a trace contain
an adjoined empty operator:
(147) V′0
V′1 •
go to the toy store (and) V′2

V′3 •
Oi buy ti (and) V′4

V′5 •
come home (and) V′6

V′7 •
Oi wrap ti up (and) V′8
Oi put ti under
the X-mas tree
This sequence may be viewed as a string of nested secondary predications, where V′8 is
a secondary predicate of V′7 and the two compose to form the predicate V′6,. where the
latter is a secondary predicate of V′5 and the two compose to form the predicate V′4.
And so on. Assuming, as before, that conjunctions form no barrier to government, the
V′s in (147) will mutually govern and permit θ-identification. The result is a complex
transitive V′ whose object is understood as something that is bought, wrapped up, and
put under the Christmas tree, and whose subject is understood as someone who goes,
buys, comes home, and so on.
The view that (147) is a secondary predication structure casts light on two important
properties of the construction. First, it suggests why the usual “parallelism” requirement
on coordination is void in these cases. As is well known, it is in general only possible to
coordinate predicates that assign the same number of thematic roles (148):
(148) a. V b. * V
V and V V and V
kick bite kick fall
<1,2> <1,2> <1,2> <1>
In the case of (147), however, this requirement is not met. Come home, for example,
contributes one θ-role, while Oi wrap up ti and Oj put tj under the X-mas tree con-
tributes two. If (147) is not a genuine instance of coordination but rather a secondary
predicate construction, then this behavior is clarified. As we have already seen (e.g., in
(112a) and (130)), with secondary predication there is no parallelism requirement on
θ-identification. Hence its absence in (147) is not anomalous.
A second consequence of this view is that we derive some idea of why “natural
courses of events” are involved. Suppose the thematic structure is as we have claimed,
where the V′s in (147) constitute successive, nested subpredications of the matrix predi-
cation. It is natural to understand these as determining subevents of the matrix clause
event. As a matter of our cognitive makeup, subordination relations between events
are doubtless limited in certain ways; we are presumably compelled to understand such
relations in terms of familiar notions like cause, purpose, temporal precedence and
inclusion, and so on. As a consequence we expect well-formed secondary predications
to be organized according to certain natural relations between events: secondary predi-
cations must conform to “natural courses of events.”40
4.3. Some Remaining Problems

Certain examples of parasitic gaps pose difficulties for the analysis developed here. In
concluding this section, we consider two important cases.
4.3.1. Subject Parasitic Gaps

English contains a class of parasitic gap structures that does not appear to fall under
the general approach pursued here. These include an empty category within a subject
NP (149a–c). There is no obvious analysis here in which the verb and subject phrase
containing the parasitic gap compose to form a complex transitive verb taking the
wh-trace as its object. Similar remarks apply to parasitic gaps contained in a preposed
adverbial (149d):
(149) a. Chiquita is a donkey that [anyone who meets e] likes.
b. Who did [Mary’s talking to e] bother most?
c. Which man did [friends of e] visit?
d. Chiquita is a donkey that [whenever you meet e] you like.
Again, barring some elaborate restructuring, it is not evident how the preposed adverb
in (149d) can be composed together with like to form a complex transitive V taking the
trace of the relativized NP as its object.
One might attempt to dismiss subject parasitic gaps and cases like (149d) as phenom-
ena fundamentally different from that discussed above, and involving fundamentally
different licensing principles. And in fact there is some plausibility to this view. Shlonsky
(1986) argues that at least certain examples of subject parasitic gaps should be analyzed
similarly to donkey anaphora cases, with (149a,d), for instance, parallel to (150a,b),
respectively:
(150) a. Anyone who meets a donkey likes it.

b. Whenever you meet a donkey you like it.
Furthermore, under the theory in Chomsky (1986a), examples like (149a–c) appear to
contravene the normal subjacency relation required to hold between parasitic and inde-
pendent gaps (Browning 1987). Nonetheless, any attempt to eliminate (149a–d), from
the present discussion must contend with the fact that there are cases similar to them
but involving a right-peripheral NP:41
(151) a. Anyone who meets e admires without qualification [a person that can swallow
their own tongue].
b. Mary’s talking to e bothered most [her ex-boyfriend from college].
c. Whenever you meet e you always like [a person with two heads and a tail].
(151a–c) have the surface appearance of NP Shift and so would appear to require treat-
ment within the same context as examples like (105a–d).
One possible approach to the data in (149) and (151) has been suggested to me by
Jane Grimshaw (personal communication). Grimshaw observes that (151a–c) might be
accommodated within the present account if they as analyzed as involving, not Heavy
NP Shift /Light Predicate Raising, but rather Right Node Raising. The latter is illus-
trated in (152):
(152) a. John likes, but Max positively idolizes [a person that can swallow their own
tongue].
b. Whenever you can, you definitely should [spend winters in Ulan Bator].
Like NP Shift, Right Node Raising yields a right-peripheral constituent in surface form.
Unlike the former, however, it appears to involve true rightward, “across-the-board
movement” to an A′-position:
(153) John likes t but Max positively idolizes t
[a person that can swallow their own tongue]
Under Grimshaw’s suggestion, (151a–c) would be analyzed similarly; for example,

(151a) would receive the structure in (154):42
(154) Anyone who meets t admires without qualification t
[a person that can swallow their own tongue]
And, by extension, cases like (149a–d) would derive by initial Right Node Raising of
wh-, with subsequent movement to CP specifier position:
(155) [Which donkey] does [anyone who meets t like t ] t
Without attempting to defend this suggestion in detail here, we may note two facts
that make it plausible prima facie. First, observe that examples like (149) and (151)
display the characteristic intonational properties of Right Node Raising. In all of these
examples we find two distinct intonational phrases separated by a heavy break after
the first trace:
(156) a. Anyone who meets t admires without hesitation t . . .
b. John likes t but Max positively idolizes t . . .
Second, the right-peripheral position of NP in (151) appears to be higher than the

VP-adjunction site assumed under standard views of NP Shift. Observe that an NP
right-adjoined to VP will fail to c-command the parasitic gap in cases like (151c):
(157) IP
PP IP
whenever you meet e NP I′
you I VP
VP NPi
always like ti a person with two heads and a tail
Here e falls outside the domain of NPi in (157), which disallows binding of the former
by the latter, even under more extended theories of c-command involving containment
in maximal projections (e.g., May 1985). Hence NPi must be attached at least as high
as IP. On a Right Node Raising view, this result is straightforward. Given that (152a)
is a sentence conjunction, it is clear that Right Node Raising must involve adjunction
at least as high as IP.43
4.3.2. Oblique Predications?

All examples of parasitic gaps analyzed above have been ones involving predication of a
direct object. As it turns out, however, parasitic gaps also occur in examples where the
apparent target of predication is a PP object:
(158) a. Who did you send a bill to without contacting e?
b. Which oven did you put the bread in after heating e?
c. Which hay did you load the truck with after bailing e?
At first glance such examples might appear to raise no special questions. Given the
analogy between parasitic gap adjuncts and secondary predications, (158a–c) will fit in
smoothly, it seems, if they are analyzed as containing oblique-oriented secondary predi-
cates. On this view, the lower portion of the VP in (158a), for example, would have the
representation in (159):
(159)
V′
V VP
e NP V′
a bill V VP
e PP V′
to who V PP
send without Oj PRO contacting tj
Here Oj would presumably be θ-identified with the role assigned by send to the indirect
object who.
Nonetheless, this analysis encounters serious problems. First, if without Oj contact-
ing tj were properly analyzed as an oblique-oriented secondary predicate, then we would
clearly expect such predicates to be available in general. Perhaps surprisingly, however,
this is not so. As (160a–c) illustrate, secondary predication of an indirect or locative
object is not possible, even when the interpretation would be a natural one:
(160) a. Maxi sent a letter to Johnj drunki/*j
(cf. Maxi sent a letter to Johnj while hei/j was drunki/j)
b. You put the breadi in the ovenj coldi/*j
(cf. You put the breadi in the ovenj when iti/j was coldi/j)
c. Felix loaded the trucki with hayj greeni/*j
(cf. Felix loaded the trucki with hayj while iti/j was still greeni/j)
Structure (159) thus appeals to a thematic possibility not elsewhere available.

An equally serious problem for (159) concerns reanalysis. We saw earlier that many
simple properties of parasitic gaps could be explained if the adjuncts containing them
obligatorily reanalyze with the verb and form a complex transitive V. This constraint
was proposed to follow from a government requirement on thematic identification.
Notice now that if (159) were correct, then no reanalysis of the lowest V′ can have
occurred in (158a). We know this from the fact that the verb raises away from the
parasitic adjunct:
(161) . . . [V′ send a bill [V′ t to who [V′ t without . . . e . . .]]]
And we know it from the fact that the lowest V′ has the wrong thematic status for
V′ Reanalysis, possessing three unsaturated arguments and not two, as required. Our
earlier reasoning about the exclusion of subject-oriented parasitic gaps should thus also
exclude oblique-oriented parasitic gaps on the structure in (159).
These points seem to weigh heavily against analyzing (158a–c) as oblique-oriented
secondary predicates. Unfortunately, however, the alternative is not at all clear. At pres-
ent the only other possibility for such examples that I can see is the (rather baroque)
structure in (162):
(162)
V′
V VP
e NP V′
who V′ PP
Oi send a bill to ti without Oj PRO contacting tj
Here who is underlyingly the object of a V′ like that appealed to in (135) above, and the
parasitic gap adjunct is once again object-oriented. Presumably, θ-identification in this
structure would involve linking the roles borne by the two operators.
At present I do not know whether such a proposal can be made to work, nor what
principles govern the formation of V′s like those assumed. At any rate it is clear from
these remarks that apparently simple examples like (158a–c) pose an important chal-
lenge to the general account we have been pursuing.
5. THE ABSENCE OF HEAVY NP SHIFT
In this paper I have argued for the reanalysis of NP Shift phenomena as instances of
predicate raising. This result, if correct, poses an interesting general question within the
syntactic framework assumed here (essentially that of Chomsky 1981). Supposing that
by now familiar examples like (163a) are indeed to be analyzed by leftward X0 move-
ment, as in (163b), and not by rightward XP movement, as in (163c), we would like to
know why the second analysis is excluded:
(163) a. Max put in his car all the boxes of home furnishings.
b. [V α ] NP t
c. [ V t . . . ] NP
Since the rule Move α applies freely, the second derivation is available in principle.
Accordingly, its absence must reflect the intervention of some definite grammatical con-
straints. We would like to know what these constraints or principles are.
At present I can see two potential answers to this question. On the one hand, the
derivation in (163c) might be excluded as part of a quite general prohibition on right-
ward movement in English. Suppose, for example, that operator-variable constructions
Oi [. . . ti . . .] are subject to the requirement that the operator Oi govern the constituent
β, which is its scope. Given the rightward direction of government in English this would
directly forbid configurations of the form [. . . ti . . .] XPi, precluding not only Heavy NP
Shift but also extraposition and Right Node Raising. This view is compatible with pro-
posals that “extraposed” relative and comparative clauses are actually base-generated
in their right-peripheral surface position (see Larson 1983). It is also compatible with
the view mentioned in note 43 that so-called Right Node Raising does not actually
involve rightward movement but rather a form of discontinuous constituency.
A second, more theory-internal answer to the question of (163c) might appeal to the
way that NP Shift interacts with basic grammatical relations on the present account.
Under traditional views of phrase structure, subject and object can be distinguished by
government relations at D-Structure. Objects (and complements) are arguments gov-
erned by the verb in underlying form, while subjects are arguments that are not:
(164) IP
NP I′
Max I VP
V NP PP
put the boxes in his car
On the present approach, however, grammatical relations are not distinguished by gov-
ernment at D-Structure, but only at a later point. Objects (and complements) are argu-
ments governed by V after raising, while the subject is that argument which remains
ungoverned after raising has occurred:
(165) VP
NP V′
Max V VP
put NP V′
the boxes V PP
t in his car
Notice now that Heavy NP Shift, if it actually existed, would exercise an important
obscuring effect on the object relation within VP. Specifically, it would result in an NP
that is notionally an object and in the linear position of objects but that is not governed
by, and hence not an object of, V. If the boxes in (165) is adjoined to the highest VP,
then it will fall outside the c-command domain of put and fail to be governed by it.
Likewise, if the boxes is adjoined to the lower VP, then government will be blocked
under the assumption that V can govern the head and specifier of an XP sister, but no
other constituents (see Chomsky 1986a).
It is plausible to think that this situation might be excluded under some strengthened
version of the Projection Principle of Chomsky (1981), which demands that selectional
properties be respected at every level of representation. Suppose, in particular, that
within VP we require α to be governed whenever the trace of any chain it participates in
is governed. This would ensure, in effect, that grammatical relations remain “univocal”
within the domain in which they are projected (VP). Movement between complement
positions would be possible, as, for example, in the account of double object con-
structions in Larson (1988). Likewise, movement of an underlying object to IP subject
position would be licensed, as in the standard analysis of passives and unaccusatives.
However, classical Heavy NP Shift would be forbidden since this involves a loss of
object status within VP.
I will not attempt to develop these proposals any further at this point. Nonethe-
less, these brief remarks do suggest that potential lines of explanation are open for the
absence of Heavy NP Shift.
NOTES
Preparation of this manuscript was supported by the MIT Center for Cognitive Science under a
*
grant from the A. P. Sloan Foundation’s particular program in Cognitive Science. I am grateful to
Elisabet Engdahl for discussion of the Swedish examples.
1. Ross (1967/1986) does not specify a particular derived structure for Heavy NP Shift, while
Bresnan (1976) and Stowell (1981) give the adjunction site as VP. Evidence is discussed in
4.3.1 showing that if NP Shift examples do indeed involve noun phrase movement, then IP
must also be admitted as a potential adjunction site.
2. More precisely, Ross’s condition (1967/1986, p. 139) proscribes rightward movement out
of [NP P NP]. Such exocentric structures have now been abandoned in favor of endocentric
constructions involving PP.
3. See Rothstein (1983), Simpson (1984), Williams (1980) and Schein (1995) among others,
for discussion.
4. Infinitival purpose adjuncts appear to pattern similarly to depictive adjuncts vis-à-vis ori-
entation and NP Shift. Thus, in (i) in order to please her mother is predicated of the subject
NP Max (we understand Max as the “pleaser”). NP Shift “around” this adjunct appears
quite marginal. In (ii) the transitive purpose clause to use in bed is predicated of the object
the lamp. Correspondingly, NP Shift around the direct object is much more acceptable:
(i) a. Max praised Mary in order to please her mother.
b. *Max praised in order to please her mother a girl who he knew from college.
(ii) a. Max bought the lamp to use in bed.
b. Max bought to use in bed a lamp with special black light attachments.
5. In Fillmore (1965), the correct surface form is derived by a rule of “Separation” that shifts
the PP rightward:
(i) VP
V′ NP
the box
PP “Separation”
V
put in his car

In the nontransformational, categorial analyses of Bach (1979) and Dowty (1978), the
peripheral position of PP arises by an operation of “Right Wrap,” which breaks up the V–
PP sequence by wrapping the verb around the object argument:
(ii) put the box in his car
Right Wrap
the box put in his car
.
. put in his car
.
.
Right Wrap is formally analogous to verb raising in analyses involving phrase structure, a
parallelism first noted and developed in Jacobson (1983, 1987).
6. Specifically, put must head a projection governed by INFL to receive tense and agreement
information. Furthermore, the object the box must be governed (and hence c-commanded)
by V in order to receive Case. The verb raises to meet these joint requirements. Likewise, the
VP specifier Max must bear Case and raises to receive it from INFL. The proposal that verbs
raise to permit Case assignment is defended by Koopman (1984), Travis (1984) and Sproat
(1985), among others. The idea that verbs raise to obtain inflection from INFL is developed
in Roberts (1985), where it is also attributed to N. Fabb.
7. This view of transitivity and V′ Reanalysis departs from Larson (1988), where α is transitive
(and undergoes reanalysis) when it selects (exactly) one internal argument —that is, when it
takes a direct object. Here α is transitive (and undergoes reanalysis) when it selects (exactly)
two arguments—that is, when it corresponds to a binary relation. These represent two views
of transitivity that have been widely debated in the literature. The former might be termed
the “grammatical” notion of transitivity, and the latter the “logical” notion of transitivity.
For a useful discussion of various approaches to transitivity see Siewierska (1984).
8. As is also discussed by Jacobson (1983, 1987), V-Raising allows for a simple treatment of par-
ticle movement. Suppose that verb-particle amalgams are entered in the lexicon as basic V′s:
(i) a. [V′ look [PP up]]
b. [V′ throw [PP out]]
c. [V′ smash [PP in]]
These structures will have the thematic status of transitive verbs, and hence will be subject
to V′ Reanalysis. If reanalysis does not apply, then V raises and the verb and particle appear
separated in surface form (iia). If reanalysis does apply, then the verb and particle raise as a
unit and the two appear adjacent in surface form (iib):
(ii) a. smash [VP the windshield [V′ t [PP in]]]
b. smash in [VP the windshield [V t]]
9. The behavior of negative polarity items provides simple evidence for the different attach-
ment of subject- versus. object-oriented secondary predicates proposed in (41). Consider
(ia); this example is ambiguous, having either of the readings: ‘I left no one with I myself in
the state of being angry at someone else,’ or ‘I left no one with that person in the state of
being angry at someone else.’ Compare now (ib):
(i) a. I left no one [angry at someone else].
b. I left no one [angry at anyone else].
Unlike (ia), (ib) is unambiguous, having only the second reading. This result is straightfor-
ward under (41). Negative polarity items like anyone must be c-commanded by an “appro-
priately negative” element at S-Structure in order to be licensed. This will be possible if
angry at anyone else is predicated of the object no one, since it will then be c-commanded
by the negative quantifier according to (41a). On the other hand, licensing of the negative
polarity item will not be possible if angry at anyone else is predicated of the subject I, since
then, according to (41b), it will not be c-commanded by no one.
10. See section 4.1.1 for more on role identification.
11. I assume a definition of c-command involving first branching nodes, essentially following
Reinhart (1979).
12. There are a number of remaining puzzles about the multiple complement data in general, and
about to . . . about constructions in particular. First, it is clear that the heaviness effect is much
stronger for NP than for PP, AdvP, or IP. Although the “c” examples are more natural than
the “b” examples in (44)–(46), the difference is not a strong one. This suggests that “heavi-
ness” may well have a grammatical component and may not be due solely to functional fac-
tors (e.g., greater ease of processing with right-peripheral components) as is usually asserted.
The fact that NP is singled out with respect to heaviness implies that Case may be involved.
Second, as Edwin Williams (personal communication) has pointed out, the degree of
deviance in the first member of each pair in (50) is noticeably less than in the second mem-
ber, with many people finding them awkward but not flatly unacceptable—particularly
when the final PP is stressed (compare (50f) to I talked about nothing TO ANYONE). I am
unable to explain this gradation of judgments at present.
13. I am grateful to Daniel Finer for drawing my attention to these facts.
14. Wexler and Culicover (1980) claim a second instance of “freezing” in examples involving
NP Shift. They state that the NP object too becomes an island for extraction:
(i) a. John mailed to Alice an expensive book about frogs.
b. ?What did John mail to Alice an expensive book about?
(cf. What did John mail an expensive book about to Alice?)
(ii) a. Felix wrote for Oscar a short article on tennis.
b. ?What did Felix write for Oscar a short article on?
(cf. What did Felix write a short article on for Oscar?)
In general, this second form of islandhood appears much weaker than that cited in the text,
with examples like (ib) and (iib) being merely awkward for many speakers. Furthermore,
the degree of deviance in the interrogative examples appears to track the degree of deviance
in the simple NP Shift counterparts. Thus (ia), with raised verb + dative, is somewhat more
natural than (iia), with raised verb + benefactive adjunct. Correspondingly, (ib) is rather
more natural than (iib). I will set aside consideration of examples like (i) and (ii) pending
clarification of whether a genuine islandhood phenomenon is actually involved here. (I am
grateful to N. Chomsky for discussion of these data.)
15. For discussion of this account of pleonastics, see Chomsky (1981, 1986b), Burzio (1986),
and Safir (1985).
16. This semantic restriction has been noted by Kimball (1973), Milsark (1974) and Szabolsci
(1986), among others.
17. Lasnik observes that V and the postverbal NP can be nonadjacent when (and only when)
the verb is be:
(i) a. There is often a problem with the water heater.
b. There are usually several hundred people at these meetings.
He suggests that in such examples the Case relation between V and the NP has been obscured
by the capacity of be (and only this verb) to raise to INFL in English. On this view the NP
is in fact adjacent to the verbal trace:
(ii) There [I′ is often [VP t a problem with the water heater]].
Lasnik (1992) notes independent support for this suggestion in the fact that when INFL is
already occupied by a modal, blocking the raising of be, adjacency between V and the NP
is again required:
(iii) *There will be often a problem with the water heater
See Shlonsky (1987) for useful discussion of these and other facts about Case assignment in
pleonastic constructions.
18. For more on the nature of this Case marking, see section 3.4.
19. See Jones (1988) for discussion of cognate objects within the general framework assumed here.
20. A natural question arises as to whether categories such as A, N, and P are also sensitive
to this correspondence—whether “transitivity effects” are observable outside the verbal
system. As pointed out to me by R. Kayne (personal communication), the equivalent of NP
Shift also appears to occur inside nominalizations:
(i) a. The destruction of Boston with rockets
b. ?*The destruction with rockets of Boston
c. The destruction with rockets of every major city on the eastern seaboard
Furthermore, although I cannot defend the claim here, I believe that certain complement
reorderings within comparative phrases are analyzable by these means.
21. Recall that the projection of oblique phrases follows the hierarchy in (69b); hence off the
coast, in the attic, and so on are realized in a position subordinate to the theme.
22. Ellen Broselow (personal communication) has pointed out to me that this analysis can
account for certain contraction data arising with NP Shift in pleonastic constructions.
Consider (i) and (ii):
(i) a. i. There is a green alien with violet eyes in the garden.
ii. There’s a green alien with violet eyes in the garden.
b. i. There is in the garden a green alien with violet eyes.
ii. *There’s in the garden a green alien with violet eyes.
(ii) a. i. There are rats with sunglasses on the table.
ii. There’re rats with sunglasses on the table.
b. i. There are on the table rats with sunglasses.
ii. *There’re on the table rats with sunglasses.
Under more standard views, contraction is blocked in the “bii” examples by a syntactic
variable, which intervenes between there and be. On the present account, this result can
be derived in either of two plausible ways. We could appeal to the fact that since be in
the garden and be on the table are lexical Vs, contraction would involve cliticizing a sub-
part of a lexical category, violating lexical integrity. Alternatively, we may observe that in
Light Predicate Raising constructions, the raised complex V invariably constitutes a distinct
intonational domain bearing stress. The latter might then be taken to block the reduction
necessary for contraction.
23. The general proposal that “syntactic word formation” is limited by what the lexicon can
independently provide was, of course, an observation frequently made within the frame-
work of Generative Semantics.
24. It is worth observing that, under these proposals, the definiteness restriction may be viewed
as a general property associated with derived transitivity and as having no special con-
nection with pleonastic constructions or those involving unaccusatives. In brief, when an
inherently intransitive predicate has its valence “boosted” to transitivity, it must determine
a Case for its object. Partitive might be then viewed as a general default Case, which carries
the semantic restriction of indefiniteness.
This idea leads us to expect definiteness effects in other pseudo-transitive constructions,
an expectation that appears correct. Note that the definiteness restriction also appears to
apply in cognate object constructions:
(i) John sneezed a mighty sneeze

some little trifling sneeze
many loud sneezes
*that sneeze
*every sneeze
*most sneezes
25. The former is true, for example, of Chamorro (Chung 1983, 1987), Hebrew (Borer 1980),
Irish (McCloskey and Hale 1984), Italian (Rizzi (1982); Burzio (1986), and Spanish (Torrego
1984). The latter is true, for example, of French (as discussed by Kayne and Pollock (1978),
Pollock (1986), Adams (1987), and Deprez (1987)) and Icelandic (as discussed by Platzack
(1987)). This second group is particularly interesting since it is precisely in the context of
“subject inversions” that null pleonastic subjects have been proposed.
26. Choe (1987) has argued that VSO surface order in Berber is underlyingly pro-V-S-O, where
pro is an empty expletive element and S is a postverbal NP. Choe suggests, furthermore, that
this analysis may apply quite generally to VSO languages—that, in the strictest sense, there
are no VSO languages at either the level of D- or S-Structure. If such a view can be sustained,
then it is plausible that the largest class of examples of subject Heavy NP Shift—those
involving VSO languages—will be analyzable along the lines suggested here. Such examples
will all involve “shift” of a postverbal NP and not a true subject.
27. The thematic operations proposed in Higginbotham (1985) are quite similar in effect to
ones proposed within the framework of Montague Grammar in the mid- to late 1970s. In
particular, θ-identification is similar to operations on functional structure involving abstrac-
tion (Dowty 1979). See also Li (1988) for recent work in a similar spirit.
28. I assume a “minimality” definition of government along the lines of Chomsky (1986a)
wherein α governs τ in the configuration [. . . τ . . . α . . . τ . . .] iff α is a lexical category,
α c-commands τ, and no closer governor δ interevenes (i.e., for δ a lexical category, if δ
c-commands τ). In configuration (i) this will allow α to govern YP, its specifier τ, and its head
μ, but no constituents of ZP owing to the presence of the closer governor μ:
X′
X YP
α SpecY′ Y′
τ Y ZP
µ ...
29. In Larson (1990) it is proposed that temporal PPs involving a clausal complement differ in
the internal syntax of the latter. It is suggested that before, after, since, and until take a CP
whose specifier contains a zero temporal operator (ia), whereas while takes a CP whose
specifier is empty (ib):
(i) a. [PP before [CP Oi [C′ . . .]]]
b. [PP while [CP . . .]]
This difference of structure accounts for an important difference of meaning between the
two. Whereas the PP in (iia) shows an ambiguity comparable to that in the when-clause of
(iic), the PP in (iib) does not:
(ii) a. John arrived before Mary said that she left.
b. John arrived while Mary said that she slept.
c. John wondered when Mary said that she left.
In (iic) it is possible to construe when with the time of Mary’s saying or the time of Mary’s leav-
ing. And similarly for before in (iia). This reading is not apparently available for (iib), however,
yielding only a pragmatically odd “while she spoke words” interpretation for the PP.
In Larson (1990) the ambiguity in (iia) is accounted for by assuming an empty operator
in (iia) parallel to when in (iic) that may originate in either of the two clauses enbedded
under before:
(iii) a. . . . before [CP Oi [Mary said [that she left] ti]]
b. . . . before [CP Oi [Mary said [that she left] ti]]
If this analysis is correct, then we might expect interference between empty operators when
before-type PPs occur in parasitic gap structures. In particular, we expect that when the
upper CP specifier position is occupied by the gap operator, no empty temporal operator
will be possible, and hence no “long-distance” temporal readings allowed.
Although the judgments are subtle, this expectation appears to be justified. Whereas in
(iv), before can be understood as associated with either the upstairs or the downstairs clause,
in (v) only the former seems possible; in my judgment, (v) must be understood as stating that
John saw Bill before the time at which Mary uttered certain words:
(iv) John saw Bill before Mary said that she visited him.
after
(v) Who did John see before Mary said that she visited e.
after
Presumably, when before’s temporal operator is blocked from appearing in CP specifier posi-
tion, this P must resort to the “operatorless” strategy of while in combining with its comple-
ment. The latter allows only the “upstairs” readings (see Larson (1990) for discussion). These
results must, of course, be regarded as tentative given the complextity of the examples and the
generally more marginal character of parasitic gaps contained in tensed clauses.
30. (118c) is taken from Lasnik and Uriagareka (1988, 75), where the observation that adjuncts
containing parasitic gaps are not frontable is attributed to E. McNulty.
31. Note that because no closer governor intervenes between N and Oi in relative clauses, no
“insubordination” and no corresponding “defect” in the N′ projection will arise.
Relative clauses and parasitic adjunct constructions differ importantly in that while the
former allow both empty and overt operators, the latter require the operator to be null:
(i) a. [ The man Oi I saw] came in.
whoi
b. An article I filed [ without Oi reading ei].
*whichi
The source of this difference is unclear; however, it is interesting to note that θ-identification
in general seems to occur only between elements sharing category features. Thus in the case
of the relative clauses we get identification between two [+N]’s (man and which), and in the
case of simple secondary predication we get identification between [+V]’s (eat and raw), or
[−N]’s (bring and in). What we do not get is identification between [+V,−N] and [−V,+N] (file
and which). It is possible that cases like the latter are confined to θ-assignment, where the
former is compelled to take the latter as an argument. If this were true, then θ-identification
in constructions like (ib) would be possible only via an empty operator, which is devoid of
categorial feature content.
32. This view of parasitic gaps interacts with earlier proposals about pleonastic there to cor-
rectly predict examples like (i):
(i) There arrived without anyone seeing e a sinister agent from Galaxy Five.
The latter will have the VP structure in (ii), where without anyone seeing e is predicated
of the postverbal NP and reanalyzes with V:
(ii) VP
NP V′
there V VP
e NP V′
a sinister agent . . . V PP
arrived P CP
without Oi anyone seeing ei
33. This analysis of parasitic gap formation shares an important feature with the account of
Kayne (1984). In the latter, the licensing of parasitic gaps involves the interaction of certain
chains or “paths” generated by a syntactic variable and a parasitic gap. A crucial aspect of
Kayne’s analysis is that the paths generated by these elements form a “connected” subtree.
On the present account, with its obligatory V′ Reanalysis, a similar (although stronger) con-
nectedness requirement is imposed; namely, the verb and adjunct clause must form a single
lexical unit.
34. Engdahl (1983) records pairs like (129a,b) as of roughly equal acceptability. I am at a loss to
explain these judgments since in my own speech (and that of others whom I have consulted)
the contrast is quite sharp. The grammaticality difference between (129a,b) has been noted
independently by Lasnik (personal communication).
35. (135) differs from structures considered thus far in that the parasitic gap occurs within a
selected complement and not an adjunct. This difference is not, however, tied to the possibility
of O making an “absolute” semantic contribution to V′. Consider (ia) with the VP structure
in (ib):
(i) a. Who did you persuade that Martha should visit e.
b. VP
NP V′
you
V VP
e
NP V′
who V CP
persuade Oi that Martha should visit ti

<1, 2, 3*>
As in (135), we have a gap-containing complement clause that receives a thematic role from
V (here the “3” role, whose assignment, or “discharge,” is marked by a*). Now note that
in this case Oi does not contribute “absolutely” to an intransitive V′—a V′ with only one
unassigned role. Instead, it identifies with the theme role of a transitive—a V with two unas-
signed roles. (ia) is thus parallel to the standard cases of secondary predicates or parasitic
adjuncts in which the operator contributes through identification.
36. More accurately, on a formal semantic treatment [V′ promise ti [Oj to take tj to the movies]]
would denote the relation holding of <x,y> on some assignment g just in case x promises
g(z) to take y to the movies, where g(z) is identical to the denotation (den) of NP. In this case
g(z) = den(the child).
37. Multiple questions appear to provide evidence for this derivation:
(i) a. *Who promised which child to take e to the movies?
b. Who promised to take e to the movies which child?
The obligatory right-peripheral position of the in situ wh-phrase which child in (ib) suggests
that extraction is from the object position of the complex predicate, and not from the “inner
object” position.
38. Ross (1967/1986) formulates the Coordinate Structures Constraint as follows (p. 98): “In
a coordinate structure, no conjunct may be moved, nor may any element in a conjunct be
moved.”
39. Lakoff (1986) records a suggestion by Pauline Jacobson that apparent violations of the
Coordinate Structures Constraint might be viewed as covert parasitic gap constructions.
The proposal developed here is in the spirit of Jacobson’s suggestion, although instead of
treating the first as an instance of the second, we are analyzing both as cases of secondary
predication.
40. See Lakoff (1986) for a variety of interesting observations on the notion “natural courses of
events” at work in these examples.
41. There is a certain amount of variation in speaker judgments with regard to (151a–c), with
some finding them quite awkward. I adopt a “worst-case” analysis of the data here, assum-
ing them to be essentially well-formed.
42. Right Node Raising generally requires a right-peripheral position for the trace in each con-
junct from which extraction occurs. This requirement is met in (154) under the assumption
that the adjunct without qualification has been reanalyzed and raised with V analogously
to what occurs in (i):
(i) I [admire without qualification] a person that can swallow their own tongue t.
The structure proposed in (153) appears compatible with the claim in Cowper (1985) that
subject parasitic gap structures behave as if there is an intervening operator or position that
is bound by the wh-phrase and that in turn binds the wh-trace and parasitic gap. In (153),
we can identify this position as that of the Right Node Raised wh- that is subsequently
extracted.
43. A number of authors have argued plausibly that Right Node Raising is not actually an
extraction rule at all but rather involves a form of discontinuous consituency in which the
“raised” expression is a simultaneous daughter of two mother nodes. McCawley (1982), for
example, analyzes Right Node Raising as involving trees with “crossing branches”; Ertes-
chik-Shir (1987) develops a related view using the analysis of “across-the-board” extraction
in Williams (1978) (see also Levine 1985, McCloskey 1986). These proposals, if correct, do
not prejudice the basic point made here, for they still permit examples like (151a–c) to be
analyzed as Right Node Raising, and examples like (149a–d) to be analyzed as extraction
of a Right Node “Raised” constituent.
It is worth noting that the view of basic clause structure advocated here largely under-
mines arguments purporting to show that Right Node Raising does not necessarily affect
constituents. Grosu (1976, 643, 644) cites (ia–c) and Abbott cites (iia–e) as exhibiting right-
peripheral elements that are not phrases (examples cited in Erteschik-Shir (1987)):
(i) a. John has sliced, and Mary also seems to have sliced, a large piece of cake with a shining
new knife.
b. Bill may present, and Mary certainly will present, a series of papers at tomorrow’s
linguistic meeting.
c. Mary may have conducted, and Bob certainly has conducted, a number of tests in the
large oval lab.
(ii) a. Smith loaned, and his widow later donated, a valuable collection of manuscripts to the library.
b. I borrowed, and my sisters stole, large sums of money from the Chase Manhattan bank.
c. Leslie played, and Mary sang, some C&W songs at George’s party.
d. Mary baked, and George frosted, 20 cakes in less than an hour.
e. John offered, and Mary actually gave, a gold Cadillac to Billy Schwartz.
Under the analysis given here, the relevant expressions are indeed all constituents—in fact,
they are all VPs. Thus, the right-peripheral expression in (iib) has the structure in (iii):
(iii) VP
NP V′
large sums of money V PP
e from the Chase Manhattan bank
REFERENCES
Abbott, B. (1976) “Right Node Raising as a Test for Constituenthood,” Linguistic Inquiry 7:
639–642.
Adams, M. (1987) “From Old French to the Theory of Pro-Drop,” Natural Language and Lin-
guistic Theory 5: 1–32.
Andrews, A. (1982) “A Note on the Constituent Structure of Adverbs and Auxiliaries,” Linguistic
Inquiry 13: 313–317.
Baker, M. (1988) Incorporation. Chicago: Chicago University. Press.
17: 347–354.
Belletti, A. (1988) “The Case of Unaccusatives,” Linguistic Inquiry 19: 1–34.
Borer, H. (1980) “Empty Subjects in Modern Hebrew and Constraints on Thematic Relations,”
in J. Jensen, ed., Cahiers Linguistiques d’Ottawa: Proceedings of the Tenth Meeting of the
Northeast Linguistic Society (pp. 25–37). Ottawa: University of Ottawa.
Bowers, J. (1973) Grammatical Relations. Doctoral dissertation, MIT, Cambridge, MA.
Bresnan, J. (1976) “On the Form and Function of Transformations,” Linguistic Inquiry 7: 3–40.
Browning, M. (1987) Null Operator Constructions. Doctoral dissertation, MIT, Cambridge, MA.
Burzio, L. (1986) Italian Syntax. Dordrecht: Kluwer.
Choe, H.S. (1987) “An SVO Analysis of VSO Languages and Parametrization: A Study of Berber,”
in M. Guerssel and K. Hale, eds., Studies in Berber Syntax (pp. 121-158). MIT Lexicon Project
Working Papers no. 14, Cambridge, MA.
Chomsky, N. (1955/1975) The Logical Structure of Linguistic Theory. New York: Plenum
Press.
———. (1965) Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
———. (1982) Some Concepts and Consequences of the Theory of Government and Binding.
——— (1986a) Barriers. Cambridge, MA: MIT Press.
———. (1986b) Knowledge of Language. New York: Praeger.
Christensen, K. K. (1987) “Modern Norwegian ingen and the ghost of an Old Norse particle,” in
R.D.S. Allen and M.P. Barnes, eds., Proceedings of the Seventh Biennial Conference of Teachers
if Scandinavian Studies in Great Britain and Northern Ireland (pp.1–17). London: University
College London.
Chung, S. (1976) “An Object-Creating Rule in Bahasa Indonesia,” Linguistic Inquiry 7: 41–87.
———. (1983) “The ECP and Government in Chamorro,” Natural Language and Linguistic
Theory 1: 207–244.
———. (1987) “The Syntax of Chamorro Existential Sentences,” in E. Reuland, and A. ter Meu-
len, eds., The Representation of (In)definiteness (pp. 191–225). Cambridge, MA: MIT Press.
Chung, S. and J. McCloskey (1987) “Government, Barriers, and Small Clauses in Modern Irish,”
Contreras, H. (1984) “A Note on Parasitic Gaps,” Linguistic Inquiry 15: 698–701.
Cowper, E. (1985) “Parasitic Gaps, Coordinate Structures, and the Subjacency Condition,”
in S. Berman, J.-W. Choe, and J. McDonough, eds., Proceedings of NELS 15 (pp. 75–86).
Amherst, MA: GLSA.
Deprez, V. (1987) “Stylistic Inversion and the Structure of COMP,” unpublished manuscript., MIT,
Cambridge, MA.
Di Sciullo, A. and E. Williams (1987) On the Definition of Word. Cambridge, MA: MIT Press.
Dowty, D. (1978) “Governed Transformations as Lexical Rules in a Montague Grammar,”
Dowty, D. (1979) Word Meaning and Montague Grammar. Dordrecht: Kluwer.
Erteschik-Shir, N. (1987) “Right Node Raising,” in M. Browning, E. Czaykowski-Higgins, and
E. Ritter, eds., MIT Working Papers in Linguistics 9, Cambridge, MA: MIT.
Filmore, C. (1965) Indirect Objects and the Ordering of Transformations. The Hague; Mouton.
Green, G. (1974) “A Syntactic Syncretism in English and French,” in B. Kachru, R.B. Lees, Y.
Malkiel, A. Pietrangeli, and S. Saporta, eds., Issues in Linguistics (pp. 257–278). Urbana: Uni-
versity of Illinois Press.
Grosu, A. (1976) “A Note on Subject Raising to Object and Right Node Raising,” Linguistic
Inquiry 7: 642–645.
Guéron, J. (1980) “On the Syntax and Semantics of PP Extraposition,” Linguistic Inquiry 11:
637–678.
Higginbotham, J. (1985) “On Semantics,” Linguistic Inquiry 16: 547–593.
Huybregts, R. and H. van Riemsdijk (1985) “Parasitic Gaps and ATB,” in S. Berman, J. W. Choe,
and J. McDonough, eds., Proceedings of NELS 15 (pp. 168–187) Amherst, University of
Massachusetts.
Jacobson, P. (1983) “Grammatical Relations,” paper presented at the 1983 Winter Meetings of the
———. (1987) “Phrase Structure, Grammatical Relations and Discontinuous Constituents,”
in G. Huch and A. Ojeda, eds., Syntax and Semantics 20: Discontinuous Constituency
(pp. 27–69). New York: Academic Press.
Jones, M. (1988) “Cognate Objects and the Case Filter,” Journal of Linguistics 24: 89–110.
Kayne, R. (1979) “Rightward NP Movement in French and English,” Linguistic Inquiry 10:
710–719.
———. (1981) “ECP Extensions,” Linguistic Inquiry 12: 93–133.
———. (1984) “Connectedness,” Linguistic Inquiry 14: 223–229.
Kayne, R. and J.-Y. Pollock (1978) “Stylistic Inversion, Successive Cyclicity, and Move NP,”
Kimball, J. (1973) “The Grammar of Existence,” in C. Corum, C. T. Smith Stark, and A. Weiser,
eds., Papers from the Nineth Regional Meeting of the Chicago Linguistic Society (pp. 262–270).
Chicago: Chicago Linguistic Society.
Koopman, H. (1984) The Syntax of Verbs. Dordrecht: Foris.
Lakoff, G. (1986) “Frame Semantic Control of the Coordinate Structure Constraint,” in A. Farley,
P. Farley, and K.-E. McCullough, eds., Proceeding of CLS 22 (Part 2), Papers from the Para-
session on Pragmatics and Grammatical Theory (pp.152–167). Chicago: Chicago Linguistic
Society.
Larson, R. (1983) Restrictive Modification: Relative Clauses and Adverbs. Doctoral dissertation,
University of Wisconsin, Madison.
———. (1988) “On the Double Object Construction,” Linguistic Inquiry 19: 335–391.
———. (1990) “Extraction and Multiple Selection in PP,” The Linguistic Review 7: 169–182.
———. (1991) “Promise and the Theory of Control,” Linguistic Inquiry 22: 103–139.
Lasnik, H. (1992) “Case and expletives: toward a parametric account,” Linguistic Inquiry 23:
381-405.
Lasnik, H. and J. Uriagareka (1988) A Course in GB. Cambridge, MA: MIT Press.
Levine, R. (1985) “Right Node (Non-)Raising,” Linguistic Inquiry 16: 492–497.
Li, Y. (1988) “On V-V Compounds in Chinese,” paper presented at the MIT Workshop on Serial
Verbs, MIT, April 11.
May, R. (1985) Logical Form. Cambridge, MA: MIT Press.
McCawley, J. (1982) “Parentheticals and Discontinuous Constituent Structure,” Linguistic Inquiry
13: 91–106.
McCloskey, J. (1986) “Right Node Raising and Preposition Stranding,” Linguistic Inquiry 17:
183–186.
McCloskey, J. and K. Hale (1984) “On the Syntax of Person-Number Inflection in Modern Irish,”
Milsark, G. (1974) Existential Sentences in English. Doctoral dissertation, MIT, Cambridge, MA.
Mohanon, K. (1982) “Grammatical Relations and Clause Structure in Malayalam,” in J. Bresnan,
ed., The Mental Representation of Grammatical Relations (pp. 504–589). Cambridge, MA:
MIT Press.
Perlmutter, D. (1978) “Impersonal Passives and the Unaccusative Hypothesis,” in J. Jaeger and
A. Woodbury, eds., Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics
Society (pp. 157–189). Berkeley, CA: Berkeley Linguistics Society.
Platzack, (1987) “The Scandinavian Languages and the Null-Subject Parameter,” Natural Lan-
guage and Linguistic Theory 5: 377–401.
Pollock, J-.Y. (1986) “The Syntax of en and the Null Subject Parameter,” unpublished manuscript.
Paris.
Rapoport, T. (1986) “Nonverbal Predication in Hebrew,” in The Proceedings of the Fifth West
Coast Conference on Formal Linguistics (pp. 207–218). Stanford, CA: Stanford Linguistics
Association.
Reinhart, T. (1979) “Syntactic Domains for Semantic Rules,” in F. Guenthner and S. Schmidt, eds.,
Formal Semantics and Pragmatics for Natural Language (pp. 107–130). Dordrecht: Kluwer.
Rizzi, L. (1982) Issues in Italian Syntax. Dordrecht: Reidel.
Ross, J. (1967/1986) Infinite Syntax. New York: Ablex.
Rothstein, S. (1983) Syntactic Forms of Predication. Doctoral dissertation, MIT, Cambridge, MA.
Safir, K. (1985) Syntactic Chains. Cambridge: Cambridge University Press.
Schein, B. (1995) “Small Clauses and Predication,” in A. Cardinaletti and M. Guasti, eds., Syntax
and Semantics, Vol. 28; Small Clauses (pp. 49–76). New York: Academic Press.
Shlonsky, U. (1986) “Donkey Parasites,” in J. McDonough and B. Plunkett, eds., Proceedings of
NELS 17 (pp. 569–579). Amherst: University of Massachusetts.
Shlonsky, U. (1987) Null and Displaced Subjects. Doctoral dissertation, MIT, Cambridge, MA.
Siewierska, A. (1984) The Passive: A Comparative Linguistic Analysis. London: Croom Helm.
Simpson, J. (1984) Aspects of Warlpiri Morphology and Syntax. Doctoral dissertation, MIT,
Cambridge, MA.
Sproat, R. (1985) “Welsh Syntax and VSO Structure,” Natural Language and Linguistic Theory
3: 173–216.
Szabolsci, A. (1986) “From the Definiteness Effect to Lexical Integrity,” in W. Abraham and S. de
Meji, eds., Topic, Focus and Configurationality (pp. 321–348). Amsterdam: John Benjamins.
Talmy, L. (1985) “Lexicalization Patterns,” in T. Shopen, ed., Language Typology and Syntactic
Description No. 3 (pp. 57–149). Cambridge: Cambridge University Press.
Taraldsen, T. (1981) “The Theoretical Implications of a Class of Marked Extractions,” in A. Belletti,
L. Brandi, and L. Rizzi, eds., Theory of Markedness in Generative Grammar, Proceedings of the
1979 GLOW Conference. (pp. 475–516). Pisa, Italy: Scuola Normale Superiore.
Torrego, E. (1984) “On Inversion in Spanish and Some of Its Effects,” Linguistic Inquiry 15:
103–129.
Cambridge, MA.
Wexler, K. and P. Culicover (1980) A Formal Theory of Language Learnability. Cambridge, MA:
MIT Press.
Williams, E. (1978) “Across-the-Board Rule Application,” Linguistic Inquiry 9: 31–43.
———. (1980) “Predication,” Linguistic Inquiry 11: 203–238.
5 Some Issues in Verb Serialization
In a variety of world languages, notions that would elsewhere be expressed through

conjunction, complementation, or secondary predication are rendered uniformly by
means of a sequence of verbs or verb phrases. This phenomenon of verb serialization is
illustrated by the sentences in (1–3), drawn from languages of West Africa:
(1) a. Kofi kɔɔe baae.
Kofi went came
‘Kofi went and came.’
b. Kofi daadaa Amma kɔɔe.
Kofi tricked Amma went
‘Kofi tricked Amma and went.’ (Akan; Schachter 1974a, 254)
(2) a. Dàdá gbé àkpótí lọ ilé ní àná.
Dada took box went home on yesterday
‘Dada took the box home yesterday.’ (Yoruba; Stahlke 1974, 271)
b. Wọ́n mu ọtí yó.
They drank wine drunk
‘They drank wine until they were drunk.’ (Yoruba; Bamgbos.e 1974, 25)
(3) a. Kɔ́kú sɔ́ àtĩ́ hò Àsíba.
Koku take stick hit Asiba
‘Koku hit Asiba with a stick.’ (Fon; Lefebvre 1991, 39)
b. Amma free Kofi baae.
Asiba called Kofi came
‘Asiba called Kofi (to come) in.’ (Akan; Schachter 1974a, 261)
c. Olú bú ọmọ náà já de.
Olu berated child the go out
‘Olu berated the child and he / the child went out.’
(Yoruba; Bamgbos.e 1974, 21)
The papers in Lefebvre (1991) offer insights into verb serialization from a variety of dif-
ferent perspectives—grammatical, comparative, and cognitive/functional. In attempt-
ing to provide some orientation for this work and for the general phenomenon, I will
arrange my remarks around two questions: first, what is the basic character of the serial
verb construction—what is its structure and thematic constitution? As we will see, the
papers in Lefebvre (1991) largely cover the spectrum of possibilities available under
current grammatical theory. Second, what analogues for verb serialization can be found
in the more familiar grammatical apparatus of English? Developing some ideas by the
major contributors, I suggest that verb serialization finds a clear echo in the secondary
predicate structures of English, and that the difference between English and a language
like Yoruba lies in the fact that secondary predicates are fundamentally nominal in the
former but verbal in the latter.
1. THE SERIALIZATION PHENOMENON
As the examples in (1–3) suggest, serial verb constructions present themselves as a noun
phrase subject followed by a sequence of verbs or verb phrases (often with accompany-
ing inflectional elements):
(4) [ S NP INFL VP1 VP2 VP3 . . . ]
A number of straightforward questions arise immediately with respect to the underlying

form of serial structures, the grammatical principles licensing them, and the parameters
responsible for their cross-linguistic variation—why some languages have serial verbs
while others do not.
1.1. Serialization Structure and Licensing Principles

Broadly speaking, three quite different proposals can be distinguished regarding the
underlying form of verb serialization. And these may be associated (although not
exactly) with specific diagnoses of the thematic relations holding among serial verbs.
First of all, the sequenced VPs might represent a basically coordinate structure as in
(5a), with all the verbal elements structurally on a par:1
(5) a. VP
Coordination
VP1 VP2 VP3 • • •
This view accords naturally with an interpretation in which the VPs represent a series of
(5) b.
successive VP1 of the matrix subject. Such an interpretation is found in examples
predications
Adjunction
VP1 VP3
VP1 VP2
Some Issues in Verb Serialization 369
like (1a,b) (repeated below), where serialization appears to express (essentially) VP
conjunction:
(1) a. Kofi kɔɔe baae.
Kofi went came
‘Kofi went and came.’

b. Kofi daadaa Amma kɔɔe.
a. Kofi trickedVP
(5) Amma went
Coordination
‘Kofi
VP1 tricked
VP2Amma and
VP3went.’ • • •
Alternatively, the VPs might represent a nested sequence of adjunctions to a main VP:2
(5) b. VP1
Adjunction
VP1 VP3
VP1 VP2
On this view, the additional VPs form a series of secondary predications with a basically
(5) c.
modificatory VP1 This idea is natural given examples like (2a,b) (repeated below), whose
status.
interpretations express roughly locative and temporal modification (respectively):
Complementation
• • • VP2
(5) a. Dàdá gbé
(2) a. • • à•kpótí
VP lọ VP3
ilé ní àná.
Coordination
Dada
VP1 took box
VP2 went VP3
home on yesterday
• • •
‘Dada took the box home yesterday.’
b. Wọ́n mu ọtí yó.

(5)
b. They drank wine
VP1drunk
Adjunction
‘They drank
VP1 wine until
VP3they were drunk.’
VP1
Finally, VP2 constitute a series of embedded verbal complements, where
the VPs might
the relation between the verbal elements is neither co-predication nor modification, but
rather selection:
(5) c. VP1
• • • VP2 Complementation
• • • VP3
Here each VP falls within the selection domain of some sister predicate. This view comports
naturally with examples like (3a,b), which are interpreted essentially as causatives:3
(3) a. Kɔ́kú sɔ́ àtĩ́ hò Àsíba.
Koku take stick hit Asiba
‘Koku hit Asiba with a stick.’ (‘Koku caused a stick to hit Asiba.’)
b. Amma free Kofi baae.
Amma called Kofi came
‘Amma called Kofi in.’ (‘Asiba caused Koku to come in by calling.’)
The contributors to Lefebvre (1991) can be seen as staking out the full range of structural
and interpretive possibilities sketched above, often with different structures proposed for
different languages.
1.1.1. Hale
In his study of Misumalpan serialization structures, Hale (1991) proposes that apparent
V-chaining constructions in these languages are underlyingly clausal adjunctions. To the
Miskitu sentence (6a), for example, Hale assigns the structure in (6b):
(6) a. Witin ai pruk-an kauhw-ri.
He me strike-OBV:3 fall-PAST:1
‘He hit me and I fell down.’ or
‘He knocked me down.’
b. IPi
IPj IPi
NPy I′j NPx I′i
witin VP Ij Ø VP Ii
ai pruk an kauhw ri
Here IPj has been adjoined to IPi, and hence although the two clauses are formally
sisters, the latter bears an asymmetric superordinate relationship to the former. This
“weak subordination” relation figures centrally in Hale’s account of the obviation and
switch reference facts of Miskitu and Ulwa.
Hale points out an important two-fold division in the types of Misumalpan clause-
chaining structures. He distinguishes a coordinate clause-chaining construction in
which the various verbs designate distinct events, and a second, “true serialization”
construction in which the various verbs jointly designate a single event. This difference
is illustrated in the two glosses for (6a); on the first (clause-chaining) reading, the hitting
and falling may represent distinct events, whereas on the second (serialization) reading,
the hitting and falling represent a single event of ‘knocking down’.
Hale notes that in Misumalpan, as elsewhere, coordinate and true serialization read-
ings are realized in indistinguishable surface forms and suggests on this basis that the
two readings are not structurally distinct. He makes the (tentative) proposal that such
differences arise from whether or not the event positions in the major predicates of the
two clauses are “identified” in the sense of Higginbotham (1985). On this view, the true
serialization interpretation of (6a) corresponds to the linked “thematic” grids shown in
(6c), while the coordinate reading simply lacks this linking:
(6) c. IPi
IPj IPi
NPy I′j NPx I′i
witin VP Ij Ø VP Ii
ai pruk an kauhw ri
<1, 2, e> <1, e>
While the empirical motivation behind Hale’s proposal is clear, there are some apparent
problems with the view. One arises in connection with examples involving negation,
such as (6d), discussed by Hale:
(6) d. Witin ai pruk-an kauhw-ras.
He me strike-OBV:3 fall-NEG (-1)
‘He hit me and I didn’t fall down.’ or=
‘He didn’t knock me down.’
As Hale notes, on its “clause-chaining” reading, the scope of negation in (6d) extends
only over the main clause IPi, while on its “serialization” reading, the scope of negation
extends over both clauses. Given the standard view of scope as a structural matter, it is
unclear how the scope of negation could vary on the two readings without concomitant
variation in structure.4
A second problem comes up in connection with Hale’s particular suggestion of “event
place identification.” In the theory proposed by Higginbotham (1985) from which this
proposal is drawn, thematic identification is crucially constrained to configurations of
sisterhood between the predicates whose roles are to be identified. In clausal adjunction
structures of the sort assumed by Hale, no such relation obtains between the relevant
predicates (pruk and kauhw in (6c)), and hence it is unclear how such a proposal is to be
executed. This last point makes clear a more general question for analyses (like Hale’s)
that assume serial verbs to be joined at a level no lower than the clause. Such a view
apparently demands a radical uncoupling of the thematic notion “single predicate” and
the structural notion “single constituent.”
1.1.2. Lefebvre
Lefebvre (1991) develops an analysis of causative serialization constructions based on
structures and principles very different from Hale’s. She is concerned, in particular, with
serial causatives in the West African Kwa language Fon involving the verb sɔ́ ‘take’. The
basic phenomenon is illustrated in (7a,b) below:
(7) a. Kɔ́kú sɔ́ àsɔ̃́˜́ yì / wá àxì.
Koku take crab go / come market
‘Koku brought (direction away / towards the speaker) the crab to the market.’
b. Kɔ́kú sɔ́ àsɔ̃´ dó távò-ǰí.
Koku take crab put table-on

‘Koku put the crab on the table.’
On the matter of structure, Lefebvre proposes a complementation analysis involving

embedded VPs, rather than an adjoined or coordinate form. (7a), for example, receives
the structure in (7c), where sɔ́ is understood essentially as a causative verb selecting a
complement with clausal meaning—in this case one expressing the proposition that
the/a crab went or came to the market:
(7) c. VP
NP V′
Kc΄kú V VP
sc΄ NP V′
àsc̃´ V XP
yì/wá àxì
As with Misumalpan “true serializations,” Fon serial predicates like sɔ́ and yì / wá are
understood to define a single event. However, Lefebvre takes the relation between these
verbs to be more intimate than linkage by θ-identification. She proposes instead that com-
plex elements like sɔ-yì / wá,
́ ‘bring away-from/to’, are formed in the lexicon by a process
that “conflates” the Lexical Conceptual Structures (LCSs) of the consituent predicates.5
To illustrate briefly with (7a), sɔ́ ‘take’ receives the LCS in (8a), and yì / wá ‘go / come’
receives the LCS in (8b); conflation combines the two, merging their shared elements, to
form the complex lexical predicate in (8c):
(8) a. sɔ́ : [x cause [ y undergo change of location]]
yì / wá :
b. [ y undergo change of location
away from / toward speaker to location y]
sɔ́-yì / wá : [x cause [ y undergo change of location

c.
away from / toward speaker to location z]]
Such predicates are then projected into syntax by means of the X-bar theory given in (9):6
(9) a. XP → SpecX′ X′
b. X′ → X YP
Under this theory, heads are permitted at most a single complement per maximal projec-
tion. This forces a binary branching structure in which the two heads of the complex
predicate (sɔ́ and-yì / wá ) are inserted into two available V head positions; the result is
(7c) (see Lefebvre (1991) for details).
Assuming that complex predicates are uniformly obtained by LCS conflation,
Lefebvre’s analysis appears compatible with the strong, and intuitively appealing,
view that predicates defining a “single event” are assigned a single representation at
some level of structure. Here the notion “single event” appears to be definable in terms
of the notion “single LCS,” and the relevant level at which this unity is represented is
the lexicon.
1.1.3. Baker
Baker (1991) adopts a position on verb serialization that combines elements of the
previous two proposals. The configuration Baker suggests is analogous to complemen-
tation, but its interpretation is rather similar to coordination-adjunction. To illustrate,
the Sranan serial verb construction in (10a) is assigned the structure in (10b):
(10) a. Kofi naki Amba kiri.

Kofi hit Amba kill
‘Kofi struck Amba dead.’
b. S
NP Infl VP
Kofi ø V′
V NP V′
naki Amba kiri
c. S
As in Lefebvre’s analysis, Baker assumes that serial constructions are dual-headed—
that the serialized
NP verbs
Infl jointlyVPconstitute a single predicate. The chief difference
between the two proposals lies in how dual-headedness is expressed. As I’ve men-
Lloyd
tioned, for ø V′ are dual-headed as a matter of lexico-semantics,
Lefebvre, serializations
but not as a matter of structure. The relevant verbs merge LCSs and form a unit in
the lexicon, but the latter Vis subsequently
NP APbroken up and realized discontinuously in
syntax. For Baker, on the other hand, no lexical relation is assumed to hold between
struck Horace dead
serial verbs prior to D-Structure, and dual-headedness is expressed directly in the
syntax. This is observed in the formally aberrant X′ configuration in (10b), where V
is permitted a complement that is nonmaximal and where the nonmaximal item is in
fact a projection of V. As a result, the indicated V′ contains two competing candidates
for head—[V naki] and [V′ kiri]. Baker proposes that both elements count as heads
for V′ and that both directly θ-mark Amba, “sharing” the direct object. Syntactic
multi-headedness, which permits object-sharing, is the distinctive property of serial
constructions on Baker’s view.
Thematically, Baker’s proposal has simple analogues in other constructions. The
basic properties attributed by Baker to serial forms like (10b) are quite similar to those
assumed by Williams (1983) for secondary predication constructions like (10c). In both
NP Infl VP
Kofi ø V′
V NP V′
cases, the verb (naki / strike) and an additional predicate (kiri / dead) directly θ-mark
(and hence “share”) an object
naki (Amba / Horace):
Amba kiri
c. S
NP Infl VP
Lloyd ø V′
V NP AP
struck Horace dead
Baker argues that the interaction between his structure, the Projection Principle (Chomsky
1981, 1986), and accepted views about Case assignment sharply constrains the class of
predicates that can appear in the serial construction. For example, assuming that roles like
agent are only assigned externally to the verb phrase, it follows that in structures like (10b),
the direct object must bear a nonagent role with respect to both the first and second verb.
This means that quite generally in serial constructions in which the initial verb is transitive
and the second intransitive, the latter must be unaccusative and not unergative.
It is important to note that on Baker’s analysis, unlike Lefebvre’s, the notion of a
complex predicate is entirely a structural one. Although the structure is dual-headed,
and although both heads contribute thematic roles, they do not “compose” or “conflate”
in any way to do so. It also appears that under Baker’s account only the way in which
θ-roles are assigned—internally versus externally, and in what order—will be relevant to
the determination of possible serial structures, and that the actual identity of the roles—
agent, theme, goal, and so on—will be irrelevant insofar as it does not bear on this issue.
This contrasts with Lefebvre’s account, in which specific lexico-semantic information is
appealed to (and potentially available) to condition conflation and hence serialization.
In view of this, Baker’s proposal appears to yield a more constrained approach to verb
serialization, appealing to no extra processes (such as conflation) and little specific lexical
information in accounting for the basic phenomenon. It is of course a separate question
as to whether such an account is empirically adequate for the range of serializations.
1.1.4. Li
Li (1991) presents an analysis of serialization that is, in a certain sense, more complex
than those offered by Hale, Lefebvre, or Baker. But it is also one treating data not
discussed by the others. Like Lefebvre, Li appeals to an account involving lexical for-
mation of complex predicates from semantic-conceptual forms. However, on Li’s view,
the formation of serial versus nonserial constructions crucially depends on behavior of
the certain “prelexical” event relations (PRs), and on how they are lexicalized vis-à-vis
verbal event-denoting concepts.7
To illustrate, Li considers verbs expressing the notion of “cutting” to involve a con-
ceptual form as in (11a), in which a prelexical predicate relating two events (MEANS)
embeds another predicate (CUT) relating two individuals:
(11) a. MEANS (e1, CUT (x1, x2))
b. cut (e1, x1, x2) (“cut 2”)
c. cut (x1, x2) (“cut 1”)

On Li’s view, the two predicates may be lexicalized together such that the event argu-
ment of MEANS is passed along to the resultant form (11b). This yields a nonserial verb
cut taking two individual arguments and an event argument; the first two correspond to
cutter and cuttee, whereas the third corresponds to an (optional) instrument argument,
expressed in English via a with-PP:
(12) S
NP VP
John V NP PP
cut the bread with a knife
cut (the knife, John, the bread)
Alternatively (11a) may be lexicalized in such a way that only the arguments of
CUT are passed to the resulting form (11c). Assuming that MEANS is still present in
“virtual” form, and given that its event argument “e1” must be satisfied, another event-
denoting expression is required. This requirement, Li assumes, must be met by the pres-
ence of another verb such as ‘take’. Hence we derive serial constructions such as (13a)
from Sranan, to which Li assigns the structure in (13b):
(13) a. Mi e teki a nefi koti a brede.
I ASP take the knife cut the bread
‘I cut the bread with the knife.’
b. S
NP V1P
mi V1′ V2P
V1 NP V2 NP
teki a nefi koti a brede
MEANS (take (I, the knife), cut (I, the bread))
In (13b) V1′ and V2P provide the two event arguments of means, which is virtually
present but not syntactically expressed. With V2P adjoined to V1′, both the former and
the latter are able to assign a thematic role to the subject noun phrase mi; the latter thus
engages in both taking and cutting.
Li’s analysis evidently resembles that of Lefebvre in using “merged” or “conflated”
conceptual representations. It differs importantly from the latter, however, in its appeal
to “prelexical representations” and in allowing substantial aspects of prelexical struc-
ture to remain “virtual” while still exercising syntactic effects. Thus, while Lefebvre’s
view involves a merging of lexical argument structures whose various parts are real-
ized discontinuously in the syntax, for Li merger occurs “prior” to the point at which
lexical argument structures are formed, and serialization involves realizing only some
part of the predicate structure. The remainder (Li’s PRs) is left thematically active but
“invisible.”
Li makes the interesting suggestion that virtual predicates like MEANS, CAUSE, and
so on are iconic in the ordering of their event arguments; this requires, for example, that
the temporal precedence of “taking” over “cutting” in (13a) be reflected in the linear
ordering of V1′ before V2P. Such iconicity has an important interaction with direction
of headedness in his account. In VO languages, Li observes, iconicity and headedness
will make compatible demands on structure; thus in serial causatives such as (14) (from
Yoruba), the ordering of the predicates ti and subu satisfies iconicity (“pushing” notion-
ally precedes “falling”), and it also satisfies X′ theory (the language is VO):
(14) Femi ti Akin subu.
Femi push Akin fall
‘Femi pushed Akin down.’
In OV languages, however, iconicity and X′ theory impose conflicting requirements,

since the former will require the (main) causal predicate to be ordered finally whereas
the latter will require it to be ordered initially. Li takes this conflict to explain the appar-
ent comparative rarity of serialization in verb-final languages.8
1.1.5. Givón
The fifth contributor, Givón (1991), is concerned with a central issue that has arisen above
a number of times: the sense in which the multiple verbs of serial verb constructions jointly
make up a single event. But whereas the other authors approach this issue structurally,
using syntactic diagnostics like extractability, or the distribution of inflection and agree-
ment, Givón deploys very different methods involving elicitation of serial and nonserial
constructions in discourse. The chief theoretical assumption underlying Givón’s analysis
is an iconicity principle relating sentence production to conceptual organization. This
“Distance Principle” (DP) is given in (15) (= (12) in Givón 1991, p. 142):
(15) The temporal-physical distance between chunks of linguistically-coded information
correlates directly with the conceptual distance between them. (142)
The specific form of the DP that Givón is interested in involves the conceptual relatedness
of events as reflected in the pause separation of the units expressing them in discourse.
The idea here is that pause separations dividing finite clauses (single event domains) in
nonserial languages should be comparable to those separating verb sequences in serial-
izing languages on the assumption that the latter define a single event.
The methodology employed by Givón involved presenting speakers of serializing
and nonserializing languages with a short movie that they were asked to describe orally.
Pause measurements are made on the recorded discourse, and probabilities were com-
puted for pauses at various points in the clause. Tok Pisin (Neo-Melanesian Pidgin)
together with two native Papuan languages (Kalam and Tairora) formed the data of the
study. In brief, the results were that pause frequencies in serial constructions (i.e., those
located between serial verbs) were very significantly lower than those associated with
finite clause breaks; indeed, the pause frequencies for serial verbs were no greater than,
or lower than, mid-clause pauses associated with lexical words. As one interpretation
of the latter result, Givón suggests that serial verb stems are in fact co-lexicalized or
grammaticalized—that is, they have become part of a larger word.
1.2. Serialization Parameters

The issue of what factors govern the occurrence of serialization cross-linguistically is
addressed explicitly by some of the contributors, and for others the general position is
inferable.
Lefebvre, for example, defends an analysis in which serialization is a fundamentally
lexical phenomenon, tied to the possibility of verb conflation with a small closed class
of verbs such as ‘take’. In such an account, serialization parameters are presumably a
matter of what general operations (like conflation) are available, what the constraints
on them are, and how broadly they apply in a language that contains them. No explicit
proposals of this kind are made by Lefebvre, and they clearly must wait upon further
development of the theory of Lexical Conceptual Structure assumed by Lefebvre as the
format for conflation operations.
Similar remarks apply to the analysis put forward by Li, in which serialization arises
according to whether and how his prelexical relations are incorporated into verb entries.
Here again, an account of cross-linguistic variation would require a fuller theory of prel-
exical relations and their lexical realization. One notable aspect of Li’s proposals is that
he does not take serialization to be a parameter in the current sense. In his discussion of
Sranan, Li notes the presence of both serial and nonserial versions of instrumental con-
structions and admits the possibility of alternate lexicalizations in the same language.
The upshot is that for Li serialization is not a property characterizing languages as a
whole, but rather a more piecemeal fact about the presence of certain lexical items with
specific argument frames.
Baker departs sharply from both of the proposals above. For him, unlike Li, serializa-
tion is a properly parametric phenomenon: serial languages are characterized by a specific
dimension of difference. Furthermore, unlike both Lefebvre and Li, Baker takes the serial-
ization parameter to be syntactic and not lexical. On Baker’s view, as we have seen, serial
languages result when a specific choice in X-bar theory is selected, that is, (16), which
allows an 1-bar projection to dominate a 1-bar projection, without adjunction:
(16) X′ → X YP X′
This allows a syntactic projection to have two heads.

The proposal in (16) evidently involves separating serial and nonserial languages in
a very “deep” way and has strong empirical and conceptual consequences. Note that
without specific stipulations, (16) leads us to expect serial languages to show serializa-
tion in all categories, and not simply in V; that is, we predict serial nouns, serial deter-
miners, serial prepositions, serial inflectional elements, and so on. To my knowledge,
this result is not attested. Furthermore, since nonserial complements are available even
in serial languages (recall Li’s point about Sranan), it follows that alongside (16) we will
have to also admit (17), where Z is potentially identical to X:
(17) X′ → X YP ZP
This will require learners to be able to distinguish secondary predication with a verbal
secondary secondary predicate (17) from “true serialization” in which the structure is
multi-headed (16). Finally, the profound difference separating serial and nonserial lan-
guages would lead us to expect profound historical discontinuity between them, with
massive reorganization of the grammar. To my knowledge this is also unattested.9
The remaining contributor, Givón, makes no specific proposals as to what condi-
tions serialization cross-linguistically; he does suggest, however, that his results show
the basic parameters of difference to be grammatical and not cognitive. That is, the
results with pause frequencies show no differences in the way that event relations are
viewed cognitively, but only in the way in which these relations are “packaged by the
grammar,” to use his terms.
2. SERIALIZATION AND SECONDARY PREDICATION
Nearly all the contributions discussed above would appear to view verb serialization as
something basically foreign to English, and without any clear analogies in its grammar.
As it turns out, however, there are a number of interesting structural and semantic simi-
larities between serial verb constructions and familiar English secondary predication
structures. And these suggest some possible alternative approaches to the phenomenon.
2.1. Interpretation
We observed earlier in connection with (1–3) that serial constructions display conjunc-
tive, modificatory, and causative readings. This kind of variation is also seen with sec-
ondary predicates. For example, subject-oriented depictive predicates like those in (18)
display readings suggestive of conjunction; (18a), for instance, is roughly synonymous
with the sentence ‘John left the party and he was angry’, and similarly (18c) is largely
synonymous with ‘Alice drove home and she was happy’:
(18) a. John left the party [angry]
b. Max arrived [ready for trouble]
c. Alice drove home [happy]
Object-oriented depictive predicates like those in (19) also show analogies to adver-
bial readings; for example, the bracketed item in (19a) can be understood equivalently
to the temporal adverbial ‘when it was raw’; similarly for the bracketed string in (19b)
and the temporal ‘when it was freshly painted’:
(19) a. Jude ate the fish [raw].
b. Felix bought the door [freshly painted].
c. Edith drank her tea [cooled with ice].
Furthermore, resultative secondary predicates like (20c) are quite similar to causative
serial verb constructions, and to adverbial adjuncts involving ‘until’ or expressing pur-
pose. Compare (20a), for instance, with ‘Carol rubbed her finger until it was raw’, and
compare (20c) with ‘Lloyd called us to come in’:
(20) a. Carol rubbed her finger [raw].
b. Black Flag kills bugs [dead].

c. Lloyd called us [in].
d. Oscar lured Eunice [away].
Much the same range in interpretation is thus found in the two construction types.
The semantics of serial and secondary predication structures are also analogous in
aspects discussed recently by Awoyale (1987, 1988). In considering a variety of serial
structures, Awoyale suggests that the semantic relations that hold between serial verbs
can actually be reduced to just two: an inclusive relation and an exclusive one. In brief,
the inclusion relation “exists between two or more predicates when the action of one
is taking place inside the domain of the other . . . the outer predicate delimits the
action of the inner verb . . .” (p. 13). On the other hand, with the exclusion relation
“the actions of the verbs are not included one inside another, but rather are separate
events” (p. 17).
Awoyale illustrates these notions pictorially with the Yoruba examples in (21) and
(22): verbs standing in the inclusion relation fall within the same circle; those standing
in the exclusion do not. Thus in (21a), under the gloss given, we understand Aje’s swim-
ming to have been bounded or delimited by his going / leaving; that is, the swimming
proceeded until Aje was gone. Similarly in (21b), Aje’s drinking of alcohol was bounded
by his state of sobriety: the drinking proceeded until Aje was intoxicated:
(21) a. Aje w`ẹ lọ
Aje swam go /away

‘Aje swam away.’
b. Aje mu ọtí yó
Aje drank alcohol be-full
‘Aje became intoxicated.’
By contrast, (22a) gives (21a) on its exclusive reading. Here the going / leaving is not under-
stood to bound Aje’s swimming; rather, the two events bear a simple (iconic) sequential
relation to each other: Aje swims and then he goes. Similarly, in (22b) there is no bounded-
ness or delimitedness between the denoted events; Aje goes and does so to run:10
(22) a. Aje w`ẹ lọ

Aje swam go /away
‘Aje swam before leaving.’

b. Aje lọ sárá
Aje go run
‘Aje went to run.’
c. Aje jókòó mu’tín jé.un
Aje sat-down drank alcohol ate

‘Aje sat, and drank, and ate.’
This central, two-fold distinction that Awoyale finds in serialization constructions
is also observed with secondary predications, as discussed in recent work by Tenny
(1987). Tenny points out that secondary predicates can be divided into essentially two
kinds: delimiting and nondelimiting. A delimiting predicate bounds the action of the
main predicate, essentially telling you when it terminates. Resultative secondary predi-
cates as in (20a–d) are thus delimiting in that the action continues (rubbing, calling, etc.)
until the object attains the state specified by the predicate (the finger is raw, we come in,
etc.). Nondelimiting predicates do not bound the action in this way and are exemplified
by depictive secondary predicates like those in (18–19). In (19a), for instance, the fish
remains raw during the course of eating, and the extent of the eating is in no way deter-
mined by the rawness of the fish. Evidently, the notion of an inclusively related serial
verb and a delimiting secondary predicate are quite similar. Likewise for the notion of an
exclusively related serial verb and a nondelimiting predicate. Fundamentally the same
aspectual distinction thus appears to characterize both construction types.
2.2. Structure
Recent views of the structure of secondary predications also appear to converge with
the structures for serializations proposed by some contributors in Lefebvre (1991). In
Larson (1989) it is suggested that English secondary predicates are uniformly daughters
of V′, occurring in either of the two configurations shown in (23a,b). The former is the
structure of subject-oriented depictive predicates. The latter is the structure of object-
oriented depictive predicates ((23b) on the reading: ‘Carol rubbed her finger while it
was raw’) and of causative/resultative predicates ((23b) on the reading: ‘Carol rubbed
her finger until it was raw’):11
(23) a. VP
NP V′
John V′ AP
V NP angry
left the party
b. VP
NP V′
Carol V VP
e NP V′
her finger V AP
rubbed raw
The intuitive idea here is that when NP receives a thematic role from a primary and a
secondary predicate, the two must appear as sisters and form a constituent that is itself
sister to NP. Thus, in (23a), John receives a thematic role from the primary V′ leave the
party and from the secondary AP angry, and hence the two appear as sisters under a V′
that is itself sister to NP. Likewise, in (23b), her finger receives a θ-role from rub and a
θ-role from raw, and hence rub and raw appear as sisters under a V′ that is predicated
of her finger.
The analyses of Lefebvre and Baker can be recast directly in terms of structures like
(23b). For example, a typical Fon take serial example like (7a) (repeated below) receives
the structure in (24):
(7) a. Kɔ́kú sɔ́ àsɔ˜́

̃ yì / wá àxì.
Koku take crab go / come market
‘Koku brought (direction away from / toward the speaker) the crab to the market.’
(24) VP
NP V′
Kɔ́kú V VP
e NP V′
àsɔ˜́ ̃ V VP
sɔ́ yì/ wá àxì
Here, much as in Lefebvre’s own account, sɔ́-yì / wá àxì forms a complex predicate
meaning ‘cause to go away from/toward the market’. This predicate selects the object
àsɔ̃´˜, ‘the crab’, to form a VP predicate meaning ‘cause crab to go away from/toward
the market’. The latter is then predicated of the subject NP Kɔ́kú. The observed surface
form results by the raising of a verbal head, just as in (23b) above (see Larson (1989)
for discussion).
This structure appears largely compatible with Lefebvre’s conflation proposal. Recall
that on her view sɔ́, ‘take’, selects an agentive subject and an object understood as under-
going a change of location. Recall further that yì / wá àxì, ‘away from/toward the market’
selects a single argument understood as undergoing a change of location. The two sister
predicates in (24) can thus be viewed as assigning a thematic role jointly to the object
àsɔ̃´˜, ‘crab’, much as the verb and AP do in the object-oriented secondary predication
structure (23b). Note, moreover, that (24) has a conceptual advantage over Lefebvre’s
own (7c), in that the former reflects “thematic constituency relations” in D-Structure—
sɔ́-yì / wá àxì forms an underlying phrase—whereas the latter does not.
A similar recasting of Baker’s proposal is possible. Thus Baker’s (10a) (repeated below)
can be reanalyzed as having the VP in (25), where naki kiri forms a complex predi-
cate strongly analogous to ‘strike dead’ in English causative secondary predication. Here
again, naki and kiri are understood as jointly predicated of the NP sister of V′—Amba:
(10) a. Kofi naki Amba kiri.
Kofi hit Amba kill

‘Kofi struck Amba dead.’
(25) VP
NP V′
Kofi V VP
e NP V′
Amba V VP
naki kiri
This proposal differs significantly from Baker’s in that, unlike (10b), (25) involves no
dual-headedness, and hence no departure from standard X-bar-theoretic assumptions.
In the complex predicate naki kiri, naki is exclusively the head verb, and kiri is a full
XP complement.
This proposal also differs from Baker’s and Lefebvre’s in its view of the relation
between “event structure” and headedness. Recall that for these authors, the multi-
headedness of a serial verb construction is taken to correspond to the fact that the verbs
in question express a single event. On the view sketched above, by contrast, the single-
event status of serializations (at least with causatives) can be attributed to the secondary
predicate’s standing in a delimiting, or “inclusive,” relation to the event expressed by
the main verb. Thus, just as the adjectival secondary predicate dead delimits the action
of swatting in John swatted the fly dead, so the verbal secondary predicate kiri delimits
the action of hitting in Kofi naki Amba kiri. In both cases a single event is involved, but
this no more results in (or requires) syntactic dual-headedness in the latter case than it
does in the former.
The claim that kiri is a full VP in (25) raises a natural question as to its internal struc-
ture. Although I cannot defend the proposal in detail here, (26), based on proposals by
Carstens (1988), represents one plausible answer:
(26)
VP
NP V′
Amba V VP
naki Oi VP
NP V′
Pro V NPi
kiri t
Here [VP kiri ] heads a maximal projection with a specifier (Pro) corresponding to the
subject of ‘kill’ and an object empty operator adjoined to the predicate. As discussed in
Larson (1989), this situation is analogous to English predications of purpose involving
objects:
(27) a. Mary [VP brought John to tease]
b.
VP
NP V′
John V IP
brought Oi IP
NP I′
Pro to tease ti
In both cases the object NP identifies the range of the empty operator, and hence ‘being
killed’ and ‘being teased’ are understood as holding of Amba and John, respectively.
Furthermore, both constructions involve some form of empty category (here labeled
simply as ‘Pro’) that is understood as bound or linked to the subject NP (Kofi and Mary,
respectively).
2.3. The Serialization Parameter Revisited

The structural and semantic analogies observed above suggest that verb serialization might
actually be a form of secondary predication similar to what is found in English. This in
turn suggests an interesting view of what the difference between serial and nonserial lan-
guages really amounts to. Notice first that while secondary predicates come in a variety
of categories in English, one predicate category is conspicuously missing: VP. Simple verb
phrases never serve as secondary predicates, no matter what their semantics:
(28) a. *John left the party [hate martinis].
b. *Alice drove the car [drink wine].

c. *Jude caught the fish [swim in river].
d. *Lloyd called us [come in].
e. *Edith drove her car [go].
The situation in English is thus as in (29):
(29)
V′ {
V (NP) NP/PP/AP
*VP
{
This pattern contrasts with that in serial languages in two important ways: first, in serial-
izing languages, notions typically expressed by AP or PP are quite pervasively grammati-
calized with intransitive verbs, and hence these minor categories are largely assimilated
to the category of VP. Second, NP secondary predicate constructions analogous to John
arrived a perfect wreck are (to my knowledge) entirely absent. The situation in serial
languages can thus be put (somewhat tendentiously) as in (30), where “PP” and “AP” are
understood as essentially a subcase of VP:
(30)
V′ {
V (NP) VP/PP/AP
*NP
{
The point of contrast here seems clear-cut: assuming serial constructions to be a form of
secondary predication, the chief difference between a “serializing language” like Yoruba
and a “nonserializing language” like English reduces to a matter of what secondary
predicate categories are allowed. More precisely, employing standard feature matrices
for the lexical categories, serializing languages have secondary predicates that are either
[–N] or [+V] (31a), whereas nonserializing languages have secondary predicates that are
either [+N] or [–V] (31b):
(31) a. +V +N b. +V +N
–V A –N –V A –N
N V N V
P P
Serial Languages Nonserial Languages
Serial languages show non-nominal secondaries, whereas nonserial languages show

nonverbal secondaries.
If correct, this result implies that the “serialization parameter” separating Yoruba
and English should involve some aspect in which verbs and nominals differ with respect
to predication. Recent proposals extending Case theory suggest one way of executing
this. Fabb (1984) and Roberts (1985) have proposed that basic notions like Case and
the Case Filter apply not only to arguments but to predicates as well. In brief, they sug-
gest that just as arguments must be marked with inflection (canonically Case) to receive
a θ-role, so predicates must be marked with inflection (canonically tense and agreement)
to assign a θ-role. Now, with simple verbal predicates, the demands of this “Inflection
Filter” apply straightforwardly—Vs must receive marking through some INFL element.
However, with nominal predicates (As) and nonverbal predicates (Ps) a genuine ques-
tion arises as to what kind of inflection should be assigned. Should minor-category
secondary predicates be considered as nominals, and hence bear Case for purposes of
the Inflection Filter? Or should they be considered as verbals and hence bear finiteness
and agreement? Suppose that languages can choose either of the two answers but must
answer uniformly. That is, they must require either that all secondary predicates bear
the tense and agreement of the head V, or else that all secondary predicates bear Case.
Assuming that nouns ([+N,–V]) can never bear tense and agreement and that verbs
([–N,+V]) can never bear Case, the result will either be a system like Yoruba in which
secondary predicates are non-nominal (a serializing language) or a system like English
in which secondary predicates are nonverbal (a nonserializing language). In this way we
would derive the “serialization parameter.”
This picture, if tenable, would be attractive in its simplicity. On such a view, the dis-
tinction between serializing and nonserializing languages would reflect neither a “deep”
difference in X-bar theory (as for Baker) nor a difference in the availability of particular
lexicalization rules (as for Lefebvre and Li), but instead a rather “shallow” difference
in how the inflectional requirements on secondary predicates are to be met. Such a
parameter could presumably be set on the basis of simple sentences involving agreement
and inflection. Whatever the prospects for these specific views, it is worth emphasizing,
in conclusion, that the general connection between serialization and secondary predi-
cation seems well worth pursuing. The correlations of form and interpretation noted
above appear substantial and strongly suggest an approach attempting to relate the two
construction types.
NOTES
1. For discussion of the coordination analysis of serial constructions see Schachter (1974a,
1974b), Bamgbose (1974), Collins (1987), and Lord (1974).
2. For discussion of the modifier analysis of serial constructions see Stahlke (1974) and
Schachter (1974b).
3. For discussion of the complementation analysis of serial constructions see Stahlke (1970)
and Schachter (1974a).
4. Extraction in other serializing languages raises a similar problem. In brief, while extraction is
generally permissible from “true serializations,” in which the verbs designate a single event,
extraction with “chaining” constructions is generally illicit. This difference is often attrib-
uted to the fact that the latter are coordinate, and hence fall under the Coordinate Structures
Constraint, whereas the latter are noncoordinate and so allow extraction (see Sebba (1987)
for recent discussion). Under a proposal that collapses chaining and serialization structurally,
this explanation is lost.
5. See Hale and Keyser (1987) for discussion of Lexical Conceptual Structures.
6. The X-bar theory in (9), embodying a “Single Complement Hypothesis,” is proposed in
Larson (1988).
7. Muysken (1987) makes the similar proposal that serial languages are distinguished from
nonserial languages in allowing only “atomic” predicates to be lexicalized.
8. Givón (1991) discusses serializing languages from Papua that are SOV. He observes, however,
that in these languages the serial clauses uniformly precede the main / finite verb; hence it is
unclear to what extent they pose a problem for Li’s generalization.
9. See Givón (1971) and Lord (1973, 1982) for discussion of the historical development of
serialization.
10. Awoyale (1988) provides very few glosses for his examples illustrating exclusion; that given
in (22c) reflects his statement in the text that such sentences are to be understood as either a
parallel / simultaneous set of events or a temporal / logical sequence of events (see pp. 17–18).
11. The VP structures in (23–25) are based on proposals in Larson (1988, 1989). The basic idea
underlying them is that subjects and complements are all initially structured in the VP in
subject-predicate form. The structural requirements on doing so (in particular, the require-
ment that maximal projections contain at most a single specifier and a single complement)
force the generation of empty head positions that are subsequently occupied by verb raising.
The VP structures in (23–25) are all to be understood as embedded under functional category
projections (inflectional elements); we ignore syntax above VP here for simplicity.
REFERENCES
Awoyale, Y. (1987). “Perspectives on Verb Serialization,” paper presented at the First Workshop
in Niger-Congo Syntax and Semantics, MIT, Cambridge, MA.
Awoyale, Y. (1988). “Complex Predicates and Verb Serialization.” MIT Lexicon Project Working
Papers 28. Cambridge, MA: MIT Center for Cognitive Science.
Baker, M. (1991). “On the relation of serialization to verb extensions,” in C. Lefebvre, ed., Serial
Verbs: Grammatical, Comparative and Cognitive Approaches (pp. 79–102). Amsterdam: John
Benjamins.
Bamgboṣe, A. (1974). “On Serial Verbs and Verbal Status,.” Journal of West African Linguistics 9:
17–48.
Carstens, V. (1988). “Serial Verbs in Yoruba,.” paper presented at the Second Workshop in Niger-
Congo Syntax and Semantics, MIT, Cambridge, MA.
Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris.
Chomsky, N. (1986). Knowledge of Language: Its Nature, Origin and Use.
New York: Praeger.
Collins, C. (1987). “Notes on Serial Verb Constructions in Ewe,” unpublished manuscript, MIT,
Cambridge, MA.
Fabb, N. (1984). Syntactic Affixation. Doctoral dissertation, MIT, Cambridge, MA.
Givón, T. (1971). “On the Verbal Origin of the Bantu Suffixes,” Studies in African Linguistics 2:
145–163.
Givón, T. (1991). “Some substantive issues concerning verb serialization: Grammatical vs. cogni-
tive packaging,” in C. Lefebvre, ed., Serial Verbs: Grammatical, Comparative and Cognitive
Approaches (pp. 137–184). Amsterdam: John Benjamins.
Hale, K. (1991). “Misumalpan verb sequencing constuctions,” in C. Lefebvre, ed., Serial Verbs:
Grammatical, Comparative and Cognitive Approaches (pp. 1–36). Amsterdam: John Benjamins.
Hale, K. and J. Keyser (1987). “A View from the Middle.” MIT Lexicon Project Working Papers 10.
Cambridge, MA: MIT Center for Cognitive Science.
Higginbotham, J. (1985). “On Semantics,” Linguistic Inquiry 16: 547–593.
Larson, R. (1988). “On the Double Object Construction,” Linguistic Inquiry 19: 335–391.
Larson, R. (1989). “Light predicate raising,” MIT Lexicon Project Working Papers 27. Cambridge,
MA: MIT. Republished in this volume.
Lefebvre, C. (1991). “Take serial verb constructions in Fon,” in C. Lefebvre, ed., Serial Verbs: Gram-
matical, Comparative and Cognitive Approaches (pp. 37–78). Amsterdam: John Benjamins.
Li, Y. (1991). “On deriving serial verb constructions,” in C. Lefebvre, ed., Serial Verbs: Gram-
matical, Comparative and Cognitive Approaches (pp. 103–136). Amsterdam: John Benjamins.
Lord, C. (1973). “Serial Verbs in Transition,” Studies in African Linguistics 4: 269–296.
Lord, C. (1974). “Causative Constructions in Yoruba,” Studies in African Linguistics, Suppl 5. 195–204.
Lord, C. (1982). “The Development of Object Markers in Serial Verb Languages,” in P. Hopper
and S. Thompson, eds., Syntax and Semantics 15: Studies in Transitivity, New York: Academic
Press.
Muysken, P. (1987). “Parameters for Serial Verbs,” paper presented at the First Workshop in
Niger-Congo Syntax and Semantics, MIT, Cambridge, MA.
Roberts, Ian. (1985). “Agreement Parameters and the Development of English Modal Auxiliaries,”
Schachter, P. (1974a). “A Non-Transformational Account of Serial Verbs,” Studies in African Lin-
guistics, Suppl. 5: 252–270.
Schachter, P. (1974b). “Serial Verbs: A Reply to a Reply,” Studies in African Linguistics, Suppl. 5:
278–282.
Sebba, M. (1987). The Syntax of Serial Verbs: An Investigation into Serialization in Sranan and
Other Languages. Amsterdam: Benjamins.
Stahlke, H. (1970). “Serial Verbs,” Studies in African Linguistics 1: 60–99.
Stahlke, H. (1974). “Serial Verbs,” Studies in African Linguistics. Suppl. 5: 271–277.
Tenny, C. (1987). Grammaticalizing Aspect and Affectedness. Doctoral dissertation, MIT, Cambridge,
Massachusetts.
Williams, E. (1983). “Against small clauses,” Linguistic Inquiry 14: 287–308.
6 Sentence-Final Adverbs and “Scope”
In linguistic theorizing, advances in syntax and semantics typically go hand-in-hand: our

grasp on the grammar of a construction goes forward with our grasp on its meaning. In
what follows, I present an illustration of this point involving sentence-final adverbs. Cur-
rent syntactic views of sentence-final adverbs include at least two diametrically opposed
accounts, each with a certain amount of empirical evidence in its favor. One is old and
well established, and has a widely accepted semantics accompanying it. The second is
much newer, and, although more in accord with current syntactic views of the clause,
it lacks a corresponding semantics. Here I argue that recent advances in the semantics
of adverbs—specifically the proposal of structured, Davidsonian event quantification—
supply the missing elements and offer a fresh picture of how the syntax and the semantics
of adverbs fit together. One consequence that emerges is a new view of certain well-
known data that have been widely analyzed as involving adverb scope. If I am right, these
phenomena and their implications have been seriously misunderstood ever since their
introduction over 30 years ago. I begin by sketching the basic problem area.
1. TWO ANALYSES OF SENTENCE-FINAL ADVERBS
Consider a simple English sentence like (1) containing two sentence-final adverbs (at
home and yesterday), and consider the simple question of how the latter are attached
in the clause.
(1) John watched TV at home yesterday
One well-known hypothesis is that adverbial phrase structure ascends rightward, with
sentence-final adverbs adjoining recursively, either to the VP or to the clause (TP) (2).
(2) [TP John [VP [VP [VP watched TV ] at home ] yesterday]]
This view has been advanced and/or assumed by a great many theorists, and also
appears in nearly all syntax textbooks. I will refer to it as the right-ascending analysis,
or, more simply, the classic analysis.
A second, somewhat more recent hypothesis holds that adverbial phrase struc-
ture descends rightward. An example of this view is offered in Larson (1988, 1989,
1991), where adverbs are generated as lowermost V-complements or specifiers and are
stranded by a verb that undergoes successive raising through a series of stacked VP
“shells” (3).
(3) [TP John [VP knocked [VP on the door t [VP at home [V′ t yesterday ]]]]]
Versions of this general position are also developed by Stroik (1990, 1992a, 1992b,
1996) and by Alexiadou (1994, 1997, 2002). I will refer to it as the right-descending
analysis.1
Given the sharply differing assumptions in these positions, one might expect deciding
between them to be a straightforward matter. Surprisingly, however, this has not been
so. Each account appears to have important arguments in its favor.2
1.1. Some Arguments for Right Descent

1.1.1. Domain Phenomena
As is well known, negative polarity items like any, anyone, any + noun, ever, and so
on require a “trigger” item standing in a structural relation to them as a condition of
well-formedness. The trigger must be an appropriately negative element like no, not,
n’t, few, rarely, and so on. The mostly widely accepted view of the structural relation is
c-command, where node α c-commands a node β in a tree iff neither α nor β dominates
the other, and β is a sister node of α or is contained in a sister node of α.
Under a right-descending analysis, well-formed examples like (4a) are accounted for
straightforwardly since the negative trigger (rarely) stands in a c-command relation to
the negative polarity item contained in the final adverb (during any of our meetings) (4b).
By contrast, under the right-ascending analysis, such examples are problematic since
the outer adverb occupies a higher position (4c) and is not c-commanded by the trigger.
(4) a. John spoke [rarely] [during any of our meetings]
b. [S John [VP spoke [VP rarely [ V′ t during any of our meetings]]]]
c. [S John [VP [VP spoke rarely] during any of our meetings]]
Domain phenomena thus offer a direct argument for right descent.3
1.1.2. Sentence-Word (Anti-)Correspondences

The right-descending analysis also provides an account of certain interesting “anti-
correspondences” between sentence form and word form observed in various languages
of the American northwest coast. Haisla, a Wakashan language, is an example. Bach
(1996) notes that in Haisla, attachment of phrases in the sentence and attachment of
affixes in the word appear to be mirror inverses. In the Haisla sentence, the general
syntactic pattern of phrases is that “the main predicate [is] followed by the nominal
arguments (subject, object, oblique object) and then by any optional adjuncts or modi-
fiers (location, time)” (p. 8). Bach gives the examples in (5) (where ‘dem’ and ‘3rem’
abbreviate “demonstrative” and “3rd person remote,” respectively).
(5) a. Duqwel qi ḡenemax̄i t’ixwax̄ i l(a) gukwdemax̄ i.
see/saw dem woman-3rem blackbear-3rem in village-3rem
‘The woman saw the blackbear in the village.’

Sentence-Final Adverbs and “Scope” 389
b. . . . qi sai k a qi
ǀ w
sa sems qix gu k ela -gax̄ aga 1 qi ’amleyaǀs-asi.
ǀ ǀ w
dem chase dem children dem dwelling -in/at dem playground-poss
. . . ‘she chased the children of these villagers at the playground.’
By contrast, within the Haisla word, the order is just the opposite: “the affixes that
correspond semantically to . . . optional modifiers are immediately adjacent to the main
functor” (p. 8). Bach’s examples are in (6).
(6) a. Kw’a’ilhnugwa
Kw’a- ilh -nugwa
sit- inside/in house -I
‘I sit down inside’
b. labetisi
la- bet -is -i
go- into enclosed space -on beach -3rem
‘she (he, it) goes into a hole on the beach’
This pattern holds not only in Haisla but in many other languages of the same geographic
region (Wakashan, Salishan, Tsimshianic) (see Anderson 1992 and Bach 1996).
These facts assume considerable importance under the so-called Mirror Principle of
Baker (1985), which expresses a fundamental, widely documented correlation between
word and sentence structure.
Mirror Principle
Morphological derivations must directly reflect derivations syntactic (and vice versa).
According to the Mirror Principle, the order of morphological composition in the word
should track the order of syntactic composition in the phrase. Hence very different pre-
dictions are made under right-ascending and right-descending theories.
Under a right-ascending (R-A) theory, nominal argument phrases (subjects, objects,
indirect objects) combine with the predicate first, followed by adjunct modifiers. The
Mirror Principle thus predicts a parallel pattern in the word: the core stem or root
should first combine with inflections or clitics marking subject, object, indirect object,
and so on, followed by affixes with adjunct or modifier relations to the core stem (7a).
By contrast, under a right-descending theory, (R-D) adjuncts “get to” the predicate first;
nominal argument phrases combine only afterwards. The Mirror Principle thus predicts
the opposite pattern in the word: stem or root followed by adjunct affixes followed by
inflections for subject, object, and indirect object (7b).
(7) a. Stem-ARG*-ADJ* Expected Haisla pattern under R-A theory ×
b. Stem-ADJ*-ARG* Expected Haisla pattern under R-D theory P

Empirically, it is the second pattern that we observe. Hence the right-descending theory
is directly supported by these facts under the Mirror Principle.4
1.2. A Potent Argument for Right Ascent

For its part, the classical analysis of sentence-final adverbs also has strong arguments in
its favor. One of the simplest, and most compelling, is offered by Andrews (1983) using
examples (8a,b).
b. John knocked on the door twice intentionally.
As Andrews notes, the two sentences differ in meaning. (8a) asserts that John did some-
thing twice, namely, knock on the door intentionally. By contrast, (8b) asserts that
John did something intentionally, namely, knock on the door twice. The difference is
truth conditional. For example, suppose John is proceeding through my neighborhood,
intending to knock once on every door. By mistake, he forgets that he has been to my
door already, and repeats himself. In this situation it seems that (8a) is true since John
intended to knock and did so twice. But (9b) is false since knocking twice was not part
of his intention.
Andrews diagnoses this meaning difference as one of scope. He suggests that the
first reading arises by twice taking scope over intentionally, and that the second read-
ing arises by the inverse scope. If this diagnosis is correct, then the standard, right-
ascending syntax, coupled with the standard semantics from Montague Grammar
(Thomason and Stalnaker 1973; Montague 1974) will explain it neatly. As we have
seen, in the standard syntax, VP adverbs are adjuncts, adjoining recursively to VP.
Outer adverbs asymmetrically c-command inner adverbs. (8a) receives the structure in
(9a), where twice is highest, whereas (8b) gets the structure in (9b), where intention-
ally is highest.
(9) a. John [VP [VP [VP knocked on the door ] intentionally ] twice ]
b. John [VP [VP [VP knocked on the door ] twice ] intentionally ]
Under the standard Montagovian semantics, VP adverbs are functors, taking (inten-
sions of) VP-denotations as arguments and returning VP-denotations as values. Outer
functors take arguments containing inner functors. If one applies this analysis to the
trees in (9a,b), the semantic representations come out as in (10a,b), where twice’ (the
interpretation of twice) has widest scope in (10a), and where intentionally’ has widest
scope, in (10b).5
twice’(întentionally’(^knocked on the door’))

(10) a.
intentionally’(^twice’(^knocked on the door’))

b.
Thus, applying the standard semantics to the standard, right-ascending syntax gives the
adverbs differential scope, as expected under Andrews’s diagnosis. The simplicity of this
picture has convinced many researchers of the need for right-ascending structure (Ernst
1994, 2001; Laenzlinger 1996; Pesetsky 1995; Cinque 1999, 2002).
1.3. A New Approach

An important feature of tightly knit arguments like the one just rehearsed is that they
cannot easily be challenged piecemeal. Since the parts fit together snugly, in general either
the argument is simply right, or else it’s all wrong—wrong in all of its parts. This makes
such arguments of great potential interest since, if they fall, much falls with them.
In the remainder of this paper, I will suggest that the scope argument given above is in
fact wrong in all of its parts. Specifically, I will argue that Andrews’s original diagnosis
was wrong and that the contrast in pairs like (9a,b) is not a matter of scope but rather
of predication, equivalent to (11a,b):
(11) a. John’s intentional knockings on the door were two (in number).
b. John’s double-knock on the door was intentional.
I will argue that the standard semantics is wrong: that sentence-final adverbs are not
scope-inducing VP-functors. Rather, they are event predicates in the sense of Davidson
(1967b), and the semantics of (8a,b) involves structured, Davidsonian event quantifica-
tion. Finally, I will argue that the standard syntax is wrong. When analyzed as struc-
tured event quantifications, (8a,b) imply a right-descending syntax under the Mapping
Hypothesis of Diesing (1992). The outcome of this challenge, I hope to show, is that a
core argument for right-ascending syntax falls, and what emerges is a new, strong argu-
ment for its competitor.
2. STRUCTURED EVENT QUANTIFICATION
Basic logic texts typically assign simple English sentences like (12a) a representation like
(12b), involving two constants and a binary relation. Davidson (1967b) proposes that
action sentences like this are not so simple and involve, in addition, a quantification
over events. Davidson’s original formulation is given in (12c) (ignoring tense), where an
existentially quantified event parameter is simply added to the relational structure of the
predicate. The widely adopted, neo-Davidsonian position is given in (12d), where the verb
is distilled into a core unary event predicate, whose participants are linked to the event by
means of conjoined binary thematic relations (see Parsons 1990 and Higginbotham 1989,
among many others, for discussion):
b. kiss(m,j)
c.
∃e[kiss(m,j,e)] (“There is a kissing of John by Mary.”) “Classical Davidson”
d.
∃e[kissing(e) & Agent(e,m) & Theme(e,j)] “Neo-Davidson”
(“There is a kissing, and it is by Mary, and it is of John.”)
Hereafter, I assume the neo-Davidsonian analysis but will sometimes abbreviate using
classical Davidsonian formulae when decomposition of the verbal relation is irrelevant.
Davidson’s key motivation for introducing events is the analysis of adverbs it offers.
Adverbs of many different kinds (manner, duration, location, etc.) are analyzed as predicates
of events, allowing a very simple account of how they are semantically integrated into the
clause (13)–(15) (again ignoring tense):
(13) a. Mary kissed John quickly. MANNER
∃e[kiss(m,j,e) & quick(e,C )]

b.
(“There is a kissing of John by Mary and it is quick (for such an action).”)
(14) a. Mary kissed John for-an-hour. DURATION
∃e[kiss(m,j,e) & for-an-hour(e)]

b.
(“There is a kissing of John by Mary and it is an hour long.”)
(15) a. Mary kissed John in-the-park. LOCATION
∃e[kiss(m,j,e) & in-the-park(e)]

b.
(“There is a kissing of John by Mary and it is in the park.”)
The Davidsonian event analysis has the interesting property that adverbial predicates
are not scopal. Unlike the standard semantics, adverbs are not analyzed as functors
applying to VP-denotations of the familiar sort. Rather they are simple conjuncts. Our
scope-like intuition that “quickly applies to kissed John” in (13a) arises from the fact
that VP denotes an event of John-kissing, and quick is true of this whole event.
Another point to note about (13)–(15) is that the event quantifications are repre-
sented as unrestricted/unstructured. The existential quantifier attaches to a “flat” struc-
ture of co-equal event conjuncts. A number of authors have argued that this aspect of
Davidson’s analysis is in need of refinement—that in certain cases event quantification
is structured into quantifier, restriction, and scope.
2.1. Focus (Herburger 2000)

Herburger (2000) offers an attractive account of focus, whose centerpiece is an appeal
to structured event quantification. Compare (12a) (repeated as (16a)) with the focused
variant in (16b). Whereas the first asserts the existence of a kissing of John by Mary,
the second is taken to divide up this information into a presupposition or background
entailment of Mary’s kissing someone, and an assertion that the individual in ques-
tion was John.6 Herburger correlates this difference in “information packaging” with
a difference in event representations. The first receives the flat, unstructured event rep-
resentation in (17a), whereas the second gets the representation in (17b), where the
nonfocused material forms a restriction on the event quantifier and the focused material
constitutes its scope.7
b. Mary kissed JOHN.
i. Presupposes/Background-entails: Mary kissed someone.
ii. Asserts: The individual was John.

(17) a. ∃e [kissing(e) & Agent(e,m) & Patient(e,j)]
“There was a kissing event with Mary as agent and John as patient”
b. ∃e [kissing(e) & Agent(e,m)] (Patient(e,j))
Q Restriction Scope
Presupposed/B-Entailed Asserted
Mary, its patient was John.
Under the assumption that the restriction on a structured quantifier is presupposed/

background-entailed and the scope is asserted, the difference in information packaging
follows directly from these representations.8 Alternative foci are accommodated along
the lines shown in (18)–(20):
(18) a. MARY kissed John.
b.
∃e [kissing(e) & Patient(e,j)] (Agent(e,m))
“For some kissing of John, its agent was Mary.”
(19) a. Mary KISSED John.
b.
∃e [Agent(e,m) & Patient(e,j)] (kissing(e))
“For some event with Mary as agent and John as patient, it was a kissing.”
(20) a. Mary kissed John QUICKLY.
b.
∃e [kissing(e) & Agent(e,m) & Patient(e,j)] (quick(e,C ))
“For some kissing with Mary as agent and John as patient, it was quick.”
Thus, if Herburger’s analysis is correct, structured event quantification is the semantic

core of focus phenomena.
2.2. Middles (Condoravdi 1989)

An even earlier appeal to structured event quantification is Condoravdi’s (1989) analy-
sis of middles, which have generic force and typically require a postverbal predicate
(an adverb or adjunct PP). As (21a–c) show, without the latter, middles are generally
perceived as odd or incomplete.
(21) a. These flowers grow ?*(quickly/in sandy soil).
b. Ballerinas dance ?*(beautifully).
c. Bread cuts ?*(easily).
Condoravdi makes a striking proposal that correlates these features. She suggests that mid-
dles involve structured, generic event quantification, in which the sentence nucleus (the verb
and its arguments) forms the restriction, and the adjunct expression constitutes the scope.
Thus (21a) is analyzed as in (22a), where Γ is a generic quantifier and where the semantic
contributions of the sentence are partitioned as shown. The “main clause” gives the restric-
tive term; its content is presupposed or background-entailed. The “adjunct” gives the scope
term; its content is asserted. (21b,c) are analyzed similarly in (22b,c), respectively.
(22) a. e[Con(f,e) & growing(e) & Theme(e,f)] (quick(e))
Q Restriction Scope
(“Generally, for contextually relevant events involving flowers in which flowers

grow, those events are quick.”)
b. e[Con(b,e) & dancing(e) & Agent(e,b)](beautiful(e))
c. e[Con(br,e) & cutting(e) & Theme(e,br)](easy(e))
This idea leads Condoravdi to a surprising conclusion, namely, that in middles, the so-
called adjunct is required because, semantically, it typically constitutes the main predi-
cation in the sentence! Far from being a modifier, the adverb or adjunct PP actually
expresses the principal assertion in the clause.9
Condoravdi’s conclusion has very interesting syntactic implications under Diesing’s
(1992) Mapping Hypothesis, which correlates hierarchical, syntactic structure with
quantificational semantic structure; generalizing slightly on her original formulation,
high material maps to the restriction and low material to the scope:
Mapping Hypothesis
Lowest material from VP is mapped to the nuclear scope. The residue is mapped
into a restrictive clause.
Note now that Condoravdi’s proposal fits the Mapping Hypothesis very neatly under a
right-descending view of adjuncts. Syntactic and quantificational parses match up with
no need for LF readjustment:
(23) VP
DP V′
these flowers V AdvP
grow quickly
Q Restriction Scope
e (Con(f,e) & growing(f,e)) (quick(e))
2.3. Because-Clauses
Condoravdi’s analysis of middles provides a first illustration of how the low position of
adverbs assigned in a right-descending theory might be understood semantically through
structured event quantification. And in fact there appear to be other cases where this
analysis is revealing. One such is the case of because-clauses, which have received two
very different semantic analyses in the literature.
2.3.1. Propositional Operators
Because-clauses (24a) have been widely analyzed semantically as propositional opera-
tors (24b), an account that fits their classic syntactic analysis as outermost TP or VP
adjuncts (shown in (24c,d), respectively) (Williams 1974; Dowty 1979; Johnston 1994).
If because-clauses apply to complete propositions, then it’s natural for them to occur
sister to elements that express complete propositions, and so on:
(24) a. John left because Mary left.

because Mary left’(^John left’)
b.
c. [TP[TP John left] [PP because Mary left]]
d. [TP John [VP[VP left] [PP because Mary left]]]
Despite its simplicity, however, the adjunct analysis leaves some puzzling, unanswered
questions. Specifically, it does not explain the information packaging of sentences with
because-clauses, and why the latter is apparently the opposite of what we find with
other adverbial clauses.
As discussed by Hooper and Thompson (1973), because-clause constructions have
the interesting property of presupposing the information expressed by the main clause
while asserting the information in the adverbial clause (25a). This contrasts sharply with
other adverbial clauses, where the converse is true. Thus Heinämäki (1978) observes
that with when/before/after-clauses, main clause information is asserted whereas adver-
bial clause information is presupposed (25b):
(25) a. John left [because Mary left].
Presupposes: John left.

Asserts:
John’s leaving was because of Mary’s leaving.
b. John left [after Mary left].
Presupposes: Mary left.

Asserts:
John’s leaving was after Mary’s leaving.
In propositional operator analyses, this property of because-clauses is either ignored or

else simply stipulated as a lexical fact about because (Johnston 1994). In fact, however,
this difference between causal and other adverbial clauses seems to be a deep one that
holds cross-linguistically.10
2.3.2. Event Relations and Quantfication

Davidson (1967a) offers an interesting alternative to the standard semantics of causal
clauses and suggests that, rather than being a propositional operator, (be)cause is in fact
a simple binary relation between events. On Davidson’s view, (26a), for example, would
be analyzed roughly as in (26b), which is read: “there is an event e that is a leaving by
John and an event e’ that caused e where e’ is a leaving by Mary.”
b.
∃e[leaving(j,e) & ∃e’[Cause(e’,e) & leaving(m,e’ )]]
The Davidsonian analysis of (be)cause has a number of salient features. For one thing,
note that because-clauses are not scopal on this view: they do not apply to, or have
scope over, VPs or TPs. Relatedly, because doesn’t relate propositions expressed by verb
phrases or sentences; rather, it relates the simple event objects that the latter describe.
Finally, and most important for our purposes, causal clauses involve quantificational
structure; because-adverbials introduce a quantification (∃e’) over events together with
the primitive cause relation.
I want to suggest that the last point provides an approach to the information packag-
ing question that eludes the propositional operator analysis. Lewis (1975) and Kratzer
(1986) have argued persuasively that if/when-clauses always restrict (explicit or covert)
adverbs of quantification, and De Swart (1993) and Johnston (1994) have extended this
general view to temporal when/before/after-clauses as well. Under the usual information
packaging of quantification, these proposals explain why the content of these adverbials
is standardly presupposed/background-entailed (27):
(27) a. John always leaves [if/when/before/after Mary leaves]
b. ALWAYS [if/when/before/after Mary leaves] [John leaves]
Q Restriction Scope
Presupposed/Background-Entailed Asserted
Suppose now we extend this idea to because-clauses, replacing the unrestricted event
quantification in (26b) with the structured/restricted event quantification shown in
(28b). Then the partitioning of information will come out correctly:
b. e [leaving(j,e)] [ e e & e e)]]

Q Restriction Scope
“For some leaving by John, it was caused by Mary leaving.”
The main clause is presupposed because it provides the restriction on the main event
quantifier. The because-clause is asserted because it constitutes the scope of the event
quantification.
The structured event quantification hypothesis fits very neatly into a right-descending
syntax under the Mapping Hypothesis, as illustrated in (29):
(29) VP
DP V′
John V PP
left because Mary left

Q Restriction Scope
e [leaving(j,e)] [ e e e e
It also permits a fresh approach to the special syntactic properties of causal modifiers. As
is well known, non-preposed causal adverbs strongly prefer an absolute sentence-final
position (cf. John will leave [tomorrow] [because Mary left] vs. ??John will leave [because
Mary left] [tomorrow]).
In a right-ascending theory, this requires an analysis as outermost adjuncts, but this
syntax is not straightforward under a propositional operator semantics since other prop-
ositional operators, for example, modals, behave differently (see Ernst 2001 for discus-
sion). In right-descending theory, the issue becomes radically recast; rather than being
outermost adjuncts, causal clauses are innermost V-complements; hence we are led to seek
some very intimate connection to the verb. Interestingly, Davidson (1967a) provides just
such a connection, arguing that the cause relation has a privileged status for events. In
brief, causal relations are for Davidson precisely what individuate and distinguish events:
events with the same causes and the same effects are the same event. If this is so, then the
intimate connection between verbs and because-clauses makes considerable sense. If verbs
are predicates of events, then because-clauses do not merely add additional information
about those events but contribute to determining what events they actually are.
2.4. Some General Lessons

We can summarize the main lessons of this section as arguing that under a semantics of
structured Davidsonian event quantification embracing the Mapping Hypothesis:
• What is called an “adjunct” may actually constitute the main predicate (middles)
• What is analyzed as scopal may be reanalyzed as predicational (because-clauses)
• What is analyzed as providing mere adjunct information may be reanalyzed as
individuating the core object of which V predicates (because-clauses)
• Right-descending syntax becomes a semantically natural structure for adverbs
(middles and because-clauses)
3. REEXAMINING ADVERBIAL “SCOPE CONTRASTS”
With these results in hand, let us now return to the analysis of sentence-final “scope”
contrasts and the crucial Andrews facts.
3.1. The Predicational Nature of Sentence-Finals Adverbs

Adverbs like intentionally in (30a) and twice in (30b) have been widely assumed to
be scopal given that intentionally resembles intend, a scopal, intensionality-inducing
element (31a), and given that twice resembles a temporal quantifier, which can take a
restriction and count over times or events (31b).
(30) a. John knocked on the door intentionally.
b. John knocked on the door twice.
{
(31) a. John intended to knock on the door .
to visit Santa Claus
{
b. Twice [when he was standing there] John knocked on the door.
In fact, however, the scopal operator status of these items is far from clear.
3.1.1. Sentence-Final Intentionally as an (Event) Predicate
English does contain clearly intensional adverbs such as purportedly, supposedly, and
allegedly. These occur sentence-initially (32a,b) and sentence-medially (32c), but never
sentence-finally (without a pause) (32d). They exhibit the usual diagnostics for intension-
ality insofar as nondenoting terms do not necessarily induce falsity (32a), substitution
can fail in their scope (32b), and indefinites in their domain can receive a nonspecific
interpretation (32c):
(32) a. Purportedly/Supposedly/Allegedly John met Santa Claus.
Purportedly/Supposedly/Allegedly John met Boris Karloff/Bill Pratt.

b.
c. John purportedly/supposedly/allegedly married a Norwegian.
d. John met Santa Claus *(,) purportedly/supposedly/allegedly.
Interestingly, intentionally, deliberately, voluntarily, and so on pattern quite differently.

These forms never occur sentence-initially (33a). They always yield falsity with nonde-
noting terms (33b), and their status with respect to other two diagnostics for intension-
ality is unclear at best (33c,d).
(33) a. *Intentionally/Deliberately/Voluntarily John insulted Boris.
b. #John met Santa Claus intentionally/deliberately/voluntarily.
c. John pushed Boris Karloff/Bill Pratt intentionally/deliberately/voluntarily.
d. John married an actress intentionally/deliberately/voluntarily.
Furthermore, there is another crucial difference: whereas the adjectival counterparts

of purportedly, supposedly, and allegedly are neither predicative nor intersective (34),
as we would expect of scopal elements, the adjectives corresponding to intentionally,
deliberately, and voluntarily are in fact both predicative and fully intersective (35):
(34) a. ??John’s actions were purported/supposed/alleged.
b. John’s painting is a purported/supposed/alleged forgery.
≠ John’s painting is purported/supposed/alleged, and it is a forgery.
(35) a. John’s actions were intentional/deliberate/voluntary.
b. That was an intentional/deliberate act of aggression against Iraq.
(That was intentional/deliberate, and it was an act of aggression against Iraq.)
c. That was a voluntary sacrifice.
(That was voluntary, and it was a sacrifice.)
Finally, Jackendoff (1972) observes that whereas sentence-medial intentionally is sub-

ject-oriented (i.e., intention is ascribed to the referent of whatever DP occupies the
subject position) (36a,b), sentence-final intentionally is not; with the latter, intention is
ascribed uniformly to the semantic agent (36c,d):
(36) a. Joe intentionally/deliberately/carefully has seduced Mary.
b. Mary has intentionally/deliberately/carefully been seduced by Joe.
c. Joe has seduced Mary intentionally/deliberately/carefully.
d. Mary has been seduced by Joe intentionally/deliberately/carefully.
I propose that these points can be drawn together, basically by accepting Jackendoff’s view
that sentence-final adverbs like intentionally are not scopal operators but rather man-
ner adverbs. Specifically, within the Davidsonian event semantics adopted here, I suggest
that intentionally (and related forms) are simple predicates, true of a particular class of
events, that is, actions (events having agents)—and that the truth of the predication car-
ries an entailment of intention. On this proposal, (37a) is interpreted as in (37b), where
intentional(e) carries the further entailment in (37c). Here the exact propositional content
of the agent’s intent is left vague, and intentionality is associated with the agent.
(37) a. John knocked on the door intentionally.
∃e [knocking(j,d,e) & intentional(e)]

b.
c. intentional(e) ⇒ the agent of e intended to bring e about
This explains the basic predicational character of intentionally, as revealed by (35), and
also why (36c,d) do not show subject orientation: intention is attributed to the agent of
e no matter how the latter is expressed syntactically.
3.1.2. Sentence-Final Twice as a (Quantity) Predicate

Twice can also, I believe, be given a predicational analysis. Twice is plainly related to the
form two, which has been analyzed not only as a quantifier (38a) but also as a quantity
predicate applying to pluralities (38b,c). It is natural to extend this idea to twice, and
to other sentence-final frequency adverbs like frequently/often and rarely/infrequently,
which correspond to the quantity predicates many and few (39):
(38) a. Two of the men were present.
b. Two men were in the garden.
c. The problems are two (in number).
(39) a. Many/Few men were in the garden.
b. Our problems are many, and our solutions are few.
Indeed, if we adopt a strictly predicational, nonquantificational view of sentence-final

frequency adverbs, then we can explain why unambiguously quantificational forms
are forbidden from final position (40), and why forms like often that are ambiguous
between a quantificational and a predicational reading when they occur mid-sentence
(41a), always lose their quantificational meaning when they appear finally (41b):
b.
{
(40) a. John knocked on the door twice/often/frequently/rarely/infrequently.
*always/*mostly.
(41) a. Texans often eat barbeque.
i. ‘Many Texans eat barbeque.’
ii. ‘In general, for a Texan, his/her barbeque-eatings are many.’
b. Texans eat barbeque often.
‘In general, for a Texan, his/her barbeque-eatings are many.’
The conclusion I draw from this is that sentence-final twice is not clearly quantifi-
cational. More specifically, I wish to propose that sentence-final twice is a quantity
predicate applying to pluralities of events—a measure adverb in the sense of Larson
(2003)—and that examples like (42a) be analyzed along the lines of (42b), which is
read as follows: “there was an event E whose subevents e were knockings on the door
by John, and that event E was two in number.”
(42) a. John knocked on the door twice.
b.
∃E [∀e[ Ee → knocking(j,d,e)] & two(E)]
3.2. Andrews’s Examples

3.2.1. Intentionally-Twice
Under the results derived above, (43a), our Andrews example with intentionally-twice, will
have the logical form in (43b), which is read as follows: “for some event E, whose subevents e
were all intentional knockings-on-the-door-by-John, E was two in number/binary.” I assume
that this is also the correct logical form for (43c), whose predicational syntax is transparent:
b.
∃E[∀e[Ee → knocking(j,d,e) & intentional(e)]] (two(E))
c. [John’s intentional knockings on the door] were two (in number).
Compare the VPs of (43a) and (43c), now assuming a right-descending syntax:
(44) VP
DP V′
John V VP
knock PP V′
on the door V VP
knock AdvP V′
intentionally V AdvP
knock twice
E [ e[Ee knocking(j,d,e) & intentional(e)]] (two(E))
Q Restriction Scope
(45) VP
DP V′
knockings on the door V AP
be two
E [ e[Ee knocking(j,d,e) & intentional(e)]] (two(E))
Q Restriction Scope
In both cases we appear to get the correct association between quantification, restric-
tion, and scope.
3.2.2. Twice-Intentionally
The second Andrews case, with twice-intentionally (46a), will have the logical form in
(46b), which is read as follows: “for some event E, whose subevents e were all knockings-
on-the-door-by-John and which was two in number/binary, it was intentional.” Again,
I assume that this is also the correct logical form for (46c), whose syntax is overtly
predicational:
(46) a. John knocked on the door twice intentionally.
b.
∃E[∀e[Ee → knocking(j,d,e)] & two(E )] (intentional(E ))
c. [John’s double-knock on the door] was intentional.
Compare the VPs of (46a) and (46c), assuming a right-descending syntax:
(47) VP
DP V′
John V VP
knock PP V′
on the door V VP
knock
AdvP V′
twice V AdvP
knock intentionally
∃E [∀e[Ee → knocking(j,d,e)] & two(E)] (intentional(E))
Q Restriction Scope
(48) VP
DP V′
-knockon the door V AP
be intentional
E [ e[Ee knocking(j,d,e)] & two(E )] (intentional(E))
Q Restriction Scope
Again, the correct information packaging seems to be derived.

These results obtained above for Andrews’s two-adverb cases appear to be exten-
sible to more complex examples and, in particular, seem to be compatible with Phillips’s
(2003) observation that in sequences of three adverbials, the “scope” relations between
the first two become fluid. Consider a pair of Philips’s examples like (49); the first, with
two adverbs, asserts that Sue’s multiple-kissing of him was intentional, not that each
of many individual kissings was intentional. By contrast, in (49b), with three adverbs,
either of these two interpretations seems possible. In the current framework we may
derive this result as in (50), where the large event E is many in number, and where either
the whole big event is intentional (50a) or where each individual subevent is (50b). This
latter possibility arises because in the three-adverb case, intentional occurs within the
scope of the distributive quantifier over individual events (∀e). This is not true in the
logical form for (49a), however.
(49) a. Sue kissed him [many times] [intentionally].
b. Sue kissed him [many times] [intentionally] [in front of the boss].
(50) a. ∃E[∀e[Ee → kissing(j,d,e)] & many(E ) & intentional(E))] (i-f-o-t-b(E))

b.
∃E[∀e[Ee → kissing(j,d,e) & many(E) & intentional(e))]] (i-f-o-t-b(E))
In brief, then, all of the crucial Andrews data, together with the more complex examples
that Phillips cites, appear to fit into a right-descending analysis.
3.3. VP-Initial Adverbs

This account raises an interesting question for VP-initial adverbs like (51a,b) and
sentence-initial cases like (51c). How should they be analyzed?
(51) a. John intentionally knocked on the door.

b. John twice knocked on the door.
Twice John knocked on the door.

c.
What I would like to (tentatively) propose is that intentionally (like similar adverbs) is
ambiguous in (51a) between a simple event predicate and a scopal operator, but that twice
in (51b,c) is a true quantifier and thus has only the reading in (52b), not that in (52a):
(52) a. ∃E [∀e[Ee → knocking(j,d,e)]] (two(E)) Quantity Predicate
b. 2e[knocking(j,d,e)] True Quantifier
If this proposal is correct, it makes several interesting predictions. First, it predicts

that (53a) will be understood as in (53b), which is paraphrased as in (53c), not as in
(53d). That is, intentionally will be understood as true of individual knocks, not of the
whole collective knocking.
(53) a. John twice knocked on the door intentionally.
b. 2e[knocking(j,d,e) & intentional(e)]
c. John’s two knocks on the door were intentional.
d. John’s double-knock on the door was intentional.
This prediction seems correct, and in fact Andrews (1983) noted it and took it as
reflecting a strong preference for a preverbal adverb to take scope over a postverbal
one. But Andrews could not explain this preference under his own syntactic assump-
tions since both representations (54a) and (54b) should be legitimate adjunction struc-
tures for him:
(54) a. [VP twice [VP [VP knocked on the door] intentionally]]
b. [VP [VP twice [VP knocked on the door]] intentionally]
If I am correct, this fact has nothing to do with scope in Andrews’s sense.

Another prediction is that in initial position, where it can be a quantifier, twice
should be able to take a when-clause as a restriction, whereas in final position,
where twice must be a quantity predicate, it should not. Again, this prediction
seems to be correct. According to my intuitions, (55a) can describe visits by Mary
to the Louvre on two separate occasions of her being in Paris. By contrast, (55b)
must describe a single occasion of Mary being in Paris, during which she two times
visited the Louvre:
(55) a. Twice when she was in Paris Mary visited the Louvre.

Can mean: ‘On two separate occasions of being in Paris, Mary visited the Louvre
during those occasions.’ (when-clause restricts twice)
b. Mary visited the Louvre twice when she was in Paris.

Must mean: ‘On one occasion of being in Paris, Mary visited the Louvre twice during
that occasion.’ (when-clause frames twice)
In the first case the when-clause functions to restrict the frequency adverb twice; in the
second case it merely frames the adverb, providing the domain of quantification.
4. CONCLUSION
In this paper I have argued for a new picture of the semantics of sentence-final adverbs,
one in which these elements are not scopal VP-functors stacked up on the edge of
the sentence or verb phrase, but rather event predicates, arranged within a structured
Davidsonian event quantification for the clause. I have shown how this quantifica-
tional structure is motivated by a number of interesting constructions (focus, middles,
because-clauses), and how it corresponds directly with a right-descending syntax for
adverbial attachment under the Mapping Hypothesis of Diesing (1992). I have further-
more tried to show how this proposal accounts for certain well-known phenomena
traditionally ascribed to differential adverb scope. Under the analysis proposed here,
the phenomena are revealed as predicational, not scopal at all. Furthermore, the cru-
cial examples are brought into very close alignment with nonadverbial cases that have
transparently predicational structure. Finally, I have briefly and tentatively considered a
number of small auxiliary predictions of the account, although I think the latter are far
from exhausting its potential consequences.
NOTES
1. Kayne’s (1994) Antisymmetry Theory is one particular instance of a right-descending the-

ory, in which right descent follows from the so-called Linear Correspondence Axiom. Kayne
(1994) does not discuss the grammar or interpretation of adverbs, however.
2. This result has led some to attempts at proposals in which both accounts are true, and adverbi-
als are given simultaneous ascending and descending analyses (Pesetsky 1995; Cinque 2002).
3. The force of this argument has been questioned on grounds that c-command can be replaced
by an alternative formal notion in the account of licensing. Ernst (1994, 2001), for example,
suggests that by substituting a structural licensing relation requiring m-command and linear
precedence by the trigger, the well-formedness of (4a) can be accommodated in a right-
ascending theory, and hence that a right-descending analysis has no advantage here.
This response tacitly assumes that c-command and m-command (plus precedence) stand
on equal footing, and hence that substituting one for the other carries no benefit or penalty.
This assumption is false, however. C-command and m-command are not on equal footing,
and theories appealing to one versus the other are not equally valued. As Reinhart (1983)
points out, in a theory of the syntax-semantics interface that embraces local compositionality,
the notion of c-command maps directly to the notion of scope: phrases have their c-command
domain as their semantic scope. This is crucial in the analysis of negative polarity items since
the latter must both be in the structural domain of a syntactic trigger and be in the semantic
scope of a negative. M-command and other, alternative licensing conditions do not deliver
this result; they do not correlate with semantic scope. For more on this see note 5.
4. Bach himself draws a very different (and rather melancholy) conclusion from these facts, namely,
that in some languages the processes building words and sentences are entirely independent. The
Mirror Principle fails to apply because there simply is no relation between the two.
Bach’s conclusion seems hasty, however. First, it does not appear to do justice to the
facts. In Haisla, word form and sentence form are not uncorrelated, as the “independent”
theory would predict. The two are inverses, and this inverse patterning seems something that
should be explained. Second, and more important for our purposes, Bach’s conclusion turns
on an assumption that is precisely at issue here: namely, that the composition of phrasal
arguments and modifiers is right-ascending.
5. Notice that for this result, a right-ascending account must assume that m-command plus
precedence do not determine scope on pain of getting exactly the wrong result: inner adverbs
taking scope over outer ones. This means that a right-ascending theory requires two distinct
notions: m-command plus precedence and c-command. This result is particular onerous for
the case of negative polarity items, which must be in the semantic scope of their syntactic
trigger; a right-ascending theory requires problematic “LF adjustment” in such cases.
6. See Jackendoff (1972), Rochemont (1985), and Rooth (1985, 1992) for representative work
on focus.
7. The focus representations in (17) and hereafter are simplified; spelled-out versions would
include the restriction as part of the assertion: ∃e [kissing(e) & Agent(e,m)](kissing(e) &
Agent(e,m) & Patient(e,j)): “there is a kissing by Mary that is a kissing by Mary of John.”
The simplification is harmless here.
8. See Diesing (1992), Heim and Kratzer (1997), and Herburger (2000) for discussion of the
presuppositional contribution of quantificational structure.
9. The adjunct predicate requirement in middles is sometimes questioned on the basis of exam-
ples like (ia,b) involving focus or (iia,b) with negation:
(i) a. FLOWERS grow.
b. Ballerinas DANCE.
(ii) a. These flowers (just) don’t grow.
b. This bread (just) doesn’t cut.
In fact, these examples pose no threat to Condoravdi’s analysis. Focused middles can be
analyzed along the lines in (iii), where (following Herburger) we appeal to structured event
quantification and where the focused item becomes the scope of the generic quantifier.
Similarly, examples with negation can be analyzed as instances of “verum focus” (focus on
truth-value) (iva–d).
(iii) a. Γe[Con(f,e) & growing(e)] (Theme(f,e))
b. Γe[Con(f,e) & Agent(b,e)] (dancing(e))
(iv) a. Whatever you do with them, these flowers don’t grow.
b. Γe[ Con(f,e)] (¬growing(f,e)).
c. Whatever you do to it, this bread doesn’t cut.
d. Γe[ Con(b,e)] (¬cutting(b,e))
10. See Sawada and Larson (2004) for recent discussion.
REFERENCES
Alexiadou, A. (1994) Issues in the Syntax of Adverbs. Doctoral dissertation, Universität Potsdam.
———. (1997) Adverb Placement; A Case Study in Antisymmetric Syntax. Amsterdam: John Ben-
jamins.
———. (2002) “The Syntax of Adverbs: Puzzles and Results,” GLOT 6(2/3): 33–54.
Anderson, S. (1992) A-Morphous Morphology. Cambridge: Cambridge University Press.
Andrews, A. (1983) “A Note on the Constituent Structure of Modifiers,” Linguistic Inquiry 14:
695–697.
Bach, E. (1996) “The Grammar of Complex Words,” in A.-M. di Sciullo, ed., Configurations. (pp. 1–36)
Somerville, MA: Cascadilla Press.
Baker, M. (1985) “The Mirror Principle and Morphosyntactic Explanation,” Linguistic Inquiry
16: 373–415.
Cinque, G. (1999) Adverbs and Functional Heads. New York: Oxford Universitry Press.
———. (2002) “Complement and Adverbial PPs: Implications for Clause-Structure,” unpublished
abstract, University of Venice.
Condoravdi, C. (1989) “The Middle: Where Semantics and Morphology Meet,” in MIT Working
Papers in Linguistics, Vol. 11, Cambridge, MA: MIT.
Davidson, D. (1967a) “Causal Relations,” Journal of Philosophy 64: 691–703.
———. (1967b) “The Logical Form of Action Sentences,” in N. Rescher, ed., The Logic of Deci-
sion and Action (pp. 81–120). Pittsburgh: University of Pittsburgh Press.
De Swart, H. (1993) Adverbs of Quantification: A Generalized Quantifier Approach. New York:
Garland.
Diesing, M. (1992) Indefinites. Cambridge, MA MIT Press.
Dowty, D. (1979) Word Meaning in Montague Grammar. Dordrecht: Reidel.
Ernst, T. (1994) “M-Command and Precedence,” Linguistic Inquiry 25: 327–335.
———. (2001) The Syntax of Adjuncts. Cambridge: Cambridge University Press.
Heim, I. and A. Kratzer (1997) Semantics in Generative Grammar. Oxford: Blackwell.
Heinämäki, O. (1978) Semantics of English Temporal Connectives. Doctoral dissertation,
University of Texas-Austin.
Herburger, E. (2000) On What Counts. Cambridge, MA MIT Press.
Higginbotham, J. (1989) “Elucidations of Meaning,” Linguistics and Philosophy 12: 465–517.
Hooper, J. and S. Thompson (1973) “On the Applicability of Root Transformations,” Linguistic
Inquiry 4: 465–497.
Jackendoff, R. (1972) Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press.
Johnston, M. (1994) The Syntax and Semantics of Adverbial Adjuncts. Doctoral dissertation,
University of California-Santa Cruz, Santa Cruz.
Kayne, R. (1994) The Antisymmetry of Syntax. Cambridge, MA MIT Press.
Kratzer, A. (1986) “Conditionals,” in A. M. Farely, P. Farely, and K.E. McCullough, eds., Pro-
ceedings of the Chicago Linguistic Society 22: Papers from the Parasession on Pragmatics and
Grammatical Theory (pp. 115–135). Chicago: Chicago Linguistics Society.
Laenzlinger, C. (1996) “Adverb Syntax and Phrase-Structure,” in A.-M. di Sciullo, ed., Configura-
tions (pp. 99–127). Somerville, MA: Cascadilla Press.
Larson, R. (1988) “On the Double Object Construction,” Linguistic Inquiry 19: 335–391.
———. (1989) “Light Predicate Raising,” Lexicon Project Working Papers No. 27. Cambridge,
MA: Center for Cognitive Science, MIT. Republished in this volume.
Larson, R. (1991) “Double Objects Revisited: Reply to Jackendoff,” Linguistic Inquiry 21: 598–632.
———. (2003) “Time and Event Measure,” in J. Hawthorne and D. Zimmerman, eds., Philosophi-
cal Perspectives, 17, Language and Philosophical Linguistics (pp. 257–258). Oxford: Blackwell.
Lewis, D. (1975) “Adverbs of Quantification,” in E. Keenan, ed., Formal Semantics of Natural
Language (pp. 3–15). Cambridge: Cambridge University Press.
Montague, R (1974) Formal Philosophy. New Haven, CT Yale University Press.
Parsons, T. (1990) Events in the Semantics of English. Cambridge, MA: MIT Press.
Phillips, C. (2003) “Linear Order and Constituency,” Linguistic Inquiry 34: 37–90.
Reinhart, T. (1983) Anaphora and Semantic Interpretation. Chicago: Chicago University Press.
Rochemont, M. (1985) Focus in Generative Grammar. Amsterdam: John Benjamins.
Rooth, M. (1985) Association with Focus. Doctoral dissertation, University of Massachusetts,
Amherst.
Rooth, M. (1992) “A Theory of Focus Interpretation,” Natural Language Semantics 1: 75–116.
Sawada, M. and R. Larson (2004) “Presupposition and root transforms in adjunct clauses,” in
M. Wolf and K. Moulton, eds., Proceedings of NELS 34 (pp. 517–528). Amherst, MA: GLSA
University of Massachusetts-Amherst.
Stroik, T. (1990) “Adverbs as V-Sisters,” Linguistic Inquiry 21: 654–661.
———. (1992a) “Adverbs and Antedecedent-Contained Deletion,” Linguistics 30: 375–380.
———. (1992b) “On the Distribution of Temporal and Locative,” The Linguistic Review 9: 267–284.
———. (1996) Minimalism, Scope and VP-Structure. Thousand Oaks, CA: Sage.
Thomason, R. and R. Stalnaker (1973) “A Semantic Theory of Adverbs,” Linguistic Inquiry 4:
195–220.
Williams, E. (1974) Rule Ordering in Syntax. Doctoral dissertation, MIT, Cambridge, MA.
Nominal Structure: Background
Linguistic theory in the 1980s witnessed two decisive developments in the analysis of
nominals. The first came in formal semantics. In 1981 Barwise and Cooper’s “General-
ized Quantifiers and Natural Language” appeared, followed closely by related work
from Higginbotham and May (1981) and significant extensions by Keenan and Stavi
(1983). The second development was in syntax. In 1987 Stephen Abney completed his
MIT thesis The English Noun Phrase in Its Sentential Aspect, proposing the so-called
DP-analysis of nominals, following related work by Szabolcsi (1983) and Fukui and
Speas (1986).
Both lines of research had enormous impact in their respective areas. Generalized
quantifier (GQ) theory prompted extensive inquiry into quantifier types, cross-linguistic
universals in determiner semantics, and quantification in categories outside the deter-
miner system (AdvPs, PPs, comparatives). It was also a crucial reference point in the
debates of the 1980s and 1990s regarding the quantificational status of indefinites and
the analysis of “donkey anaphora,” following work by Kamp (1981) and Heim (1982).
Equally, the DP hypothesis prompted an explosion of work, not only in the syntax of
nominals proper and its relation to sentence structure, but in the broader domain of
so-called “functional categories,” which multiplied very rapidly in the ensuing period
and which today form an important part of modern syntactic theorizing. Nonetheless,
despite these significant developments, no integration of the two lines of inquiry was
attempted at the time. Indeed, to the best of my knowledge, the basic question was never
posed, to what extent is the GQ semantics for determiners compatible with the syntactic
picture of DP introduced by Abney?
At MIT in the late 1980s, I was well placed to appreciate the relation, and ulti-
mately the tension, between these two research programs. I had been a PhD student
at the University of Wisconsin–Madison and had attended the seminars in which
the GQ analysis of natural language determiners had first been publicly presented.
Robin Cooper had been my thesis adviser and Jon Barwise a member of my thesis
committee. At MIT, Stephen Abney was my first PhD student, and working through
his thesis material had afforded me a close look at the syntactic issues involved. By
1988 I had concluded that the semantic and syntactic lines of research, although
superficially convergent in their view of D as the head of the nominal, were, in
fact, incompatible at a deeper level. Specifically, GQ theory clashed with a central
assumption about D made by Abney (and many others subsequently), an assump-
tion Abney extended to functional categories generally, that is, that D is without
semantic content—specifically, that D lacks argument structure. By contrast, GQ
theory implied a rich view of argument structure for D, which in turn suggested a
very different picture of DP than the one that has been widely assumed in syntactic
theory following Abney’s work.
1. THE DP–IP ANALOGY
A central claim of Abney (1987) is that nominals are headed by their determiner element
D, and hence belong to the category DP. This proposal is meant to capture, among
other things, certain important distributional similarities between sentences and nomi-
nalizations, recognized in the generative tradition since at least Lees (1968).1 By the
late 1980s, clauses had widely come to be regarded as projections of their inflectional
element I(nfl), which incorporated tense and agreement.2 Abney argued that if D plays
a role in the nominal similar to I in the clause, then a strikingly parallel view of their
syntax becomes possible. The subject of IP in sentential structure and the “subject” of
DP in nominalizations, and the predicates that they relate to (VP and NP, respectively),
all become symmetric counterparts (1a,b):
(1) a. IP b. DP
DP I′ DP D′
John I VP John’s D NP
[TNS] complete the plan [AGR] completion of the Plan
The IP(TP)–DP analogy proved irresistible for many researchers working on nominal
structure and exerted an enormous influence on thinking in the field thereafter.3
A key feature of Abney’s (1987) analysis of clauses and nominals is its assimilation of
D and I to the class of functional elements, which are to be distinguished from familiar
lexical elements like nouns, verbs, and adjectives. Abney characterizes functional ele-
ments by a range of diagnostic criteria:
• They belong to closed lexical classes.

• They are generally phonologically and morphologically dependent.
• They permit only one complement, usually not an argument.
• They are usually inseparable from their complement.
• They lack descriptive or semantic content.
These criteria are not of equal significance, however. Abney (1987, p. 65) states: “The final
characteristic . . . is in some sense the crucial characteristic.” The core fact about functional
categories is thus that they lack the kind of semantic structure that would support their
projection like normal, lexical elements. Instead, their integration into syntax requires a
separate notion of selection—“functional selection”—not based on thematic role assign-
ment, valence, or any of the notions familiar from the projection of semantically rich lexical
elements like verbs. For Abney, D and DP are simply not to be understood in such terms.
2. GQ THEORY: THE RELATIONAL VIEW OF DETERMINERS
Abney’s view of D as a semantically empty “functional element” clashes sharply with

that of GQ theory, which is based on the the “relational view of determiners”:
Relational View of Determiners:
Determiners express relations among predicate meanings.

This idea descends from Aristotelian logic and is advanced in a particularly clear way
by Frege (1953), who suggests that in quantified examples like (2a), all can be seen as
expressing a subordination of the concept ‘whalehood’ to that of ‘mammalhood’. On
this view, the logical form of (2a) is approximately as in (2b), where ALL corresponds
to the subordination relation. (2a) is thus true just in case being-a-whale is subordinate
to being-a-mammal so that all individuals possessing the first property possess the
second.
(2) a. All whales are mammals.
b. ALL - ‘whalehood’ - ‘mammalhood’
GQ theory reformulates Frege’s picture in extensional terms, replacing concepts with

sets and replacing concept relations with set relations. Thus ‘whalehood’ and ‘mammal-
hood’ are replaced with the corresponding sets of individuals (3a), and the subordina-
tion relation between concepts is replaced with the subset relation between sets (3b).
This yields (correct) truth conditions for (2a) as in (3c):4
(3) a. {x: x is a mammal}, {x: x is a whale}
all (X,Y ) iff Y ⊆ X

b.
c. {x: x is a whale} ⊆ {x: x is a mammal}
This analysis can be extended to a wide range of determiners (4). And indeed the exten-
sion even includes quantifiers like most (4f), which are well known to escape first-order
treatment (Barwise and Cooper 1981):
(4) a. some(X,Y ) iff Y ∩ X ≠ Ø
no(X,Y ) iff Y ∩ X = Ø
b.
the(X,Y ) iff Y ⊆ X & |Y | = 1

c.
both(X,Y ) iff Y ⊆ X & |Y | = 2

d.
neither(X,Y ) iff Y ∩ X = Ø & |Y | = 2

e.
most(X,Y ) iff |Y ∩ X | > |Y − X |

f.
2.1. Determiners and Valence

Under the relational analysis, determiners are semantically contentful and indeed can be
viewed as possessing argument structure and valence, counterpart to verbs. Typical Ds
like all or most express binary relations—they select two arguments (cf. 3b and 4a–f).
Essentially, they have the status of transitive predicates, albeit transitive predicates of
sets rather than transitive predicates of individuals.
The notion of valence in D can be generalized further. Following an old proposal by
Postal (1969), pronouns can be regarded as intransitive determiners—Ds that take only
a single set argument. Assuming pronouns to bear an index (n) that is mapped by an
assignment function (g) to an individual, a pronoun is true of all sets X of which g(n)
is a member (5):
(5) Intransitive Ds (Pronouns)
HEn(X) iff g(n) ∈ X
Likewise, it is coherent to talk about ditransitive determiners. Keenan and Stavi (1983)
offer as cases in point comparative determiners like more-than and as-many-as and deter-
miners with exception phrases like every-except or none-but. Intuitively, a sentence like
(6a) with more-than expresses a relation between three sets—smokers (X), men (Y), and
women (Z). The sentence is true if the size of Y ∩ X (the men who are smokers) is greater
than the size of Z ∩ X (the women who are smokers). Similarly, the truth conditions of
(6b) involve three sets: smilers (X), boys (Y), and the set containing John and Max ({j,m}).
The sentence is true if Y − {j,m} (the boys minus John and Max) is a subset of X (the
smilers), and if {j, m} does not intersect X (if John and Max are not smilers themselves).
(6) a. More men than women smoke.
Every boy except John and Max smiled.

b.
We can formalize these observations for ditransitive determiners as in (7a–d), where

X is the set contributed by the main predicate, Y is the set contributed by the nominal
complement of D, and Z is the set contributed by the than- or except-phrase:
(7) Ditransitive Ds
a. MORE-THAN(X,Y,Z) iff |Y ∩ X| > |Z ∩ X|
b. AS-MANY-AS(X,Y,Z) iff |Y ∩ X| ≥ |Z ∩ X|
c. EVERY-EXCEPT(X,Y,Z) iff ALL(X, (Y−Z)) & NO(X,Z)
d. NO-EXCEPT(X,Y,Z) iff NO(X, (Y−Z)) & ALL(X,Z)
2.2. Thematic Roles for Determiners

The semantic parallels between V and D can be further extended to notions of thematic
roles and thematic hierarchy, as discussed in Larson (1991, republished in this volume).
Canonically, verbs describe events, with concepts like agent, theme, goal, and so on
representing the recurring semantic/functional roles that verbal arguments play with
respect to those events. These roles also seem to be implicated in recurrent syntactic
generalizations, for example, the fact that agents are typically realized as subjects, that
agents and themes typically show evidence of being projected higher in syntactic struc-
ture than oblique elements, and so on.
Thinking analogically with determiners, it seems the latter canonically express quan-
tifications, with concepts like restriction and scope representing the two main recurring
semantic/functional roles that set arguments play in quantification. Semantically, the
restriction fixes the domain of quantification, whereas the scope determines what is
true of the individuals in that domain. Syntactically, restriction and scope also appear
relevant in mapping the parts of DP. The former role is mapped to the NP complement
of D. The latter role is associated with a main clause predication.
Given these points it seems plausible to consider the notions of scope and restriction
as the counterparts of verbal thematic roles: θRESTRICT and θSCOPE. Expanding on this,
since triadic determiners allow for a third set argument, typically introduced by an
oblique, preposition-like element such as than, as, or except, it also seems plausible to
recognize certain oblique thematic roles for predicate arguments, beyond θRESTRICT and
θSCOPE. For want of a better term, we might simply group these roles together as θNOBLIQUE
(“nominal oblique”), counterpart to the oblique thematic roles found in the verbal sys-
tem. The resulting picture is therefore as in (8):
(8) θ-roles and Thematic Hierarchy for V and D:
V: θAGENT > θTHEME > θOBLIQUE
D: θSCOPE > θRESTRICT > θNOBLIQUE
3. PROJECTING DP
The strong resemblance between V and D at the semantic-thematic level suggests an

approach to nominal syntax quite different from the one pursued in Abney (1987), which,
given its insistence on the “functional” status of D, requires nonthematic (and frankly
stipulative) notions like “functional selection” as a basis of structure projection. As dis-
cussed in the General Introduction, the Larson (1988) account of VP projection was a
category-neutral one. Given a set of thematic roles arranged in a hierarchy, there is essen-
tially no barrier to projecting a head assigning those roles in the same way that verbal
heads are projected. In my fall 1988 lectures at MIT, I experimented with the projection
of DP and DegP along precisely these lines, employing the thematic roles for quantified
elements posited above and the hierarchy in (8). The results were subsequently written up
as “The Projection of DP (and DegP)” (Larson 1991), included in this volume.
Without rehearsing the details, VP and DP were argued to exhibit close structural
parallelism. Specifically, pronouns (he, them, etc.) were analyzed as the counterparts of
unergative verbs, taking only a scope argument (9a,b) as their subjects:
(9) a. VP b. DP
DP V Pro D
John laughed he
Standard binary quantifiers (all, most) were counterpart to transitive verbs, taking both
a scope argument and a restriction (10a,b):
(10) a. VP b. DP
DP V′ Pro D′
John V DP D NP
saw Mary all women
Finally, comparatives (more) and quantifiers with exception phrases (every-except) were
counterpart to ditransitive verbs, taking a scope argument, a restriction, and an oblique
complement (11a,b):
(11) a. VP b. DP
DP V′ Pro D′
John V VP D DP
put DP V′ more NP D′
the key V PP men D PP
put on the table more than women
Each of the DP structures involves a null anaphoric element Pro in subject position whose
value was analyzed as given by the sister of the quantified DP at Logical Form(LF), after
Quantifier Raising has applied. Larson (1991) argues that the postulation of Pro is nec-
essary to resolve an antinomy of θ-theory that would otherwise arise under GQ theory,
wherein D selects a predicate whose head simultaneously selects DP.5 The existence
of this external subject argument for determiners entails the existence of DP shells for
ditransitive determiners equivalent to VP shells for ditransitive verbs. Assuming put and
more in (11) to both determine three thematic roles, only two of these can be assigned
within a single maximal X-bar projection VP/DP. Furthermore, the thematic hierarchy
requires the oblique arguments to be projected lowermost (cf. (8), above). A “minimal
structural elaboration” is therefore necessary to accommodate the final argument in
specifier position, with subsequent raising of the head (V/D). In the case of the verb put,
this final argument is the agent phrase John. In the case of the determiner more, the final
argument is the scope phrase Pro. The results are thus entirely parallel.
As discussed in detail in the General Introduction, the original shell-theoretic view of
projection involving templatic X-bar theory and the accompanying notion of “minimal
structural elaboration” are not formulable within current syntactic views. Nonetheless,
the updated framework for shell projection outlined above appears adequate to revise
the analysis of DP as well. In the remainder of this section I briefly sketch how this
revision might proceed, reasoning in parallel with our considerations from the verbal
domain. I close in section 4 with some intriguing general questions, including ones aris-
ing with respect to DP interpretation.
3.1. θ-Features and θ-Agreement: Determiner–Restriction Term

We earlier explored the general view of Hornstein (1999) that θ-roles be formalized as
θ-features and that θ-role assignment be understood as θ-feature agreement under the analysis
of agreement developed by Pesetsky and Torrego (2004). On analogy with verbal θ-features
like [ag] and, [th], suppose then that we have quantificational θ-features [scp] (“scope”),
[res] (“restriction”), [nob] (“nominal oblique”), and so on. Assume also that the selection
relation between a determiner D and its NP complement is θ-feature agreement (12):
(12) DP
every man
[RES[ ]] AGREE [RES[ ]]
...
Reasoning as in the verbal case, where θ-features are interpretable on the arguments of
a head, we conjecture the situation in (13a), since the nominal (NP) is an argument of
the head D under GQ semantics. Compare (13b):
(13) a. DP b. VP
every man kiss Mary

[uRESval[1]] [iRES[1]] [uTHval[1]] [iTH[1]]
... AGREE! ... AGREE!
(13a) also adopts the view that [res] is valued on the determiner head, just as [th] is
valued on a transitive verb with an internal argument (13b). I will further extend the
parallel between [res] in D and [th] in V by assuming that [res] is in fact the default
valued θ-feature on any D-head bearing it.
3.2. θ-Features and θ-Agreement: Determiner–Scope Term

Every is not an intransitive determiner taking only a restriction. Like all Ds, every
also has a scope argument; that is, the θ-set of every includes [scp]. As noted above,
in Larson (1991) the scope argument of a determiner is assumed to be syntactically
realized by a special anaphoric element Pro, whose value is given by the sister of the
quantified DP at LF, after DP has raised by Quantifier Raising.
Suppose we wish to merge Pro with structure (13) as its scope term. Pro should
bear the interpretable θ-feature [iscp]. If the constraint on bearing at most one val-
ued feature holds uniformly for all heads, and hence for D as well as V, then (14)
faces the same problem we encountered earlier in the verbal domain. If we merge
Pro directly, the result is an unvalued θ-feature, and hence an interface-illegible
object:
(14) DP
Pro D′
[iSCP[2]]
every man
UNVALUED! [uSCP[2]] [iRES[1]]
[uRESval[1]]
Suppose then that we have access to light d, fully analogous to little v, with the follow-
ing properties:
Light d:
Bears a strong D feature. Bears an EPP feature. Bears one valued occurrence of a
θ-feature unvalued on a D with which it was co-selected.
This permits a numeration for every man that includes a light d carrying [uscpval[ ]]. We
can merge this element with (13a); D raises and d-D agree, yielding (15).
(15) dP
d DP
AGREE!
d every every man
[uSCPval[2]] [uSCP[2]] [uSCP[ ]] [iRES[1]]
[uRESval[1]] [uRESval[1]]
Pro can now merge with (15), satisfying d’s EPP feature. Unvalued [iscp[ ]] on Pro
probes valued [uscpval[2]] on d and agrees (16).
(16) dP
Pro d′
[iSCP[2]]
d DP
AGREE!
d every every man
[uSCPval[2]] [uSCP[2]] [uSCP[ ]] [iRES[1]]
[uRESval[1]] [uRESval[1]]
The result is now an LF-legible feature and an interface-legible DP structure. Observe

that (16) is isomorphic to the transitive vP/VP structure (17), differing only in category
and the specific features and lexical items involved:
(17) vP
John v′
[iAG[2]]
v VP
AGREE!
v kiss kiss Mary
It’s worth noting that issues concerning [scp] and intransitive DP projection arise
in the same way discussed earlier for [ag] and intransitive VP projection. Recall that if
valued [ag] is allowed on a lexical V, then simple unergative structures like (18a) will be
possible.6 However, if valued [ag] is forbidden on a lexical verb head, then projection of
unergative verbs must always appeal to little v (18b):
(18) a. VP b. vP
John sneeze John v′

[iAG[1]] [uAG[1]] [iAG[1]]
v sneeze
[uAG[1]]
v sneeze
Just so, if valued [scp] is allowed on lexical D, then simple pronoun DP structures like
(19a) will be possible. But if valued [scp] is forbidden on a lexical determiner head, then
projection of pronouns must always appeal to little d (19b):7
(19) a. DP b. dP
Pro him Pro d′

[iSCP[1]] [uSCP[1]] [iSCP[1]]
d him
[uSCP[ ]]
d him
[uSCPval[1]] [uSCP[1]]
As in the vP/VP case, I leave the choice between (18a) and (18b) an open question.
3.3. θ-Features and θ-Agreement: Determiner–Oblique Argument

Consider the determiner more, which GQ theory analyzes as ditransitive under the
proposals in (7) (Keenan and Stavi 1983). Assume more to bear the three θ-features in
(20), with [res] the default (and only) valued θ-feature, as usual:
(20) more {[uscp[ ]], [uresval[ ]], [unob[ ]]}
We now face issues identical to those facing us with the ditransitive verb give. Direct
merge of the nominal oblique argument of more (which must occur first according to
the thematic hierarchy) will not yield a valued feature (21):
(21) DP
more women
[uSCP[ ]] [iNOB[1]]
[uRESval[ ]]
[uNOB[1]] UNVALUED!
However, if the possibilities open to us in the DP and VP domains are the same, then
we can value this feature through an oblique preposition. Suppose the preposition than,
which is lexically governed by more, bears the relevant oblique θ-feature. Then the two
can merge, valuing [nob] (22a); the PP can then in turn merge with more, yielding a
valued, interpretable oblique θ-feature (22b):
(22) a. PP b. DP
than women more PP

[uNOBval[1]] [iNOB[1]] [uSCP[ ]]
[uRESval[ ]] than women
AGREE! [uNOB[1]] [uNOBval[1]] [iNOB[1]]
AGREE!
The rest of the derivation proceeds as before. The restriction argument merges directly
and unproblematically since [res] is valued on more. The scope argument Pro is valued
by little d, which also raises D, yielding the final DP word order more men than women.
(23) dP
Pro d′
[iSCP[3] ]
d DP
d more men D′
[uSCPval[3]] [uSCP[3]] [iRES[2]]
[uRESval[2]] more PP
AGREE! [uNOB[1]] [uSCP[ ]]
[uRESval[2]] than women
AGREE! [uNOB[1]] [uNOBval[1]] [iNOB[1]]
Again the resulting nominal structure for the ditransitive determiner is identical to that
projected for a verbal ditransitive like give, up to category differences and the specific
features and lexical items involved. Compare (23) with (24), given earlier:
(24) vP
John v′
[iAG[3]]
v VP
v give Fido V′
[uTHval[2]] give PP
[uTHval[2]] to Mary
3.4. Multiple Oblique Arguments and Recursion

If the D-system is allowed a “pure” little d, as in the V-system, then recursion with
rightward descent will also be possible. Consider the DP in (25), containing two relative
clause modifiers:
(25) every man that I know that you met.
Suppose, as discussed in the introduction to the section on complex predicates, that

we are permitted to augment the basic θ-set of a predicate head to include additional,
optional θ-features. In the verbal system these are typically features corresponding to
“circumstantial adjuncts” like benefactives, instrumentals, and locatives. In the deter-
miner system, the counterpart is restrictive modifiers. Let’s assume, therefore, that in
addition to a basic collection of θ-features for every like (26a), we are permitted aug-
mented sets like (26b) and (26c), which contain a new oblique θ-feature that I will
simply label [rmod] for “restrictive modifer”:8
(26) a. every {[uscp[ ]], [uresval[ ]]}
every {[uscp[ ]], [uresval[ ]], [urmod[ ]]}

b.
every {[uscp[ ]], [uresval[ ]], [urmod[ ]], [urmod[ ]]}

c.
Assuming [rmod] to be ranked low on the thematic hierarchy like other oblique argu-
ments, the relative clauses will combine with the determiner every first:9
(27) DP
CP D′
that I know every CP

[iRMODval[2]] [uAG[ ]]
[uTHval[ ]] that you met
[uRMOD[2]] [iRMODval[1]]
AGREE! [uRMOD[1]] AGREE!
An attempt now to merge the restriction argument (man) directly with the structure
in (27) will yield a minimality violation, as discussed earlier. Agreement between the
restriction argument and every must cross a closer θ-feature-bearing element (the CP
that I know) that also stands in an agreement relation to the head (28):
(28) DP
man D′
[iRES[3]]
CP D′
that I know every CP

[iRMODval[2]] [uSCP[ ]]
[uRESval[3]] that you met
X AGREE! [uRMOD[2]]
AGREE! ...
If, however, “pure little d” is available in the determiner system, we can merge it, raise
the D head, and then merge the restriction argument without violating minimality on
agreement. The agreement relations will have been separated in the necessary way (29):
(29) dP
man d′
[iRES[3]]
d DP
every d CP D′
[uSCP[ ]]
[uRESval[3]] every CP
that I know
AGREE! ... [uSCP[ ]]
[uRESval[ ]] that you met
...
The remainder of the derivation, adding the scope argument through a little d voice
head, is then unproblematic and yields the final word order every man that I know that
you met (30):
(30) dP
Pro d′
[iSCP[4]]
d dP
d d man d′
[uSCPval[4]]
every d d DP
[uSCP[4]]
. .. every d CP D′
[uSCP[ ]]
AGREE! that I know every CP
...
[uSCP[ ]]
... that you met
In the presence of a pure light determiner head, recursion on shells thus becomes avail-
able in DP in full analogy with VP.
3.5. Voice Alternation in DP

As noted in Larson (1991), postulation of a Pro subject in all DPs has strong implica-
tions for the analysis of prenominal genitive constructions like (31a–d):10
(31) a. John’s briefcase
b. John’s picture
c. John’s grandmother
d. John’s completion of the plan
Recall in particular Abney’s (1987) assimilation of nominals like (31d) to clauses (IPs),
with the possessive element claimed to occupy a subject-like position. Szabolcsi (1989,
1992, 1994) develops this analogy further with Hungarian examples like (32), where
the possessive item co-occurs with a definite article. Szabolcsi analyzes the DP as clausal
and the article as counterpart to a complementizer.
(32) (a) Mari kalap-ja-i
(the) Mari hat-poss-pl-2sg
‘Mari’s hats’
The analysis of DP structure developed above is incompatible with the sentential

analogy pursued by Abney, Szabolcsi, and indeed many others. If the thematic “subject”
of DP is always the scope argument Pro, then the possessive phrase cannot be paral-
lel to a subject. By extension, the Hungarian definite article cannot be parallel to a
complementizer.
Larson (1991) proposes that PP genitives and prenominal genitives in the nominal
system are the parallels of PP datives and double object datives in the verbal system.11
Specifically, double object datives and prenominal genitives both involve inversion, as
shown in (33b)/(34b).
(33) a. VP b. VP
DP V′ DP V′
Mary V VP Mary V VP
gave DP V′ gave DP V′
Fido V PP John V′ DP
gave to John V DP Fido
gave John
PP Dative Double Object Dative
(34) a. DP b. DP
Pro D′ Pro D′
D D D DP
the NP D′ THE DP D′
book DP PP John’s D′ NP
the of John’s D DP book
THE John’s
PP Genitive Prenominal Genitive
If this association is correct, then prenominal genitives represent a form of voice

alternation within DP.
Within the current system, the counterpart of (34a) is parallel to the derivation dis-
cussed earlier with ditransitive D more: we employ the P option for valuing the lower
complement of D, which I assume to be a definite determiner THE, unpronounced in
English but overt in Italian, Hungarian, and other languages, and bearing the genitive
θ-role [GEN] (35):
(35) dP
Pro d′
[iSCP[3]]
d DP
d the book D′
[uSCPval[3]] [uSCP[3]] [iRES[2]]
[uRESval[2]] the PP
AGREE! [uGEN[1]] [uSCP[]]
[uRESval[2]] of John’s
AGREE! [uGEN[1]] [uGENval[1]] [iGEN[1]]
Compare this with our earlier ditransitive vP/VP(24), repeated below as (36):
(36) vP
John v′
[iAG[3]]
v VP
v give FIdo V′
[uTHval[2]] give PP
[uTHval[2]] to Mary
By contrast, John’s book is an applicative structure and receives an A-Shift derivation

as in (37), where we value [gen], not via a preposition, but by selecting a little d voice
head bearing [ugenval]. The latter raises the genitive argument (John’s) from its low
merge position to the Spec of dP:
(37) dP
Pro d′
[iSCP[3]]
d dP
d d John’s d′
[uSCPval[3]] [iGEN[1]]
d THE d DP
AGREE! [uGENval[1]] [uSCP[3]]
[uRESval[2]] d THE book D′
[uGEN[1]] [uGENval[1]] [uSCP[1]] [iRES[2]]
[uRESval[2]] THE John’s
[uGEN[1]] [uSCP[ ]] [iGEN[1]]
[uRESval[ ]]
[uGEN[1]]
Compare this with an applicative/double object derivation in the verbal domain (38).
The parallelisms are exact:
(38) vP
John v′
[iAG[3]]
v vP
v v Mary v′
[uAGval[3]] [iGL[1]]
v give v VP
AGREE! [uGLval[1]] [uAG[3]]
[uTHval[2]] v give Fido V′
[uGL[1]] [uGLval[1]] [uAG[1]] [iTH[2]]
[uGL[1]] [uAG[ ]] [iGL[1]]
[uTHval[2]]
[uGL[1]]
Thus, under both the Larson (1991) analysis and the updated version of it sketched
here, prenominal genitives are not parallel to subjects in the clause but rather to (indi-
rect) objects: they are internal arguments of D.12
4. IMPLICATIONS
The results discussed here for nominals (DPs) have a variety of implications. As noted
in Larson (1991), just as θ-roles like THEME, EXPERIENCER, and GOAL appear to
apply beyond the verbal domain and to figure, for example, in the projection of adjec-
tives, so roles like scope and restriction appear to apply beyond the nominal domain.
Conceptually, it seems that quantificational θ-roles/θ-features should factor in the
projection of any quantificational element of any category, including quantificational
adverbs (always, often, sometimes), quantificational prepositions (before, after), and
degree quantifying elements like equatives, comparatives, superlatives, and too/enough-
constructions. Larson (1991) explores the projection of DegP along similar lines to DP,
and these proposals seem fully updatable in the way sketched above.13
The current results suggest a further, far more radical conclusion as well. In dis-
cussing verbal semantics in the General Introduction, we contrasted three different
approaches and their consequences for how one views selection. Under what I called the
standard analysis, a predicate is interpreted as denoting a function (39a), its arguments
are interpreted as denoting individuals (39b,c), and the former combines with the latter
by function-argument application (39d):
(39) a. kiss ⇒ λyλx[kiss(x,y)]
John ⇒ john
b.
Mary ⇒ mary
c.
John kiss Mary ⇒ λyλx[kiss(x,y)](mary)(john) = kiss(john,mary)

d.
This theory motivates a classical view of selection as semantically based, deriving from
the argument-taking properties of the verbal head.
Under one version of the Neo-Davidsonian approach to verbal semantics, what I
called the P-analysis, essentially the same situation holds (40), despite extra-semantic
event structure introduced internally to the interpretation of V:
(40) a. kiss ⇒ λyλx∃e[kiss(e) & Agent(e,x) & Theme(e,y)]

John ⇒ john
b.
Mary ⇒ mary
c.
John kiss Mary ⇒ λyλx∃e[kiss(e) & Agent(e,x) & Theme(e,y)](mary)(john) ≡

d.
∃e[kiss(e) & Agent(e,john) & Theme(e,mary)]
Again selection can be seen as semantically based, issuing from the argument-taking
properties of the verb.
However, under a second version of the neo-Davidsonian approach, which I
termed the K-analysis, a radically different picture obtains. V doesn’t combine with
its syntactic arguments by function-argument application, as in (39), nor provide
the thematic relations in which they stand, as in (40). Rather, V denotes a simplex
event predicate (41a), its arguments “come with” their thematic relations already
attached (41b,c), and the sentence interpretation is assembled by means of predicate
conjunction (41d):
(41) a. kiss ⇒ λe[kiss(e)]
John ⇒ λe[Agent(e,john)]
b.
Mary ⇒ λe[Theme(e,mary)]
c.
John kiss Mary ⇒ ∃e[kiss(e) & Agent(e,john) & Theme(e,mary)]

d.
This last approach entails the view that selection is a purely syntactic matter, and not
semantically based at all. It also underwrites a syntactic analysis in which θ-features are
interpretable on arguments. Whereas a bare proper noun (Mary) can be analyzed as
denoting a bare individual (mary), as in classical accounts (42a), a proper noun bear-
ing an interpretable θ-feature can be analyzed in terms of that individual standing in a
thematic relation to an event (42b):
(42) a. Mary ⇒ mary
b. Mary ⇒ λe[Theme(e,mary)]
[ith]
Thus the K-analysis provides a semantic interpretation for the notion “θ-feature inter-
pretable on an argument.”
Under the classical GQ semantics that we have assumed to underlie nominal pro-
jection, this clean match-up between feature distribution and interpretation does not
hold. GQ takes quantificational heads to denote functions (43a) and their arguments to
denote sets (43b,c) that combine with it by function-argument application (43d):
(43) a. every ⇒ λYλX [EVERY(X,Y)]
bird
b. ⇒ λy[bird(y)]
flies
c. ⇒ λx[flies(x)]
every bird flies ⇒ λYλX [EVERY(X,Y)](λy[bird(y)])(λx[flies(x)]) ≡

d.
EVERY(λx[flies(x)], λy[bird(y)])
In other words, this semantics is counterpart to (39) insofar as it locates selection in

the semantics of the D head. Accordingly, it will not underwrite our assumptions about
θ-feature distribution and in particular θ-feature interpretability on arguments. Under
standard assumptions, predicates like bird and flies denote sets of individuals, and sets
of individuals are what generalized quantifers combine with. Hence placing an interpre-
table instance of a θ-feature in the syntax will correspond to no semantic effect. If bird
denotes the set of birds, so must bird bearing [ires] (4):
(44) a. bird ⇒ λy[bird(y)]
bird ⇒ λy[bird(y)]
b.
[ires]
Thus, although we can distribute (valued and interpretable) θ-features among deter-
miner heads and their arguments so as to replicate verbal projection under our syn-
tactic assumptions, classical GQ semantics doesn’t itself directly support these feature
distributions, and hence doesn’t underwrite the syntax they yield. It doesn’t provide a
semantics for the notion “θ-feature interpretable on an argument of D.”
These points suggest a radical revision of the semantic theory of generalized quanti-
fiers that would “Davidsonianize” them in a way parallel to the move from (39) to (41).
Specifically, rather than interpreting quantifiers as relations between sets, we might take
them instead to denote unary predicates of states—quantificational states—in which
their set arguments participate via binary thematic relations. Thus, in place of the stan-
dard relational picture in (45a), we might seek something like (45b), where every bird
flies is rendered as saying that there is a quantificational “every-state” s such that the
set of fliers bears the scope relation to s and the set of birds bears the restriction rela-
tion to s. The contributions of the separate phrases would be as in (45c–e), once again
assembled by predicate composition:
(45) a. EVERY(λx[flies(x)], λy[bird(y)])
b.
∃s[every(s) & Scope(s, λx[flies(x)]) & Restriction(s, λy[bird(y)])]
every ⇒ λs[every(s)]
c.
bird
d. ⇒ λs[Restriction(s, λy[bird(y)])]
flies ⇒ λs[Scope(s, λx[flies(x)])]

e.
This revision would permit a view of θ-feature interpretability on quantifier arguments

that is fully parallel to that of verbal arguments. Once again, we could take adding an
interpretable feature to the phrase to correspond to adding a certain thematic relation
to the interpretation of the bare argument (46a,b) (cf. 39b and 41b):
(46) a. bird ⇒ λy[bird(y)]
bird ⇒ λs[Restriction(s, λy[bird(y)])]

b.
[ires]
These conclusions appear to bring us full circle in our syntax-semantics deliberations

on VP-DP. Starting from a certain view of verbal semantics (the P-analysis of neo-
Davidsonianism) we developed a syntactic theory of θ-features, θ-feature agreement,
and syntactic projection that updates the Larson (1988) shell analysis of VP. Extending
this syntactic treatment to nominals, following in the tracks of Larson (1991), offers
an intriguing approach to DP structure radically different from, and more semantically
plausible than, views pursuing the DP/TP analogy. At the end of the day, however, we’ve
seen that to make the syntactic extension to DP (and other quantificational categories)
truly complete, it appears necessary to revisit GQ semantics itself, revising it along
radically neo-Davidsonian lines and bringing it in line with the semantics for the verbal
domain adopted earlier.
A neo-Davidsonian GQ semantics raises many difficult and intriguing questions,
such as how familiar notions like scope are to be captured and, indeed, how the basic
notion of a quantificational state in which sets participate is to be understood in the
first place. Writing down formulas like (45) and (46) is not, after all, the same thing
as explaining what they mean. I must leave these questions to be taken up elsewhere,
however.
NOTES
1. See Larson (in preparation) for an account of the historical development of these ideas.
2. The idea that inflection is the head of the clause was first proposed (to my knowledge) by
Jeanne (1978). Following the work of Pollock (1989), the category I was fractionated into
other constituent functional categories such as T and Agr. Current theory basically replaces
I with T and regards the clause as TP.
3. See Giusti (1997) and Bernstein (2001) for useful summary discussion.
4. Note that in (3a–c) the nominal (whale) corresponds to the Y argument of ALL, whereas
the predicate (is a mammal) corresponds to the X argument. This reflects the syntactic fact
that in a standard structure like [DP D NP], the nominal (NP) is the internal argument of the
determiner (D). Compare the situation with a transitive verb like eat in its standard structure
[VP V DP]. The internal argument of eat—its DP object—corresponds to the y argument in
the usual relational representation EAT(x,y) (ia–d):
(i) a. All (X,Y ) b. EAT (x,y).

c. DP d. VP
D NP V DP

5. See Larson (1991c) for details.
6. See the General Introduction, specifically the discussion of (55) and (56).
7. Again, if all pronouns could be argued to involve an underlying complement, along the
lines of We Americans discussed in Postal (1969), then unergative D would simply reduce
to the transitive case. See Vassilieva and Larson (2005) for further discussion of pronoun
complementation.
8. Technically, (26a–c) are multisets because they allow two instances of the same object within
their membership.
9. As a simplification, the relative clauses in (27) are represented as bearing interpretable and
valued θ-features; that is, as having that same status as PPs, which contain both valued and
interpretable θ-feature instances within them (cf. (22a)). For present purposes I pass over
the question of how the [rmod]-feature is distributed within relative CPs, including whether
there are separate [irmod] and [urmodval] instances, as in the PP case.
10. Strictly speaking, postulation of a Pro subject holds for all quantified DPs. I am here taking
the classical line that the definite determiner involving possessive DPs is quantificational in
all cases.
11. This idea has been recently taken up by Kupula (2008) in a nonderivational applicative
framework.
12. The implications of this analysis are particularly radical in the case of nominalizations. On
the view adopted in Larson (1991) and developed further in Larson (in preparation), not
only is genitive the voters’ in (ia) structurally nonparallel to nominative the voters in (ib), but
of Mary in the nominal is not parallel to Mary in the clause. In the sentential example, Mary
is the object of elect and is selected by it. In the nominal example, of Mary is an argument
of the determiner (THE) and bears no direct selection relation to election.
(i) a. The voters’ election of Mary
b. The voters elected Mary.
The resemblance between the two cases is thus largely illusory. Se Larson (1991) for further
details.
13. It seems at least possible that these two distinct and non-overlapping collections of roles/
features and associated hierarchies exhaust the domain of thematic concepts.
REFERENCES
Abney, S. (1987) The English Noun Phrase in Its Sentential Aspect. Doctoral dissertation, MIT,
Cambridge, MA.
Bernstein, J. (2001) “The DP Hypothesis: Identifying Clausal Properties in the Nominal Domain,”
in M. Baltin and C. Collins, eds., The Handbook of Contemporary Syntactic Theory.
(pp. 536–561). Malden, MA: Blackwell.
Frege, G. (1953) The Foundations of Arithmetic. Trans. J. L. Austin. New York: Philosophical
Library. Originally published 1884.
Fukui, N. and M. Speas (1986) “Specifiers and Projection,” in N. Fukui, T. Rapoport, and
E. Sagey, eds., Papers in Theoretical Linguistics (pp. 85–127). Cambridge, MA: MIT Working
Papers in Linguistics.
Giusti, G. (1997) “The categorial status of determiners,” in Liliane Haegeman, ed., The New
Comparative Syntax (pp. 124–144). New York: Longman.
Heim, I. (1982). The Semantics of Definite and Indefinite Noun Phrases. Doctoral dissertation,
University of Massachusetts, Amherst.; Published 1989, New York: Garland.
Higginbotham, J. and R. May (1981) “Questions, Quantifiers and Crossing,” The Linguistic
Review 1: 41–80.
Hornstein, N. (1999) “Movement and Control,” Linguistic Inquiry 30: 69–96.
Jeanne, L. (1978) Aspects of Hopi Grammar. Doctoral dissertation, MIT, Cambridge, MA.
Kamp, H. (1981) “A Theory of Truth and Semantic Representation,” in J.A.G. Groenendijk
et al., eds., Formal Methods in the Study of Language; Mathematical Centre Tracts 135
(pp. 277–322). Amsterdam: Mathematical Centre.
Kupula, M. (2008) Adnominal Possession and Ditransitives. Doctoral dissertation, Stockholm
University.
Larson, R. (1991) “The Projection of DP and DegP Structure,” unpublished manuscript, State
University of New York—Stony Brook Republished in this volume.
———. (1988) “On the double object construction,” Linguistic Inquiry 19: 335–391. (Republished
in this volume.)
———. (in preparation) VP and DP.
Lees, R. (1968) The Grammar of English Nominalizations. The Hague: Mouton.
Pollock, J.-Y. (1989) “Verb Movement, Universal Grammar and the Structure of IP,” Linguistic
Inquiry 20: 365–424.
Postal, P. (1969) “On So-Called ‘Pronouns’ in English,” in D. Reibel and S. Schane, eds., Modern
Studies in English (pp. 201–224). Engelwood Cliffs, NJ: Prentice-Hall.
Szabolcsi, A. (1983) “The Possessor That Ran Away from Home,” The Linguistic Review 3:
89–102.
———. (1989) “Noun Phrases and Clauses: Is DP Analogous to IP or CP?”, unpublished ms., New
York University, New York.
———. (1992) “Subordination: Articles and Complementizers,” in I. Kenesei and Cs. Pléh, eds.,
Approaches to Hungarian 4 (pp. 123–137). Szeged: JATE.
———. (1994) “The Noun Phrase,” in I. Kenesei , ed., Syntax and Semantics 27: The Syntactic
Structure of Hungarian (pp. 179–275). New York: Academic Press.
Vassilieva, M., and R. Larson (2005) “The Semantics of the Plural Pronoun Construction,” Natural
Language Semantics 13: 101–124.
7 The Projection of DP (and DegP)
Recent versions of the Extended Standard Theory (e.g., Chomsky 1981, 1986)
have proposed that syntactic structures are largely projectable from the properties
of their constituent lexical items—in particular, from their thematic properties. In
this paper I argue for an extension of this approach to nominal structure. My focus
will not be on nominals related in an obvious way to verbal projections (1a), but
rather on ones involving simple determiners, possessives, and restrictive modifiers
(1b–d):
(1) a. The enemy’s destruction of the city
b. Every flower
c. John’s book
d. The man that I met
The proposed analysis draws crucially on three lines of semantic and syntactic
research: (a) the relational view of determiners, under which elements like some,
every, the, and so on. correspond to binary relations between properties or con-
cepts; (b) the “DP Hypothesis,” under which noun phrases (so-called) are actually
projections of their constituent determiners; and (c) the view of constituent struc-
ture proposed in Larson (1988a), according to which X-bar projections assume a
uniform, recursive transitive form. As I show, this account illuminates a variety of
familiar questions about the form of noun phrases and suggests a return to certain
“classical” transformational proposals about the relation between nominals and
modifiers.
In section 1, I sketch the relational view of determiners, and in section 2, I show its
basic implications for noun phrase structure under the “head-raising” syntax in Larson
(1988a, 1989). In section 3, I examine nominals containing restrictive relatives and
PPs and argue, in effect, for a restoration of the “Article-S” analysis of Smith (1964).
According to the latter, restrictive modifiers in NP are complements of the determiner
and not adjuncts of the nominal. Section 4 next considers the structure of posses-
sive nominals, including possessives that are intuitively linked to relative constructions
(John’s book/the book that John has) and possessives that involve argumental relations
(the city’s destruction/the destruction of the city). I suggest a derivational analysis of
such pairs analogous to the derivational relation holding between oblique and double
object forms in examples like John gave Mary a book/John gave a book to Mary.
Finally, in section 5, I briefly explore an extension of these views to the structure of
degree phrases and comparatives.1
1. THE SEMANTICS OF DETERMINERS
In logic texts, sentences like those in (2a,b) are standardly represented in the format of
“unrestricted quantification” shown in (3a,b) (respectively):
(2) a. All whales are mammals.
b. Some man arrived.
(3) a. ∀x[whale(x) → mammal(x)]
b.
∃x[man(x) & arrived(x)]
On this view, quantificational determiners correspond roughly to operators combining

with a single (possibly complex) unary predicate. Thus in (3a) ∀ combines with the
complex unary predicate “if-a-whale-then-mammal,” and so on.
As is well known, however, the format of unrestricted quantification appears unsat-
isfactory, on a number of counts, for representing natural language quantification. One
problem is that the syntax departs sharply from that of natural language: (3a,b) involve
truth-functional connectives that do not appear to be present in (2a,b); moreover, (3a,b)
have fundamentally the structure of conjunctions, something not obviously true of
(2a,b). A second, more serious difficulty is that sentences involving certain quantifiers
can be shown to have no unrestricted representation. Thus, it can be proven formally
that first-order representations analogous to (3a,b) simply cannot be given for sentences
containing most, many, and few, such as (4a,b):2
(4) a. Most people think that dinosaurs were cold-blooded.
b. Few cats reject tuna fish.
Thus there is an expressive limitation on unrestricted quantification that is apparently

exceeded by natural language.
1.1. The Relational View of Determiners

Given these results, there has been considerable interest in recent years in the analy-
sis of natural language quantification as involving generalized or restricted quantifiers
(Rescher 1962; Barwise and Cooper Davies 1981 Higginbotham and May 1981 Keenan
and Stavi 1983;Wiggins 1980). The approach is based on a single, very simple idea that
may be stated as follows:
Relational View of D
Determiners express relations among predicate meanings.
This idea descends from the Aristotelian tradition in logic but is also advanced by
Frege (1953), who suggests that in quantified examples like (2), the element all
expresses a relation between between concepts. In particular, Frege proposes that all
expresses subordination of the concept ‘whalehood’ to that of ‘mammalhood’. On this
view, the “logical form” of (2a) is something like (5a), where ALL corresponds to the
The Projection of DP (and DegP) 429
subordination relation. That is, (2a) is true just in case being-a-whale is subordinate to
being-a-mammal so that all individuals possessing the first property possess the second.
In a similar way, (2b) may be taken to have the logical form in (5b), where SOME is the
“nonexclusion” relation. That is, (2b) is true just in case being-a-man and arriving are
nonexclusive properties.
(5) a. ALL(‘whalehood’, ‘mammalhood’)

b. SOME(‘man’, ‘arrive’)
According to the relational analysis, then, determiners are semantically similar to tran-
sitive predicates such as touch; but whereas the latter express relations between two
individuals, such as Mary and John, the former express relations between two concepts.
1.2. Determiners as Set Relations

Frege’s basic idea can be spelled out precisely by construing “property,” “subordina-
tion,” “nonexclusion,” and so on in set-theoretic terms. Suppose common nouns and
verb phrases are viewed as corresponding semantically to sets of individuals:
(6) a. whale => {x: x is a whale} b. mammal => {x: x is a mammal}
man
c. => {x: x is a man} d. arrive => {x: x arrives}
Then determiners can be interpreted as expressing relations of quantity between such

sets. The “subordination” relation ALL can be spelled out in terms of the number of
individuals in the common noun set (Y) that are not in the verb phrase (X) (7a). Like-
wise, the “nonexclusion” relation SOME can be spelled out in terms of the number of
individuals that are in both (7b):
(7) a. ALL(X,Y) iff | Y − X | = 0
b. SOME(X,Y) iff | Y ∩ X | ≠ 0
Given (7a), All whales are mammals will be true if and only if the set of whales contains
no members not in the set of mammals. Given (7b), Some man arrived will be true just
in case the set of men and the set of arrivers have a nonempty intersection. These are
the correct results.
This general picture extends naturally to a variety of other determiners, including
most, which was problematic for unrestricted quantification:
(8) a. NO(X,Y) iff | Y ∩ X | = 0
b. MOST(X,Y) iff | Y ∩ X | > | Y − X |
c. TWO(X,Y) iff | Y ∩ X | = 2 (and similarly for other numeral determiners)
d. THE-TWO(X,Y) iff | Y − X | = 0, where | Y | = 2
(similarly for other numeral determiners of the form “the-n,” for some n)
e. BOTH(X,Y) iff THE-TWO(X,Y)

f. NEITHER(X,Y) iff | Y ∩ X | = 0, where | Y | = 2
g. THE(X,Y) iff THE-ONE(X,Y)
In all such cases, the determiner expresses a relation of quantity between the extension
of a common noun (Y), traditionally referred to as the restriction, and the extension of
a verb (or other predicate) phrase (X), traditionally referred to as the scope. D specifies,
in effect, how many things satisfying the restriction Y are true of the scope X.3
2. IMPLICATIONS FOR NOMINAL SYNTAX
Under the relational analysis, determiners possess argument structure and lexical prop-
erties much like other predicate expressions. This view has interesting consequences for
the syntax of nominals under the Extended Standard Theory of Chomsky (1981, 1986),
which hypothesizes an intimate connection between argument structure and form.
2.1. The Relational View and DP

As we have noted, the relational view of determiners treats D as a predicate, which
selects its sister noun much like a transitive verb selects an object. This semantic analysis
appears to fit naturally with the theory of nominal syntax proposed by Abney (1987)
and Fukui and Speas (1986), wherein Ds are heads that take their nouns as complements
(9a). In fact, the relational view appears to fit much better with the DP analysis than
with the traditional NP picture (9b), which expresses no selection relation between D
and N:
(9) a. DP b. NP
D NP Det N
the man the man
Nonetheless, despite the obvious attractions of connecting the relational semantics and
DP, the assimilation is not completely straightforward.
Abney (1987) classifies D as a “functional category,” a group of forms bearing little
or no semantic content on his view. He analogizes D in DP to I in IP, along the lines
shown in (10a,b), analyzing John as the subject of DP in (10a) just as John is the subject
of IP in (10b).
(10) a. DP b. IP
DP D′ DP I′
John D NP John I VP
s′ completion of the plan [TNS] complete the plan
These proposals are simply not tenable under the relational analysis, however. First, as
we have seen, the relational analysis does not take Ds to be semantically empty; rather,
they express relations of quantity between sets. Furthermore, D is not analogous to I
under the relational analysis, but rather to a predicate category such as a V. Indeed, as
we will discuss in detail below, the general semantics for Ds offers a natural classifica-
tion of these elements into monadic, dyadic, and triadic forms, much as one finds with
verbs. Basic determiners (every, some, the, etc.) correspond to dyadic predicates (11a),
pronouns (she, him, they, etc.) correspond to monadic predicates (11b), and complex
Ds like more-than and every-except correspond to triadic predicates (11c,d):
(11) a. EVERY(X,Y) iff | Y – X | = 0 “dyadic D”
b. HEn(X) iff g(xn) ∈ X “monadic D”
c. MORE-THAN(X,Y,Z) iff | Y ∩ X | > | Z ∩ X | “triadic D”
Finally, under the relational view, an expression like John could not possibly constitute
the subject of D in a semantic sense. Recall that the external “subject” argument of D
(X) is semantically a predicate, one given by the syntactic constituent comprising the
scope of the DP (usually the main predicate of the sentence). This is not compatible
with a structure like (10a), which makes a name the subject of DP and which makes no
provision for the scope argument of D.
In general, then, although a joining of the relational analysis and DP is attractive,
this move doesn’t appear to be possible under the original intuition of Abney (1987)
and Fukui and Speas (1986) that D is a functional category comparable to I. Instead, the
natural correspondence offered by the relational semantics is between D and V.
2.2. A Theory of Structural Projection

I wish to offer an alternative account of DP projection, one that is (I believe) more
clearly compatible with the relational analysis of determiners. This account is based
on the theory of argument projection in Larson (1988a), which includes the following
principles:
(12) a. XP → YP X′
b. X′ → X ZP
(13) If β is an argument of α, then β must be realized within a projection of α.
(14) Roles determined by a predicate are projected according to the thematic hierarchy
ΘAGENT > ΘTHEME > ΘGOAL > ΘOBLIQUE, such that if Θ1 > Θ2, then the
argument to which Θ1 is assigned c-commands the argument to which Θ2 is assigned.
(12) is a restricted version of X-bar theory embodying a “Single Complement Hypoth-

esis.” Under the latter, maximal projections are limited to one specifier and one comple-
ment per phrase. (13) and (14) give principles for the realization of arguments vis-à-vis
their selecting head, specifying the location of these arguments and their relative hierar-
chical organization, respectively.
To illustrate these principles briefly, consider first the transitive verb kiss, which
assigns an agent and a theme role. (12)–(14) determine a VP headed by kiss as in (15).
This structure conforms to the restricted X-bar theory; furthermore, all arguments of
V are contained within a projection of V; finally, the argument bearing the agent role is
projected into a position c-commanding the argument bearing the theme role, in compli-
ance with (14) and the fact that ΘAGENT > ΘTHEME.
(15) VP
DP V′
John V DP
kiss Mary
Ditransitive put, which assigns agent, theme, and location, represents a more compli-
cated case. Assuming ΘAGENT > ΘTHEME > ΘLOC, we project a minimal VP as in (16), contain-
ing arguments corresponding to ΘTHEME and ΘLOC, with the former higher than the latter:
(16) VP
DP V′
salt V PP
put on the fish
This structure leaves ΘAGENT unassigned, and no position for its bearer. In Larson (1988a)
it is proposed that this circumstance licenses the “VP shell” in (17a), which contains a
higher specifier for the agent and an empty verbal head position. The surface word order
derives by raising the verb form to [V e] (17b):
(17) a. VP b. VP
DP V′ DP V′
John V VP John V VP
DP V′ put DP V′
salt V PP salt V PP
put on the fish put on the fish
2.3. Projecting DP
The theory sketched above can be extended to DPs under the assumption that they
are projections of their constituent determiners. To do so, however, we must first
settle some important preliminary questions about the thematic hierarchy and the
status of the scope argument for DP. We then turn to the projection of DPs of vari-
ous types.
2.3.1. The Thematic Hierarchy in D
A crucial element in the theory of projection given above is the assumption of a thematic
hierarchy. The principle in (14) orchestrates the projection of verbal arguments by map-
ping relative prominence on the thematic hierarchy into relative structural prominence
as defined by c-command. Larson (1988, 1989) assumes the specific thematic hierarchy
argued for by Carrier-Duncan (1985) and M. Baker (1988), among others, that is:4
ΘAGENT > ΘTHEME > ΘGOAL > ΘOBLIQUE
Under the latter, agent phrases are always projected into structure higher than other
arguments, themes are projected higher than everything except agents, and so on.5
Whatever the correctness of this hierarchy for projection of verbal arguments in VP,
it should be clear that it cannot help us with the projection of DP. There simply is no
sense in which the set arguments (X,Y) of D under the relational analysis can be thought
of as playing roles like agent or theme in DP. These concepts seem to be irrelevant. What
then are the appropriate notions?
Proposals in this area must be regarded as highly tentative since the terrain is almost
entirely unexplored. To my knowledge, application of thematic theory to nominals
has so far been confined entirely to nominal gerunds and derived nominals like John’s
destroying of the evidence and John’s destruction of the evidence, which show an obvi-
ous connection to verbal forms (John’s destroying the evidence, John destroyed the evi-
dence). Nonetheless, it is possible to reason by analogy to some extent. Canonically,
verbs describe events, and notions like agent, theme, goal, and so on represent recurring
semantic/functional roles that verbal arguments play in those events. Thinking analogi-
cally, we observe that determiners express quantification, and notions like restriction and
scope represent two main recurring semantic/functional roles that set arguments play in
quantification. Semantically, the restriction sets the domain of quantification, whereas
the scope determines what is true of those individuals. Syntactically, restriction and scope
are also plainly relevant in mapping the parts of DP. The former role is mapped to the NP
complement of D. The latter role is associated with a main clause predication.
Given these points, I suggest an approach employing the two basic roles ΘSCOPE and
ΘRESTRICT, which are ordered as such and play a part roughly similar to ΘAGENT and ΘTHEME
for V in a canonical VP. Thus, the scope argument is projected into Spec DP, and the
restriction argument is projected lower down, inside D’. Below we will introduce vari-
ous additional “oblique” arguments of D, including comparison phrases, exception
phrases, and various forms of adjuncts. Summarizing, then, I propose the following
thematic hierarchy in DP, where “noblique” stands for nominal obliques:
ΘSCOPE > ΘRESTRICT > ΘNOBLIQUE
It follows from this hierarchy that the scope argument will always be projected highest,
that the nominal restrictor will be projected higher than everything except the scope
argument, and so on.
2.3.2. The Scope Argument X

Having established a rudimentary thematic hierarchy for determiner argument projection,
we must return more carefully to the nature of the scope argument. In our semantic discus-
sion of examples like All whales swim, we have analyzed D as relating sets X, Y. The Y
argument was given by the noun and corresponded to the restriction on the determiner.
The X argument was given by the predicate and corresponded to its scope. On this view,
the main predicate appears to function directly as one of the arguments of D.
(18) a. whales => {x: x is a whale} b. swim => {x: x swims}
c. ALL(X,Y) iff | Y – X | = 0
All whales swim is true iff | {x: x is a whale} – {x: x swims} | = 0

d.
Consider now the two candidate structures in (19a,b) in light of our principle (13)
requiring all arguments of a predicate α to be realized within a projection of α:
(19) a. VP b. DP
DP V′ D′ VP
D NP swim D NP swim
all whales all whales
Structure (19a) correctly expresses the projection requirements of V: all whales, the
agent of swim, is realized within a projection of swim (VP). However, (19a) fails to
express the projection requirements of D. Although swim is an apparent argument of
D under the relational view, it fails to be realized within a projection of D. By contrast,
structure (19b) has the opposite problem. Here the projection requirements of D are
correctly expressed: swim is an argument of all and occurs within a projection of all
(DP). But (19b) fails to express the projection requirements of V since all whales occurs
outside VP. The problem is clear-cut. If DP is an argument of V and VP is simultaneously
an argument of D, how can we find a structure that meets their joint requirements under
the locality constraint on θ-role assignment (13)?
A related question arises with sentences containing a quantified object (20a). Here
the scope argument is a set of individuals ({x: John respects x}) that is not given by any
surface constituent (20b). A familiar view is that the quantified DP undergoes covert
raising and adjoins to a containing category; the structural residue of movement (John
respects ti) then determines the scope argument of D (20c).
(20) a. John respects [DP all whales].
John respects all whales is true iff | {x: x is a whale} – {x: John respects x} | = 0
b.
c. [DPi all whales] [John respects ti ]
Here again we may ask how the locality requirements of D are met given that its apparent
scope argument (John respects ti) does not occur within DP. But notice a further question
as well. Since all whales is a complement of respect in (20a), it should constitute a phrase
under X-bar theory, and hence should contain all its arguments within it. But how can this
be if the scope argument of all is not present until after all whales has undergone raising,
as in (20c)?
The answer to these questions I wish to suggest is that the syntactic scope argument
of D is never in fact an overt predicate in the clause—neither the surface one given by
VP (swim), nor a derived one created by movement (John respects ti). Rather, the scope
argument of D is an independent, inaudible, pro-predicate element Pro, licensed by D
and projected in Spec of DP, under the hierarchy ΘSCOPE > ΘRESTRICT (21a). I suggest that
the semantic value of this Pro argument is determined configurationally at the level of
Logical Form(LF). Specifically, Pro gets its value from the derived predicate that is the
structural sister of DP at Logical Form (LF) (21b).
(21) a. [ Pro [ D′ D NP ]]
DP
θ SCOPE θ RESTRICT
b. [ Pro [ D′ D NP ]] [ XP . . . t i . . . ]
DPi
GETS ITS VALUE FROM
To illustrate these ideas with a concrete case, consider again our example all whales
swim. Under the proposals just stated, this sentence is projected initially with the
structure in (22a) (irrelevant details suppressed). VP contains all whales in specifier
position, satisfying the local projection requirements of swim. By contrast, DP con-
tains the pro-predicate Pro in Spec position, satisfying the requirements of all and
completing its argument projection. DP subsequently undergoes raising at the level
of LF as in (22b). At LF, Pro’s value is identified by the the TP ti swim, the structural
sister of the raised DP (22c). Thus Pro comes to denote the set {x: swims(x)}, the
desired semantic result.6
(22) a. TP b. TP
...VP... DP1 TP
DP V′ Pro D′ ...VP...
Pro D′ swim D NP DP V′
D NP all whales t1 swim

Quantifier
all whales Raising
c. TP
DP1 TP
Pro D′ ...VP...
D NP DP V′
all whales t1 swim
{x: swims(x)} Pro Construal

The same analysis applies straightforwardly to examples with a quantified DP object,
such as (20a).
This account answers our two questions directly. With example (18a), we see
(contrary to initial impressions) that swim is not in fact an argument of all, and
hence not required to occur within a projection of DP. Rather, Pro is the true scope
argument of D; swims simply identifies Pro’s value. Similarly, in example (20a), John
respects ti is not an argument of all, and hence need not be formed at the point where
the DP all whales is projected. Rather, the scope argument is Pro, which is pres-
ent when [DP all whales] is formed but whose value is only determined at LF, after
the quantified object has raised and the derived predicate John respects ti has been
formed. In essence, then, Pro, answers our two questions by separating the thematic
domains of D and V, relating them only in an indirect way, through the assignment
of its value.
The account also explains another, otherwise puzzling fact, what we might call the
“categorial neutrality” of D′s scope argument. Quantifiers have been argued to be able to
adjoin to any category of phrase XP, taking XP as their scope (Stowell 1981). On a theory
in which XP constitutes the direct argument of D, this implies that any category of phrase
can be the scope argument of D—in other words that, D exercises no syntactic selection.
This situation is at least anomalous. Other predicates typically do exert categorial selec-
tion on their arguments, and D itself limits its restriction argument to NP. The analysis in
(21)/(22) resolves this puzzle: D does indeed exercise selection on its subject, constraining
it to be Pro. The appearance of categorial neutrality arises from the fact that various dif-
ferent types of phrase can function as antecedents to Pro, fixing its value.
2.3.3. Dyadic Ds
Earlier we suggested that determiners, like verbs, can be divided semantically into monadic
(intransitive), dyadic (transitive), and triadic (ditransitive) forms, according to whether
they take one, two, or three predicate arguments. By far the most common case seems to
be the dyadic-transitive one, illustrated by determiner relations like (23a–d). These take a
restriction argument Y and a scope argument X and map to the general structure in (24),
where the NP complement denotes the former and Pro in Spec denotes the latter:
(23) a. ALL(X,Y) iff | X − Y | = 0
b. SOME(X,Y) iff | X ∩ Y | ≠ 0
c. NO(X,Y) iff | X ∩ Y | = 0
d. MOST(X,Y) iff | X ∩ Y | > | X − Y |
(24) DP
Pro D′
D NP
all
some
no
most
In this structure, all positions made available by the X-bar theory in (12)—Spec, head,
and complement—are realized in a single projection.
2.3.4. Monadic Ds
The case of monadic, intransitive Ds is plausibly represented by the class of pronouns,
which Montague (1974) analyzes (in effect) as restriction-less quantifiers. Montague
assigns pronoun meanings according to a scheme equivalent to (25), which involves the
single scope argument X. Under this scheme, the pronoun he1, for example, is true of
those sets containing the individual g(x1) under some assignment g:
(25) For any assignment g, HEn(X) iff g(xn) ∈ X
Montague’s semantics can be mapped to the syntax in (26), where the pronoun is analyzed
as a determiner (following Postal (1969)) and where Pro constitutes D′s sole argument:
(26) DP
Pro D′
he1
Note that this structure treats pronouns specifically as “unergative determiners” inso-
far as their one argument is an underlying subject.7 This point might lead us to expect
parallelisms between unergative Ds and Vs. (27a–c) show that unergative verbs have the
property of licensing “cognate objects,” dummy complements that (in bare form) add
no truth-conditional content to VP but at most serve to convey emphasis. Interestingly,
pronouns have the property of licensing “emphatic reflexives,” dummy anaphors that
also make no truth-conditional contribution but serve to emphasize or intensify:
(27) a. [VP laughed [a laugh]] Cognate objects
b. [VP coughed [a coughed]]
c. [VP smiled [a smile]]
(28) a. [DP he [NP himself]] Emphatic reflexives
b. [DP she [NP herself]]
c. [DP they [NP themselves]]
Larson (1988a) notes the special status of transitive structures under the X-bar theory
in (12) and proposes that cognate object formation represents a way of “filling out”
the basic transitive frame with complement material. If this line of reasoning is correct,
we might expect parallel processes in other categories, with other unergative heads.
Emphatic reflexives are a potential candidate in the domain of DP; they might be ana-
lyzed, in effect, as cognate complements of D.
2.3.5. Triadic Ds
Finally, consider triadic, or three-argument, Ds. We suggested complex determiner con-
structions like (29) and (30) as representatives of this case:
(29) a. more women than men
b. *no/three women than men
(30) a. every boy but/except Bill
b. no boy but/except Bill
c. *each/some/three/most/many boy(s) but/except Bill
As noted by Keenan and Stavi (1983), examples like these exhibit a dependency between
the boldfaced elements. (29) shows that the determiner more licenses a comparison
phrase following N, whereas other determiners do not. (30) shows that the universal
determiners every, all, and no license an exception phrase following N, whereas other
determiners (including universals like each) do not.
The dependencies in (29) and (30) can be analyzed as arising out of the basic seman-
tics of the determiners in question. Following Keenan and Stavi (1983), a straightfor-
ward analysis of (29) is that more-than expresses the three-place relation in (31a), with
the set argument Z provided by the than-phrase. (31b–d) illustrate how truth conditions
with more-than might be computed in a simple case, where the Z argument is supplied
first (31b), followed by the restriction Y (31c) and the scope X (31d):
(31) a. MORE-THAN(X,Y,Z) iff | Y ∩ X | > | Z ∩ X |
More than women

b. iff | Y ∩ X | > | {z: woman(z)} ∩ X |
More men than women

c. iff | {x: man(x)} ∩ X | > | {z: woman(z)} ∩ X |
More men than women smoke iff

d.
| {y: man(y)} ∩ {x: smokes(x)} | > | {z: woman(z)} ∩ {x: smokes(x)} |
Although I will not try to defend the proposal in detail here, I suggest a similar
approach to the exceptive constructions in (30). Specifically, I propose that the universal
determiners licensing exception phrases are subject to a lexical alternation, which raises
their valence from two to three, and that the exception phrase supplies the third argu-
ment to the augmented determiner.8 The basic idea is sketched in (32). Thus, for certain
universal determiners D, including every, all, and no (but not each, both, all three, etc.),
the grammar makes available an augmented three-place relation D-except′(X,Y,Z),
where X is the scope set, Y is the restriction set, and Z is a set given by the exception
phrase. The semantics of D-except′(X,Y,Z) is stated in (32b), where ±Cond(X,Z) is a
relation whose content depends on whether D is positive (every, all) or negative (no).
For D positive, Cond(X,Z) is | Z ∩ X | = 0. For D negative, Cond(X,Z) is | Z ∩ X | ≠
0. (33) and (34) show how truth conditions with exception phrases are computed in
simple examples:
(32) a. D ∈ {every, all, no}
b. D-except′(X,Y,Z) iff D′(X, (Y − Z )) and ± Cond(X,Z)
(33) a. EVERY(X,Y) iff | Y − X | = 0
b. EVERY-except(X,Y,Z) iff | (Y − Z ) − X | = 0 and | Z ∩ X | = 0

every except Bill
c. iff | (Y − {bill}) − X | = 0 and | {bill} ∩ X | = 0
every boy except Bill

d. iff | ({y: boy(y)} – {bill}) – X | = 0 and | {bill} ∩ X | = 0
every boy except Bill smokes iff

e.
| ({y: boy(y)} − {bill}) − {x: smokes(x)} | = 0 and | {bill} ∩ {x: smokes(x)} | = 0
(34) a. NO(X,Y) iff | Y « X | = 0
b. NO-except(X,Y,Z) iff | (Y – Z) ∩ X | = 0 and | Z ∩ X | ≠ 0
no except Bill
c. iff | (Y – {bill}) ∩ X | = 0 and | {bill} ∩ X | ≠ 0
no boy except Bill iff | ({boys} – {bill}) ∩ X | = 0 and | {bill} ∩ X | ≠ 0

d.
no boy except Bill smokes iff

e.
| ({y: boy(y)} – {bill}) ∩ {x: smokes(x)} | = 0 and | {bill} ∩ {x: smokes(x)} | ≠ 0
The semantic analyses in (31) and (32) make than-phrases and exception phrases
arguments of their associated determiner; thus the relevant Ds become three-place. This
in turn makes their syntactic projection similar to that of verbs like put, discussed ear-
lier. Assuming ΘSCOPE > ΘRESTRICT > ΘNOBL, we project the minimal DP in (35), containing
arguments corresponding to ΘRESTRICT and ΘNOBL, with the former higher than the latter:
(35) DP
NP D′
men D PP
more than women
As in the case of put, this structure leaves a thematic role unassigned (Θscope), and no
position for its bearer. We therefore license a “DP shell” in (36a), containing a higher
specifier for the scope argument and an empty D head position. The surface word order
derives by raising more to [D e], stranding the than-phrase (36b):
(36) a. DP b. DP
Pro D′ Pro D′
D DP D DP
NP D′ more NP D′
men D PP men D PP
more than women more than women

Notice that this derivation directly accounts for the discontinuous dependency holding
between more and than. The former underlyingly governs the phrase headed by the lat-
ter, a relationship that is broken up by the subsequent raising of D.
A parallel derivation can be given for exceptive constructions under the semantics
proposed above. Under the assumption that the exception phrase represents the first
argument of D, we create an initial DP as in (37a). The need to integrate the scope
argument then licenses a higher DP shell as in (37b). Finally, D raises to the empty head
position, stranding the exception phrase (37c):
(37) a. [DP boy [D′ every [PP except Bill]]]
b. [DP Pro [ D′ e [DP boy [D′ every [PP except Bill]]]]
c. [DP Pro [ D′ every [DP boy [D′ t [PP except Bill]]]]]
As above, the discontinuous dependency holding between D and PP is directly accounted

for under this derivation.9
3. MODIFIERS IN DP
The analogy between DP and VP claimed in this analysis has interesting implications
for the syntax of modifiers.
3.1. Adverbs and Adverbials as V-Complements

Larson (1988a, 1990a) proposes a theory of verbal modifiers, including adverbs and
adverbials, that departs significantly from conventional views. A common proposal
is that VP-modifiers adjoin on the right in VP, so that rightmost modifiers are highest
(38a). Larson (1988a, 1990a) proposes that adverbs descend to the right, so that right-
most modifiers are lowest (38b).
(38) a. VP b. VP
VP DP DP V′
V′ yesterday John V VP
John V DP met V′
met Bill Bill V DP
met yesterday
The structure in (38b) follows from the thematic hierarchy assumed, which ranks
oblique phrases, such as manner, locative, and temporal modifiers, lower than agents,
themes, or goals:
ΘAGENT > ΘTHEME > ΘGOAL > ΘOBLIQUE

The low insertion for adverbs typically triggers VP-shell projection and subsequent
verb raising, as in (38b). An interesting result of this theory is that verbal modifiers are
actually complements of the V head, and in fact closer complements of V than subjects,
objects, or indirect objects.
Larson (1990a) offers several arguments for the low position for modifiers, includ-
ing the fact that adverbs on the right typically behave as if they are in the domain of
other VP elements, including objects. For example, consider the facts that adverbials
containing negative polarity items can be licensed by an affective object (39). Assuming
a restrictive theory of negative polarity item licensing based on c-command, this result
follows under a structure like (38b), where the object c-commands the adverbial. It does
not follow not under (38a), however.
(39) a. John met few friends [any day this week].
b. Alice speaks few languages [with any fluency].
c. Gwen does few things [because anyone asks her to]
Another argument concerns the existence of verb-adverbial idioms like (40a–c), which
suggest a form of discontinuous dependency between the boldfaced elements:
(40) a. [VP treat John with kid gloves] (“treat carefully”) MANNER
b. [VP rub John the wrong way] (“bother”) MANNER
c. [VP put John on the spot] (“confront”) LOCATION
d. [VP kill John with kindness ] (“be very solicitous toward”) INSTRUMENT
Such items receive a very natural analysis in terms of V-Raising, where the semantic unit
constituted by the idiomatic elements corresponds to an underlying syntactic unit that
is broken up by subsequent movement (41):
(41) [VP . . . e [VP John [ V′ treat [with kid gloves ]]]]
3.2. Relative Clauses as D-Complements

The general head-raising analysis, and the treatment of discontinuous dependencies,
suggests a way of reviving some old but intuitively appealing views about the grammar
of relative clauses. In the history of transformational grammar, there have been three
main approaches to relative clause syntax. One is the NP-S analysis of Ross (1967),
according to which relative clauses are adjoined to the maximal nominal phrase (42a).
The second is the NOM-S analysis of Stockwell, Schacter, and Partee (1970), according
to which relative clauses are adjoined to a smaller nominal phrase inside NP (42b). The
third, and oldest, proposal is the Article-S analysis of Smith (1964), according to which
relative clauses are not modifiers of the noun at all, strictly speaking, but are instead
constituents of the determiner (42c).
(42) a. The NP-S Analysis b. The NOM-S Analysis

NP NP
NP S Det NOM
D N that I saw the NOM S
the man man that I saw
c. The ARTICLE-S Analysis

NP
Art N
Art S man
the that I saw
The nominal modifier analyses (42a,b) have so far received the widest support in the
literature, with textbooks (C. Baker 1978) and professional articles (Partee 1976)
framing the question of relative clause structure as a choice between the two. Among
these structures, certainly the least frequently defended is the Article-S analysis. The
reasons are fairly clear. The latter is plainly the most “abstract” of the three accounts
insofar as its structure does not match surface word order (in English, at any rate).
This abstractness also makes it the most complex, since it necessitates some kind of
extra movement operation in order to derive the correct surface forms. Nonetheless,
the Article-S analysis also has a certain attraction insofar as it appears to shed light
on certain interesting data that are not easily accommodated in the nominal modifier
accounts.
Kuroda (1969) points out that indefinite nouns like way can co-occur with a bare
demonstrative D, but not with a bare definite article (43a,b). Interestingly, when the
article is accompanied by a restrictive adjective or a relative clause, the result improves
dramatically (43 c,d). In effect, the + modifier appear to “add up” to a determiner like
that. Kuroda observes a similar dependency with the pair in (44), where the presence/
absence of negation in the relative correlates with the appropriateness of an indefinite
versus a definite D (respectively):
(43) I earned it a. that way.
b. *the way.
c. the old-fashioned way.
d. the way that one should.
(44) He greeted me with a. the/*a warmth I expected.
b. *the/a warmth I hadn’t expected.

Jackendoff (1977) makes virtually the same point with proper nouns, observing that
although the latter reject a bare definite article, a relative clause or other restrictive
modifier (AP, PP) renders the construction acceptable (45):
(45) a. *the Paris
b. the old Paris
c. the Paris that I love

d. the Paris of the twenties
What such examples appear to show is a form of discontinuous dependency holding

between the determiner and the restrictive modifier, whether relative clause, attributive
adjective, or PP.
The basic constituency of the Article-S analysis provides a natural account of these
kinds of dependencies in terms of selection between D and its sister modifier. By con-
trast, under the nominal modifier theories (42a,b), the explanation must presumably be
more complicated.10
3.2.1. A D-Raising Analysis

The pattern of dependency seen above with D and a relative clause modifier resembles
that noted earlier with V and an adverbial modifier in our idiom cases (46):
D NP RC
(46) a.
V NP AdvP
b.
This suggests a similar approach. Suppose we treat relative clauses (and other restric-
tive modifiers in DP) as a form of determiner complement, governed by our thematic
hierarchy for D and instantiating a lower thematic role than ΘRESTRICT. For concreteness,
I will label this role “ΘRMOD,” for restrictive modifier:
ΘSCOPE > ΘRESTRICT > ΘRMOD
Then the inclusion of a relative clause modifier in a DP headed by a dyadic D will result
in the minimal DP projection being filled by the arguments expressing ΘRESTRICT and
ΘRMOD. This will trigger DP shell projection to accommodate the scope argument (Pro),
and subsequent D-Raising. The resulting structure (47a) is parallel to the adverbial case
discussed earlier (38b), repeated here as (47b).
(47) a. DP b. VP
Pro D′ DP V′
D DP John V VP
the NP D′ met DP V′
way D CP Bill V DP
the that one should met yesterday

The head-raising analysis can accommodate facts originally taken to argue for the
NP-S and NOM-S analyses. Consider the example in (48a), for instance. It displays
apparent conjunction of a constituent that includes the noun and relative clause but
excludes the determiner (48b). The acceptability of such examples can be taken to argue
for the NOM-S analysis, as discussed by C. Baker (1978).
(48) a. All students who voted for Clinton and faculty who voted for Perot showed up.
b. All [[students who voted for Clinton] and [faculty who voted for Perot]]
Under the head-raising analysis proposed here, this example can be analyzed as a case of
inner DP conjunction, with across-the-board D movement along the lines indicated in (49).
(49) DP
Pro D′
D DP
all DP and DP
NP D′ NP D′
students D CP faculty D CP
all who voted for Clinton all who voted for Perot
On this view, examples like (48a) become analogous to cases of apparent nonconstitu-
ent coordination of objects and modifiers in VP, such as (50a). In Larson (1988a) these
are analyzed as inner VP conjunctions, with across-the-board V movement (50b):
(50) a. Max met Bill yesterday and Sue Tuesday.
b. [ V′ met [ VP [ VP Bill [ V′ t [ DP yesterday ]]] and [ VP Sue [ V′ t [ DP Tuesday]]]]]
Consider also the example in (51a), which displays apparent conjunction of a con-
stituent that includes the determiner and noun but excludes the relative clause. Its
acceptability can be taken to argue for the NP-S analysis.
(51) a. All students and many faculty who voted for Clinton showed up.
b. [[All students] and [many faculty]] who voted for Clinton
Under the head-raising analysis, this example can be analyzed as a case of outer DP con-
junction, with Right Node Raising of the relative clause to the right edge of DP.11 The
analysis of Right Node Raising is controversial;12 however, if this operation is viewed as
across-the-board movement of D, then the representation of (51a) is approximately as
in (52), where D-raising has occurred separately in each of the conjoined DPs:
(52) DP
DP CP
DP and DP who voted for Clinton
Pro D′ Pro D′
D DP D DP
all NP D′ many NP D′
students D CP faculty D CP
all t many t
In short, then, standard conjunction facts taken to argue for nominal modifier accounts
are also compatible with the D-Raising view.
3.2.2. Some Semantics

The analysis of relative clauses as D-complements invites natural questions about the
computation of meaning for a structure like (47a). In giving a semantics, I will adopt
the general proposal of Keenan and Stavi (1983) that determiner + relative clause com-
binations comprise a form of complex D. Specifically, I will treat relative clauses as
combining with dyadic determiners to form new, complex dyadic determiners. This
view is made concrete in the rule (53); a sample application is given in (54a–e) for the
sentence every boy that swims jogs:
(53) Let Δ be a determiner projection denoting a determiner relation D(X,Y),
where Y has the role θRESTRICT. Let CP be a relative clause denoting the set R.
If Δ is D, then [dʹ D CP ] denotes the relation D′(X,Y), where Y has the
role θRESTRICT and D′(X,Y) iff D(X,(Y∩R)). If Δ is D′, then [DP CP D′ ] denotes
the relation D′(X,Y), where Y has the role θRESTRICT and D′(X,Y) iff D(X,(Y∩R)).
(54) a. EVERY(X,Y) iff | Y − X | = 0
b. EVERY’(X,Y) iff | (Y ∩ R ) − X | = 0
every that swims

c. ⇒ | (Y ∩ {r: swims(r)}) − X | = 0
every boy that swims

d. ⇒ | ({y: boy(y)} ∩ {r: swims(r)}) − X | = 0
every boy that swims jogs ⇒ | ({y: boy(y)} ∩ {r: swims(r)}) − {x: jogs(x) | = 0
e.
Basic every expresses a dyadic relation EVERY between two sets X and Y, where Y is the
restriction (54a). (53) entails, in effect, that combining every with a relative CP creates
the new dyadic determiner EVERY’, defined as in (54b), where the restriction argument
of EVERY’ is specified as the intersection of the relative clause denotation R with the
original restriction of EVERY. The relative clause supplies the value of R (54c); after-
wards, the nominal restriction and scope arguments combine, respectively (54d,e).13
Although nonstandard, this analysis of relative clauses has precedent. Bach and Coo-
per (1978) propose a Montague Grammar semantics for relatives based on determiner
translations like (55b); this may be compared to the more standard Montague Grammar
translation in (55a) (which ignores intensions):14
(55) a. every ==> λQλP∀x[Q(x) → P(x)]

every ==> λQλP∀x[[Q(x) & R(x)] → P(x)]
b.
The crucial feature of (55b) is the inclusion of a distinguished variable R, whose value
is supplied by a relative clause and whose denotation is intersected with that of the
restriction set Q. In effect, Bach and Cooper (1978) offer an Article-S semantic analy-
sis, composing relative clause denotations with determiner denotations, analogously to
what is proposed here.
RELATIVE CLAUSES AS ARGUMENTS?

It is interesting to compare a representation like (55b) with one like (56), in which the restric-
tion variable R not only is present in the interpretation of D but is also abstracted over:
(56) every ==> λRλQλP∀x[[Q(x) & R(x)] → P(x)]
(55b) represents every as a binary determiner that always contains a restriction R on its
quantificational domain. The value of R is presumably determined by context or by an
overtly occurring restriction phrase, such as a relative. By contrast, (56) analyzes every as
a true ternary determiner, which requires an additional syntactic restrictor argument to
yield a binary D. The correct analysis of a given D as in (55b) or (56) is presumably a mat-
ter of whether the determiner in question genuinely requires a syntactic restrictor element.
Some interesting observations by Vendler (1967) suggest that interpretations like (56) may
be justified for definite determiners. Consider (57) and (58), based on Vendler’s examples.
(57) a. I see a man. The man wears a hat.
b. I see a man. The man I see wears a hat.
c. I see a man. The man you know wears a hat.
(58) a. I see a rose. The rose is lovely.
b. I see a rose. The rose I see is lovely.
c. I see a rose. The red rose is lovely.15
(57a) contains a bare definite description that is naturally understood along the lines of
(57b). Both examples present discourse that is “continuous” in Vendler’s terms: the indi-
vidual introduced by the indefinite DP is understood as the same one picked up by the
definite. Interestingly, as Vendler points out, (57c) is not continuous in the same sense.
The individuals picked out with the definite and indefinite are not naturally understood
as the same. The difference appears to be induced by the relative clause you know in the
second clause. Analogous points apply to (58).
Vendler interprets these results as supporting the view that “the definite article in front of a
noun is always and infallibly the sign of a restrictive adjunct, present or recoverable “(p. 46).16
In modern terms, a definite D selects a restrictive modifier. (57a) is analyzed as containing an
elliptical or “deleted” relative clause equivalent to (57b), allowing continuity. By contrast, in
(57c) the overt relative in effect “saturates” the relative clause required by the; hence (57c)
cannot be understood equivalently to (57b), and hence there is no continuity.17
The continuity phenomenon distinguishes the from other quantifiers. Observe that
although (59a) is naturally read as continuous, with the linguists referring to the linguists
I met, this is not true in (59b). Most linguists is not naturally read as referring to most
linguists that I met. To obtain this interpretation, an explicit definite is required (59c):18
(59) a. I met some linguists. The linguists were educated in California.
b. I met some linguists. Most linguists were educated in California.
c. I met some linguists. Most of the linguists were educated in California.
These observations are very naturally interpreted in our terms by saying that whereas
other determiners combine with restrictive modifiers via the rule in (53), the definite
determiner actually selects a restrictive modifier as an argument, as part of its basic lexi-
cal semantics. That is, the should be interpreted via the relation in (60a), which replaces
our earlier (8g); (60b) is the equivalent in a conventional Montague Grammar–style
notation:
(60) a. THE(X,Y,R) iff | (Y ∩ R) – X | = 0, where | (Y ∩ R) | = 1
the ==> λRλQλP∃y∀x[[[Q(y) & R(y)] ≡ y = x] & P(x)]

b.
Thus Vendler’s view that the always occurs with a restrictive modifier, overt or covert,
fits in well with the notion of relative clauses as D-complements.19
ORDERING OF RELATIVE CLAUSES AND EXCEPTION PHRASES

The analysis of relatives offered here yields an account of the ordering of relative clauses
in relation to other oblique elements. Consider the facts in (61a,b), which show that
the exception phrase must occur rightmost (under a normal intonation for the DP).
The reverse order is awkward to unacceptable. Since rightmost phrases are lower and
combine earlier with D in this framework, the strongly preferred order in (61a) suggests
that the exception phrase should project lower and combine with every before the two
combine with the relative (61c).
(61) a. Every boy that you saw except John
b. ??Every boy except John that you saw
c. [DP Pro [D′ every [DP boy [ D′ t [DP that you saw [ D′ t [PP except Bill]]]]]]]
This result follows under the view that every-except is a triadic determiner, as proposed
in (32)–(33). This entails that every-except must first combine with its third argument
before it can be subject to the rule in (53), which combines relatives only with Ds
denoting dyadic determiner relations. Every-except does not become dyadic until after
its third argument is saturated. So the exception phrase must combine first, that is, at
lowest point in the tree.
There is one interesting qualification of the facts in (61) concerning heaviness effects.
Consider the dialogue in (62), where the relative clause is given heavy stress to convey
contrastive emphasis; or consider the example in (63), where the relative is quite long:20
(62) A: What was every boy except John wearing?

B: Well, every boy except John THAT I SAW was wearing a kilt.
(63) Every boy except John [that was roaming the highlands of Scotland] was wearing a kilt
In both cases, the otherwise disfavored order—exception phrase followed by relative

clause—seems to improve.
The view that I wish to support, ultimately, is that (61a,b) represent the true rela-
tionship between the relative and the exception phrase, and that (62)–(63) represent
forms derived by movement from the equivalent of (61a). To motivate this proposal,
we examine a broader, parallel phenomenon that arises in the context of multiple rela-
tive clauses.
3.3. Multiple Relatives

Multiple relatives receive very different representations in the nominal modifier
analyses versus the D-Raising account proposed here. On the former, examples like
(64a) stack upward to the right (64b), whereas on the latter they branch downward
(64c).
(64) a. The woman who I like who I invited (came to the party).
b. DP c. DP
D NP Pro D′
the NP CP D DP
who I invited the NP D′

NP CP
woman who I like woman D DP
the CP D′
who I like D CP
the who I invited
Note an apparent difference of scope with respect to the two relatives. In the former
tree, who I like is the first restrictor of the nominal, with who I invited restricting the
result. In the latter, who I invited restricts the nominal first (according to our semantics),
with who I like following.
Under a standard semantics using set intersection like the one adopted here,
differences of relative scope with relative clauses are not truth-conditionally significant.
This is because set intersection is a commutative and associative operation. If we have
a noun set N, and two relative clause sets R1 and R2, then the result of intersecting N
with R1, and the result of that intersection with R2, will always be the same as the result
of intersecting N with R2 the result of this intersection with R1:
(65) (N ∩ R1) ∩ R2 = (N ∩ R2) ∩ R1
In many cases multiple relatives do not appear to show relative scope, so that inverting
the order of multiple relatives carries no difference in meaning. For example, when defi-
nite descriptions like (66a,b) are used referentially21 and read with neutral intonation,
permutation of relatives has no apparent effect; both pick out the same individual:
(66) a. the man [that you met] [that I talked to]
b. the man [that I talked to] [that you met]
In other cases, however, particularly when emphasis is added, there does appear to
be a scopal difference, even if not one that is truth-conditionally significant. Consider
(67a,b), where capitalization indicates stress or emphasis:
(67) a. Every woman who enjoys books WHO READS SHAKESPEARE (will like this).
b. Every woman who reads Shakespeare WHO ENJOYS BOOKS (will like this).
(67a) is naturally understood as saying something like this: “among women who enjoy
books, every one who reads Shakespeare will like this.” By contrast, (67b) conveys:
“among women who read Shakespeare, every one who enjoys books will like this.”
The force of this difference is clear in question-answering contexts like (68), where the
questioner sets up the domain as women-who-enjoy-books, and hence the answerer
must quantify over this same domain:
(68) A: Which women who enjoy books will like this?
B: Every woman who enjoys books WHO READS SHAKESPEARE will like this!
#Every woman who reads Shakespeare WHO ENJOYS BOOKS will like this!
Summarizing informally, woman who enjoys books is perceived as a semantic constituent

in (67a), and woman who reads Shakespeare is perceived as a constituent in (67b).
When these results are matched up against the structures in (64), the stacking theo-
ries of relative clauses appear to be favored. (64b) seems to project the correct constitu-
ency relations whereas (64c) does not. In fact, however, I think that an appropriate
structure can be assigned under the D-Raising theory once another parallelism between
VP and DP is acknowledged.
3.3.1. Light Predicate Raising in VP

Pairs like (69a,b), involving permutation of an object and other VP material, have been
widely analyzed in the literature as the product of a movement operation that shifts the
object rightward (69c), adjoining it at the edge of VP. This operation typically involves
phrases that are phonologically “heavy” in relation to the material shifted over; hence
it is referred to as “Heavy NP Shift”:
(69) a. John gave a picture of Mary to Bill.
b. John gave to Bill a picture of Mary.
c. John [ VP gave t [ PP to Bill]] [ DP a picture of Mary]
Larson (1988a, 1989) proposes an alternative analysis of this phenomenon using VP

shell structures. The basic idea is that examples like (69b) are not instances of right-
ward movement of a heavy nominal, but rather leftward movement of a light predicate.
Accordingly, the phenomenon is rechristened “Light Predicate Raising.” The key ingre-
dient of this account is a reanalysis rule that permits thematically transitive phrases—
XPs with two unassigned thematic roles—to be categorially reanalyzed as X0s. Reanaly-
sis allows the entire transitive phrase to undergo head raising.
To illustrate, consider (70a), which is similar to our earlier (17a) but contains a
heavy object all the salt he had. The sentence receives the underlying VP shell structure
in (70b). Since the verb put is ditransitive, when it combines with a location phrase the
resulting V′ (put on the fish) is thematically transitive, with the two thematic roles θagent
and θtheme unassigned. This entails that V′ can undergo V′ Reanalysis, as shown in (70d).
Once put on the fish is reanalyzed as a head, this element can raise around the object,
resulting in a right-peripheral position for the object (70d):22
(70) a. John put all the salt he had on the fish.
b. [VP John [V′ e [VP the salt he had [ V′ put on the fish]]]]
c. VP
V′
John V VP Unmarked Word Order
e DP V′
all the salt V PP
he had
put on the fish
d. VP
DP V′
John V VP
V′ Reanalysis +
put on the fish DP V′ V ‘Light Predicate Raising’
all the salt e

he had
Although the exact nature of V′ Reanalysis is somewhat unclear in Larson (1988a),
one property it clearly must have is that it does not yield X0s that are opaque to further
syntactic rules, including further head raising. This is clear from the need for V to raise
out of V′/ V0 for inflection in examples like John was putting on the fish all the salt he
had. The progressive verb form (putting) must assume a local relation with progressive
be for agreement.23 V′ Reanalysis must therefore be viewed as producing X0s only in
the limited sense of items that can be treated as unprojected. Reanalysis clearly does not
produce a morphological “word.”
3.3.2. Light Predicate Raising in DP?

Although reanalysis and Light Predicate Raising were originally introduced in the con-
text of VP, we have seen that notions like monadic/intransitive and dyadic/transitive
can be carried over from VP to DP. A transitive V-predicate is one with Θ-roles like
ΘAGENT and ΘTHEME to assign. A transitive D-predicate is one bearing Θ-roles like ΘSCOPE
and ΘRESTRICT. Accordingly, there seems to be no barrier to a more general notion of X’
Reanalysis, allowing any transitive X’ to reanalyze as X0.
To illustrate this extension, consider again our multiple relative example every
woman who enjoys books WHO READS SHAKESPEARE. Suppose this example has
the underlying form in (71a), in which the outer relative is projected higher than the
inner one. Every denotes a binary determiner relation and, under our semantics (53), so
does the result of combining every with a relative clause. Thus every who enjoys books
also denotes a binary determiner relation. Since [D′ every who enjoys books] is themati-
cally transitive, it is subject to reanalysis as a head (71b).
(71) a. DP
Pro D′
D DP
e NP D′
woman D DP
e CP D′
WHO READS D CP
b. DP SHAKESPEARE
every who enjoys books
Pro D′
D DP
every NP D′
woman D DP D' Reanalysis
every CP D
WHO READS D CP
SHAKESPEARE
This allows it to raise as a unit around the CP to its left (72):

(72) DP
Pro D′
D DP
e NP D′
woman D DP
every who enjoys books CP D
t
WHO READS SHAKESPEARE
Finally, the D head raises out of the reanalyzed portion to the higher empty D position
(73), yielding the desired surface order every woman who enjoys books WHO READS
SHAKESPEARE:
(73) DP
Pro D′
D DP
every NP D′
woman D DP
CP D
WHO READS SHAKESPEARE t

The same general derivation will account for our earlier examples (62) and (63) (repeated
below) in which the order of the relative clause and exception phrase is opposite to what
we expect:
(62) A: What was every boy except John wearing?
B: Well, every boy except John THAT I SAW was wearing a kilt.
(63) Every boy except John [that was roaming the highlands of Scotland] was wearing a kilt.
These examples can be analyzed as deriving by D′ Reanalysis of the lower D′ containing
the determiner + the exception phrase. This complex D is then raised, with the deter-
miner later raising on its own (74):
(74) DP
Pro D′
D DP
every NP D′
boy D DP
D PP CP D
t except John that I saw t

Prenominal Relatives
It is natural to ask what blocks the raising of the entire reanalyzed D′ into prenominal
position, producing the ungrammatical (75). In fact, the same question arises with the
simplest examples of relatives (76a). D′ Reanalysis might lead us to expect the ungram-
matical (76b), where a reanalyzed D′ raises around NP (76c):
(75) *every who enjoys books woman WHO READS SHAKESPEARE
(76) a. every woman who enjoys books
b. *every who enjoys books woman
c. [DP Pro [D every who enjoys books] [DP [NP woman] t]]
Examples like (75) and (76b) are plausibly ruled out by the same general constraint
on prenominal items observed in cases like (77) and (78). It has been widely observed
that prenominal modifiers must typically occur head-adjacent to the nouns they modify.
Complements of prenominal adjectives are thus excluded since they prevent adjacency
(77ai–ci). Either the complement must be “extraposed” rightward (77aii–cii), or else the
entire adjective + complement must occur postnominally (77aiii–ciii). Similar remarks
apply to the PPs in (78):24
(77) a. i. *a similar to Bill man
ii. a similar man to Bill
iii. a man similar to Bill
b. i. *a fun for children game
ii. a fun game for children
iii. a game fun for children
c. i. *an unfortunate for Max complication
ii. an unfortunate complication for Max
iii. a complication unfortunate for Max
(78) a. i. *an at two o’clock meeting
ii. a meeting at two o’clock

b. i. *a nearby the park restaurant
ii. a nearby restaurant (*the park)
iii. a restaurant nearby (the park)
Relative clauses are widely analyzed as CPs, with heads initial in their phrase in
a language like English (79a). Consider then a raised, reanalyzed D′ structure like
(79b). If the whole complex D is considered as the prenominal element, then it will
clearly fail the requirement of head-adjacency: D is headed by every, and the latter
is not adjacent to woman. Similarly, if the relative CP is considered as the pre-
nominal element, it will also violate the requirement since C is not head-adjacent
to woman:
(79) a. [CP who C [IP t enjoys books]]
b. [DP . . . [D every [CP who C t enjoys books]] woman t ]
It follows, then, that raising of a complex D′ into prenominal position will always
be excluded, and hence stranding derivations of the kind in (73) and (74) will be
required.
3.4. Other D-Modifiers

The account of relative clauses proposed above can be extended to other categories of
postnominal nominal modifiers, including postnominal PPs and APs like those illus-
trated in (80):25
(80) a. the man [PP at the podium] [PP in a grey suit]
b. three women [AP present] [AP capable of lifting a sofa]
c. every book [PP on the shelf] [AP published since WWII]
These can be analyzed as projected into low positions as complements of D, with sub-
sequent raising of the determiner (81):
(81) [DP Pro [ D′ every [DP book [ D′ t [DP [PP on the shelf] [ D′ t [AP published since WWII]]]]]]]
The intersective semantics for relative clauses can likewise be directly extended to
these categories. We simply generalize our rule to cover all predicative XPs of this
kind:
(53’) Let Δ be a determiner projection denoting a determiner relation D(X,Y), where Y has
the role θRESTRICT. Let XP be an AP, PP, or relative CP denoting the set R. If Δ is Δ, then
[D′ D XP ] denotes the relation D′(X,Y), where Y has the role θRESTRICT and D′(X,Y) iff
D(X,(Y ∩ R)).
If Δ is Δ′, then [DP CP D ] denotes the relation D′(X,Y), where Θ has the role ΘRESTRICT
and D′(X,Y) iff D(X,(Y ∩ R)).
3.4.1. Prenominal APs

These results raise an interesting question as to how prenominal adjective modifiers
should be analyzed. The adjectives in (82a–c) combine with their nominal in a way
equivalent to that of a relative clause. All involve an intersective semantics:
(82) a. the tall woman
(cf. the woman who is tall)
b. every beautiful house
(cf. every house that is beautiful)
c. three blind mice
(cf. three mice that are blind)
Under the general analysis pursued here, there appear to be few options. If we attempt
to treat prenominal adjectives as base-generated in the D-projection, along the lines of
(83), then we must analyze them as items that can be combined with D between the
scope and restriction arguments:
(83) DP
Pro D′
D DP
three D′
blind D NP
three mice
Achieving this is not straightforward, however. A crucial element in our approach to

intersective D-modifiers is that they interact semantically with transitive determiner rela-
tions—in essence, they take transitive Ds and form larger, complex transitive Ds (recall
(53) and (53’) above). This in turn requires that the restriction phrase (NP) not be com-
bined with D at the point where the modifer is added in. Evidently, this requirement is not
met in (83); here D has already combined with NP at the point where AP is encountered.
If we cannot utilize the equivalent of (53) or (53′), the only obvious alternative for
generating (83) is to treat prenominal APs as arguments of D in their own right, assign-
ing them a thematic role lying between ΘSCOPE and ΘRESTRICT:
ΘSCOPE > ΘX > ΘRESTRICT
But this move is also problematic. Prenominal adjectives are optional DP elements, and
although there is no problem making them arguments of D like relative clauses, it does
seem quite strange to locate an optionally assigned thematic role (ΘX) between two
obligatorily assigned thematic roles (ΘSCOPE, ΘRESTRICT). Furthermore, it is well known
that prenominal adjectives are iterable, so that we can get a number of such elements
together (84):
(84) a. three German mice
b. three blind German mice

c. three grey blind German mice
d. three furry grey blind German mice
e. three small furry grey blind German mice
f. three excellent small furry grey blind German mice
On the approach being considered, this would seem to entail expanding the thematic
hierarchy to include a number of roles between ΘSCOPE and ΘRESTRICT, all of which must
be optional:
ΘSCOPE > ΘX1 > ΘX2 > ΘX3 > ΘX4 > ΘX5 > ΘX6 > ΘRESTRICT
Note that none of these issues arises in our approach to relative clauses and other post-
nominal restrictive modifiers, such as PP and AP. The latter were not analyzed as argu-
ments of D, and not as part of the nominal thematic hierarchy, but rather as elements
that were (recursively) added in by a process forming complex Ds.
If a base-generated approach to prenominal adjectives is problematic, an attractive
alternative is to adopt some version of the proposal by Smith (1964) and Jacobs and
Rosenbaum (1968) that prenominal adjectives originate as postnominal modifiers and
obtain their surface position by movement (85):
(85) DP
Pro D′
D DP
e D′
mice D AP
three blind
Under this approach, the specific problems raised above for (83) disappear. However,
at least two new issues arise. First, we require an account of the precise mechanism by
which adjectives generated postnominally are advanced to prenominal position. This
account must accommodate the familiar fact that adjectives in prenominal position
appear to obey certain (universal) restrictions on order of occurrence, which, for exam-
ple, rule out combinations like those in (86) (uttered with neutral intonation) (Dixon
1977; Hetzron 1978; Sproat and Shih 1991):
(86) a. ?*three blind small mice
b. ?*three grey small blind mice
c. ?? three small blind furry grey mice
d. *three German small furry grey blind mice
e. ?*three furry excellent small mice

Second, such an analysis must deal with the fact that not all prenominal adjectives
have the intersective semantics found with relative clauses and, postnominal PPs and
APs. Cases like (87)–(89) are familiar examples:
(87) a. Olga is an alleged dancer.
(cf. *Olga is a dancer who is alleged.)
b. Alice is an imagined werewolf.

(cf. ≠ Alice is a werewolf who is imagined.)
c. Boris is a supposed perpetrator of a crime.
(cf. *Boris is a perpetrator of a crime who is supposed.)
(88) a. Olga was a reluctant dancer.
(cf. ≠ Olga is a dancer who is reluctant.)
b. Boris was a willing perpetrator of a crime.
(cf. ≠ Boris was a perpetrator of a crime who was willing.)
(89) a. Olga is a beautiful dancer.
(cf. ≠ Olga is a dancer who is beautiful.)
b. Kathrin is a skillful manager.
(cf. ≠ Kathrin is a manager who is skillful.)
c. Peter is an old friend.
(cf. ≠ Peter is a friend who is old.)
I believe that both of these issues can be dealt with satisfactorily, and the picture in (85)
maintained; however, justifying this claim would require extensive additional discus-
sion, which I put aside for development elsewhere.
4. GENITIVES
The postulation of a Pro subject in all DPs has strong consequences for the analysis of
prenominal genitive constructions like (90a–d):
b. John’s picture
c. John’s grandmother
d. John’s completion of the plan
As noted earlier, Abney (1987) assimilates the structure of genitive DPs to clauses (IPs), with
the possessive element occupying a subject-like position; recall (10a,b) (repeated below):
(91) a. DP b. IP
DP D′ DP I′
John D NP John I VP
’s completion of the plan [TNS] complete the plan
Szabolsci (1983) further develops the clausal analogy with Hungarian examples like
(91), in which the possessive item co-occurs with a definite article. Szabolsci analyzes
the latter as counterpart to a complementizer; compare (92a,b):
(91) (a) Mari kalap-ja-i
(the) Mari hat-poss-pl-2sg
‘Mari’s hats’
(92) a. DP b. CP
Spec D′ Spec C′
D (N+I)P C IP
(a) DP (N+I)′ that NP I′
Mari-NOM kalap-ja-i Mari-NOM ran
The analysis of DP structure developed here does not comport with the basic sentential
analogy. As we have seen, on the current account the highest argument position in a
quantified DP—its thematic “subject”—is always the scope argument Pro. The posses-
sive item therefore cannot be structurally parallel to a subject, and, by extension, the
Hungarian definite determiner cannot be parallel to a complementizer. In place of the
sentential picture, a rather different analogy suggests itself.
4.1. Possessive Ds as Triadic Predicates

Genitive DPs are familiar as definite nominals (McCawley 1988; Neale 1990). Suppose
we view Hungarian as displaying the “true shape” of the genitive DP, where the head is
a definite determiner and where the genitive-marked possessor occurs below the definite
D. As a first approximation, we might propose the analysis in (93) for John’s briefcase,
where the possessor (John) and possessed (briefcase) are both arguments of a definite
determiner (THE), which raises:
(93) a. [DP Pro e [DP D′ THE briefcase]]]
b. [DP Pro THE [DP D′ t briefcase]]]
In Hungarian this definite determiner would be phonetically realizable, whereas in English

(as in many other languages) it would be necessarily covert.
Notice that on this proposal, genitive nominals become a form of triadic DP con-
struction, in which the two lower arguments of D (John and briefcase) stand in a pos-
sessive relation. This situation is interesting given the general parallelism between DP
and VP that arises in our theory. Larson (1988a, 1990a, 1991) proposes an analysis
very similar to (93) for a class of triadic VP constructions in which the two lower
arguments of V stand in a possessive relation: namely, double object structures. An
example like Mary gave John a briefcase, for instance, gets an analysis approximately
as in (94a,b):
(94) a. [VP Mary e [VP John [ V′ gave a briefcase]]]
b. [VP Mary gave [VP John [ V′ t a briefcase]]]
What these points suggest, then, is that rather than viewing genitive nominals as clause-
like, with the possessor analogous to a subject and the definite determiner parallel to
C, we might instead see them as VP-like, with the possessor analogous to an object and
the definite D parallel to V. Specifically, we might analyze genitive nominals as the DP
equivalents of double object constructions in the verbal domain.
4.1.1. Prepositional Datives and “Dative Shift” in VP

The analogy between prenominal genitives and double object constructions can be
developed further, through a more refined view of the latter. Larson (1988a, 1990a,
1991) proposes that prepositional datives like (95a) involve a relatively transparent
source, in which the goal argument is projected lower than the agent and theme (in
accordance with the thematic hierarchy in (14)), and where the dative verb raises
(95b):
(95) a. Mary gave a briefcase to John.
b. VP
DP V′
Mary V VP
give DP V′
a briefcase V PP
give to John
By contrast, double object constructions have a more complex derivation, which

involves a modern version of “Dative shift”. An example like (96a) is assigned the
underlying VP in (96b), where the goal (John) is initially projected into a low posi-
tion, without the preposition to that would normally accompany it, and where the
theme is a V-bar adjunct.26 Absence of the Case marking provided by to triggers
NP Movement. The result is (96c) (where the verb has also raised to the higher V
position).
(96) a. Mary gave John a briefcase.
b. VP c. VP
DP V′ DP V′
Mary V VP Mary V VP
e DP V′ give DP V′
e V′ DP John V′ DP
V DP a briefcase V DP a briefcase
give John give John
As discussed in Larson (1988a, 1990a, 1991), this account respects a strong theory of
projection, in which the thematic hierarchy is directly reflected in the relative heights of
arguments. Hence the goal argument (John) starts out lower than the theme argument
(a briefcase) in initial structure. But it also allows for the important observation (due
to Barss and Lasnik (1986)), that in a double object construction the goal argument
appears to c-command the theme argument at surface form. This result is achieved by
raising the goal to the higher position.
4.1.2. Postnominal Genitives and “Genitive Shift” in DP

This analysis of prepositional datives, double object structures, and their relationships
can be extended directly to postnominal and prenominal genitives, following the basic
analogy suggested above. Postnominal genitive constructions like (97a) can be assigned
a relatively transparent initial structure as in (97b), where the genitive PP is treated as
an oblique modifier and projected lower than the scope and restriction arguments of D,
in accordance with the hierarchy discussed earlier. The definite determiner subsequently
raises, yielding the correct surface order.27
(97) a. the briefcase of John’s (that Alice lost/on the desk/taken)
b. DP
Pro D′
D DP
THE NP D′
briefcase D PP
THE of John’s
Postnominal genitives thus become the DP equivalents of the prepositional dative con-
structions.
By contrast, prenominal genitives receive a more complex movement derivation,
involving what we might call “genitive shift.” Example (98a) is assigned the underlying
DP in (98b), involving the covert definite determiner THE. Here again, the possessor
(John’s) is projected into a low position, but now without the preposition of. The restric-
tion argument (briefcase) is projected as a higher D′ adjunct. Absence of the Case mark-
ing provided by of once again triggers movement. Finally, THE raises to the higher D
position, yielding (98c):
b. DP c. DP
Pro D′ Pro D′
D DP D DP
e DP D′ THE DP D′
e D′ NP John’s D′ NP
D DP briefcase D DP briefcase
THE John’s THE John’s
As in the double object derivation, this account maintains a strict mapping between the
thematic hierarchy and the relative structural height of arguments. Thus the possessor
argument (John’s) is projected lower than the restriction argument (briefcase) in both
the prenominal and postnominal constructions. But this account also allows for the fact
that in the prenominal genitive construction, the possessor c-commands the restriction
argument at surface form, as shown by examples like (99a,b) involving negative polar-
ity items and anaphors. This result is achieved by raising the possessor to the higher
DP-Spec position.
(99) a. no one’s picture of anything
(cf. *anyone’s pictures of no one)
b. their pictures of each other
(cf. *each other’s pictures of them)
4.2. Consequences and Comparisons

The analysis sketched above entails that prenominal genitives always achieve their sur-
face position by movement and that the genitive DP always originates as an (oblique)
argument of DP. Let us examine these points more carefully, considering two basic
classes of prenominal genitives identified in the literature: so-called lexical, or nonthe-
matic genitives, in which DP plainly does not bear a θ-role assigned by N, and thematic
genitives, in which DP at least appears to bear a role assigned by N.
4.2.1. Nonthematic Genitives

Nonthematic genitives include examples like (100a–d), where, in each case, the only the-
matic role assigned by N (briefcase, arm, accessories, afternoon) is the usual one going
to its external argument. As many have noted, the exact relation between possessor and
possessed is typically vague in these cases, and not confined to ownership. Thus John’s
briefcase can refer to one he owns, but it can also refer to one near him, one he was talk-
ing about, and so on. Similarly Mary’s arm can refer to her own limb, but also one she is
holding on to, or one lying before her on a dissection table. And so on.
(100) a. John’s briefcase (is on the veranda).
b. Mary’s arm (is tanned).

c. Men’s accessories (are in the next aisle).
d. Jill’s afternoon (was hectic).
On the present account, examples like (100a–d) derive by movement from the position
of a postnominal genitive of-PP (101):
(101) a. [DP the briefcase of John’s ]
b. [DP THE John’s briefcase ______ ]
A movement account of nonthematic prenominal genitives is not new. Ross (1967,

1981), Chomsky (1970), Stockwell, Schacter, and Partee (1970), and McCawley (1988)
all offer analyses that include the equivalent of (101a,b) at some derivational stage. For
example, McCawley (1988) proposes the steps in (102a–c), where the genitive originates
as the predicate of a copular relative clause (102a) that is subsequently reduced (102b),
and where the genitive either combines with of (102ci) or fronts to prenominal position
(102cii). Stockwell, Schacter, and Partee (1970) observe that the predicate genitives like
(102a) show essentially the same range of readings found in (nonthematic) prenominal
genitives and, hence a derivational relation between them seems semantically sound:
(102) a. the briefcase [which is John’s]
b. the briefcase [John’s] (from (102a) by Relative Clause Reduction)
c. i. the briefcase [of John’s] (from (102b) by of-insertion)
ii. John’s briefcase ___ (from (102b) by fronting DP’s)
McCawley’s analysis anticipates the one proposed here on several important points.
In McCawley’s account, as in ours, the prenominal genitive derives by fronting from
the postnominal position of a genitive of-PP. Furthermore, for McCawley postnominal
genitives occupy the same position as possessive relatives, a parallelism that also holds
in our account, where relative clauses (including possessive relatives) and possessive PPs
like of John’s are analyzed as oblique D-arguments. The main divergence between the
proposals is the assumption that nonthematic prenominal genitives literally derive from
possessive relatives. That apart, the analyses are very similar.
It is interesting to note in this context that many languages show a formal similarity
in the marking of relative clause and genitive constructions. The Australian languages
Dyirbal and Gumbaingar, discussed by Dixon (1966), illustrate this phenomenon. As
(103a,b) illustrate, the nu suffix appearing on the verb in Dyirbal relatives (103a) also
occurs in Dyirbal genitives (103b):
(103) a. yibi yara-ngu njalnga-ngu djilwa -nu -ru bura-n.
woman-nom man-erg child-erg kick -rel -erg see-tns
‘The man who had been kicked by the child saw the woman.’
b. njalnga guda-ngu yara -nu ndjin-du badja-n.
child-nom dog-erg man -rel erg bite-tns
‘The man’s dog bit the child.’
Dixon (1966) argues that this fact is not coincidental: that Dyirbal and Gumbaingar
possessive nominals actually derive from possessive relatives and that their shared mor-
phology reflects this shared derivational history. Dixon’s results (and the general con-
vergence between relative clauses and genitives) also appear compatible with the weaker
proposal made here, that genitive DPs are generated in the same position as relatives
(without being literally derived from the latter).28
4.2.2. Thematic Genitives

The “genitive shift” analysis appears more problematic for thematic genitives like those
in (104) and (105), where the possessive-marked DP appears to bear a thematic role
assigned by N. Thus, in (104), John seems to receive an agent role from N, and in (105)
John appears to receive a theme role from N (at least on one reading).
(104) a. John’s examination of the plan
(cf. John examined the plan.)
b. John’s selection of the winner
(cf. John selected the winner.)
(105) a. John’s election
(cf. They elected John.)
b. John’s grandmother
(cf. the grandmother of John)
c. John’s picture
(cf. a picture of John)
Such facts naturally suggest analyses in which John is an underlying argument of N. For
example, Chomsky (1970, 1981), Anderson (1983/1984), Kayne (1984), and Giorgi
and Longobardi (1990) (among others) take the possessive DPs in (104) to be base-
generated in the subject position of the nominal, parallel to the subject position of a
clause (106). (105a–c) are taken to derive by movement of the theme argument of N to
subject position, much like what occurs in a clausal passive (107).29
(106) a. [John’s selection of the winner]
b. [John selected the winner]

(107) a. [ the election of John ]
b. [ John’s election ____ ]
c. John was elected ___
On the analysis entertained here, these proposals are not available, however. Assuming
genitive DPs to be derived uniformly, none of the prenominal genitives in (104) or (105)
is base-generated; all undergo movement. Furthermore, none originates as an argu-
ment of N; instead, all are generated initially as oblique D-arguments. Under our own
assumptions about locality of θ-role assignment, this view appears to entail that with
relational nouns and in nominalizations, prenominal genitives do not in fact receive a
θ-role directly from N, but rather via some other, more indirect mechanism.
4.2.3. The Semantics of Thematic Genitives 30

Interestingly, Grimshaw (1990) has drawn essentially the same conclusion. In a thor-
oughgoing study of nominalizations, Grimshaw argues that, despite appearances, rela-
tional nouns and nominalizations never assign thematic roles directly to prenominal
genitives, and hence possessives are never parallel to verbal arguments in this respect.
Grimshaw terms such phrases “argument adjuncts,” a label expressing their paradoxi-
cal status as appearing to bear a thematic role assigned by N, but fully optional like
unselected adjuncts, and unlike true subjects.
More recently, Burton (1995) has advanced an attractive semantic proposal that imple-
ments Grimshaw’s conclusions. Following a number of authors, including Higginbotham
(1983), Partee (1983/1997), and Williams (1985, 1987), Burton assumes that possessives
are headed by a definite determiner containing a free variable R over relations.31 In geni-
tives containing a nonrelational noun, like (108a), R is determined deictically (108b).
b. [the x: briefcase(x) & R(x,John)]
Letting the value of R vary with context, John’s briefcase may thus denote the briefcase
that John owns, the briefcase he is holding, the briefcase sitting on the desk in front of
him, and so on.
By contrast, in genitives containing a relational noun, like (109a), the value of R is
determined in one of two different ways. One way is through contextual determination,
as before. John’s wife thus denotes the individual who is a wife of someone and who
John stands in some contextually given relation to (109b). Although not the normal
understanding with relational nouns, this reading can be made pragmatically acces-
sible. Imagine a diplomatic visit by a male dignitary from a foreign country allowing
polygamy. During the visit, each of the diplomat’s wives is assigned her own individual
security agent. John is one of these agents. In such a situation, John’s wife might natu-
rally denote the wife that John has been assigned to protect; R is contextually fixed as
the assignment relation, and so on.32
(109) a. John’s wife
b. [the x: ∃y[wife(x,y) & R(x,John)]]

Along with the deictic reading, there is also a (much more favored) “thematic” reading
where John’s wife is understood to denote the individual that John is married to. Burton
derives this reading in an interesting way, proposing that it arises, in effect, by a form
of "inner anaphora," in which the relational noun serves as the antecedent of R (110a)
and determines its value (110b).33
(110) a. [the x: y[ wife(x,y)] & R(x,John)]]

antecedes
b. [the x: y[ wife(x,y)] & wife(x,John)]]
c. [the x: wife(x,John)]
As Burton observes, the complex expression (110a/b) can be shown to be semantically

equivalent to the simpler (110c), but note an important difference between the two.
Under (110a), John is not a direct argument of the relational noun wife. Rather, it is
an argument of the relation R provided by the definite determiner. John comes to be
understood as a semantic argument of wife through an indirect chain: John is an argu-
ment of R whose value is given by the relational N wife.
Burton’s semantics appears to be fully compatible with the syntactic results derived
above, according to which prenominal possessors are never direct arguments of N,
even when N is relational. Rather, the possessor is an argument of the R variable in D,
which gets its value through N. This proposal appears to be generalizable to all rela-
tional nouns, and to nominalizations as well, along the lines of (111c), using the event
semantics of Davidson (1967).34
(111) a. Nero’s destruction of Rome
b. [the e: x[ destruction(e,x,Rome)] & R(e,Nero)]
antecedes
c. [the e: x[ destruction(e,x,Rome)] & destruction(e,Nero,Rome)]
d. [the e: destruction(e,Nero,Rome)]
Again (111c) is equivalent to the simpler (111d), but once again under (111b), Nero is
not a direct argument of the nominalized form destruction. Rather, it is an argument of
the determiner—specifically, its relation R, whose content is determined by the noun.
Under these proposals, then, the second major assumption of the genitive shift
account appears to be sustainable. Indeed, the analysis seems to accord naturally with
Grimshaw’s (1990) conclusions regarding the “argument adjunct” status of prenominal
genitives in the context of relational nouns and nominalizations.
5. DEGREE PHRASES AND DEGP
The general approach to DP structure pursued here can be directly extended to other
functional categories whose semantics is relational and quantificational. Degree modi-
fiers of adjectival expressions are a potential case in point.
5.1. Comparatives as Quantificational

Predicative adjectives are often analyzed in logic books as simple unary predicates of
individuals (112); however, examples like (113) and (114) show this view to be sim-
plistic. (113a) appears to attribute, not simply intelligence, but a certain degree of intel-
ligence; that is, the adjective appears to relate individuals and degrees (113b). Likewise,
(114a,b) appear to compare degrees of intelligence.
(112) a. Lester is smart.
b. Smart(l)
(113) a. Lester is that smart.
b. Smart(l,d)
(114) a. Lester is smarter than Kenton.
b. Kenton is as smart as Lester.
There is evidence that certain degree-modified adjectival expressions are quantifica-

tional in nature. For example, equative comparatives like (115a) can be directly para-
phrased with expressions independently argued to be quantificational (Larson 1987),
such as the free relative construction in (115b).
(115) a. Lester will grow as big as Kenton grew.
b. Lester will grow however big Kenton grew.
Furthermore, as first noted by Russell (1905), comparatives appear to participate in famil-

iar de dicto/de re ambiguities, widely taken to be a matter of scope. For example, (116a) is
ambiguous between two readings, one of which is sensible (116b) and one of which is not
(116b). This ambiguity can be analzyed in terms of two positions for the degree-modified
adjectival expression taller than he is, along the lines in (117a,b). In the first, the adjectival
phrase is scoped outside the propositional attitude verb, yielding the sensible reading. In
the second, it takes scope inside thinks, yielding the nonsensical reading.35
(116) a. Maryann thinks Kenton is taller than he is.
b. Kenton’s height is such that Maryann thinks Kenton is taller than that.
c. #Maryann thinks Kenton is taller than himself.
(117) a. [taller than he is ]i Maryann thinks [ Kenton is ti ]
b. Maryann thinks [ [ taller than he is ]i Kenton is ti ]
Finally, comparatives exhibit the Antecedent-Contained Deletion phenomenon, widely

associated with quantificational structures. Larson (1988b) notes that the boldfaced
phrase in (118a), containing an elliptical VP, is a complement of the verb last. As dis-
cussed by Sag (1976), Williams (1977), and May (1985), proper recovery of the elided
material in such circumstances requires the null VP to escape the matrix VP at LF. This
result is achieved if the comparative AP is quantificational and undergoes raising (118b)
with subsequent reconstruction of the missing VP (118c):
(118) a. John’s party will VP last as long as Bill’s party will VP ]].
b. [AP as long as Bill’s party will VP ] John’s party will VP last t ]
c. [AP as long as Bill’s party will VP last t ] John’s party will VP last t ]
RECONSTRUCTION
Ross (1973), Seuren (1973), and Klein (1980) offer an attractive quantificational
semantics for comparatives and equatives, in which adjectives are understood as apply-
ing to pairs of individuals and degrees. Comparatives are analyzed as involving existen-
tial quantification over the degree element (119)–(120), whereas equatives are analyzed
as involving universal quantification (121):
(119) a. Kenton is taller than Lester.
b.
∃d[¬tall(l,d) & tall(k,d)]
(120) a. Lester is less tall than Kenton.
b.
∃d[tall(k,d) & ¬tall(l,d)]
(121) a. Lester is as tall as Kenton.
∀d[tall(k,d) → tall(l,d)]
b.
Note crucially that the notion of “degree” invoked in these formulae is not that of point-
on-a-scale, but rather that of vector: a “directed interval” or “extent.” So understood,
when an adjectival predicate is true of an individual to a degree/extent d, it is true of that
individual to all lesser extents/degrees d′ as well. For example, if Kenton, a chimpanzee,
is tall to degree/extent 5 feet, then he is also tall to degrees/extents 4.5 feet, 4.0 feet, 3.5
feet, and so on.36
(122) 5 ft.
4.5 ft. If Kenton is tall to the degree/extent 5 feet, then he is
tall to all lesser degrees/extents as well.
Kenton
With degrees conceived in this way, (119b)–(121b) seem to correctly express the
truth conditions of (119a)–(121a), respectively. If there is a degree of height that Lester
lacks and Kenton has, there can be no degree of height that Lester has and that Kenton
lacks. Thus, Kenton must be taller than Lester (119a). Likewise, if every degree of height
that Kenton possesses, Lester possesses as well, then Lester must be (at least) as tall as
Kenton (121b). As noted by Klein (1980), this analysis captures certain intuitively cor-
rect inferences as a simple matter of first-order logic. For example, (123a), the negation
of (121a), does not merely imply that Lester and Kenton are of different heights but
specifically that Lester is shorter (123b). This follows directly under the Ross/Seuren
analysis, since the negation of the logical form of the first (124a) entails the logical form
of the second (124b):
(123) a. Lester is not as tall as Kenton. ⊢
b. Lester is less tall than Kenton.

(124) a. ¬∀d[tall(k,d) → tall(l,d)] ⊢
b.
∃d[¬tall(l,d) & tall(k,d)]
5.2. Degree Morphemes as Relational

Larson (1988b) offers a precise compositional semantic analysis of degree morphemes that
adopts the basic Ross/Seuren proposal. Adjectives like red are analyzed as of type <d,<e,t>>—
functions from degrees to functions from entities to truth-values. Comparative degree mor-
phemes -er/more and less are analyzed as triadic relations that combine with an adjective and
two term phrase denotations. The basic analyses of the comparative morphemes -er/more
and less are given in (125a) and (126a), respectively, where Q and P are variables of the type of
DP denotations (<<e,t>,t>) and where A is a variable over adjective denotations (<d,<e,t>>).
Examples are given schematically in (125b–d) and (126b–d). In brief, the degree morpheme
first combines with the comparative complement (the than-phrase element), then combines
with the adjective, and finally combines with the subject.
(125) a. -er/more ⇒ λQλA λP ∃d[¬A(d)(Q) & A(d)(P )]
b. Lester is taller than Kenton.
c. λQλA λP ∃d[¬A(d)(Q) & A(d)(P )](Kenton’)(tall’)(Lester’)
∃d[¬tall’(k,d) & tall’(l,d)]

d.
(126) a. less ⇒ λQλA λP ∃d[A(d)(Q) & ¬A(d)(P )]
b. Lester is less tall than Kenton.
c. λQλA λP ∃d[A(d)(Q) & ¬A(d)(P )](Kenton’)(tall’)(Lester’)
d.
∃d[tall’(k,d) & ¬tall’(l,d)]
The equative degree morpheme receives a similar analysis. As is assigned the interpreta-
tion in (127a), expressing a three-place relation between one DP meaning, an adjective
meaning, and a second DP meaning. An example is given schematically in (127b–d):
as ⇒ λQλA λP ∀d[A(d)(Q) → A(d)( P )]

(127) a.
b. Lester is as tall as Kenton.
c. λQλA λP ∀d[ A(d)(Q) → A(d)(P )](Kenton’)(tall’)(Lester’)
∀d[tall’(k,d) → tall’(l,d)]
d.
Larson (1998b) shows how these proposals can be extended to clausal comparative
and clausal equative complements (like taller than Kenton is and as tall as Kenton is),
and how they can be mapped into a quantificational syntax, in which comparative and
equative phrases like taller than Kenton or as tall as Lester raise and take scope, leaving
a trace in their base position.37
It is revealing to compare this analysis of quantificational degree morphemes with
that of quantification determiners. (128) shows existential some, analyzed as a binary
relation between sets of individuals (128a), and applied in an example (128b–d). (129)
gives the parallel points for the universal determiner every.
some ⇒ λQλP∃x[Q(x) & P(x)]

(128) a.
b. Some man smiles.
c. λQλP∃x[Q(x) & P(x)](man’)(smiles’)
∃x[man’(x) & smiles’(x)]

d.
every ⇒ λQλP∀x[Q(x) → P(x)]

(129) a.
b. Every man smiles.
c. λQλP∀x[Q(x) → P(x)](man’)(smiles’)
∀x[man’(x) → smiles’(x)]
d.
Evidently, the parallels between degree elements and determiners are quite close on this
account, with both receiving a relational analysis. Notice also that the nominal element
in DP and the adjectival element in DegP play very similar semantic roles. In the former,
NP functions to restrict a quantification over individuals, and hence receives the role
ΘRESTRICT. In the latter, AP functions to restrict a quantification over degrees—specifically,
degrees of A as the latter applies to the nominal arguments. Hence AP is also a natural
candidate for the role ΘRESTRICT in our analysis. Likewise, both categories involve a scope
element. In Larson (1988b), the element represented by the variable P in (125)–(127)
functions very much like the element represented by the variable P in (128)–(129). Both
receive their values from the phrase to which DP or DegP adjoins at LF. Hence these ele-
ments are both candidates for the role ΘSCOPE.
5.3. DegP Syntax

The semantic parallels between Deg and D match recent parallel syntactic accounts of
these elements. Like determiners, degree elements were earlier analyzed as specifiers of
a lexical phrase: just as Dets were viewed as specifiers of NP, Degs were analyzed as
specifiers of AP (130a) (Bowers 1975; Jackendoff 1977). More recently, however, Abney
(1987) and Corver (1990) have argued that Deg, like D, heads its own phrasal category,
DegP (130b):
(130) a. AP b. DegP
Deg A′ Deg AP
that/very tall that/very tall

Note that the latter fits smoothly with our relational analysis, in which a degree element
takes an adjectival phrase as its semantic complement. Indeed, our triadic compara-
tive and equative Degs can be directly analogized to the triadic determiner relations
discussed earlier.
Recall that the universal determiners every, all, and no license an exception phrase
following N, whereas other determiners (including universals like each) do not (131a).
We analyzed the relevant instances of every and no as ternary determiners that select
the exception phrase as an initial complement. The latter is stranded by subsequent
D-Raising, which produces a discontinuous dependency (131b):
(131) a. Every/No/*Each/*Some boy but/exceptJohn was present.
b. [DP Pro every [ DP [NP man] t [PPbut/exceptJohn ]]]
Something quite similar is motivated for DegP. Comparative and equative degree mor-
phemes exercise a well-known selection relation on the element introducing the compara-
tive/equative complement (Bowers 1975); -er/more and less require a complement intro-
duced by than (132a), whereas as requires a complement introduced by as (132b).
(132) a. That car is more/less expensive than/*as the one I bought (is).
b. Bill is as tall as/*than Harry (is).
On the basis of this, we might suggest a similar syntactic analysis, which I will sketch
briefly.
Suppose that comparatives and equative degree elements combine with their argu-
ments in the order specified by the semantic analyses in (125)–(127), which, as we have
noted, express essentially the same thematic hierarchy introduced for DP (viz., ΘSCOPE >
ΘRESTRICT > ΘOBLIQUE). Thus, Deg first combines with the comparative/equative comple-
ment, then with the adjective phrase (AP) that functions as the restriction, and then with
a subject of the semantic type of DP that constitutes the scope.
Under our assumptions about syntactic projection, Deg and its first two arguments
project within a minimal phrasal projection DegP containing no position for the subject
argument (133a). This situation prompts the projection of an additional DegP “shell” to
accommodate the subject (133b), which is analyzed as a null element (Pro) whose value
is fixed at LF after DegP is assigned scope, in parallel to the DP case. Deg raises to the
empty Deg position, achieving the correct surface ordering of elements.
(133) a. DegP b. DegP
AP Deg′ Pro Deg′
smart Deg PP Deg DegP
as as Kenton as AP Deg′
θRESTRICT > θ OBLIQUE smart Deg PP
as as Kenton
θ SCOPE > θ RESTRICT > θ OBLIQUE

Comparatives receive the same analysis, except that the complement involves than, and,
in cases like taller than Kenton, there is presumably an extra step in which the head of
AP raises to Deg to support the bound morpheme -er (134a,b).
Pro Deg′ Pro Deg′
Deg DegP Deg DegP
-er Deg′ A Deg AP Deg′

tall Deg PP tall -er tall Deg PP
-er than Kenton -er than Kenton
5.3.1. Too and Enough Constructions

The syntactic analysis offered here for comparatives may be extended to too and enough
constructions, which are known to exhibit similar properties (135a,b).
(135) a. Kenton is too large [to lift].
b. Kenton is strong enough [to lift Lester].
As discussed by Jackendoff (1977), Guéron and May (1984), Baltin (1987), and Nikifori-
dou (1987), the infinitive following AP is licensed only by the too or enough morpheme,
and unavailable without it. Furthermore, even when the too and enough morphemes
occur without an overt infinitive, the latter is understood through context (136):
(136) A: Kenton is too large.
B: Too large to do what?
A: Too large to lift.
It is natural to view these facts in terms of the head-complement relation; specifically,

too and enough select the infinitive, as well as the AP. We might then project them into
DegP structures analogous to those of comparatives. Too would undergo simple raising
(137a), whereas enough would combine raising with the cliticization operation already
observed with -er (137b).38
Pro Deg′ Pro Deg′
Deg DegP Deg DegP
too Deg′ A Deg AP Deg′
large Deg CP strong enough strong Deg CP
too to lift enough to lift Lester

A confident analysis in these terms evidently awaits an explicit semantic analysis. How-
ever, it seems at least clear what direction such an analysis should take: too and enough
should be analyzed as relational Degs quantifying over degrees, with AP providing the
restriction on this quantification.
5.3.2. Light Predicate Raising in DegP?

In addition to extensions to too and enough, it is also interesting to consider phenom-
ena discussed earlier in connection with VP and DP, and ask whether the equivalents
for DegP may also occur. Specifically, we noted that Light Predicate Raising, in which
a head and its complement jointly raise around a higher specifier, seemed to occur in
both (138a,b).
(138) a. [VP John [gave to Mary] all the books in his bookcase t ].
b. [DP every boy [ t except John] that we talked to t ].
Consider in this light the pair of sentences in (139), which appear to be synonymous,
and in which a complex degree modifier appears to the left of the adjective high.
(139) a. The porch was as high as ten feet.
b. The porch was as much as ten feet high.
Under the proposals made above, (139a) would be analyzed as involving a simple DegP,
in which Deg first selects an equative complement and subsequently raises away from
it (140a,b):
(140). a. [DegP e [DegP [AP high] as [PP as ten feet]]
b. [DegP as [DegP [AP high] t [PP as ten feet]]
The derivation of (139b) might be taken as similar, but with Deg and its complement
undergoing “Deg’ Reanalysis” and raising together around high (141). Evidently, the
possibility for this depends on the presence of the adjectival much, which appears to
satisfy certain requirements of as that would not otherwise be met (*as as ten feet
high):
(141) a. DegP
Pro Deg′
Deg DegP Unmarked Word Order
e AP Deg′
high Deg PP
as (much) as ten feet
b. DegP
Pro Deg′
high Deg PP
as (much) as ten feet
b. DegP
Pro Deg′
Deg DegP
Deg’ Reanalysis +
AP Deg′ Deg “Light Predicate Raising”
as much as ten feet
high e
I will not attempt to develop these proposals further, but it should be clear that the rela-
tional semantic analysis of DegP, taken together with the theory of projection developed
here, offers rich possibilities for syntactic analysis.
6. CONCLUSION
Following work by Szabolsci (1983) and Abney (1987), many researchers have pursued
the idea that clauses (CP/TPs) and nominals (DPs) are parallel in structure. Despite its
overwhelming popularity, however, this view is not well supported by semantic analysis.
Indeed, under generalized quantifier theory (Barwise and Cooper 1981; Keenan and
Stavi 1983), which provides the basis of nearly all recent work on quantification, C/T
and D have little in common.
In this paper, I have discussed the syntactic projection of DP from the standpoint of
generalized quantifier theory and have argued that, under the latter, the most appropri-
ate analogy is not between DP and CP/TP, but rather between DP and VP. Specifically,
I have suggested (i) that DP can be understood as projecting arguments according to
a thematic hierarchy that is parallel to (but different in role-content from) that found
in VP; (ii) that Ds sort themselves into intransitive, transitive, and ditransitive forms,
much like Vs; and (iii) that nominal modifiers, including relative clauses, project in
the DP very much like adverbial elements in VP. A surprising consequence of this view
concerns prenominal genitives, which have (since at least Chomsky 1970) been taken
to be sentence-like in many cases. I have suggested that, on the view argued for here,
prenominal genitive constructions are fundamentally parallel to double object forms in
the VP, arising by a form of “genitive shift.”
Finally, I have briefly sketched how, under a quantificational/relational analysis of
degree elements, conclusions about DP might be extended to DegP. If correct, these
proposals suggest that many of the putative parallels between DP and CP/TP claimed
over the last 15 years merit serious rethinking.
NOTES
1. Material in this paper was presented at MIT (1988),at CUNY (1991), and in the 1991
Syntax Colloquium series at University of Indiana–Bloomington. I am grateful to audience
members for helpful comments and suggestions, including Bob Fiengo, Janet Fodor, Steve
Franks Jerry Katz, Richard Kayne, David Pesetsky, Ken Safir, and Laurie Zaring.
2. See Barwise and Cooper (1981) for discussion of the non–first-order character of most and
related quantifiers.
3. The relational view, which treats quantifiers as full-fledged lexical items with independent
meaning, has a number of virtues beyond expressive richness. Importantly, it permits us to talk
about lexico-semantic properties of specific determiners, such as (in)definiteness and direc-
tional entailingness, and universal properties of determiners, such as conservativity. For more
on this see Barwise and Cooper (1981), Keenan and Stavi (1983), and Larson (1990b).
4. The equivalent is adopted in Relational Grammar in the form of a Relational Hierarchy (see
Perlmutter 1981; Perlmutter and Rosen 1984).
5. The term QOBLIQUE is a cover term embracing the various adverbial elements. In Larson
(1990a) it is suggested that these might be further ordered as:
QMANNER > QLOC > QTEMP > QCAUSE/PURPOSE
See section 3 for discussion of adjunct projection in DP.
6. The standard Montague Grammar analysis of quantification takes (the equivalent of) DP to
apply to its scope S as function to argument (i) (ignoring intensions):
(i) [[ [ DPi S ] ]]M,g = [[ DP ]]M,g(λxi [[ S ]]M,g)
Under the syntax proposed here, Pro denotes a distinguished variable R over characteristic
functions, and we specify:
(ii) [[ [ DPi S ] ]]M,g = 1 iff [[ DP ]]M,g’ = 1, where g’ is that R-variant of g such that g’(R) = (λxi [[ S ]]M,g)
7. It is natural to ask whether there are also “unaccusative determiners.” Presumably this
would be a D whose scope arose in the position of the restrictive term. To answer this ques-
tion would require a better understanding than I have at present of the hierarchy by which
D arguments are projected.
8. The dyadic-triadic alternation proposed here may be viewed as roughly analogous to that
found with the verb write. The latter has a dyadic form write(x,y) that means (approxi-
mately) ‘x produces y by writing characters’. But write also has a triadic form write(x,y,z)
that means (approximately) ‘x communicates with z using y produced by writing characters’.
This alteration might be analyzed as the result of a regular lexical valence alternation process.
9. Lappin (1988) presents (ia–b) as arguing against a discontinuous analysis of exceptives like
that proposed here:
(i) a. Bill saw no student but John, although Mary saw several.
b. No students other than the radicals participated in the demonstrations, although
many supported it.
Lappin analyzes several in (ia) as an elliptical NP containing an N’ anaphorically dependent
upon the N’ in no student but John. He states: “The second conjunct of [(ia)] can only [my
emphasis] be understood as asserting that Mary saw several students other than John” (p.
987). Lappin concludes that student but John must be a constituent in order to furnish an
appropriate antecedent. Similar reasoning is applied to (ib). I have two points in response.
First, I do not share Lappin’s judgments, and I have found no one else who does. For
myself and every speaker I have consulted, the only reading of (ia) is one in which Mary
saw several students, which may or may not have included John. In other words, for these
speakers, and for me, (ii) is perfectly coherent.
(ii) Bill saw no student but John, although Mary saw several, including John.
Exactly the same results obtain with (ib). Many refers to “many students,” not “many stu-
dents other than the radicals.” The analysis proposed here predicts these judgments; since
student but John and students other than the radicals are not constituents, they should not
offer natural antecedents.
Second, Lappin’s analysis of the phenomenon in (ia,b) as ellipsis is highly questionable,
as opposed to the alternative (by Hoeksema (1984)) that several, many, none, and so on.
are pronominal determiners, and the relevant relation one of simple anaphora. Consider
(iiia–c):
(iii) a. There were men present and women present. Many were under the age of twenty.
b. I bought three presents for Alice and two gifts for Sue. Several were under $20 in price.
c. Several Korean nouns and Japanese adjectives were analyzed. None were dismissed.
Clearly, many in (iiia) can be read as referring to “men or women,” despite the fact that it
has no explicit N’ antecedent that picks out this set. Similar remarks apply to (iiib,c). This
argues for anaphora, not ellipsis.
10. Bach and Cooper (1978) offer a Montague Grammar–style compositional semantics for the
NP-S relative clause syntax that assigns determiner interpretations containing a variable R
for the meaning of the relative.
(i) a. λQλP∀x[[Q(x) & R(x)] → P(x)]
b. λQλP∃x[[Q(x) & R(x)] & P(x)]
This represents, in effect, an Article-S analysis since the underlying composition is between
the determiner and the relative clause. For more see section 3.2.2.
11. The analysis of examples like (51a) as Right Node Raising constructions is first proposed
(to my knowledge) by McCawley (1981).
12. See McCawley (1982) and McCloskey (1986) for discussion.
13. A more traditional Montague Grammar version of (53) employing Montague’s Intensional
Logic is given below, where (following Dowty, Wall, and Peters (1981)), the categorial defi-
nition of determiners is T/CN, and where RC is the category of wh-relative clauses:
SRC If δ ∈ PT/CN and φ ∈ PRC, then F1000,n(δ,φ) ∈ PT/CN, where F1000,n(δ,φ) = δ^φ
TRC If δ ∈ PT/CN and φ ∈ PRC, and δ and φ translate into δ’ and φ’, respectively, then
F1000,n(δ,φ) translates into λQ[δ’(^λxn[Q{xn} & φ’])]
14. Larson (1982) extends Bach and Cooper’s account of determiners and relative clauses to
the relation between tenses and temporal adverbial clauses in analyzing certain readings of
Warlpiri adjoined relatives.
15. The adjective red in this example is to be understood restrictively. Under a nonrestrictive
reading, the discourse can be continuous.
16. Vendler’s sentence actually continues "attached to the noun," but he seems to mean this only
in the general sense of accompanying the noun, rather than as a definite syntactic proposal
of NP–relative clause constituency. In any case, his observations are most compatible with
the view expressed in the text.
17. Vendler states: “[(57a)] is continuous. The is the sign of the deleted clause (whom) I see. In [(57c)],
the possibility of this clause is precluded by the presence of the actual clause (whom) you know.
The in [(57c)] belongs to this clause and any further restrictive clauses are excluded. Conse-
quently, there is no reason to think that the man you know is the same as the man I see” (p. 53).
18. Note that although every differs from most in allowing its restriction to be understood via a
preceding clause, it still seems to differ from the insofar as an overt relative doesn’t produce
discontinuity. Thus it appears possible to understand Every linguist you know in (ib) as
referring to every linguist whom you know and whom I met:
(i) a. I met some linguists. Every linguist was educated in California.
b. I met some linguists. Every linguist you know was educated in California.
To my knowledge, such “Vendler effects” with quantifiers have not been explored systemati-
cally in the literature.
19. In a very general sense, the definite determiner might be viewed as analogous to a verb like
word, which appears to take a manner adverbial as its complement, whereas other verbs of
similar meaning are merely compatible with such a modifier:
(i) a. John worded the letter *(carefully).
b. John wrote the letter (carefully).
In a similar way, the takes a restrictive modifier as its complement, whereas other determin-
ers are merely compatible with such a modifier.
20. I am grateful to C. de Cuba for this example.
21. The term “referential use” is adapted from Ludlow and Neale (1991), where it refers to the
use of a description in which the speaker has singular grounds for his/her assertion, and
where the proposition that he/she intends to convey is also singular. See Ludlow and Neale
(1991) for details.
22. Reanalysis is conceived as a consequence of the X-bar theory in (13), which creates an asso-
ciation between the notions of transitive predicate and head. The idea is that phrases that are
thematic transitives (i.e., having two unassigned θ-roles) can be reanalyzed as structural transi-
tives (i.e., X0s). The Light Predicate Raising analysis is explored in detail in Larson (1989).
23. The point is even clearer in a pair like (ia,b). In Larson (1989), presentational there examples
like (ia) are analyzed as deriving through Light Predicate Raising:
(i) a. There [V′/V was in the room] a tall, dark stranger.
b. Was there [V′/V t in the room] a tall, dark stranger?
The fronting of the verb in (ib) indicates that V (be) must be able to raise out of a reanalyzed
V′ even in surface form.
24. Apparent prenominal PPs like (ia–c) are plausibly analyzed as some form of compound
formation, whose surface head-initial structure is not visible to the syntax:
(i) a. an under the counter deal
b. this over the counter medication
c. three off the wall ideas
25. There is an interesting, well-known question as to whether postnominal PPs and APs are
uniformly bare categories or contain "hidden" relative clause structure, at least in certain
instances (i):
(i) a. the books printed
(cf. the books that have been printed)
b. the books in print
(cf. the books that have been in print)
One potential consideration (noted by Hudson 1973) is the fact that these elements accept
PP modifiers whose presence otherwise requires an aspectual verb (ii)–(iii):
(ii) a. the books printed since 1980
b. the books in print since 1980
(iii) a. *?These books were printed/in print since 1980.
b. These books have been printed/in print since 1980.
The need for have illustrated in (iii) argues for the presence of a concealed have in (ii), hence
a concealed relative.
26. In Larson (1988a) it is proposed that the dative preposition (to) normally accompanying the
goal phrase is “absorbed” by give in the double object construction, equivalently to the Case
absorption that occurs in a passive. Lack of Case marking on the goal triggers NP Move-
ment. After raising of give, the lower V′ reanalyzes as V, yielding a structure equivalent (up
to the layered V-trace) to (i):
(i) [VP Mary gave [VP John [V′ t a briefcase]]]
27. Postnominal genitives with a definite determiner, like that in (97a), have the often-noted
property of requiring a restrictive modifier (relative clause, PP, or postnominal adjective) in
order to occur smoothly (Lyons 1986). On an account where relative clauses are arguments
of definite Ds (recall section 3.2.2), this might be understood as follows: whereas definite
D typically allows the deictic determination of its restrictive argument, the presence of the
postnominal genitive blocks this possibility, forcing structural realization. This proposal
appears sensible given the semantics for thematic genitives adopted below, in which the
definite D in genitives contains its own relational variable R, whose value is deictically
or structurally determined. In essence, structural (nondeictic) determination of R forces
structural (nondeictic) determination of the restrictive argument as well. This leaves open
the question of why prenominal genitives show the opposite requirement: why an overt
restrictive element is blocked in the latter case (*John’s briefcase that Alice lost). I have no
proposal to make at this point.
28. Another claim of the present analysis is that prenominal genitive DPs occupy an (indirect)
object-like position in DP. Consider the fact that for many speakers (including myself) the
verb award allows to-datives, double objects, and with-PPs. In the latter, with seems to be
associated with the possession relation holding between Mary and the prize:
(i) a. John awarded the grand prize to Mary.
b. John awarded Mary the grand prize.
c. John awarded Mary with the grand prize.
Rothstein (1988) observes that with-PPs also appear to play a specifically possessive role
within DP. She notes that although (iia,b) look superficially similar, the locative PP can be
paraphrased with a copular relative clause, but the with-PP requires a possessive relative
(iii):
(ii) a. The plate on the table
b. The plate with the gold rim
(iii) a. The plate that is on the table
b. The plate that has the gold rim
(cf. *The plate that is with the gold rim)
Given this, it is interesting to observe certain possession/position alternations in DP remi-
niscent of those found with award. For example, consider (iv), which exhibits an of-variant
(iva) parallelling the to-dative, a prenominal variant (ivb) parallelling the double object
structure, and a with-variant (ivc):
(iv) a. the gold rim
{
of the plate
that the plate has
b. the plate’s gold rim
c. the plate with the gold rim
29. Compare also analyses like Siegel (1974) and Drescher and Hornstein (1979) in which the
postnominal genitives are derived from prenominal forms by rightward movement.
30. This section was written after the remainder of the paper (including section 5). It was not
until Burton (1995) (discussed below) that a semantics became available to execute the con-
clusions drawn at the end of 4.2.2, that is, that so-called thematic genitives do not receive a
θ-role directly from N.
31. Cooper (1979) develops a closely related proposal in which definite descriptions contain a
free property variable Π, whose value is fixed by context and that can be elaborated as a
variable over relations. Burton’s analysis can be considered an extension of Cooper’s general
proposal to the specific case of possessive definites.
32. Note that reading (109b) is compatible with the woman in question being John’s wife. Imag-
ine a bridge party for married couples in which husbands and wives are paired as partners
by drawing lots. By chance John is paired off with the woman to whom he is in fact married.
In these circumstances, the sentence John’s wife is his wife is not redundantly true.
33. On reading ((110b)/(110c)), the sentence John’s wife is his wife is redundantly true.
34. In prose, (111c) may be read: “the event e such that e is a destruction of Rome by some x,
and e is a destroying of Rome by Nero.” See also Higginbotham (1983) for an alternative
version of this proposal.
35. See also Postal (1974) and Drescher (1977).
36 This notion of degree is familiar in contexts like carnival rides, with signs saying “you must
be this tall to go on this ride.” “This tall” is understood as expressing a degree or extent;
anyone measuring that height or higher is understood to have that extent of height.
37. In brief, and updating Larson (1988b) somewhat, adjectives occur in DegPs that take DPs as
their subjects (see below for discussion of DegP). The derivation for a basic case like Kenton
is tall goes as in (ia–d), which analyzes the sentence as true just in case Kenton has some
(contextually relevant) degree of tallness.
(i) a. [AP [A tall]] ⇒ λ 𝒫∃d[𝒫(tall’(d))]
b. λ 𝒫 ∃d[𝒫 (tall’(d))](λP[P(Kenton’)])
c. ∃d[λP[P(Kenton’)](tall’(d))]
d. ∃d[tall’(d)(Kenton’)]
DegP traces are then assigned the Intensional Logic-translation in (iia) where Pi is a vari-
able of type <e,t> a set of individuals. Clausal comparative complements are analyzed as
abstracting over this variable, yielding a set of sets (iib,c) and making them appropriate
arguments for degree morphemes:
(ii) a. λ𝒫[𝒫(Pi)]
b. [CP Kenton is [DegP t ]] ⇒ λ𝒫[𝒫(Pi)](λP[P(Kenton’)]) ⇒ Pi(Kenton’)
c. [PP than [CP Kenton is [DegP t ]]] ⇒ λPi[Pi(Kenton’)]
The same assumptions can be used to interpret the DegP trace left by the raising of compara-
tives and equatives. See Larson (1988b) for details.
38. Nikiforidou (1987) considers several different constituencies for too and enough construc-
tions, including ones similar to (137a,b), in which too and enough combine initially with
an infinitive that subsequently extraposes rightward. Nikiforidou questions this analysis for
enough on grounds that it would require an added positioning rule for the morpheme. The
point is correct so far as it goes, but the operation seems to be exactly the same one required
for -er. So it is not clear that any additional cost is incurred. Nikiforidou’s own proposal
(following a suggestion by Fillmore) is that too and enough constitute “valence-changing”
morphemes, which combine with A, altering its selectional requirements to include an infini-
tive argument. It is difficult to evaluate this proposal in the absence of an accompanying
semantics. While the structures in (137) are also given without an explicit semantics, the
general character of the latter is nonetheless clear, as discussed below.
REFERENCES
Abney, S. (1987) The English Noun Phrase in Its Sentential Aspect. Doctoral dissertation, MIT.
Anderson, M. (1983/1984) “Prenominal Genitive NPs,” The Linguistic Review 3: 1–24.
Bach, E. and R. Cooper (1978) “The NP-S Analysis of Relative Clauses and Compositional Seman-
tics,” Linguistics and Philosophy 2: l45–l50.
Baker, C. L. (1978) Introduction to Generative-Transformational Syntax. Englewood Cliffs, NJ:
Prentice-Hall.
Baker, M. (1988) Incorporation. Chicago: University of Chicago Press.
Baltin, M. (1987) “Degree Complements,” in G. Huck and A. Ojeda, eds., Syntax and Semantics.
Vol 20: Discontinuous Constituency (pp. 11–26). New York: Academic Press.
17: 347–354.
Bowers, J. (1975) “Adjectives and Adverbs in English,” Foundations of Language 13: 529–562.
Burton, S. (1995) Six Issues to Consider in “Choosing a Husband.” Doctoral dissertation, Rutgers
University , New Brunswick, NJ.
Carrier-Duncan, J. (1985) “Linking of Thematic Roles in Derivational Word-Formation,”Linguistic
Inquiry 16: 1–34.
Chomsky, N. (1970) “Remarks on Nominalization,” in R. Jacobs and P. Rosenbaum , eds., Read-
ings in English Transformational Grammar (pp. 184–221). Boston: Ginn.
———. (1986) Knowledge of Language. New York: Praeger.
Cooper, R. (1979) “The Interpretation of Pronouns,” in F. Heny and H. Schnelle, eds., Syntax and
Semantics. Vol. 10: Selections from the Third Gröningen Roundtable (pp. 61–92). New York:
Academic Press.
Corver, N. (1990) The Syntax of Left Branch Extractions. Doctoral dissertation, Tilburg Univer-
sity , The Netherlands.
Davidson, D. (1967) “The Logical Form of Action Sentences,” in N. Rescher, ed., The Logic of
Decision and Action (pp. 81–120). Pittsburgh, PA: University of Pittsburgh Press.
Davies, M. (1981) Meaning, Quantification, Necessity. London: Routledge and Kegan Paul.
Dixon, R.M.W. (1966) “Relative Clauses and Possessive Phrases in Two Australian Languages,”
Dixon, R.M.W. (1977) “Where Have All the Adjectives Gone?” Studies in Language 1: 19–80.
Dowty, D., R. Wall, and S. Peters (1981) An Introduction to Montague Semantics. Dordrecht: Kluwer.
Drescher, E. (1977) “Logical Representation and Linguistic Theory,” Linguistic Inquiry 8: 351–378.
Drescher, E. and N. Hornstein (1979) “Trace Theory and NP Movement Rules,” Linguistic Inquiry
10: 65–82.
Frege, G. (1953) The Foundations of Arithmetic. Trans. J. L. Austin. New York: Philosophical
Library. Originally published 1884.
Fukui, N. and M. Speas (1986) “Specifiers and Projection,” in N. Fukui, T. Rapoport, and E. Sagey,
eds., Papers in Theoretical Linguistics (pp. 85–127). Cambridge, MA: MIT Working Papers
in Linguistics.
Giorgi, A. and P. Longobardi (1990) The Syntax of Noun Phrases. Cambridge: Cambridge Uni-
versity Press.
Grimshaw, J. (1990) Argument Structure. Cambridge, MA: MIT Press.
Guéron, J. and R. May (1984) “Extraposition and Logical Form,” Linguistic Inquiry 15: 1–31.
Hetzron, R. (1978) “On the Relative Order of Adjectives,” in H. Seiler, ed., Language Universals
(pp. 165–184). Tübingen: Narr.
Higginbotham, J. (1983) “Logical Form, Binding and Nominals,” Linguistic Inquiry 14: 469–708.
Higginbotham, J. and R. May (1981) “Questions, Quantifiers and Crossing,” The Linguistic
Review 1: 41–80.
Hoeksema, J. (1984) “Partitives,” unpublished manuscript, University of Pennsylvania, Philadelphia.
Hudson, R. (1973) “Tense and Time Reference in Reduced Relative Clauses,” Linguistic Inquiry
4: 255–256.
Jackendoff, R. (1977) X-bar Syntax. Cambridge, MA: MIT Press.
Jacobs, R. and P. Rosenbaum (1968) English Transformational Grammar. Waltham, MA: Ginn.
Kayne, R. (1984) “Unambiguous Paths,” in Connectedness and Binary Branching, Studies in Gen-
erative Grammar 16, (pp. 129–164). Dordrecht: Foris.
Klein, E. (1980) “A Semantics for Positive and Comparative Adjectives,” Linguistics and Philoso-
phy 4: 1–45.
Kuroda, S. Y. (1969) “English Relativization and Certain Other Related Problems,” in D. Reibel
and S. Schane, eds., Modern Studies in English: Readings in Transformational Grammar
(pp. 264–287). Englewood Cliffs, NJ: Prentice-Hall.
Lappin, S. (1988) “The Semantics of ‘Many’ as a Weak Determiner,” Linguistics 26: 977–998.
Larson, R. (1982) “A Note on the Interpretation of Adjoined Relative Clauses,” Linguistics and
Philosophy 5: 473–482.
Larson, R. (1987) “‘Missing prepositions’ and the analysis of English free relative clauses,” Lin-
guistic Inquiry 16: 239–266.
———. (1988a) “On the Double Object Construction,” Linguistic Inquiry 19: 335–391.
———. (1988b) “Scope and Comparatives,” Linguistics and Philosophy 11: 1–26.
———. (1989) “Light Predicate Raising.” MIT Lexicon Project Papers 27. Cambridge , MA: MIT.
———. (1990a) “Double Objects Revisited: Reply to Jackendoff,” Linguistic Inquiry 21: 589–632.
———. (1990b) “Semantics,” in: D. Osherson, ed., Invitation to Cognitive Science. Vol. 1: Lan-
guage, Cambridge. MA: MIT Press.
———. (1991) “Promise and the Theory of Control”, Linguistic Inquiry 22: 103–139.
Ludlow, P. and S. Neale (1991) “Indefinite Descriptions: In Defense of Russell,” Linguistics and
Philosophy 14: 171–202.
Lyons, C. (1986) “The Syntax of English Genitive Constructions,” Journal of Linguistics 22:
123–143.
May, R. (1985) Logical form; its structure and derivation. Cambridge,MA: MIT Press.
McCawley, J. (1981) “The Syntax and Semantics of English Relative Clauses,” Lingua 53: 99–149.
———. (1982) “Parentheticals and Discontinuous Constituent Structure,” Linguistic Inquiry 13:
91–106.
———. (1988) The Syntactic Phenomena of English. Chicago: University of Chicago Press.
McCloskey, J. (1986) “Right-Node Raising and Preposition-Stranding,” Linguistic Inquiry 17:
183–186.
Montague, R. (1974) “The Proper Treatment of Quantification in Ordinary English,” in Montague
(1974) Formal Philosophy (pp. 247–270). New Haven, CT: Yale University Press.
Neale, S. (1990) Descriptions. Cambridge, MA: MIT Press.
Nikiforidou, K. (1987) “Constituency Considerations in Too and Enough Constructions,”
paper presented at the 62nd Annual Meeting of the Linguistic Society of America, San
Francisco, CA.
Partee, B. (1976) “Some Transformational Extensions of Montague Grammar,” in B. Partee, ed.,
Montague Grammar (pp. 51–76). New York: Academic Press.
———. (1983/1997) “Uniformity vs. Versatility: The Genitive, a Case Study,” Appendix to Theo
Janssen (1997) “Compositionality,” in J. v. Benthem and A. ter Meulen, eds., The Handbook
of Logic and Language (pp. 464–470). Cambridge , MA: MIT Press.
Perlmutter, D. (1981) Studies in Relational Grammar 1. Chicago: University of Chicago Press.
Perlmutter, D. and C. Rosen (1984) Studies in Relational Grammar 2. Chicago: University of
Chicago Press.
Postal, P. (1969) “On So-Called ‘Pronouns’ in English,” in D. Reibel and S. Schane, eds., Modern
Studies in English (pp. 201–224). Engelwood Cliffs, NJ: Prentice-Hall.
———. (1974) “On Certain Ambiguities,” Linguistic Inquiry 5: 367–424.
Rescher, N. (1962) “Plurality Quantification,” Journal of Symbolic Logic 27: 373–374.
Ross, J. R. (1967) Constraints on Variables in Syntax. Doctoral dissertation, MIT , Cambridge, MA.
———. (1973) “A Proposed Rule of Tree-Pruning,” in D. Reibel and S. Schane, eds., Modern Stud-
ies in English (pp. 288–299). Engelwood Cliffs, NJ: Prentice-Hall.
———. (1981) Infinite Syntax. New York: Ablex.
Rothstein, S. (1988) “Conservativity and the Syntax of Determiners,” Linguistics 26: 999–1019.
Russell, B. (1905) “On Denoting,” Mind 14: 479–493.
Sag, I. (1976) “Deletion and logical form,” Doctoral thesis. Cambridge, MA: MIT.
Seuren, P. (1973) “The Comparative,” in F. Kiefer and N. Ruwet, eds., Generative Grammar in
Europe. (pp. 528–564). Dordrecht: Reidel.
Siegel, D. (1974) Topics in English Morphology. Doctoral dissertation, MIT, Cambridge, MA.
Smith, C. (1964) “Determiners and Relative Clauses in Generative Grammar,” Language 40:
37–52.
Sproat, R. and C. Shih (1991) “The Cross-Linguistic Distribution of Adjective Ordering Restric-
tions,” in C. P. Georgopoulos and R. L. Ishihara, eds., Interdisciplinary Approaches to Lan-
guage: Essays in Honor of S.-Y. Kuroda (pp. 565–593). Dordrecht: Kluwer Academic.
Stockwell, R., P. Schacter, and B. Partee (1970) The Major Syntactic Structures of English. New
York: Holt, Rinehart and Winston.
Stowell, T. (1981) Origins of phrase structure. Doctoral dissertation, MIT.
Szabolsci, A. (1983) “The Possessor That Ran Away from Home,” The Linguistic Review 3:
89–102.
Vendler, Z. (1967) Linguistics in Philosophy. Ithaca, NY: Cornell University Press.
Wiggins, D. (1980) “‘Most’ and ‘All’: Some Comments on a Familiar Program and the Logical
Form of Quantified Sentences,” in M. Platts, ed., Reference, Truth and Reality (pp. 318–346).
London: Routledge and Kegan Paul.
Williams, E. (1977) “Discourse and logical form,” Linguistic Inquiry 8: 101–139.
———. (1985) “PRO and the Subject of NP,” Natural Language and Linguistic Theory 13: 277–295.
———. (1987) “Implicit Arguments, the Binding Theory, and Control,” Natural Language and
Linguistic Theory 15: 151–180.
Index
Illustrations and tables are indicated with italic formatting.

A-bar extraction 61–2 Alsina, A. 61
Abbott, B. 218, 363 alternations:award-present 300; complement
Abney, Stephen 407–8, 418, 430, 457, 201–9; dative 65–7, 86–96, 127–8;
469, 473 dyadic-triadic 474; oblique-applicative
accusatives: bare 156; double form 57, 66, 66, 96, 126; possessive 126;
57; multiple Cases 114–16; non- pronoun 343, 343–4, 344; spray-
prepositional Cases 115, 115–20, load 90, 114, 131, 201, 203, 210,
118, 119, 120; objects 172; see also 213–14, 292, 300; trace 343, 343–4,
unaccusatives 344; transitivity 320–5, 322, 323;
across-the-board: extraction analysis 226; VP-internal 285–8, 286; see also voice
movement 146, 146, 351, 444, 444 alternation
actions 50, 52 alternative projection 205–9
active-passive relationships 67, 81 ambiguity 46, 261, 360; deontic versus
active transitives 14, 103 epistemic 257
adjectives: closed-scale 282; open-scale 282; The American Heritage Dictionary of the
prenominal 454–7, 456 English Language (AHDEL) 52–3, 128
adjuncts 217; of arguments 464–5; bounding A-moveability 183
effects with 306–7; circumstantial 416; A-movement 37, 92, 127, 149, 154–5, 158;
control 3, 3, 8; control-oriented 343; Extended Standard Theory and 87;
infinitival purpose 356; orientation OBL  2 as 88–90, 89, 90; Pylkkänen
of 314, 314–15, 315; parasitic and 76
constructions 361; projection of 89, 91, A′-movement 284–5, 339; asymmetry and
93, 103, 103; sentence-final adverbs 304, 304–9; bounding effects with
and 30; VP 272 adjuncts 306–7; Indonesian language
advancement 153; relational 183; 3 → 1 and 307–9; parasitic gaps and 334–6;
161–4, 185 P-stranding and 304–6
adverbs and adverbials 182, 391–2; frequency Anagnostopoulou, E. 109–11, 123, 133
399; scope contrasts 397–404; as analysis: across-the-board extraction 226;
V-complements 440, 440–1; verb- of applicatives 71–6, 71–6; Article-S
adverbial idioms 441; VP-initial 402–3; analysis of relatives 11, 427, 441–3,
see also sentence-final adverbs 442, 446; AspP 105–6; bar-level 291; of
affectedness 71–2, 171, 187, 209–10 complementations 372; concordial 119,
affective element 139 125; derivational 67–71; of DPs 407;
Africanist descriptive tradition 37 D-raising 443–5; event 217; extraction
agency 15–16, 98 226; functional composition 299; K
Agent Postposing 89, 89 18–20, 19, 20, 22, 29, 31, 33, 422;
agreement 21, 93, 98; Case theory and nonderivational 71–80; null pleonastic
234; of θ-features 21–2; transitivity 334; one-step 161; P 18, 18, 21, 422,
of 99, 101–2; upward 29; see also 424; prepositional incorporation
θ-agreement 70; right-adjunction 299; rightward
AHDEL see The American Heritage downward branching 192; semantic
Dictionary of the English Language 16–18, 17; of small clauses 79, 222;
Alexiadou, A. 388 standard 421; stranding 131; subatomic
allow 256–8, 257 18; sublexical 18; two-step 161
482 Index
anaphora 35, 138, 163–4, 239, 244; control structural 9, 29; goal-theme 92; Heavy
and 260; cross-sentential 195; donkey NP Shift and 304–9; single complement
351, 407; dummy 437; inner 465; hypothesis and 175; subject-object 9;
intersentential 194; orientation of 182 of syntactic domain 138–40, 139, 140;
ancient Greek (language) 161–2 theory 404
Anderson, S. 201, 463 augmentation 211–13
Andrews, A. 274–5, 307, 390–1, 400–3 augmented determiners 438
antecedent condition 306 Australian languages 462–3
antecedent-contained deletion 84, 224, 466 Austronesian languages 83
anti-correspondences 388–90 availability verbs 320–2, 327
Antisymmetry Theory 1 award-present alternations 300
Aoun, Joseph 140, 180, 222 Awoyale, Y. 379
A-over-A principle 342
apparent object-control double object verbs Babby, L. 172–4
251–8; allow 256–8, 257; ask 253–4; Bach, Emmon 1, 4, 260–1, 357, 404, 446,
teach 255–6 474–5; Chomsky and 31; double object
applicative constructions 60, 60–6, 66, 95, constructions and 233; generalization
168, 183; asymmetric 130; languages of 241–2, 242; Heavy NP Shift and
with 150; morphological form of 65–6; 309; promise and 230, 233, 242–5;
semantic range of 62–5; symmetric 130 Right Wrap and 181; V-raising and 143
applicative-oblique alternation 66, 66, 96, 126 background-entailed content 394
applicatives: analysis of 71–6, 71–6; Bahasa Indonesian (language) 169, 307–9, 315
benefactive 62; heads 20, 71; high Baker, C. 444
72, 74; languages and 83; low 72, Baker, Mark 70–1, 105, 180, 258, 377,
74; morphemes and 110; nondative 381, 433; accusative Case and 109;
107, 107–8, 108; oblique-applicative applicative constructions and 61, 65–6;
alternation 66, 66, 96, 126; obliques dative projection and 37; datives and
and 66–86; relating 66–86; stimulus 92; head-to-head movement and 144;
64; structures 102; thematic hierarchy P-incorporation and 312; Travis and
and 132 132; UTAH and 150, 198; on verb
applicative shift (A-shift) 101, 102, 105–6; serialization 373–4
nondative applicatives and 107, 107–8, ballistic motion verbs 40–1
108; voice alternation and 100–2 Baltin, M. 471
applied affix 169 Bantuists 186
applied object 61 Barbosa, Pilar 126
arguments: adjuncts of 464–5; asymmetry of bare accusatives 156
36; augmentation of 211–13; demotion bare phrase structure theory 298
of 91, 93, 151, 183; determiner-oblique bar-level reanalysis 291
415, 415–16, 416; external 324; Barss, Andrew 35–6, 137, 138–9, 194–5,
extra 20; internal 324; introducing 244–5, 317; goal-theme asymmetries
20; inversion of 104; multiple oblique and 92; Larson and 191
26–9, 27, 28, 29, 416–18, 417, Barwise, J. 407
418; predicate-oblique 25–6, 26; base object 61
projection of 175–7, 431; realization because-clauses 394–7; event relations and
of 175–7, 176, 245; relative clauses quantification 395–7; propositional
as 446–7; subordination of 176; X operators and 395
433–6, 434, 435; see also A-movement; Beck, S. 40
A′-movement Belletti, A. 163, 179, 184–5, 221, 321, 330–1
Aristotelian logic 409, 428 benefactives 125; augmentation of 211–12;
Aronoff, Mark 223 Halkomelem applicatives 62
Arregi, E. 37 beneficiaries 209–10
Article-S analysis of relatives 11, 427, 441–3, Benveniste, E. 132
442, 446 Berber (language) 144, 359
A-shift see applicative shift Birner, B. 81
ask 253–4, 261, 262 Bleam, T. 37, 110, 121, 128–9, 132
AspP analysis 105–6 bounding effects, with adjuncts 306–7
asserted content 394 Bowers, J. 1, 33, 35, 37, 222, 226, 230, 345
asymmetry 292; A′-movement and 304, Bresnan, J. 61, 304, 306, 356
304–9; of applicative constructions British English 127
130; of arguments 36; in datives 35–6; Broselow, Ellen 359
domain 191–6, 208, 317; generalized Bruening, Benjamin 85, 131
Index 483
Brugè, L. 120 children, dative constructions by 186
Brugger, G. 120 Cho, S. 117, 118
Burton, S. 464–5, 477 Choe, H. S. 180, 359
Burzio, L. 162, 320–1, 330, 359; chomeur status 89
generalization of 159, 169–70, 186, Chomsky, Noam 89, 137–8, 180, 258, 360,
322, 331 430, 462–3; anaphor binding and
239; A-over-A principle and 342;
candidate derivations 11 Bach and 31; bare phrase structure
Carlson, Greg 180 theory and 298; dative complements
Carrier-Duncan, J. 176, 433 and 181; dative structures and 140–3;
Carstens, V. 382 D-structure and 233, 309; ECM
Cascade Theory 1, 33 and 279; EPP and 119; equidistance
Case assignment 67, 164, 259, 259, 321, 357, and 93; expletives and 320–1; Full
374; Burzio’s generalization and 169– Interpretation Principle and 321;
70; cross-linguistic variation and 108; functional head movements and
to double objects 157, 157–9, 158; 226; generative grammar and 1;
multiple accusative 114–16; V-raising governing category and 146–7;
and 12 Heavy NP Shift and 267; little v
Case marking 37, 89–90, 149–50, 209; dative and 24–5, 98; L-marking and 91,
movement and 165–6; ECM 257–8, 215; LPR and 271, 296; oblique
262, 277–9, 283; Japanese and 126; constructions and 105; passivization
NP movement and 459; in passives 172 and 84; Projection Principle and 153,
Case theory 94, 100, 108–25, 133, 183, 260; 356; question formation and 86;
absorption and 105; agreement and radical shift in theory and 197–8;
234; datives and 112–13; direct Case structural accusative Case and 114–15;
184; with double objects 108–11; V-complement constructions and 266;
features of 113–25; filter of 161; V′ Reanalysis and 288–91; X-bar
inherent Cases 112, 122–5, 123, structure and 174; X-bar theory and
124; non-case transmitters 331; non- 12–13, 92, 290
prepositional accusative 115, 115–20, Christensen, K. 155, 305
118, 119, 120; objective Case 158; Chulsu (language) 79
resistance 239; serialization and 384; Chung, S. 186, 307, 309
structural 112, 114–16, 158 circumstantial adjuncts 416
Categorial Grammar 32, 243 classic transformational relationships 67
category neutrality 30, 33, 436 clauses: because 394–7; chaining 370–1;
causal modifiers 396 comparative 31; if/when 396; reduced
causation 15–16 277, 277–83; relative 31, 361, 446–8;
causative serialization 371 small 79, 277; when/before/after 395–6
caused-motion: constructions 60; idioms 41, clitic-doubled forms 110–11, 121
48, 76–7; verbs 44, 45, 57 cliticization 161, 185
caused possession 39, 43, 46, 60, 60, 64, 77; closed-scale adjectives 282
encoding 40, 42–3, 47; give and 48, cognate objects 179, 437; formation of 324,
48–55, 49; idioms 41; interrogatives 327
and 44; meaning and 51, 54, 57; Collins, C. 86, 195
polysemy and 45 communications 50, 52
C-command 7, 75, 113–14, 239–40, 274, comparatives 411, 471; clauses 31; as
404, 441; anaphor binding and 163; quantificational 466–8
asymmetric 36, 111, 153, 200; dative- complement alternations: alternative
marked possessor 129; of direct object projection and 205–9; LPR and 203–5;
192, 216; little v and 29; negative NP movement and 201–3
polarity and 139, 215, 221; probe-goal complementation 236; analysis of 372; of
relations and 27; QR and 225; syntactic datives 137, 146
domain and 138–40, 194 complements 286; control 3, 3, 8;
Cebuano (language) 185 D-complements 441–8; embedded
CED see Condition on Extraction Domains verbal 369; interrogative 254; multiple
CFC see complete functional complex 358; oblique 30; position 215;
Chaga (language) 72, 74 realization of 252; single complement
Chamorro (language) 359 hypothesis 6, 174–5, 175, 431;
Cheng, Lisa 221 structure 217; V-complements 266,
Chichewa (language) 61, 63–5, 79, 95, 169 397, 440, 440–1
Chierchia, G. 261 complete functional complex (CFC) 146–7
484 Index
complex coordinations 347–50, 348, 349, 350 languages without double object forms
complex predicates 3, 8, 265–98; Heavy NP 57–60
Shift as LPR 265–8; interface logic of dative movement: Case marking and
LPR 269–71; structural implications of 165–6; recoverability and 166–9;
LPR 271–83 unaccusativity and 169–71
complex telic predicate formation 281–3 dative projection 39, 40, 42, 43; applicative
compositionality 75 constructions 60–6; double object
computational engines 70–1 construction and 39–42; idioms and
concealed imperatives 256, 262 47–60; oblique locatives and 39–47;
concealed questions 254–5 oblique possessives and 42–7; relevant
concordial analysis 119, 125 information for 37–9
concordial inflection 119, 119 datives 35–125, 36; applicative constructions
Condition on Extraction Domains (CED) 215 60–6; Case theory and 112–13;
Condoravdi, C. 393–4, 405 children and 186; complementation of
conflation 329, 381 137, 146; double object construction
conjunction 146; reduction of 225 and 39–42; idioms 47–60; morphology
constraints 24–5, 116; coordinate structures 123, 125; oblique 103, 140; oblique
348–9, 362, 385; on dative movement locatives and 39–47; oblique
165–71; on dative shift 171, 209–15; possessives and 42–7; prepositional
minimality 27–9, 99; as presentationals 35, 459–60; relating applicatives and
328–30; surface realizational 56 obliques 66–86; summary of 125;
construal 256, 261; object 250 unaccusatives 164
consumption: of media 299; verbs of 280 Dative Shift 35–7, 42, 90, 165, 173, 233,
contrasts 55 244, 246; affectedness and 209–10;
Contreras, H. 216, 225, 339 beneficiaries and 209–10; Case
control 239–58; adjuncts 3, 3, 8; alternative assignment to double objects 157–9;
account of 242–5; anaphora and 260; constraints on 171, 209–15; dative
apparent object-control double object complementation and 137; for-datives
verbs and 251–8; complements 3, 3, 8; and 210–15; Extended Standard
generalizations of 239–42; infinitival Theory and 149; morphology
structures 345; object control and 156–7; motional to-datives and
verbs 241; objections and 247–58; 210–15; OBL  2 86–8; as passive
obligatory structures 261; projection 150, 150–3, 151, 152; properties of
of structures 245–7, 246, 247; 159; semantic 68, 68–70, 69; subject
relations 4; shifting 247–51, 261–2; suppression and 156–7; syntactic 67–8;
subject control verbs 241 in VP 459–60
Cooper, R. 407, 446, 474–5, 477 dative structure 1, 140–9; consequences of
coordinate structures: constraints of 348–9, 145–7; heavy NP shift and 147–9;
362, 385; violations of 348 V-raising and 143–5
coordination: complex 347–50, 348, 349, Davidson, D. 217, 391–2, 395–6, 399, 465
350; nonconstituent 217–19, 219, D-complements, relative clauses for 441–8
225–6 definiteness restrictions 330–2
Corpus Jurus Secundum 224 DegPs see degree phrases
Corver, N. 469 Deg’ Reanalysis 472, 472, 473
Couquaux, D. 11 degree elements 31; LPR in 472–3
Cowper, E. 362 degree morphemes, as relational 468–9
cross-linguistic questions 172–4 degree phrases (DegPs) 1, 411, 465–73;
cross-linguistic variation 108 comparatives as quantification 466–8;
cross-sentential anaphora 195 degree morphemes as relational 468–9;
Cuervo, M. 121, 128–9 LPR in 472–3; modifiers in 440–57;
Culicover, P. W. 35, 294–5, 318, 358 projection of 427–73; syntax of 469–73
curried functions 187 deletion: antecedent-contained 84, 224, 466;
Czepluch, H. 109, 181, 185 left peripheral 225; recoverability of
166–9, 223
Danish (language) 173–4, 182 delimiting secondary predicates 380, 382
dative alternation 65–7, 127–8; A-movement Demonte, V. 37, 110, 121, 132
and 88–90; dative shift and 86–8; other demotion 89; of arguments 91, 93, 151, 183;
derivational accounts 92–6, 92–6; 2-to-3 259
problems and questions with 90–2 den Dikken, M. 37, 94–6, 106, 109
dative idioms 47–60; double object forms deontic ambiguity 257
without oblique variants 54–7; Department of Linguistics and Philosophy 1
Index 485
De Pedro Munilla, M. 128–9 dative structure and 140–9; derived
derivational analysis: computational engines status of 81–6; domain asymmetry and
and 70–1; semantic dative shift 68–70; 191–6; extra elements of meaning in
syntactic dative shift 67–8 38; gapping and 217–21, 221; indirect
De Swart, H. 396 passives and 159–64, 160; languages
determiner-oblique argument 415, 415–16, without 57–60; linear order and 191–6;
416 modifiers and 215–17; morphology and
determiner phrases (DPs) 1; analysis of 407; 156–7; nonconstituent coordination
DP-IP analogy 408, 408; hypothesis and 217–19, 219; nonderivational
427; LPR in 451, 451–2, 452; analyses of 71–80; oblique locatives and
relational view and 430, 430–1; see 39–42; without oblique variants 54–7;
also DP projection projection of raising structures 174–80,
determiner-restriction term 412–13 174–80; structure of 149–59; subject
determiners 11, 11, 31, 408–11; augmented suppression and 156–7; syntax of 233–9;
438; ditransitive 410; relational view v-raising and 196–201
of 428–9; semantics of 428–30; in set double object idioms 49–54, 77
relations 429–30; thematic roles for double object productivity 168
410–11; unaccusative 474; unergative double object verbs: apparent object-control
437; universal 470; valence and 409–10 251–8; promise as 230–3
determiner-scope term 413, 413–15, 414, 415 Dowty, David 1, 67, 181, 233, 261, 309, 314,
detransitivization 241 357; argument asymmetry and 36;
Dickens, Charles 131 dative movement and 165; dative shift
Diesing, M. 270, 391, 394, 404 and 68–70, 74; Montague Grammar
differential object markers 120–1 and 3; Right Wrap and 5, 9; V-raising
direct Case 184 and 143
direction 196, 221; of relation 32 DP see distance principle
directionality 259 DP-IP analogy 408, 408
direct objects (DObjs) 3, 3 DP projection 411, 411–21, 412, 427–73,
direct θ-marking 198, 222 443–5, 455; DegPs and 465–73;
Di Sciullo, A. 319 determiner-oblique argument and 415,
discontinuous idioms 3, 3, 8, 142 415–16, 416; determiner-restriction
discourse, linearity in 194–6, 195 term and 412–13; determiner-scope
Discourse Representation Theory 195 term and 413, 413–15, 414, 415;
dispersion, verbs of 224 determiner semantics and 428–30;
distance principle (DP) 376; see also Minimal genitives and 457–65; modifiers in
Distance Principle 440–57; multiple oblique arguments
ditransitives 45, 180, 324; determiners 410; and 416–18, 417, 418; nominal syntax
locatives 129; oblique constructions and 432–40; recursion and 416–18;
100; V′ Reanalysis and 324 voice alternation in 418–21
ditransitive verb phrases (DVPs) 3 DPs see determiner phrases
ditransitive verbs (DTVs) 69 D-raising analysis 443–5
Dixon, R. M. W. 462–3 Drescher, E. 477
DObjs see direct objects Dryer, M. 222
domain asymmetry 208, 317; linearity D-structure 152, 233, 309, 313, 321;
in discourse 194–6, 195; linearity projection of raising structures and
versus hierarchy 191–3, 192, 193; 174; projection principle and 88;
restrictiveness and 193–4, 194 relative subordination of arguments
dominance-based condition 221 and 176; θ-roles and 197; UTAH and
donkey anaphora 351, 407 150, 199, 199–200; V-raising analysis
Donohue, M. 83 and 196; weak crossover effect and
double accusative form 57, 57 138; X-bar theory and 179
double object constructions or forms 35–7, DTVs see ditransitive verbs
68, 72, 121, 137–80, 181, 244–5; dual-headed serialization 373–4
asymmetries of syntactic domain 138– dualism 124
40; Case assignment to 157, 157–9, 158; dummy anaphora 437
Case theory and 108–11; complement dummy element 320
alternations and 201–9; consequences Dutch (language) 173–4, 226
of 153–5; constraints on dative shift and DVPs see ditransitive verb phrases
209–15; cross-linguistic questions 172–4; dyadic Ds 436–7
dative projection and 39–42; datives and dyadic-triadic alternation 474
39–42; dative shift as passive 150–3; Dyirbal (language) 462–3
486 Index
ECM see Exceptional Case Marking Finer, Dan 223, 258, 358
ECP see Empty Category Principle flat structures 92
EI see event identification focus 392–3, 405
embedded verbal complements 369 Fodor, Janet 473
Emonds, Joseph 4, 5, 109, 142 Fon (language) 371
empathic reflexives 437 force 232–3, 236–8, 236–9, 242
empty categories 320 for-datives 210–15
Empty Category Principle (ECP) 173, formal features 21
221, 313 for-phrase 10, 10, 33, 72
empty head positions 222 Frampton, John 205
empty subject 259 Franks, Steve 473
empty verbs 13–15, 13–16 Freeze, R. 94, 96, 132
Endo, Yoshio 180, 258 freezing 341, 358; effects 84, 110, 131,
Engdahl, Elisabet 284, 305, 334, 335, 361 222, 294, 296; Heavy NP Shift and
The English Noun Phrase in Its Sentential 318–19, 319
Aspect (Abney) 407 Frege, G. 409, 428–9
enough constructions 471–2, 477 Freidin, Robert 172, 173, 174
entailments 248–51 French (language) 173–4, 328, 330, 334, 359
epistemic ambiguity 257 frequency adverbials 399
EPP see Extended Projection Principle from-phrase 10, 10, 33
equidistance 93 Fukui, N. 176, 407, 430
ergative splits 184 Full Interpretation Principle 321
Ernst, T. 274, 404 functional categories 407, 430
Erteschik-Shir, N. 226, 363 functional composition analysis 299
Eskimoan languages 183 functional elements 408
European languages 60 functional heads 20; movements of 226
European Portuguese (language) 45, 57–9, functional selection 408
128–9
event identification (EI) 73, 73–4 gapping 217–21, 221, 225–7; contrasts 216;
events: analysis of 217; natural course pseudogapping 220
of 348, 350; natural theory of generalizations 239–42, 335–6; of Bach
composition of 72, 76; place 241–2, 242; of Burzio 159, 169–70,
identification of 371; prelexical 186, 322, 331; of Visser 240, 240–1,
relations of 374; relations of 395–7; 241, 260
single 374; structured quantification Generalized Phrase Structure Grammar 6, 33,
of 269, 275–7, 294–5, 391–7; 181, 183
structure of 382 generalized quantifier (GQ) 11, 407–11, 415,
Exceptional Case Marking (ECM) 257–8, 422–4, 473
262, 277–9, 283 “Generalized Quantifiers and Natural
exception phrases 411; ordering of 447–8 Language” (Barwise and Cooper) 407
exclusive relation 379 generalized structural asymmetry 9, 29
existence verbs 320–2 generative semantics 359; neo-generative
expletives 320–1, 334; thematic status of semantics 76, 76–80, 77, 79, 130–1;
283–4 style models 38
Extended Projection Principle (EPP) 12, 12, genitives 457–65; consequences and
105, 119, 149, 414; A-movement and comparisons of 461–5; morphology
88; little v and 99 of 123; nonthematic 461–3;
Extended Standard Theory 5–6, 61, 87–9, possessive Ds as triadic predicates
427, 430 458–61; postnominal 460, 460–1,
external merge 23, 27, 105 476, 477; prenominal 418–19, 457;
extra argument 20 shift 460–1, 461, 463, 473; thematic
extraction analysis 226 463–5, 477
Gerdts, D. B. 83, 84, 109, 131
Fabb, N. 357, 384 German (language) 172
falsity 398 Germanic languages 173
Farkas, Donka 250, 258 gestures 50, 52
Farrell, P. 109 Giorgi, Alessandra 164, 180, 182, 463
Feldman, H. 162 give 48, 48–54, 49; raising structures for 177,
Fiengo, Bob 473 177–9, 178, 179
Fillmore, C. 31, 181–2, 309, 356 Givón, T. 185, 376–7, 378, 385
Index 487
goal augmentation 212–13, 223 Hoeksema, J. 220, 227
goal-theme asymmetries 92 Hoekstra, T. 260
Goldberg, A. 40 Hooper, J. 395
governing category 146–7 Hornstein, N. 21, 224, 412, 477
government-binding theory 158, 320 Huang, C.-T. J. 215, 224
GQ see generalized quantifier Hudson, R. 225
Greek (language): ancient 161–2; modern Hungarian (language) 418–19, 458, 458
110–11, 122 Huybregts, Riny 258, 339
Green, G. 48, 50, 55, 57, 59, 167, 170; French
and 328 Icelandic languages 86, 119, 182, 359
Grimshaw, Jane 213, 254, 351, 464 identity: mapping 67–8; sloppy 216
Gropen, J. 213 idioms 127; caused-motion 41, 48, 76–7;
Grosu, A. 218, 363 dative 47–60; discontinuous 3, 3, 8,
Guéron, Jacqueline 180, 471 142; double object 49–54, 77; oblique
Gumbaingar (language) 462–3 47–54; verb-adverbial 441
if/when-clauses 396
Haegeman, L. 127, 183 ill-formedness 112, 300, 341–2, 344
Haisla (language) 388–9, 404 imperatives, concealed 256, 262
Hale, K. 186, 195, 215, 370–1 impersonal constructions 334
Halkomelem (language) 90, 109–10, 114; inchoative predicates 325
benefactive applicatives of 62; stimulus inclusive relation 379, 382
applicatives of 64 incomplete expressions 33
Harada, N. 37, 111, 124, 125, 129 indeterminacy 296–8
Harley, H. 40, 76–7, 79–80, 109, 127–31 indirect objects (IObjs) 3, 141, 238
head-adjacency 454 indirect passives 159–64, 160
Head Driven Phrase Structure Grammar indirect θ-marking 198, 222
21, 33 Indonesian (language) 169, 307–9, 315
head-raising syntax 427 infinitival control structures 345
heaviness effects 448 infinitival purpose adjuncts 356
Heavy NP Shift 153–4, 179, 182–3, 194, inflection 424; concordial 119, 119
299, 303, 450; absence of 354–6, inflectional filter 181, 384
355; adjunct orientation and 314, informational restrictedness 81–4
314–15, 315; A′-movement and 304, information packaging 269, 392
304–9; bounding effects with adjuncts inherent Cases 112, 122–5, 123, 124
306–7; consequences of 311–19; inner anaphora 465
dative structure and 147–9; freezing innermost V-complements 397
and 318–19, 319; Indonesian and inner object 141, 347
307–9, 315; Larson on 258; LPR and inner subject 144, 181, 222, 234, 310
203, 265–8, 267, 304–19; preposition instantiate structure 49
stranding and 312–13; P-stranding intensional logic 475, 477
and 304–6; reconsidered 309–19, 310, intentionality 85, 397–402, 400, 401, 402
311; from subject position 332–4; V′ interface: pragmatics-grammar 38;
Reanalysis and 147–9, 148; XP shift validity 274
and 315–18, 316 interface legibility 98, 114; syntactic
Hebrew (language) 334, 359 object 23
Hegarty, Michael 221 interface logic 285; of LPR 269–71
Heim, I. 32, 224, 407 internal argument 324
Heinämäki, O. 395 internal θ-role 290, 290, 300
Herburger, E. 269, 270, 392–3 interpretable features 113, 113, 116
Hestvik, A. 155 interrogatives 44; complements 254
hierarchy 196; linearity versus 191–3, 192, intersentential anaphora 194
193; relational 87, 187, 474; of intransitive verb phrases (IVPs) 3, 272
θ-features 24–5; see also thematic intrasentential binding 194
hierarchy introducing arguments 20
Higginbotham, James 180, 360, 371, 407, inversion: of argument 104; relation 75;
464; Categorial Grammar and 32; LF of subject 359
and 223; role identification and 337–8; IObjs see indirect objects
θ-grid and 17 Irish (language) 144, 359
high applicatives 72, 74 Italian (language) 164, 359, 419
high for-phrases 72 IVPs see intransitive verb phrases
488 Index
Jackendoff, R. 11, 30, 223–4, 398, 443, 471; 172; Germanic 173; Greek 110–11,
double objects and 191–8, 201, 204–5, 122, 161–2; Gumbaingar 462–3;
207–12, 215, 218, 221; Larson and 90 Haisla 388–9, 404; Halkomelem 62,
Jacobs, R. 456 64, 90, 109–10, 114; Hebrew 334,
Jacobson, Pauline 10, 266–7, 271, 288–9, 359; Hungarian 418–19, 458, 458;
293, 309, 362; Generalized Phrase Icelandic 86, 119, 182, 359; Irish 144,
Structure Grammar and 6, 181, 183; 359; Italian 164, 359, 419; Japanese
Heavy NP Shift and 299; Montague 59, 82–3, 93, 111–13, 122, 124, 126,
Grammar and 2, 5; thematic status of 129–30, 161–2, 258; Kalam 376;
expletives and 283; V-raising and 143 Kinyarwanda 63–4, 88, 169, 185;
Jaeggli, O. 156, 184 Korean 59, 77–9, 109, 115–16, 122,
Japanese (language) 59, 93, 111–13, 122, 125–6; Miskitu 370, 370; Misumalpan
129–30, 258; Case marking and 126; 370, 372, 372; modern Greek 110–11,
informational restrictedness and 82–3; 122; natural 321; nonserializing 384;
3 → 1 advancement in 161–2 Norwegian 155, 183, 305; Papuan
Jeanne, L. 424 376; Portuguese 45, 57–9, 128–9;
Johns, A. 222 Romance 173–4; Romanian 111–12,
Johnson, Kyle 40, 180, 182, 187 122, 132; Russian 172, 174; Salishan
Johnston, M. 396 389; Spanish 46, 112, 120–2, 128, 174,
359; Sranan 373, 373, 375; Swedish
Kaiser, E. 82–3 305; Tairora 376; Tok Pisin 376;
Kalam (language) 376 Tsimshianic 389; Tukang Besi 83; Ulwa
Kamp, H. 407 370; Wakashan 388–9; Warlpiri 185;
K analysis 18–20, 19, 20, 22, 31, 33, 422 Welsh 144; West African 367; Yoruba
Katz, Jerry 473 368, 379, 384
Kayne, Richard 38, 133, 306, 361, 404, Lappin, S. 474
463, 473; accusative Case and 114; Larson, Richard K. 8, 10–12, 37, 52, 90,
affectedness and 187; Antisymmetry 230, 412, 424; ambiguity and 360;
Theory of 1; NP shift and 358; applicative analysis and 71; argument
preposition stranding and 173–4; small augmentation and 211; argument
clauses analysis and 222 projection and 431; A-shift and 102,
Keenan, E. 183, 407, 410, 438, 445 105–6; Baker and 92, 109, 198; Case
Kinyarwanda (language) 63–4, 88, 169, 185 assignment and 259; chomeur status
Kitagawa, Y. 176 and 89; complement alternations and
Kiyosawa, K. 83, 84, 131 201; concordial inflection and 119;
Klein, E. 467 datives and 209–10; degree morphemes
Ko, H. 94 and 468; determiners and 474; domain
Koopman, H. 259, 357 asymmetry and 191; double object
Korean (language) 59, 77–9, 109, 115–16, constructions and 233; dualism and
122, 125–6 124; empty head positions and 222;
Koster, J. 260 gapping and 219; Heavy NP Shift
Kratzer, A. 20, 32, 73, 396 and 203, 258, 309; infinitival control
Krifka, M. 18, 33, 40, 279–80 structures and 345; intentionality and
Kupula, M. 425 85; Japanese and 112, 124, 129–30;
Kuroda, S. Y. 176, 442 linearity versus hierarchy and 191–3;
LPR and 285–7; modifiers and 215;
Ladusaw, William 180 nonconstituent coordination and 217,
Lakoff, G. 347 225; nonthematic resultatives and 300;
languages: ancient Greek 161–2; applicative parasitic gaps and 285; prepositional
83; with applicative constructions 150; datives and 459; recursivity and 15;
Australian 462–3; Austronesian 83; RG and 87; scope antinomies and
Bahasa Indonesian 169, 307–9, 315; 275; scope freezing effect and 84,
Berber 144, 359; British English 127; 222; secondary predicates and 380;
Cebuano 185; Chaga 72, 74; Chamorro sentence-final adverbial adjuncts and
359; Chichewa 61, 63–5, 79, 95, 169; 30; spray-load verbs and 110; structure
Chulsu 79; Danish 173–4, 182; without projection and 196–7; thematic
double object forms 57–60; Dutch hierarchy and 7, 433; θ-features and
173–4, 226; Dyirbal 462–3; Eskimoan 24; θ-theory and 16; transitivity and
183; European 60; Fon 371; French 322, 357; UTAH and 198–200; verbal
173–4, 328, 330, 334, 359; German modifiers and 440; voice alternation
Index 489
and 103; V-raising and 196; V′ locatives: construction of 206; modification of
Reanalysis and 268, 288–94 369; oblique 39–47, 103; with/of 104
Lasnik, Howard 35–6, 138–9, 194–5, 317, locative verbs 39; transitive 129
321, 361; dative shift and 153; double logical form (LF) 223–4, 299, 335, 412, 428,
object structures and 92, 137, 244–5 435
LCSs see lexical conceptual structures Logical Form-legible features 23
Lebeaux, David 84, 202 Longobardi, P. 463
Lees, R. 408 low applicatives 72, 74
Lefebvre, C. 368, 370–3, 375, 377, 380–1, low for-phrases 72
385 LPR see Light Predicate Raising
Left Concatenation 4, 10, 12, 69, 266 Ludlow, P. 475
left peripheral deletion 225 Lujan, Marta 180
Levin, B. 41, 43–5, 48, 55, 64, 77, 128, 184
Lewis, D. 396 McCawley, J. 226, 363, 462
lexical conceptual structures (LCSs) 372–3, McConnell-Ginet, S. 182
377 McNulty, E. 361
lexical conditioning 210–12 Manzini, M. R. 260
lexical decompositions 39, 39 mapping: hypothesis 270–1, 271, 391, 394,
Lexical Functional Grammar 61 394, 396, 396–7, 404; of identity 67–8
lexical integrity 294, 295 Marantz, Alec 70–3, 107–8, 167, 180, 186,
lexical marking (L-marking) 91, 215 312; applicative-oblique alternation
LF see logical form and 66; Chomsky and 141; Extended
Li, Y.-H. A. 222, 374–6, 377, 385 Standard Theory and 61
licensing 351, 361, 438; polarity 192; verb maximal phrases 291, 297
serialization and 368–77 May, R. 222, 224, 407, 466, 471
Light Predicate Raising (LPR) 1, 203, 204, Mchombo, S. 61
208, 303–56, 451, 452; absence of M-command 192, 274–5, 404
Heavy NP Shift and 354–6, 355; MDP see Minimal Distance Principle
A′-movement and 304–9; complement measure adverbials 117
alternations and 203–5; in DegPs media consumption 299
472–3; in DPs 451, 451–2, 452; meng- (Bahasa Indonesian) 307–9, 315
Heavy NP Shift and 265–8, 267, merge 25, 99, 102; external 23, 27, 105; Right
304–19; interface logic of 269–71; Wrap and 4, 4–5, 5
modifiers and 272–7; parasitic gaps metaphorical motion verbs 41, 41
and 284–5, 334–54; in pleonastic metonymic extension 53
constructions 319–34; reduced clauses middle formation 88
and 277, 277–83; revisited 285–96; middles 393–4, 405
structural implications of 271–83; Minimal Distance Principle (MDP) 4, 35, 230,
thematic status of expletives and 239–45, 254
283–4; in VP 449–51; VP-internal Minimalist Program 21
alternations and 285–8, 286; V′ Minimalist Theory 93
Reanalysis and 288–96, 450; as X′ minimality 95, 105, 123, 360, 360, 417;
indeterminacy 296–8 constraint of 27–9, 99
light v 23, 28–9, 33 minimal structural elaboration 8, 10, 10, 412;
linear correspondence axiom 404 X-bar theory and 13
linear order: linearity in discourse 194–6, 195; mirror principle 389–90, 404
linearity versus hierarchy 191–3, 192, Miskitu (language) 370, 370
193; restrictiveness and 193–4, 194 Misumalpan (language) 370, 372, 372
Linguistic Society of America Institute 35 Mittwoch, Anita 257, 258, 262
Linguistic Society of America Summer Institute Miyagawa, Shigeru 111
35, 258 modern Greek (language) 110–11, 122
little v 24–6, 97, 98–106, 123, 297–9; agentive modifiers 215–17, 222; causal 396; in DPs
114; causation and 15–16; C-command and DegPs 440–57; in LPR 272–7;
and 29; K analysis and 20; recursivity verbal 440
and 29–30; structural accusative Case monadic Ds 437
and 115; unergatives and 33; voice monosemous caused-motion verbs 45
alternation and 97 monotransitive pairs 131
L-marking see lexical marking Montague, Richard 187, 437
locality 7, 93, 95 Montague Grammar 2–6, 265, 274, 277, 390,
location 196 446–7, 474–5; asymmetry and 36;
490 Index
dative alternation and 67; modifiers nonserializing languages 384
and 272; Right Wrap and 8–12, 88; nonstrandability condition 305
unary function structure and 32 nonthematic genitives 461–3
morphemes: applicative 110; degree 468–9; nonthematic resultatives 283, 300
passive 157; valence-changing 477 Norwegian (language) 155, 183, 305
morphology 37; applicative constructions and noun phrases 427
65–6; dative 123, 125; genitive 123; NP movement 153, 160, 201, 202, 212–14,
object and 319, 331; participial 156; 235, 238; Case marking and 459;
passive 88, 156–7, 163 complement alternations and 201–3;
Moshi, L. 61 government-binding account of 158;
motional to-datives 210–15 rightward 304; subject position
motion verbs 40–1, 41, 44, 45, 47 and 151
multiple accusative Cases 114–16 NP-S analysis 442
multiple complement data 358 NP shift 285, 303–4, 330–1, 331, 358; see
multiple oblique arguments 26–9, 27, 28, 29, also Heavy NP Shift
416–18, 417, 418 null causative verbs 13–16
multiple relatives 448, 448–54 null pleonastic analysis 334
multisets 424 null suppletion 65
Munro, P. 254 numeration 28, 30
Muysken, P. 385
Oba, Y. 37, 94–6, 106, 132
Nakanishi, K. 82 object construal 250
natural course of events 348, 350 object control verbs 241
natural language 321 objective Case 158
natural theory of event composition 72, 76 object-oriented depictive predicates 378
Neale, S. 475 Object Preposing 89, 89
negation 244 object shift 93
negative polarity 273–4, 357, 388, 404, OBL → 2 89, 90; as A-movement 88–90;
441; C-command and 139, 215, 221; dative shift and 86–8
licensing 192; outer object item 244 obligatory control structures 261
negative quantifiers 244 obligatory V′ Reanalysis 340–3, 341
Neijt, A. 218, 225, 226 oblique-applicative alternation 66, 66, 96, 126
neo-Davidsonian semantic composition 18–20, oblique-copular possessives 106
29, 80, 269, 422, 424 oblique-double object relationship 68
neo-generative semantics 76, 76–80, 77, 79, oblique locatives 103; double object
130–1 constructions and 39–42; oblique
Neo-Melanesian Pidgin 376 possessives and 42–7
neutrality 11, 16; category 30, 33, 436 obliques: applicatives and 66–86; arguments
Nikiforidou, K. 471, 477 26–9, 416–18; complements 30;
nominalizations 464 construction of 92, 105; datives 103,
nominal structures 407–24; DP-IP analogy 140; ditransitive constructions 100;
and 408, 408; GQ theory and 408–11; grammatical relations of 87; idioms
implications of 421–4; projecting DP 47–54; object forms 165; predications
and 411–21; recursion and 416–18 352–4, 353, 354
nominal syntax: projecting DP and 432–40; Oehrle, R. T. 35, 88, 139, 161, 187, 248, 300
relational view and 430, 430–1; Oh, Sunseek 258, 261
structural projection and 431–2 one-step analysis 161
Nom-S analysis 442 open-scale adjectives 282
non-case transmitters 331 Ormazabal, J. 37, 46
nonconstituents 118; coordination of 217–19, outer object negative polarity item 244
219, 225–6 outer pronoun effect 223
nondative applicatives 107, 107–8, 108
nondelimiting secondary predicates 380 P analysis 18, 18, 21, 422, 424
nonderivational analysis, of double object Papuan languages 376
constructions 71–80 parallelism 349–50, 357, 437, 459, 462
nonexclusion relation 429 parasitic adjunct constructions 361
non-literal give expressions 50 parasitic gaps 284–5, 298, 334–54; adjuncts
non-prepositional accusative Cases 115, 361; complex coordinations and
115–20, 118, 119, 120 347–50; consequences of 344–50;
nonreflexive object nominals 308 oblique predications and 352–4, 353,
Index 491
354; with promise 345–7; RNR and predicate-object features 22, 22–4, 23
351, 351–2, 352; as secondary predicates predicate-oblique arguments 25–6, 26
336, 336–44, 337; subject 350–2 predicates and predication 16; complex 3,
Parsons, Terence 18, 217 8, 265–98; inchoative 325; oblique
Partee, B. 67–70, 201, 242–5, 260–1, 352–4, 353, 354; raising 325–30;
462, 464 relations of 4; sentence-final verbs and
partially decompositional model 14, 15–16 397–400; subatomic analysis of 18;
participial morphology 156 subject-oriented depictive 378; subject-
particle movement 357 predicate features 23, 23–5, 24, 25;
passives 90, 97–100, 103–4, 116; active- transitive 149, 179, 319, 409; unary
passive relationships and 67, 81; Case 19; see also light predicate raising;
assignment to double objects 157–9; secondary predicates and predication
Case marking in 172; dative shift as predictability 223
150, 150–3, 151, 152; derivation 36, prelexical event relations (PRs) 374
37; indirect 159–64, 160; morphemes prelexical representations 375
157; morphology 88, 156–7, 163; prenominal adjectives 454–7, 456
properties of 159; pseudo-passives 131; prenominal genitive constructions 418–19, 457
subject suppression and 156–7; voice prenominal relatives 453–4
alternation and 97–100 prepositions 105; datives 35, 459–60;
passivization 84, 161, 240, 247 incorporation analysis of 70
Perlmutter, D. 162, 187, 320 preposition stranding (P-stranding) 173–4,
persuade 232–3, 236–8, 236–9, 242, 258–62 312, 312–13, 313; absence of 304–6
Pesetsky, David 118–20, 131, 180, 258, 261, presentational constructions 326–8;
473; accusative Case and 109; Cascade constraints as 328–30
Theory of 1, 33; idioms and 142; presupposed content 394
neo-generative semantics and 76–7; primary object 61
psych verbs and 185; θ-features and probe-goal relations 27, 93
21–2, 412 productivity, double object 168
Peterson, D. 84 projection: of adjuncts 89, 91, 93, 103, 103;
phase-state transitives 280 alternative 205–9; of arguments 175–7,
Phillips, C. 402 431; of control structures 245–7, 246,
phrasal verb 141 247; EPP 12, 12, 88, 99, 105, 119,
phrase structure 4 149, 414; pleonastic 325–30; structural
physical motion verbs 41, 41 21–9, 196–7, 431–2
physical states 50, 52 projection of datives 39, 40, 42, 43;
Pica, P. 182 applicative constructions 60–6; double
P-incorporation 312 object construction and 39–42; idioms
Pinker, S. 40, 186 and 47–60; oblique locatives and
pleonastic constructions 359; definiteness 39–47; oblique possessives and 42–7;
restrictions and 330–2; Heavy NP Shift relevant information for 37–9
and 332–4; LPR in 319–34; predicate projection of DPs 411, 411–21, 412, 427–73,
raising and 325–30; projection of 443–5, 455; DegPs and 465–73;
325–30; transitivity alternation and determiner-oblique argument and 415,
320–5 415–16, 416; determiner-restriction
polarity: licensing 192; negative 139, 192, 215, term and 412–13; determiner-scope
221, 244, 273–4, 357, 388, 404, 441 term and 413, 413–15, 414, 415;
Pollock, J.-Y. 226, 424 determiner semantics and 428–30;
polysemy 42, 45, 46 genitives and 457–65; modifiers in
Portuguese (language) 45, 57–9, 128–9 440–57; multiple oblique arguments
possession 59–60; loss of 41; transfer of and 416–18, 417, 418; nominal syntax
74, 248, 250, 262; see also caused and 432–40; recursion and 416–18;
possession voice alternation in 418–21
possessives: alternation 126; Ds as triadic projection of raising structures 174; argument
predicates 458–61; oblique 42–7; realization and projection 175–7,
oblique-copular 106 176; illustration for 177–9; single
possessor raising structures 116–17 complement hypothesis 174–5, 175; V′
Postal, P. 187, 409, 424 Reanalysis and 179–80
postnominal genitives 460, 460–1, 476, 477 Projection Principle 88, 153, 242, 356, 374
postverbal position 330 promise 230–58, 234, 235, 236, 258–62,
pragmatics-grammar interface 38 260; alternative account of control and
492 Index
242–5; apparent object-control double reciprocals 138
object verbs and 251–8; as double recoverability of deletion 166–9, 223
object verb 230–3; generalizations of recursion 26–30, 27, 28, 29, 33, 416–18
control and 239–42; objections and recursivity 15, 29–30; derivation 12;
247–58; parasitic gaps with 345–7; potential 9
persuade and force versus 236–8, reduced clauses 277, 277–83; complex telic
236–9; projection of control structures predicate formation and 281–3; telicity
and 245–7; shifting control and 247–51 and 279–81
pronominal objects 163 referential use 475
pronouns 161; alternation of 343, 343–4, reflexives 138; empathic 437; object
344; meanings of 437; outer pronoun nominals 308
effect 223 registration markers 169
property 429 regrouping 329
propositional nucleus 279 Reinhart, T. 140, 216, 274, 404
propositional operators 395 relational advancement 183
proto-agent 261 Relational Grammar (RG) 87, 90, 102–3, 153,
proto-patient 261 185, 474; applicative constructions and
proto-roles 261 61; datives and 125
PRs see prelexical event relations relational hierarchy 87, 187, 474
pseudocleft paradigms 209 relational view of determiners 408–11
pseudogapping 220 relation inversion 75
pseudo-passives 131 relative clauses 31, 361; as arguments 446–7;
P-stranding see preposition stranding for D-complements 441–8; ordering of
psychological states 50, 52 447–8
psych verbs 162–4, 164, 178, 185, 221 relatives: Article-S analysis of 11, 427, 441–3,
purely formal role 24 442, 446; multiple 448, 448–54;
purpose adjuncts, infinitival 356 prenominal 453–4
Pylkkänen, Llina 20, 72, 74–6, 130–1 relative scope 84, 131
relative subordination, of arguments 176
Q-float see quantifier float relativization 55
QR see quantifier raising relativized UTAH 200
quantification 395–7, 474; comparatives as remuneration phrase 26
466–8; degree morphemes and 469; restrictedness: informational 81–4;
restrictedness of 84–6; structured quantificational 84–6
event 269, 275–7, 294–5, 391–7; restrictions 430; definiteness 330–2;
unrestricted 428 determiner-restriction term 412–13
quantifier float (Q-float) 126, 131 restrictiveness 193–4, 194
quantifier raising (QR) 225, 299 resultatives 3, 3, 8; constructions 281–3,
quantifiers 31, 411; generalized 11, 407–11, 300; nonthematic 283, 300; secondary
415, 422–4, 473; negative 244; predicates 378, 380
passivization and 84 result state semantics 279
quantization 280–1 RG see Relational Grammar
question formation 55 Richards, N. 40, 127, 128
right-adjunction analysis 299
R-A see right ascent right ascent (R-A) 387, 389–90, 397, 404
Radford, A. 127 Right Concatenation 4, 12, 69
raising structure projection: argument right descent (R-D) 33, 394, 404; anti-
realization and projection 175–7, correspondences and 388–90; domain
176; illustration for 177–9; single phenomena and 388
complement hypothesis 174–5, 175; V′ Right Node Raising (RNR) 226, 355, 362–3,
Reanalysis and 179–80 363, 444; constituency and 218;
raising structures, possessor 116–17 parasitic gaps and 351, 351–2, 352
Randall, J. 213 rightward downward branching analysis 192
Rapoport, T. 328 Right Wrap 29, 31, 68–9, 183, 266–7, 357,
Rappaport Hovav, M. 41, 43–5, 48, 55, 357; merge and 4, 4–5, 5; Montague
64, 77 Grammar and 8–12, 88; V-raising and
R-D see right descent 5, 181
realization: of arguments 175–7, 176, 245; of Rizzi, L. 163, 179, 185, 221
complements 252; surface realizational RNR see Right Node Raising
constraints 56 Roberts, Ian 144, 181, 259, 384
Index 493
Rochemont, M. 295 shell structures 9–12; empty verbs and
Roeper, Tom 258 13–15, 13–16; problems with 12–13;
role identification 337–40, 338, 339, 349, 353 properties of 9–12; reconstructing
Romance languages 173–4 16–31; revised account of 29, 29–31,
Romanian languages 111–12, 122, 132 30, 31; semantic composition and
Romero, J. 37, 46 16–20; structure projection and 21–9;
Rosenbaum, P. 4, 35, 239, 456 θ-features and 21–9; θ-theory and
Ross, J. R. 195, 356, 441, 462, 467–8; 16–20; verb phrase 6–8
complex coordinations and 347; LPR shifting control 247–51, 261–2
and 303–5 Shimizu, M. 112, 161, 162
Rothstein, S. 476 Shlonsky, U. 351
Rouveret, A. 185 Siegel, D. 477
rule-to-rule hypothesis 2, 2 single argument requirement 3
Russell, B. 32, 221, 466 single complement hypothesis 6, 174–5,
Russian (language) 172, 174 175, 431
single event 373
Safir, Ken 326, 328, 473 sloppy identity 216
Sag, I. G. 222, 225, 466 small clauses: analysis of 79, 222; construction
Salishan (language) 389 of 277
Schacter, P. 441, 462 Smith, C. 11, 427, 441, 456
Schein, B. 314 snowballing derivation 288
Schneider-Zioga, P. 131, 202, 222 Solan, L. 224, 225
scope 277; Andrews and 400–2; antinomies sound symbolism 53
275; contrasts of 397–404; determiner- Spanish (language) 46, 112, 120–2, 128,
scope term 413, 413–15, 414, 415; 174, 359
freezing effect 84, 110, 131, 222; relative Speas, M. 176, 407, 430
84, 131; sentence-final adverbs and special control behavior 230
397–400; VP-initial adverbs and 402–3; special external argument 324
X argument and 433–6, 434, 435 special internal argument 324
scrambling 94, 94, 105 Sportiche, D. 140, 176
secondary object 61 spray-load alternations 90, 114, 210, 213–14,
secondary predicates and predication 314, 292, 300; NP movement and 201, 203;
345, 349, 369; delimiting 380, 382; scope freezing and 131
interpretation of 378–80; nondelimiting spray-load pairs 131, 202–3, 293
380; parasitic gaps as 336, 336–44, spray-load verbs 90, 110, 193, 202, 214–15,
337; serialization and 378–85; structure 223
of 380, 380–3 Sproat, R. 357
semantic analysis 16–18 Sranan (language) 373, 373, 375
semantic composition 2; neo-Davidsonian S-structure 137–8, 178; negative polarity and
18–20, 29, 80, 269, 422, 424; standard 215, 244; parasitic gaps and 335, 341;
semantic analysis 16–18; structure promise and 236, 236–7, 237, 244;
projection 21–9; θ-features 21–9 UTAH and 200
semantic conflation 329 standard analysis 421
semantic dative shift 68, 68–70, 69 standard semantic analysis 16–18, 17
semantic kick 131 Standard Theory 159
sentence-final adverbs 387–404; adjuncts Stavi, Y. 407, 410, 438, 445
and 30; Andrews and 400–2; new Stillings, J. 225
approach to 391; predicational nature stimulus applicatives, Halkomelem 64
of 397–400; right ascent and 390; right Stockwell, R. 441, 462
descent and 388–90; structured event Stowell, Tim 11, 180, 242–5, 260–2,
quantification and 391–7; VP-initial 347, 356; Case resistance and 239;
adverbs and 402–3 ECM and 279; promise and 230–1;
sentence-word anti-correspondences 388–90 P-stranding and 304, 306
separation 356, 356 stranded direct object 219
serialization 367–85; licensing principles of stranding analysis 131
368–77; parameters of 377–8, 383–5; Stratal Uniqueness Law 89
secondary predication and 378–85; strict subject orientation 182
structure of 368–77 Stroik, T. 388
set relations, determiners in 429–30 structural Cases 112; accusative 114–16;
Seuren, P. 467, 468 objective 158
494 Index
structural projection 21–9, 196–7, 431–2 thematic status, of expletives 283–4
structured event quantification 269, 275–7, thematic transitivity 289, 296
294–5, 391–7; because-clauses 394–7; θ-agreement: determiner-oblique argument
focus and 392–3; middles and 393–4 and 415, 415–16, 416; determiner-
subatomic analysis 18 restriction term and 412–13;
subject-object asymmetry 9 determiner-scope term and 413,
subject-oriented adjuncts 343 413–15, 414, 415
subject-oriented depictive predicates 378 θ-assignment hypothesis 150
subject-predicate features 23, 23–5, 24, 25 θ-features 21–9, 22, 39, 97–8, 104, 125,
subjects: control verbs 241; empty 259; inner 424; agreement of 21–2; determiner-
144, 181, 222, 234, 310; inversions of oblique argument and 415, 415–16,
359; orientation of 164, 182; parasitic 416; determiner-restriction term and
gaps 350–2; suppression of 156–7; 412–13; determiner-scope term and
surface 163 413, 413–15, 414, 415; hierarchy
subject-to-object raising 88 of 24–5; multiple oblique arguments
subject-to-subject raising 88 26–9, 27, 28, 29; predicate-object 22,
sublexical analysis 18 22–4, 23; predicate-oblique argument
subordination 429; of arguments 176; 25–6, 26; recursion and 26–9, 27, 28,
weak 370 29; subject-predicate 23, 23–5, 24, 25
suppressed subjects 151 θ-grid 17, 290
surface realizational constraints 56 θ-identification 372
surface subjects 163 θ-marking 198, 222
Swedish (language) 305 θ-role identification 337–40, 338, 339,
symbolism, sound 53 349, 353
symmetric applicative constructions 130 θ-theory 16–20, 98, 176, 178, 239, 247, 289;
syntactic atom 319 neo-Davidsonian semantic composition
syntactic composition 2 18–20; standard semantic analysis
syntactic dative shift 67–8 16–18, 17; structure projection 21–9;
syntactic domain 191; asymmetries of 138–40, X-bar theory and 149
139, 140 Thomason, R. 230, 242–5, 260–1
syntactic word formation 359 Thompson, S. 395
Syntax Colloquium 473 Thráinsson 86
Syntax Workshop 258 3 → 1 advancement 185; in ancient Greek
Szabolcsi, A. 407, 418, 458, 473 language 161–2; in Japanese language
161–2; psych verbs 162–4, 164
Tairora (language) 376 3 → 2 advancement 153
Talmy, L. 328 through-movement 307
Taraldsen, T. 334 Tok Pisin (language) 376
target position 89 too constructions 471–2, 477
teach 255–6 Torrego, E. 21, 22, 118, 119, 120, 412
telicity 279–81; complex telic predicate to-the-possession-of(x,y) 74–5
formation and 281–3 trace alternation 343, 343–4, 344
temporal modification 369 transfer of possession 74, 248, 250, 262
Tenny, C. 223 transformational relationships 67
thematically monotransitive 311, 456 transitives: active 14, 103; locatives 129;
thematic constituency relations 381 phase-state 280; predicate 149, 179,
thematic genitives 463–4, 477; semantics of 319, 409; verbs 2; see also ditransitives
464–5 transitive verb phrases (TVPs) 3, 3, 267–8
thematic hierarchy 7, 11, 287, 417, 432–3, transitivity 307, 315, 336, 357; of agreement
459–61; alternative projection and 99, 101–2; alternation 320–5, 322,
205–9; applicatives and 132; argument 323; effects 358; reexamined 291–3,
realization and 176; determiners and 292, 293; thematic 289, 296
410; inner subject and 222; Larson transmission verbs 40–1, 41
and 24; modifiers and 217; promise Travis, L. 132, 259, 357
and 246; relational hierarchy and 187; triadic Ds 437–8
transitivity alternation and 323; UTAH true serialization 370, 372, 377, 385
and 200 truth-value 2, 2
thematic relations 38–9 Tsimshianic (language) 389
thematic roles 350; assignment of 151; for Tsujioka, T. 111
determiners 410–11 Tukang Besi (language) 83
Index 495
TVPs see transitive verb phrases 40–1, 41; unaccusative 100; see also
2 → 1 advancement 153 V-complements; V-raising
2-to-3 demotion 259 verb serialization 367–85; licensing principles
two-constituent test 225 of 368–77; parameters of 377–8,
two-step analysis 161 383–5; secondary predication and
378–85; structure of 368–77
Ulwa (language) 370 Vergnaud, J.-R. 185
unaccusatives 320, 324, 324–6, 325, 326, verum focus 405
331; dative 164; determiners 474; Visser, F. T., generalization of 240, 240–1,
formation of 88; verbs 100 241, 260
unaccusativity 169–71 voice 186; heads 25
unary function structure 32 voice alternation 96–125, 103, 104, 106;
unary predicate 19 A-shift and 100–2, 101, 102; Case
unergatives 14, 180, 324, 327; determiners theory and 108–25; in DPs 418–21;
437; little v and 33 little v and P 97; observations and
ungrammaticality 154 comparisons 102–8; passives 97–100
Uniformity of Theta Assignment Hypothesis VP-adjuncts 272
(UTAH) 150, 196, 198–201, 199, 222 VP-functors 272, 272
uninterpretable features 113, 113, 118 VP-initial adverbs 402–3
universal determiners 470 VP-internal alternations 285–8, 286
universal grammar 105 VP-preposing 307
universal schematism 174 V-raising 6, 12, 143, 143–5, 144, 145,
University of California, Irvine 258 267; across-the-board 146, 146;
University of Indiana, Bloomington 473 nonconstituent coordination and 217;
University of Wisconsin, Madison 5 passives and 104; psych verbs and
unmarked word order 472 178; radical shift in theory of 197–8;
unrestricted quantification 428 Right Wrap and 5, 181; strict subject
unsaturated expressions 33 orientation and 182; structure projection
unvalued features 21, 21, 113, 113, 118 and 196–7; UTAH and 198–201
upward agreement 29 V′ Reanalysis 179–80, 268, 310–13, 354,
Uriagareka, Juan 361 357; ditransitives and 324; examined
UTAH see Uniformity of Theta Assignment 288–96; Heavy NP Shift and 147–9,
Hypothesis 148; LPR and 288–96, 450; obligatory
340–3, 341
valence 409–10 V-visibility 181
valence-changing morphemes 477
valued features 21, 21, 113, 113, 116 Wakashan (language) 388–9
Van Riemsdijk, H. 183, 339 Ward, B. 81
V-chaining constructions 370 Warlpiri (language) 185
V-complements 440, 440–1; constructions weak crossover effect 138
266; innermost 397 weak subordination 370
Vendler, Z. 446–7, 475 Webster’s Seventh New Collegiate Dictionary
verb-adverbial idioms 441 (WSNCD) 50, 52–3, 127
verbal modifiers 440 Wechsler, S. 281–2
verb-particle compounds 167 Weinberg, A. 224
verbs: apparent object-control double object well-formedness 404
251–8; of availability 320–2, 327; of Welsh (language) 144
ballistic motion 40–1; caused-motion West African languages 367
44, 45, 47; of consumption 280; of Wexler, K. 294, 318, 358
dispersion 224; division of 43; double Wh-clefting 307
object 230–3, 251–8; DTVs 69; DVPs when/before/after-clauses 395–6
3; empty 13–15, 13–16; of existence Wh-movement 284, 294–6 304, 306, 349, 349
320–2; locative 39, 129; metaphorical Wilkins, W. 35
motion 41, 41; of motion 40–1, 41, Williams, Edwin 319, 358, 464, 466;
44, 45, 47; null causative 13–16; object across-the-board extraction and 226;
control 241; phrasal 141; phrase shells anaphora and 260; internal θ-role
6–8; psych 162–4, 164, 178, 185, 221; and 290
raising 5, 5, 12, 28, 143–5; spray- with/of-locatives 104
load 90, 110, 193, 202, 214–15, 223; WSNCD see Webster’s Seventh New
subject control 241; of transmission Collegiate Dictionary
496 Index
X argument 433–6, 434, 435 alternation in 103; V′ Reanalysis and
X-bar theory 197, 313, 384, 427, 431, 179, 288
437, 475; chomeur status and 89; X′ indeterminacy 296–8
Chomsky and 12–13, 92, 290; XP shift 204, 315–18, 316
direction and 221; EPP and 12; X-raising 12
minimal structural elaboration
compatible with 13; projection Yoruba (language) 368, 379, 384
of control structures and 245–7;
serialization and 377; structure Zaring, Laurie 473
and 174–5, 195; θ-theory and Zubizarreta, M.-L. 156
149; transitivity and 322–3; voice Zushi, M. 37, 93–4, 111, 124, 132

Untitled

Uploaded by

Copyright:

Available Formats

Untitled

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Untitled

Uploaded by

Copyright:

Available Formats

On Shell Structure

Richard K. Larson is Professor of Linguistics at Stony Brook University, US.

Edited by Carlos P. Otero, University of California, Los Angeles, USA

1 Partitions and Atoms of Clause 10 Theoretical Comparative Syntax

4 Essays in Syntactic Theory 14 Between Syntax and Semantics

15 Regimes of Derivation in Syntax

1 On the Double Object Construction 137

2 Double Objects Revisited: Reply to Jackendoff 191

3 Promise and the Theory of Control 230

4 Light Predicate Raising 303

5 Some Issues in Verb Serialization 367

6 Sentence-Final Adverbs and “Scope” 387

7 The Projection of DP (and DegP) 427

1.1. Montague Grammar

(1) Syntax Semantics

a. John walks walk’(John’)

b. John kisses Mary kiss’(<John’, Mary’>)

Taken on its own, rule-to-rule correspondence imposes few constraints on either

(2) a. Syntax b. Semantics

John kisses Mary (kiss’(Mary’))(John’)

(3) Syntax Semantics

John give Fido to Mary ((give’(Mary’))(Fido’))(John’)

(4) a. Control Complements b. Complex Predicates (Resultatives)

John persuade Mary to leave John hammer the metal flat

c. Control Adjuncts d. Discontinuous Idioms

1.2. “Right Wrap” and Merge

(5) a. persuade Mary to leave b. hammer the metal flat

c. buy the book to read d. put Mary on the spot

(6) a. kisses John Mary b. CP

(7) a. John give Fido to Mary b. S

1.3. Questions and Puzzles

In a paper on the double object construction (Larson 1988), I attempted to for-

(8) a. XP SpecXʹ Xʹ

b. Xʹ X ZP (Single Complement H

(= (76) in Larson 1988, p. 381)

(9) P1: If is a predicate and is an argument of , then must be realized within a

projection of . (Larson 1988, p. 382)

(10) a. Thematic Hierarchy

put on the table put on the table

(14) a. Control Complements b. Complex Predicates (Resultatives)

persuade PRO to leave hammer flat

c. Control Adjuncts d. Discontinuous Idioms

buy OPi PRO to read ti put on the spot

2.1. Properties of Shell Structures

talk about Mary tell about Mary

Similarly, as discussed in detail in Larson (1991a,c), example (21b) can be analyzed as in

(24) a. TP EPP Projection of Empty Spec

3. EMPTY V AS NULL CAUSATIVE?

BE/GO on the table BE/GO on the table

CAUSE + ?? = wipe CAUSE + ?? = talk

put on the table “put” put on the table

(29) a. John bought [Fido] [from Mary] [for $200].

4. RECONSTRUCTING SHELL THEORY

4.1. θ-Theory and Semantic Composition

4.1.1. The Standard Semantic Analysis

(30) John gave Mary Fido.

(31) The Standard Analysis

(33) e[ give’(e) & Agent(e,j) & Goal(e,m) & Theme(e,f)]

One possible syntax-semantics mapping for such a representation is (34), which I

(34) The P Analysis

1 On the Double Object Construction 137

2 Double Objects Revisited: Reply to Jackendoff 191

3 Promise and the Theory of Control 230

4 Light Predicate Raising 303

5 Some Issues in Verb Serialization 367

6 Sentence-Final Adverbs and “Scope” 387

7 The Projection of DP (and DegP) 427

(8) a. XP SpecXʹ Xʹ

b. Xʹ X ZP (Single Complement H

ears a strong V feature, an EPP feature, and a valued occurrence of a θ-feature

d. John sent each parent the other’s daughter/