THE SYNTAX AND SEMANTICS

OF INFINITIVES OF RESULT
IN ENGLISH

MJ.WHELPTON, B.A., M.PHIL. (OXON)

This thesis is submitted to

the Committee for Comparative Philology and General Linguistics,
University of Oxford,
in part requirement for the degree of
Doctor of Philosophy
(Trinity 1995)
 The syntax and semantics of infinitives of result in English
MJ.Whelpton (St. Peter's College, Oxford)
D.PM1. (Trinity 1995)

This thesis concerns infinitives of result in English, examples of which are

given under (a-c).
(a) John designed a battery to operate at high temperatures.
(b) John designed a battery to win a prize.
(c) John designed a battery for the competition, only to discover that he was
ineligible.
The infinitive in (a) is called a Purpose Clause; the infinitive in (b) is called a Rationale
Clause; and the infinitive in (c) is called a Telic Clause. These infinitives are optional
modifiers of the verbs with which they occur. I argue, however, that important
characteristics of their form and interpretation follow naturally if the infinitives are
assumed to have argument structures which must be incorporated into the argument
structure of the verb they modify.
In Chapter 1,1 introduce the constructions and offer a general discussion of the
problems posed by modifiers in a study of natural language semantics, arguing that
these problems are best addressed by viewing modifiers as predicates of a standard
sort. This establishes a central hypothesis of this thesis, namely that the syntax and
semantics of modification should be incorporated into the theory of argument structure,
or "theta theory".
In Chapter 2,1 offer a summary of key works in the literature in two areas:
theories of argument projection by Williams and Higginbotham and studies of
infinitives of result by Faraci, Bach, Chierchia, and Jones. In Chapter 3, I apply a
range of syntactic tests to establish the structural relation of the infinitives to the
sentences they modify. I argue that the infinitives are embedded at an increasing
distance from the modified verb, from which important characteristics of their
distribution follow.
In Chapter 4,1 turn to the internal syntax of the constructions and consider the
nature and distribution of the gaps that occur within each infinitive and the manner in
which those gaps are interpreted. I relate key characteristics of the infinitives to the
nature of their argument grids. I also address constraints on interpretation introduced by
pronominal binding in the infinitives.
In Chapter 5,1 discuss a variety of problems on the syntax-semantics interface
involving argument structure and control. The issues raised move from the relatively
syntactic (phrase structure accounts of Purpose Clause antecedence and the possibility
of event control for the Rationale and Telic Clauses) to the philosophical semantic
(causal efficacy and the nature of resultant states in the interpretation of the Purpose
Clause).
I then offer a full sample analysis of a sentence modified by all three infinitives
of result. My conclusion points again to the importance of argument structure in an
analysis of infinitives of result in particular and of modification in general.
 ACKNOWLEDGEMENTS
This doctorate is an early submission prompted by the offer of a lectureship. It could
not have come together without the help of many friends and colleagues. I owe a chief
debt of gratitude to Professor James Higginbotham who has guided and encouraged me
throughout the last two years; our arguments have proved a great source of inspiration
and enlightenment I would also like to thank Dr David Cram and Dr Gillian Ramchand
for their help, support, and insight, which they have always made available, and at very
short notice.
I would like to thank all my friends for affection, diversion, and dinners, and
for standing by me when heat and deadlines were too much; and my family without
whose support I could never have come this far.
One year of this research was supported by the British Academy, and the
Graduate Fund of St. Peter's College has been a consistent source of aid.
 THE SYNTAX AND SEMANTICS OF
INFINITIVES OF RESULT
IN ENGLISH

CHAPTER 1: The Introduction

1.1: Introduction
1.2: The constructions
1.3: The semantics of modification
1.4: Argument positions and the theta-role

CHAPTER 2: The Literature

2.1: Two theories of argument projection
2.1.1: Williams (1994) and related works
2.1.2: Higginbotham (1985; 1989)

2.2: Some analyses of infinitives of result

2.2.1: Farad (1974)
2.2.2: Bach (1982)
2.2.3: Chierchia(1989)
2.2.4: Jones (1991)

CHAPTER 3: The External Syntax of Infinitives of Result

3.1: The tests themselves
3.2: The Purpose Clause
3.3: The Rat:onale Clause
3.4: The Telic Clause
3.5: Some other observations
3.6: Further observations: external syntax and conditions on argument
projection
CHAPTER 4: The Internal Syntax of Infinitives of Result
4.1: The distribution of gaps
4.1.1: The Rationale and Telic Clauses
4.1.2: The internal syntax of the Purpose Clause
4.1.2.1: Gaps in the Purpose Clause
4.1.2.2: The Purpose Clause and predication

4.2: Internal syntax and conditions on argument projection

4.3: Gaps, pronouns, and binding
4.3.1: Pronominal binding in the Rationale Clause
4.3.2: Pronominal binding in the Telic Clause
4.3.3: Pronominal binding in the Purpose Clause
4.3.4: Classifying the "resumptive infinitive"

CHAPTER 5: Argument structure and control

5.1: The Purpose Clause: objects, Themes, and resultant states
5.1.1: Conditions on Purpose Clause antecedence
5.1.2: Some problems
5.2: The Rationale Clause: intentions and control
5.3: The Telic Clause: evaluating the consequences
,!'.•

5.4: The typology of Result Predicates and in order insertion

5.5: A sample analysis

CONCLUSION

BIBLIOGRAPHY
 CHAPTER 1: The Introduction

Section 1.1: Introduction

The infinitive has long been an object of interest in generative linguistics, posing
some of the most complex and interesting problems of form and interpretation for the
theory of grammar. Interest has tended to focus on infinitives which occur as the
Complements of verbs such as promise and persuade, as in (1-2),

[1] Sven promised Ute to visit Germany.

[2] Sven persuaded Ute to visit Iceland.

and analyses of these constructions have tended to focus on the problem of referential
control: how is it that we know that in (1) it is Sven who is to do the visiting, where in
(2) it is Ute? In this thesis I will be concentrating on a rather different group of
infinitives which raise a number of interesting issues concerning the relation between
syntax and semantics, issues which can to a large extent be set aside in (1-2). Examples
of the infinitives in question are given under (3-5).

[3] Sven brought a saga along e to read e to his friends.

[4] Sven worked at the newspaper e to earn money.
[5] Sven left Oxford forever only e to be sent back on an assignment by the
newspaper.

The first construction has been discussed extensively in the literature. Following
standard usage, I will call it a Purpose Clause in the general sense of Jones (1991: 26,
fn 18). The construction in (4) has been discussed less often and has received a
systematic treatment principally in Faraci (1974) and Jones (1991); I call this
construction a Rationale Clause after Faraci (1974). The construction in (5) has
generally gone unnoticed in the generative literature, although Faraci (1974: 45, fn 3)
does mention it as a deviant example of a Rationale Clause. The construction has been
recognised in traditional grammar, although the classification is rather tentative: Quirk et
1
al (1985: 629; f 8.132) refer to it as one of the "somewhat vaguer 'outcome1 disjuncts".
The use of the term disjunct implies that the infinitive is being classed as a peripheral
modifier distinct from the infinitives in (3-4), although the nature of the peripherality is
not formally defined. In general though the infinitive in (5) has been lumped with the
infinitive in (4) and taken as an exceptional case. The general sentiment is well-
expressed by Duffley (1992: 21) who provides a functionalist approach to the to-
infinitive.

The existence of two different ways of actualizing the potential

meaning of to in discourse is clearly seen in sentences which can be
ambiguous:
(6) I got up and opened my mouth to commence a downright torrent
of abuse.
(E.Bronte 1847: 236)
Here to commence could mean either "in order to commence" (implying
that the person changed his mind before actually saying anything) or "and
commenced" (implying realization).

It will be argued in this thesis that the difference in interpretation to which Duffley
refers correlates systematically with a number of features of its syntactic distribution
and semantic interpretation which indicate that it is a distinct and independent
construction from both the infinitives in (3) and (4): we do not merely understand the
presence or absence of intention as the discourse suits but attribute different types of
structure and interpretation to sentences which are interpreted in the one way or the
other.
For expository convenience, Jones (1991: 25) terms the infinitives in (3-4)
purpose constructions; not only does this imply that the analysis is intended to apply to
all constructions (including the finite) which carry purpose connotations, but the term
purpose is itself clearly inappropriate for the infinitive in (5), whose reading is purely
factive. On the other hand, all of the infinitives in (3-5) are concerned with the notion of
result in one way or another: in (3), the saga "ends up" in a place where it is available
for Sven to read it to his friends; in (4) Sven's working is intended to have the result of
 earning him money; and in (5) the unexpected result or outcome of Sven leaving Oxford
forever is that he must return. I will therefore adopt the term infinitives of result as an
umbrella term for these constructions; it should be noted, however, that in traditional
grammar the term is used to apply only to the infinitive in (5).
This thesis will have two main aims:

[6] to introduce into the discussion concerning the two infinitives of purpose a third
related construction, the Telic Clause;
[7] to provide an account of the syntax and semantics of these infinitival modifiers
in terms of a general theory of argument projection (or theta theory).

It should be noted from the outset that I will be excluding certain closely related
constructions from discussion. First, I will not address the correspondences between
the Purpose Clause which occurs as a modifier of the verb and that which is optionally
subcategorised by adjectives such as easy, in the classic alternation noted by Chomsky.

[8] Nick is easy e to please e.

[9] Nick is eager e to please.

Second I will exclude from discussion those Purpose Clauses which occur with verbs
such as use and choose. It is true that these infinitives are syntactically optional and that
they are semantically related to the notion of purpose. However, it seems that they may
act as genuine implicit arguments to these verbs, as observed by Bach (1982). Certainly
the verb use entails that the object used is being used for some purpose, to serve some
function: unlike the object of a verb modified by a Purpose Clause, the object of use
does not become available to serve some purpose by virtue of its participation in the
event of using; rather it is serving that purpose simply by virtue of being the object of
that event of using. It is less clear whether the infinitive which occurs with choose is
simply implicit: this would imply that the verb choose always implies some purpose for
the object chosen, but this need not be the case. If your grandmother offers you one of
her chocolates you may choose one because you do not want to offend her but you need
not intend to eat it or do anything with it at all. Nevertheless there remains a distinction
3
between the infinitive which occurs with choose and the infinitive which occurs as a
true Purpose Clause. In the Purpose Clause reading an object comes to have a property
by virtue of its participation in the event: if I bring a bottle opener along to open the
wine with then it is implied that the bottle-opener becomes useful for that purpose by
virtue of having been brought along. But with choose an object is assumed to have a
property which already makes it useful for that purpose and the object is chosen
precisely because of that usefulness: i.e. the event of choosing is precipitated by the
object having that property, rather than the object gaining that property by virtue of
being chosen. In this thesis, I will be concerned only with infinitives which are clear
cases of optional modification.
Section 1.2: The constructions

The infinitive of result which has aroused most interest in the literature is the
Purpose Clause, examples of which are given under (1-3).

[1] Sven handed Ute the ice-cream e to put e in the freezer.

[2] Sven opened the freezer for Ute to put the ice-cream in e.
[3] Sven sent Ute into the kitchen e to put the ice-cream in the freezer.

Both Faraci (1974) and Jones (1991) distinguish the infinitives in (1-2) from the one in
(3). Faraci calls the infinitive in (3) an Objective Clause and the infinitives in (1-2)
Purpose Clauses; Jones calls the infinitives in (3) Subject-gap Purpose Clauses and the
infinitives in (1-2) Object-gap Purpose Clauses. Given that principles of syntactic
distribution and semantic interpretation which apply to the infinitives in (1-2) generally
also apply to the infinitive in (3), I will follow Jones in labelling both infinitives
Purpose Clauses and use the qualification "subject-gap" versus "object-gap" to
distinguish them where necessary. The Purpose Clause can be identified in three ways:
(i) it is always controlled by the direct object or Theme of the main verb, so in (1) the
object gap is controlled by the Theme, ice-cream, in (2) it is controlled by the Theme,
freezer, and in (3) it is controlled by the Theme, Ute; (ii) a gap in the clause is
obligatory (see Section 4.1.2.1); and (iii) the infinitive cannot be prefaced by the phrase
in order, as shown in (4-6) .

[4] *Sven handed Ute the ice-cream in order e to put e in the freezer.
[5] *Sven opened the freezer in order for Ute to put the ice-cream in e.
[6] *Sven sent Ute into the kitchen in order e to put the ice-cream in the freezer.

(6) is only well-formed in the rather odd interpretation where Sven sends Ute into the
kitchen so that he, Sven, can put the icecream in the freezer. This is in fact a Rationale

Clause reading.

Thanks to Joan Maling for pointing out the robustness of these judgements with the strictly
subcategorising verb, put.
 Examples of Rationale Clauses are given under (7-8).

[7] David spoke to Damian in Welsh (in order) e to keep the gossip secret.
[8] David spoke Welsh (in order) for Damian to feel more at home.

Given the possibility of inserting the phrase in order before this infinitive, Jones calls
these infinitives In Order Clauses. There is evidence, however, that the phrase in order
is not a simple marker of the Rationale Clause (see Section 5.4 for detailed discussion).
I have therefore decided to adopt Farad's original term, referring as it does to a stable
notional property of the infinitive which is uncontroversial. I will continue to assume,
however, that the preface in order is associated strongly with the Rationale Clause,
modulo the reservations expressed in Section 5.4.
Rationale Clauses may be distinguished in four ways: (i) they need not contain
any gap, as in (8); (ii) where there is a gap it must be in subject position, as in (7); and
(iii) the gap must be controlled by the Agent (in (7), it is David who intends to keep the
gossip secret, not Damian); (iv) they can generally be prefaced by the phrase in order,
as in (7) and (8), modulo the provisos mentioned in Section 5.4. Notice that it really is
the Agent rather than merely the subject of the sentence which controls the gap; whether
the Agent is the subject, as in (11), a by-NP adjunct, as in (12), or an implicit
argument, as in (13), it is always the controller.

[11] The terrorists stole the plutonium (in order) e to intimidate Western
governments.
[12] The plutonium was stolen by the terrorists (in order) e to intimidate Western
governments.
[ 13] The plutonium was stolen (in order) to intimidate Western governments.

Examples of Telic Clauses are given under (14-16).

[14] John entered the house, e to discover that everyone had already left.
[ 15] The leaves were blown by the wind from the trees, e to fall in piles at the end of
the garden.
[ 16] John bought the house, only for his partner to discover that it had rising damp.
The Telic Clause is similar to the Rationale Clause in containing only an optional subject
gap. It may be distinguished, however, in three ways: (i) the infinitive is factive (the
infinitival event actually occurs); (ii) the gap is controlled by the syntactic subject
(regardless of active or passive voice): notice in (14) that it is the Agent John (the
syntactic subject) who acts as the subject of discover, where in (15) it is the leaves
(which are syntactic subject), not the Agentive wind (now merely an adjunct), which
acts as the subject of fall', (iii) the Telic Clause may be prefaced by only where the
infinitival event is an outcome which is somehow unexpected.
In this analysis, I will assume that each construction has the properties it does
because it has a head which conditions its form and intepretation. The head has an
argument structure, just like any other predicative head, and the interpretation of the
modifier with respect to the modified sentence is a reflex of the argument structure of
this predicative head, as is the interpretation of the matrix verb (and its arguments) with
respect to the infinitival verb (and its arguments). I will adopt the view of argument
structure developed by Higginbotham (1985; 1989). Following Stowell (1981),
Higginbotham assumes that lexical heads come with a grid specifying the number of
their arguments. The grid is a piece of syntactic information associated with a label in
the syntax and provides the starting point for the translation rules to the semantics. A
verb like hit might therefore have a syntactic argument grid, as in (17), interpreted as in
(18).

[17] hit <1,2>

[18] hit(x,y)

This tells us that hit must occur with two phrases in the syntax and that it must be
interpreted with respect to two individuals in the semantics. I assume that amongst the
information associated with these grids is some specification of the semantic types over
which the variable positions range. So, as suggested, the verb hit is associated with
information which tells us that both its arguments are drawn from the domain of
individuals. A verb such as believe on the other hand will specify that one of its
7
 arguments is an individual where the other is a proposition (with respect to which the
belief is held).

[19] believe <1,2>

[20] believe(x,Ap)

In the tradition of Davidson (1966), I will assume that all verbs have an argument
position ranging over events2 , where events are taken to be ontological primitives of the
semantics.
Given this approach, the question is what information is associated with the
argument grid of each infinitive of result. Say that the head of the Purpose Clause is the
predicate P, that the head of the Rationale Clause is the predicate /?, and that the head of
the Telic Clause is the predicate T, then each of these predicates will have a syntactic
argument grid. Let us consider the content of these argument grids.
It seems clear that the Purpose Clause mediates the relation between a verb and
its object and that it concerns the function that the object of the verb is to perform.

[21] John designed the battery.

[22] John designed the battery to operate at high temperatures.

Consider (21). To design a battery is to assign to some object a particular internal

structure such that it can serve a particular function, in this case the function of a
battery. In (22), the object is still being assigned an internal structure; and that internal
structure still allows it to perform the function of a battery. Now, however, our
understanding of the structure and the function which it is intended to perform are more
constrained. The battery is being attributed a structure such that it will be able to operate
at high temperatures. The action in (21) is one of designing; the action in (22) is of a
particular type of designing.

2
Higginbotham in fact assumes that all lexical items have an event position available.
8
 We might therefore suggest that the Purpose Clause is headed by a predicate
which predicates a relation between an entity, an event, and a property, as in (23).

[23] P(y, e, AXz F(z))

This says that there is a relation between an entity, an event, and a property, such that
the entity is involved in the event for the sake of its having the property. If we take the
sentence in (22), it will then have a logical translation as in (24), which can be
paraphrased as in (25).

[24] (for x=a battery) design (John, x) & P(x, e, A^z (Be1) operate (z, e'))
[25] "the eventuality of the battery being designed by John is for the sake of its
having the property of operating at high temperatures"

Notice that this predicate is in fact teleological in a quite general sense. It can apply to
inherent natural functions of objects, as in (26), to designed or constructed functions of
objects, as in (27), as well as to much more general cases, as in (28).
[26) (a) Grass is green to photosynthesise more efficiently.
(b) (for x=grass) GREEN (x, e) & P(x, e, AXz (Be1) photosynthesise (z,
e'))
(c) "the eventuality of the grass' being green is for the sake of its having the
property of photosynthesising more efficiently"

[27) (a) This car is contoured to accelerate more quickly.

(b) (for x=car) CONTOURED (x, e) & P(x, e, AA,z (Be1) accelerate (z, e1))
(c) "the eventuality of the car's being contoured is for the sake of its having
the property of accelerating more quickly"

[28) (a) John bought some cream cakes at the shop to eat with his grandmother.
(b) (for y=cream cakes) buy (John, y, e) & P(y, e, AXz (Be1) eat (z, e'))
(c) "the eventuality of the buying of the cakes is for the sake of their having
the property that John eats them with his grandmother"

The point of these examples is that an event can qualify for modification by to if the
object can be seen as having a particular property by virtue of its relation to the event. In
cases where objects have natural or designed properties, this follows straightforwardly;
9
 but an event can also qualify if it is interpreted in the scope of an Agent's intention.
Cream cakes may not be for eating with one's grandmother in any sense, but if we
attribute the relevant beliefs and intentions to the Agent, then relative to those beliefs
and intentions that is exactly what the cream cakes are for. Intended attribution of a
property therefore becomes a special subcase of the Purpose Clause reading. And note
that that does leave certain types of event out of the domain of the Purpose Clause, as
illustrated by the contrast in (29-30).

[29) The engineers created a channel to drain the lake water away with.
[30) *The earthquake created a channel to drain the lake water away with.

The earthquake's creation of a channel is neither an eventuality which naturally

attributes a property to the object, nor is it saved by any intentional reading. The crucial
point in each case is that the interpretation is secured (or restricted) on the basis of the
meaning of the predicate P itself: we know that the it is the object that has the property,
that it has the property by virtue of its relation to the event, and that its so having the
property is what the event is for, by virtue of the fact that that is what the predicate P
actually means (a meaning which will be spelt out in more detail in its conditions on
reference).
Unlike the Purpose Clause, the Rationale Clause does not seem to be involved
with the nature of the Theme of the construction, nor does it deal with the attribution of
properties. The Rationale Clause concerns the intention of the Agent of the main event.
We might therefore see the Rationale Clause as headed by a predicate which relates an
entity, an event, and a proposition, as in (31).

[31] R(x, e, Ap)

This predicate is a relation between an entity, an event, and a proposition, such that the
entity acts in the event with the intention that p (i.e. that the proposition be true).
Finally, there is the Telic Clause, as in (32).

10
 [32] John designed a battery for the competition, only to discover that he was
ineligible.

This construction does not involve Themes and their properties or Agents and their
intentions. It concerns outcomes, expected and unexpected, of main clause events. In
fact, it involves a strong sense of inevitable consequence. In the light of examples such
as (33) it is tempting to call these infinitives, "clauses of destiny".

[33] Christ entered Jerusalem in triumph, only to die in shame five days later.

The predicate which heads the Telic Clause therefore seems to be genuinely and purely
telic. I will treat it as a relation between an ordered pair of events, as in (34).

[34] T(<e!,e2>)

This predicate is a relation between an ordered pair of events, such that the first event
finds its conclusion (expected or unexpected) in the second.
Consider now the sort of translation which will be produced in this analysis.
The root predicate in (21) is design which is a relation between two entities and an event
such that a designs b in e. Consider a Purpose Clause modification as in (35).

[35]
design (a, b, e) & P (V, e1, AXz (Be") operate (z, e"))

The purpose predicate, P, forms a property argument from the infinitival verb which it
then directly theta-marks (in a manner to be discussed, see Section 4.1.2.1); it then
theta-identifies its event and entity arguments with the event and second entity argument
of the verb. This will give a translation as in (36) after existential closure.

1 1
[36] (Be) (3b: battery (b)) design (John, b, e) & P(b, e, AXz (3e") operate (z, e"))

(36) might be paraphrased as: there is an event of designing, e, of an entity, b, by an

entity, a; and that design event is for the sake of the entity, b, having the property of
operating at high temperatures.
Now consider a Rationale Clause modification as in (37).

[37]
design (a, a battery, e) & R(x,'e', A(3e") win (PRO, a prize, e"))

The rationale predicate, R, forms a proposition from the infinitival verb which it then
directly theta-marks; it then theta-identifies its event and entity arguments with the event
and first entity argument of the verb. This will give a translation as in (37) after
existential closure.

[38] (3e) (3a: John (a)) design (a, a battery, e) & R(a, e, A(3e") win (PRO, a prize,
e"))

(38) might be paraphrased as: there is an event of designing, e, of a battery by an

entity, a; and the entity, a, intends by that act of designing to make it the case that PRO
wins a prize. The value of PRO will be derived under a syntactic principle of control
(see Section 4.1.1).
Finally, consider a case of Telic Clause modification as in (39).

[39]
design (John, a battery, e) & T(<el, e2>) & discover (x,Ap, e1)

The telic predicate, T, simply identifies its second event argument with the event
argument of the second verb; it then identifies its first argument with the event argument
of the first verb. Existential closure will then apply to both event variables
12
(straightforwardly accounting for the factivity of the constructions), giving the
translation in (40).

[40] (3e) (3ef) design (John, a battery, e) & discover (PRO,Ap, e1) & T(<e, e'>)

(40) might be paraphrased as: there is an event of designing, e, by John of a battery;

and there is an event of discovering, e1 , by PRO that p; and e' is inevitable outcome of
e.
The analysis of infinitives of result set out above is in effect an attempt to
integrate the syntax and semantics of modification into theta theory. This raises a
number of important questions concerning the nature of modification and the content of
the notion "theta role", to which I now turn.

13
Section 1.3: The semantics of modification

The problem posed by these constructions is how we are to integrate them into
the sentences which they modify and, fundamentally, what precisely it means to
"modify" a sentence. Our intuitions on the subject are vague but highly suggestive: a
modifier is an optional piece of information added to the basic sentence; it tells us more
about what is going on by supplementing information already provided by the verb,
usually without changing the verb, and so on. Different analyses appeal to different
aspects of these intuitions, but in formal terms views on modification have radical
implications for the way the grammar is set up. I will assume that there are three basic
ways of approaching the problem: the natural augmentation analysis; the higher-order
function analysis; and the predicate of events analysis.
The natural augmentation analysis (McConnell-Ginet (1982); see also Chierchia
(1989)) characterises certain types of modifier as extra arguments of the verb: certain
verbs have the option of expanding their argument structures to take extra arguments. It
is argued that in the manner reading of the adverb rudely, in (1), the verb itself theta-
marks the adverb and gives it its interpretation.

[1] Sven left rudely.

Leave is a verb which ir. this case relates an individual to a manner on condition that the
individual performs the leaving in that manner. This "transitive" leave is derived from
intransitive leave by a lexical rule; the rule is intended to be constrained by an appeal to
the "natural" meaning of a predicate, i.e. leaving is something which you can naturally
do in a particular way and therefore the lexical rule will allow you to derive from a verb
leave another verb which means leave-in-a-particular-way. This is therefore a sort of
natural polysemy analysis. The analysis is in fact limited to a certain class of
"modifiers" which are intimately associated with verb. As a general theory of
modification it is unattractive because the number of "natural" augmentations to

14
argument structure is potentially infinite: we can naturally specify all sorts of
information on places, times, causes, consequences etc but we do not wish to have a
100-place predicate or lexical rules offering unbounded lexical polysemy. The
arguments are most convincing for those adjuncts which are unique and change the
meaning of the predicate in some way.
This approach seems to get the selectional properties of modifiers, including
infinitives of result, the wrong way around; in the case of infinitives of result it would
also require two theories of modification: one for the Purpose Clause which is
syntactically closely associated with the verb and another for the Rationale Clause and
Telic Clause. Consider (2-4).

[2] Sven brought Sigrun along to the party to introduce to his friends.
[3] Sven climbed the glacier to impress Sophia.
[4] Sven left Oxford forever, only to be sent back on an assignment by his
newspaper.

In (2) we are required to understand that Sigrun is brought along to the party for some
purpose which she is to serve: she is to be introduced to Sven's friends. In (3) we are
required to understand that Sven climbed the glacier with a specific purpose in mind,
namely impressing Sophia. In (4) we are required to understand that Sven's leaving of
Oxford had the unexpected outcome that he in fact returned. Yet these aspects of the
interpretation of (2-4) have nothing to do with the verbs themselves. There is nothing in
the interpretation of bring that entails that the bringing have a purpose. We have
perfectly acceptable examples of the verb bring in sentences which clearly imply no
intention of any sort.

[5] Sven has brought his cold along with him.

However, when a Purpose Clause is used to modify the sentence we must have an
intentional reading.

[6] Sven has brought his cold along with him to give to everyone.
15
In (6) we must take the bringing to be a malicious act intended to harm others. The
reading is strictly tied to the presence of the infinitive itself. If we take the Purpose
Clause to form part of the argument structure of the verb then we must take bring to be
ambiguous between a verb bringl which means "to bring with a purpose" and a verb
bring? which means "to bring accidentally". This sort of ambiguity analysis does not
seem to account naturally for the fact that the relevant reading is obligatory when the
infinitive is present and is optional otherwise. A similar point can be made for the other
infinitives of result.
There is nothing in the interpretation of climb that entails that climb be
undertaken with a particular intention in mind. Even if Sven undertakes the climb
volitionally he may do so for no particular reason: the infinitive itself requires that the
verb be understood as having an intentional Agent. Most clearly of all there is nothing
in the interpretation of leave that entails the unexpected outcome of a return. In the
case of the Telic Clause it is quite clear that the infinitive gives us some additional piece
of information concerning the outcome of the event which is entirely independent of the
content of the modified event itself. It is interesting to note that Jones (1991) amongst
others supports McConnell-Ginet's view of a natural augmentation of meaning on the
basis of distinctions between the Purpose Clause and Rationale Clause with respect to
phenomena such as iterability and preposing. However, the Telic Clause patterns with
the Purpose Clause and not the Rationale Clause in these respects, yet it is clearly not an
argument of the verb, being neither VP-internal in the syntax nor in any non-vacuous
way based on an "extention" of what the modified verb means.
As the discussion has implied, if we are thinking of selectional restrictions it
seems that the important relation is in the other direction: if something is done with a
purpose then the matrix event must be the sort of event that will attribute a purpose to
the object in question; if something is done with an intention it must be the sort of thing
that can be done with an intention in mind; and if an event has an unexpected conclusion
it must the sort of event where such a reversal is possible. If the infinitives are
16
unselected by the verbs they modify and possibly even select properties of those verbs
in some sense then the interpretation of sentences containing infinitives must be reliant
in some way on the features of the infinitives themselves. Notice that in other types of
modification this is intuitively clear.

[7] Sven bought an apple in the shop.

It is clear that the relation between the buying of the apple and the shop is established in
some way by the preposition in independently of what it means to buy an apple and
independently of what it means to be a shop. However, the infinitives of result are often
discussed as if the problem of interpretation is related exclusively to the problem of
linking an argument position of the infinitival verb with an argument position of the
matrix verb.

[8] Sven bought an apple e to give e to Sigrun.

Clearly an account of the infinitive must establish that the object gap in (8) is
coreferential with the object, an apple, and that the subject gap is coreferential with the
subject, Sven. Yet the problem of interpretation for these infinitives remains: the
infinitive in and of itself introduces a particular type of interpretation, as discussed
above. The question is how the infinitive introduces this interpretation itself. Both the
higher-order function analysis and the predicate of events analysis provide an answer to
this question.
Consider first the higher-order function analysis. In this view the modifier is
interpreted as a function from one interpretation (the unmodified) to another
interpretation (the modified), where the function is linked to a particular lexeme or
construction. This analysis flows naturally from assumptions concerning argument
structure and projection in the tradition of Montague (1974). The Montague tradition is
founded on the assumption that there is a strict isomorphism between syntactic and
semantic structure (every syntactic category maps onto a semantic type and every
17
syntactic rule maps onto a semantic rule) and that argument discharge proceeds solely
by function-argument application. As in traditional logic, the theory assumes only two
primitives in the semantics: terms and truth values; all other expressions are functions
from the one to the other. For instance, the verb love is a two-place predicate, as in (9);
say that the proper names Eloise and Abelard are terms, then the predicate might apply
to these terms to yield the proposition in (10), which is true if Eloise loves Abelard and
false otherwise.

[9] love(x,y)
[10] love(Eloise Abelard)

Given the isomorphy of syntax and semantics, the syntax must be set up to map directly
to the semantics. There is however the well-established fact that consituents in the
syntax are hierarchically organised, with the subject dominating all other arguments, as
in (11).

11] _____

Following Schoenfinkel (1924) and Curry (1963), the n-place functions are then treated
as a series of one-place functions from an n-place function to an n-1 place function. A
verb is then a function from terms to truth values (NPs to Ss) which takes one term at a
time until a truth value is yielded. In this view, there must then be a strict divide
between arguments (the terms to which the function applies) and true modifiers because
the modifiers do not and cannot affect the argument structure of the verbs they modify.
Say a phrase modifies an intransitive verb which is interpreted as a one-place function
from terms to truth values: that verb clearly does not apply to the modifier as an
argument because the subject is interpreted as the subject of the verb; the modifier must
18
therefore introduce its own interpretation without actually altering the value of the
function it modifies. It is therefore a function from a one-place predicate to a one-place
predicate: a higher-order function.
Consider a standard case of modification as in (12) and (13).

[12] Sven walked.

[13] Sven walked quickly.

Assume walk to be a 1-place predicate true of an x such that x walks. Quickly in this
view will be a higher order function from 1-place predicates to 1-place predicates; the
new 1-place predicate, walk-quickly, will be true of an x iff x walks-quickly. The
crucial thing to notice about this analysis is that walk-quickly is a new predicate whose
interpretation is secured by semantic postulate. Strictly there is no reason for there to be
a predictable relation between the input to the function, walk, and the output, walk-
quickly, because the meaning is entirely dependent on the character of the function
itself. We cannot prove for example that if Sven walked-quickly then Sven walked,
because we cannot be certain of what the relation between walking and walking-quickly
is. This relation between the input and output of the function must be stated
independently 3 .
Notice that there are cases where we might wish this failure of entailment to
hold, namely in cases of metaphorical extension, where the meanings are related but the
trivial entailment described above does not hold. Consider for instance the sentences in
(14-15).

[14] Sam pushed Emily to join the club.

[15] Sam pushed Emily.

The meanings of push in (14) and (15) are clearly related, but the use of the infinitive in
(14) forces the abstract metaphorical reading of push, meaning urge. (14) certainly does

3I take these observations from lectures delivered by James Higginbotham at the University of Oxford during
1994-5.
19
 not entail (15). It is just this sort of unpredicable relation which is predicted by the
function analysis of modifiers but which is clearly inappropriate for modifiers such as
quickly where the adverb's contribution is clearly isolable and the relevant entailment
holds trivially. Notice also that it is not entirely clear that we would really want this
function analysis even for the contrast in (14-15) because this is probably best analysed
as a case of genuine lexical polysemy, resulting from metaphorical extension, i.e. that
there is a verb push! which is true of x and y iff x shoves y; and there is verb pustr
which is true of x, y, and z iff x urges y to do z. Notice that a lexical ambiguity analysis
also predicts that the interpretation of the infinitive will be lexically idiosyncratic.
It is this problem concerning the relation between the interpretation of the
modified and the unmodified constituent that is addressed in the predicate of events
analysis. Events are taken to be primitives of the model, in fact individuals of an
abstract kind. The verb establishes a relation between individuals and an event;
modifiers then add additional pieces of information concerning that event. In the famous
example of Davidson (1966) under (16) it is therefore assumed that we have three
conjoined predicates, as in (17).

[16] John buttered toast in the kitchen at midnight.

[17] butter(John,toast,e) & in(e,the kitchen) & at(e,midnight)

(17) tells us that there is an event of John buttering toast AND that it (the event of
buttering) was in the kitchen AND that it was at midnight. Notice that this trivially
entails the sentences in (18-20a) whose logical forms are given in (18-20b).

[ 18] (a) John buttered toast in the kitchen.

(b) butter(John,toast,e) & in(e,the kitchen)

[19] (a) John buttered toast at midnight.

(b) butter(John,toast,e) & at(e,midnight)

[20] (a) John buttered toast,

(b) butter(John,toast,e)

20
Modifiers are then taken as adding to the meaning of the sentence by simple
conjunction, where the event is taken to fall in the intersection of the sets denoted by
each predicate . The question is then how these logical forms are derived from the
syntax in a compositional theory. Clearly, function-argument application will be
inadequate for providing these translations, if for no other reason than that they are
recursive.
Higginbotham (1985; 1989) argues that natural language does in fact have more
than one principle of argument discharge and that one of the tasks of linguistic
semantics is to define a small and explicit set of principles of argument discharge which
will apply generally within the grammar. He argues that there are in fact four such
principles (see Section 2.1.2). The one which will interest us here is theta identification:
this allows the values assigned to the argument positions of two syntactic sisters to be
identified, with the argument position of the head filtering up to the mother node (and
the non-head position being discharged). On this account the patterns shown in (16-20)
follow trivially.

21]
VP<E>

John buttered the toast in the kitchen at midnigh

*^
As Shalom Lappin has pointed out to me (personal communication), a simple intersective account fails for a
certain class of adverbs, namely the "quantificational adverbs" discussed in Lewis (1975).

(a) John rarely climbed the mountain to impress Mary.

(b) John rarely climbed the mountain.

In an intersective account, (a) entails (b), but of course it is not the case that if John rarely climbs the
mountain with a particular intention, then he rarely climbs the mountain in general. Within the Davidsonian
tradition (cf Higginbotham 1989) it is assumed that such adverbs introduce implicit quantification over
events relative to a background, as in (c).

(c) [Rarely e:B(e)] F(e)

As Higginbotham has pointed out to me (personal communication), in this view such adverbs do not pose a
problem for extensionality, and indeed the quantification is a purely extensional context relative to the
background B for e.
21
 By theta identification the value of e is assigned to e'\ syntactically the position <F> is
discharged but the head position, <E>, is not and filters up to the mother. Similarly the
value of e is assigned to e", discharging the syntactic position, <E">; <E> filters up.
The translation provided will then be as in (22).

[22] (Be) butter(John,the toast,e) & in(the kitchen,e) & at(midnight,e)

The modifiers are optional conjuncts and the entailments follow straightforwardly.
These observations follow straightforwardly for the infinitives of result.
Consider (23).

[23] John designed a battery [pc to operate at high temperatures] [RatC to

prize] [TC only to discover that he was not eh'gible for the competition]

It is a trivial fact about (23) that it entails the sentences in (24-29).

[24] John designed a battery to operate at high temperatures to win a prize.

[25] John designed a battery to operate at high temperatures.
[26] John designed a battery.
[27] John designed a battery to win a prize.
[28] John designed a battery only to discover that he was not eligible for the
competition.
[29] John designed a battery to win a prize only to discover that he was not eligible
for the competition.

Given the basic analysis in Section 1.2 of the infinitives of result, the translation of (23)
would be as in (30).

[30] (Be 1 ) (Be7) design (John, a battery, e 1 ) & to 1 (a battery, e 1 , AXa (Be3) operate
(a,e3)) & to2 (John, e 1 , A(3e5) win (PRO, a prize, e5)) & to3 ^e^e7^ &
realise (PRO, Ap, e7)

It is a trivial fact about the representation in (30) that it entails each of the conjuncts
which it contains, i.e. (30) entails (31-34).
22
 [31] design (John, a battery,
[32] P(a battery, e 1 , AXa (3e3) operate (a,e3))
[33] R(John, e 1 , A(3e5) win (PRO, a prize, e5))
[34] T(<e! ,e7>) & realise (PRO, Ap, e7)

Notice also that (30) will entail any proposition identical to itself but with one or more
of the conjuncts dropped, i.e. it implies that the interpretation of each infinitive of result
is not dependent on the presence of one of the others: notice that it is intuitively clear
that (23) entails (27-29) and so this also is correct. The event analysis thus applies
naturally and straightforwardly to modification by infinitives of result.
It should be noted, however, that the event analysis is not an addition to the
Montague view, simplifying its account of modification: the Davidsonian approach
attempts a simplification and restriction of the semantic apparatus at all levels. In the
Montague tradition we find the elaboration of higher-order functions and the
introduction of type-shifting functions to guarantee compositionality. The Davidsonian
tradition attempts to reduce the semantic apparatus to (first-order) predicate-argument
application, ordinary quantification, and truth functions, the core apparatus already
required in the Montague approach. The Davidsonian approach, as a complete theory of
natural language semantics, is therefore more restrictive than its Montagovian rival.
This restrictiveness must of course be paid for in ontology: it is only by assuming
events as primitives of the model that the other reductions are possible. Much work has
therefore been spent on showing the naturalness and necessity of assuming events as
primitives: I refer the reader to Davidson (1966), Parsons (1990), Higginbotham
(1983a; 1985; 1989); and Schein (1992) for detailed discussion on this issue. This
thesis is intended as a contribution to the investigation of the role of events in natural
language semantics.
I therefore assume that the Davidsonian analysis of modifiers in terms of
predicates of events is the correct one in an analysis of infinitives of result. The
infinitives of result are each headed by a predicate whose argument grid contains an

23
event position which will be identified with the event position of the matrix verb. As
will be seen in the sections which follow, I exploit this notion of argument structure as
a series of variable place-holders ranging over particular semantic types to derive
important properties (both syntactic and semantic) of the constructions. However, such
a view of argument grids is not compatible with all views of theta theory which have
been developed. I now turn therefore to the question of the nature of argument positions
and their relation to thematic roles.

24
Section 1.4; Argument positions and the theta-role

I am therefore adopting the standard assumption within the Principles and

Parameters tradition (following Stowell (1981)) that lexical heads are listed in the
lexicon with an argument grid which specifies an addicity for the head and whatever
additional information must be stipulated with regard to these positions. I depart from
the general convention in assuming that modifiers also have heads (sometimes lexical
heads) and that these heads also have argument grids; it is by virtue of these argument
grids that the modifier phrase can be integrated into the interpretation of the sentence in
the relevant manner. Taking this step, however, of generalising the theory of argument
grids, requires a particular view of these grids and the information that they contain. In
particular, it requires certain assumptions concerning the nature of theta-roles.
Assessing the way that these assumptions relate to others in the literature is complex
given that the language of thematic relations is generally adopted but without any
commitment to the assumptions of those who are viewed as its principle exponents,
namely Gruber (1965) and Jackendoff (1972; 1983; 1990; 1992) (and, orthogonally but
in the same spirit, Fillmore (1966) in his work on "deep case").
In the view I am adopting here, which follows Dowty (1989), the meaning of a
word may be characterised as the set of lexically specified entailments: those things
which must be true whenever a proposition yielded by the predicate is itself true, where
these entailments are accessed immediately and without difficulty by the speaker of the
language as a part of his knowledge of the meaning of the word in question. Consider
the predicate, love.

[1] love <1,2>

[2] love(x,y)

The argument grid of love contains two positions and it translates as a relation between
two individuals. Love is a two-place predicate which is true of two individuals, x and
y, if x loves y, and false otherwise; we know as a direct part of our knowledge about
25
the predicate love that the first individual must be a sentient being, that that being must
experience an emotional reaction with regard to some stimulus, and that that emotional
reaction must be of a particular kind, i.e. love; we also know that the second object
must be related to the first in such a way as to be the cause of the emotional experience
in that individual (although that it need not stimulate that reaction intentionally or
perform an action which does stimulate the reaction: it is simply a "trigger" of some
sort). These entailments might be represented as conditions on the reference of love, of
the sort given informally in (3).

[3] love(x,y) iff x is sentient & x experiences an emotional reaction & the
emotion is love & y stimulates the emotion in x etc.

' i, *

Every lexical item will then be provided with a syntactic argument grid, as in
(1), a translation of that grid, as in (2), and conditions on its reference, as in (3). The
problem with this view is that there will be no fixed way of relating predicates which
seem to form lexical classes with important reflexes in syntactic behaviour. For instance
the class of experiencer verbs are taken to exhibit peculiar syntactic behaviour cross-
linguistically, but in the view offered so far there is no way of making relations between
verbs in a particular class necessary facts about the form or interpretation of the
lexemes. The problem is intuitively apparent if you consider the meaning of hate. It too
is a relation between two individuals, x and y, although it is true only if x hates y. If we
consider the nature of the entailments of hate, however, we find that they are extremely
similar to those of love. Hate also requires that x be sentient, that x experience an
emotional reaction, that y stimulate that emotional reaction in x; it differs only in the
entailments which concern the nature of the reaction itself: namely, that the emotion
must be that of hatred not of love. The similarity between the two predicates follows
only if we offer conditions on the reference of hate which take the form of a lexical
decomposition, as for love in (3).

[4] hate(x,y) iff x is sentient & x experiences an emotional reaction & the
emotion is hatred & y stimulates the emotion in x etc.
26
We might then take it that particular types of syntactic behaviour correlate with
particular sets of lexical entailments. As Dowty (1989) observes, not any old set of
lexical entailments is relevant: the verbs jog and walk both entail that the entity of which
they are true must have legs, i.e. the intersect of the entailments lexically specified by
jog and those lexically specified by walk will include the entailment that x have legs.
However, this is not the sort of entailment set which is generally considered to be
syntactically significant. Further, the types of entailments which are considered to be
syntactically significant seem to come in clusters. So the relevant observation with
regard to (3) and (4) is that both love and hate entail that x is sentient AND that x
experiences an emotional reaction AND that y stimulates the emotion in x, etc: this is a
significant set of entailments, what Dowty (1989) calls L-Thematic Role Types. We
might therefore argue that thematic roles are relations which themselves have these
clusters of entailments as conditions on their reference.

[5] Experiencer(x,y) iff x is sentient & x experiences an emotional reaction & y

stimulates the emotion in x etc.

Both love and hate will then select for this relation as a condition on their reference,
along with other lexically idiosyncratic pieces of information.

[6] love(x,y) iff Exp(x,y) & the emotion is love & ...
[7J hate(x,y) iff Exp(x,y) & the emotion is hatred & ...

In a Davidsonian analysis such decompositions become particularly attractive as the

theta-role can be seen as specifying a relation between some entity and the event in
which it participates of a particular type.

[8] love(x,y) iff love(e) & Exp(x,e) & Stim(y,e) & ...
"love is true of x and y iff the event is an event of loving and x is involved in the
event as an Experiencer and y is involved in the event as a Stimulus and..."

27
 One important point to note concerns implicit arguments.

[9] Sven ate some Kentucky Fried Chicken.

[10] Sven ate.

In (10) there is an existential presupposition to the effect that there is some object which
bears the same relation to the event of eating as some Kentucky Fried Chicken does in
(9): i.e. Sven eats something. It would be possible to assume that this gives evidence
that positions in an argument grid need not be overtly discharged but may (under
certain, presumably lexical, conditions) remain implicit. Thus in a theory using thematic
labels, the distinction between (9) and (10) would be as in (11-12) (I use the notation of
Williams (1994)).

[11] Sven1 ate some Kentucky Fried ChickenJ.

(Agi.ThJ)

[12] Sveni ate.

(Agi,Th)

I will however assume that where a position is present in an argument grid it must be
discharged in the standard manner. This will mean that the intransitive use of eat will
have to be derived by a lexical rule which reduces the number of argument positions of
the verb by one. This view gains some support from the fact that the interpretation of
intransitive eat with respect to its implicit argument is more constrained that the
interpretation of transitive eat with respect to its internal argument: I can assert that
someone eats almost anything (he ate sawdust, he ate fire etc); but the intransitive use
must mean that someone has eaten something edible and in fact generally means that he
has had a meal of some sort (eaten a portion of food of an appropriate type given the
context). The problem is then of relating the meanings of transitive and intransitive eat.
This however is straightforward in a lexical decomposition: we simply say that the
lexical rule does not remove the thematic relation in the translation but supplements it
with some further lexically specific relation, C.
28
[13] eat(x,y,e) <-» eat(e) & Ag(x,e) & Th(y,e) & ...
[14] eat(x,e) <-» eat(e) & Ag(x,e) & Th(y,e) & C(y,e) & ...

I make this assumption because it appears that the Purpose Clause cannot take
an implicit argument as an antecedent, as pointed out in Jones (1991: 38).

[15] (a) We've been hiring guards.

(b) We've been hiring guards to look after the children.

[16] (a) We've been hiring all week.

(b) *We've been hiring all week to look after the children.

I take it (in line with the general position) that the Purpose Clause involves predication
which will require some form of coindexing or identification of an argument position in
the Purpose Clause with the object argument of the main verb (see Section 4.1.2.2 for
detailed discussion). If implicit arguments are syntactically available then one would
expect (16b) to be well-formed (as far as I can see this is a problem for Jones' theory
although he does not mention it). By assuming that implicit arguments are present only
in the lexically specified conditions on reference and are not present in the argument
grid, the restriction (16b) follows automatically: there will be no appropriate position
for the argument of the Purpose Clause to theta identify with.
Thematic relations have a further significance for an analysis of infinitives of
i

result in that they are seen as providing important constraints on the reference of the
predicate itself. Chierchia (1989) suggests that the antecedent of the Purpose Clause
must be a Theme. In this view, such a restriction would be treated as a condition on the
reference of the predicate: the Purpose Clause is a relation between an entity, an event,
and a property such that the entity's participation in the event in a Theme-like way is for
the sake of its having the property.

[17J P(x,e,A^z F(z)) iff Theme(x,e) & ...

29
In fact I will argue in Chapter 5 that thematic relations are inadequate as a
characterisation of the linguistically significant constraints on the interpretation of the
infinitives, even for the Purpose Clause. Constraints of a rather different kind will be
required, which involve relations between events, objects and properties, and
individuals and intentions. The view taken here is however naturally open to these types
of restrictions as thematic relations are not necessarily privileged as conditions on
reference.
This view is different in crucial ways from analyses which presuppose that the
theta-role is in some sense a "special" entity with its own characteristics and content.
Assume for instance that love had a syntactic argument grid as in (18).

[ 18] love <EXPERffiNCER, STIMULUS>

Here the verb's argument structure is represented as a list of roles: prinicples of

argument projection must therefore reduce to the assignment of these roles to the
relevant phrases; and the notion of (natural language) predicate must reduce to an
assigner of such roles. This is rather different to the view given above that the positions
in an argument grid were simply minimally differentiated positions, interpreted as
variables for the arguments of logical predicates. The positions in the syntactic
argument grid now have a privileged status: they are roles or syntactic atoms of a
paricular type which must be assigned and organised by some set of syntactic
principles. The importance of this point can be seen when we turn to infinitives of result
(and modifiers in general): it is clear that the Telic Clause does not "assign theta-roles"
in the traditional sense.

[19] The side-door on the plane was opened [only for Marcus to realise that his
parachute wasn't fastened properly].

30
The Telic Clause here bears some relation to the sentence, which it was argued above
should be seen as a type of argument relation, namely that the Telic Clause is headed by
a predicate true of events and predicates itself of the matrix event (open(e)) by the
principle of theta-identification. This is an argument relation; it is not however a
thematic relation of any sort. If the argument positions in a syntactic argument grid are
limited to thematic labels then we are forced away from seeing modification in terms of
event predication and hence miss the generalisations concerning systematic entailments
and optionality discussed in Section 1.3.
Notice that this follows even in a "bare theta theory" such as the one we find in
Williams (1994; 1995). As he observes, his theory does not require any particular
definition of the content of theta-roles, merely a list of minimally differentiated roles, as
in (20).

[20] love <9 1 ,02>

However, the argument positions are still "theta-roles" understood as syntactic atoms of
a privileged type. It is possible that this is a notational mirage and that the representation
in (20) is intended to be equivalent to the notation in (1). However, this appears not to
be the case. Williams (1994: 45) in fact makes precisely the distinction between
standard argument projection and modification which necessarily follows if theta-roles
are seen as syntactically real, privileged entities.

The relation of the item [the modifier] to the modifiee is structurally

identical to the theta relationithat is, absolute locality is required... For
these reasons, we will want to subsume this relation under theta theory.
On the other hand, we also want to distinguish this relation from
theta relations, for the reason given above: this relation is "transparent" to
theta relations in that for a given functor/, the object, shown in {42},
(42) [fX]

has exactly the same potentiality for theta relations that X had. For
example, slowly arrest Bill has one external argument, just as arrest Bill
does.
Clearly, the functor relation is a semantic relation, akin to the theta
relation. However, rather than filling theta roles, a functor maps theta
31
 assignors into theta assignors. And, as in the case of theta role assignment,
the relations are both asymmetric: one item is the functional element, and
the other is its argument.

This is in fact a statement of the higher-order function analysis of modification which

was argued to be inadequate in Section 1.3. The verb has an argument structure which
can be discharged only under a certain type of assignment (here assignment of a theta-
role); modifiers clearly do not do this and therefore they must be a higher-order function
from theta-assigners to theta-assigners. This position is forced by the assumption that
theta roles are syntactic objects of a particular type. Notice that nothing has been said
about their interpretation. It may be that the semantics interprets the syntactic theta-
label in exactly the manner described above: i.e. as a condition on the reference of
variable positions. It may be that this syntactic theory of thematic relations interfaces
with a semantic theory such as that of Chierchia (1989), where theta-roles are seen as
partial functions from information lists to particular elements on that list. Some such
statement needs to be made whatever view of syntactic argument positions is
adopted.
The question is then whether rejecting the relatively flexible position in (1) for
the position in (20) will actually gain anything linguistically, given that it will lose the
more convincing view of modification. This is not at all clear. The central problem with
thematic relations has always been defining what they actually mean. We have said
following Dowty (1989) that the meaning of thematic roles might be seen as clusters of
privileged lexical entailments (L-Thematic Role Types); it is significant, however, that
Dowty seems unconvinced that defining such privileged sets is possible, given the
variability in the situations to which the same theta-role is often applied. Ladusaw and
Dowty (1988: 63) comment:

It is possible that if one were to catalog all of the entailments associated

with the arguments of predicates and then to consider the intersection of
those entailments for all of the arguments which are called "Agents" some
non-empty set of entailments would result. We doubt that such a fixed set
of roles could be defined in this way.

32
 One possible reason for viewing the theta-role as special comes from the
phenomenon of control by implicit arguments (cf the discussion in Williams (1994:
208ff)).

[21 ] John performed his first operation yesterday.

[22] John underwent his first operation yesterday.

Assuming that the noun operation is understood in parallel to the verb operate (on) we
can assume that it has an argument structure as in (23).

[23] operation <AGENT, PATffiNT>

It then seems that the verbs perform and undergo lexically specify with which role of
their complement their subject is coindexed: the subject of perform always controls the
Agent of the complement, i.e. in (21) it is John who does the operating; the subject of
undergo on the other hand controls the Patient of the complement, i.e. in (22) it is John
who is operated on. This however is no reason for assuming privileged thematic labels
in the syntax as in (23). As Dowty (1989: lOOff) points out in a discussion of Higgins
(1979), these "control" facts may be viewed as lexically specified entailments of the
verb.

[24] Ve Vx [perform(x,e) <-» Agent(x,e) & ...]

"for all e and all x, x performs e if and only if x is the Agent of e"

[25] Ve Vx [undergo(x,e) <-> Patient(x,e) & ...]

"for all e and all x, x undergoes e if and only if x is the Patient of e"

That is, the empirical observations concerning (21-22) follow without taking the view
that positions in argument grids are restricted to privileged thematic labels. Williams

33
also observes, however, that there seem to be binding effects in these instances of
implicit control, as in (26-27) (Williams 1994: 210f, his { 1 1 } and { 13}).

[26] *John took his picture.

(AgL) (Agi.

[27] John took his own picture.

(Agi...) (Agi.Thi)

The verb take is like the verb perform in that the Agent of take is understood as co-
referential with the Agent of its complement. The noun picture has an argument
structure including an Agent (the individual taking the picture) and a Theme (the
individual pictured). It seems that where a pronoun is used as in (26), the Agent of take
may not be coreferential with the Theme of picture', where the anaphoric construction,
his own, is used, however, coreference is allowed. Williams suggests that this indicates
that the binding theory should in fact be relativised to thematic roles rather than the
phrases to which they are assigned. This view might be correct. The important point to
note however is that relativising the binding theory to argument structure requires only
differentiated argument positions (Williams' "bare theta theory") not argument positions
which are specifically understood as "thematic". The fact that the subject of take is
always coreferential with the Agent argument of its complement will follow by lexical
entailment as for perform. The crucial restriction is that two positions in an argument
grid may not be coindexed unless one is assigned to an anaphor. The restriction might
as well be represented by (28) as by (29).

[28]
[29] *<Agi,Thi>

The infinitives of result themselves provide no real support for a specific class
of thematic relations. In theories which concern themselves only with the problem of
antecedent selection for the gap in the Purpose Clause, the relevant generalisation seems

34
to be that the controller is a Theme: however, as Bach (1982) first pointed out, the real
restriction seems to apply to event structure, Purpose Clauses modify events with
"resultant states". In Section 5.1.1 I argue that the notion of Theme is parasitic of the
notion of resultant state. The Rationale Clause might conceivably be argued to take an
Agent antecedent, but this seems to be far too strong (see Section 5.2); in fact the
Rationale Clause seems to modify only events which are "manipulable", where the
subject can be construed as manipulating the event with an intention in mind. The Telic
Clause clearly does not select for thematic relations at all. In the theory adopted here,
whatever restrictions do actually occur as part of the meaning of an infinitive of result
will be expressed as a condition on the reference of the predicate of result itself: the
verbs which allow modification by this predicate must then be compatible with these
particular restrictions. It therefore seems that no advantage in an account of the
intepretation of the infinitives of result is to be gained by restricting argument positions
to thematic labels; taking the argument positions to be free variable positions on the
other hand allows the flexibility required for a unified view of argument projection
(without a fundamental distinction between functors and theta-assigners) as well as the
flexibility required to explore precisely what sort of generalisations do seem to be active
in constraining the intepretation of natural language predicates. In viewing the problem
of explaining the interpretation of the infinitives of result in this light I am in effect
following Bach (1982: 52) in pursuing an investigation of natural language metaphysics
in which the nature of the conceptual underpinnings of natural language interpretation
are explored. It seems that this exploration (for the infinitives of result at least) is
restricted rather than enhanced by a theory which concedes a special place to relations
which are "thematic".
In fact, a serious commitment to a thematic semantics in the Principles and
Parameters tradition is rare. The actual distinction between the theories of Williams and
Higginbotham in this respect is extremely small. Both follow the general tendency in
the Principles and Parameters tradition to treat theta-roles as significant only in so far as

35
they represent syntactic positions which can be used in constraining possible structures.
As Ladusaw and Dowty (1988: 63) observe:

Despite Chomsky's [(1981: 35)] invocation of the wrok of Gruber,

Jackendoff, Fillmore, and Davidson and the suggestive use of the 0, the 9-
criterion and 9-roles themselves are a principally diacritic theory: What is
crucial in their use in the core of GB is whether an argument is assigned a
6-role or not, which limits possible structures and thereby constrains the
application of rules.

The language of theta theory is something of a notational mirage. However, the

implications that its use can have for the way a theory is developed is clearly seen in
Williams' distinction between theta-assigners and functors. Higginbotham calls his
principles of argument discharge thematic transactions but the notion of theta role as
such is taken less seriously. All that is required in his view is a number of differentiated
positions which can be translated as variables in the usual manner. I will adopt this
theory as it allows precisely the flexibility that is required to produce an integrated
theory of argument projection and modification.

36
 CHAPTER 2: The Literature

Section 2.1: Two theories of argument projection

In the following sections I will discuss two theories of argument projection

which have proved important in a discussion of the syntax and semantics of
modification. Although similar in many respects, the theories differ significantly in the
way they approach the problem of "argument projection". The different conceptions of
the problem consequently pull the analyses in different directions. The tension is
particularly interesting given that the main work on Purpose Clause modification (Jones
(1991)) adopts the theory of Williams, where I will be adopting the theory of
Higginbotham.
!$>

Section 2.1.1: Williams (1994) and related works

A principle aim of Williams' work is the reduction of a seemingly disparate set

of data including NP-movement and certain cases of control to a limited number of
principles of theta-role assignment. These principles, consitituting the Theta Calculus,
involve a number of strictly local relations between the argument positions of heads,
mediated by syntactic structure. Following Stowell (1981), Williams assumes that a
lexical head has as part of the syntactic information associated with it a specification of
its argument structured He assumes that these positions are minimally differentiated
labels, whose notional content is not relevant to the core of his theory. As observed in
Sections 1.4, however, the idea that these labels represent relations of a particular type
is important to distinctions between standard predicates such as verbs and functors such
adverbials. This theory of minimally differentiated labels Williams calls Bare Theta
Theory. A predicate such as hit will therefore have the argument grid in (1), which
might be more fully specified as in (2).

37
[i]
[2] hit(Ag,Pa)

He assumes that the argument structure (similar to a traditional valency) is an

unordered list of roles with the exception that the subject theta-role is distinguished in
some manner in the lexicon: in the discussion that follows I will use bold font, as in
(3).

[3] hit(Ag,Pa)

The external argument is taken to be the "head" of the argument structure. The central
innovation in Williams' theory is the assumption that it is the theta-roles themselves
which refer rather than the NPs to which they are assigned. Argument projection then
involves the coindexing of theta-roles with the phrases to which they are assigned; the
binding theory is taken to apply to binding relations between thematic roles rather than
between the NPs.
Given that all lexical categories can act as predicates, it is assumed that all lexical
categories (including nouns) have an argument structure. Even simple nouns have an
argument, an external argument, R, indicating their reference. In a simple sentence such
as the one in (4) we have the argument structures as in (5).

[4] John hit Bill.

[5] (a) hit(AgX,Pay)

(b) John(R!)
(c) Bill(RJ)

Argument projection then proceeds by coindexing the external referential role of the
noun with the relevant theta-role of the verb, as in (6).

[6] John hit Bill

(Ri=x) (Agx,PaX) (RJ=y)

38
This coindexing proceeds under certain strict syntactic conditions and Williams
explicitly defines three types of coindexing that are allowed, as under (7).

[7] (a) Internal 9-role assignment

(b) External 6-role assignment (or vertical binding)
(c) Predication

Consider the application of the principles in (7) in a traditional phrase structure.

[8]

NP1

(Ri) (Agi,GoJ,Thk)

The principle in (7a), Internal 0-role Assignment, discharges (all) internal arguments by
coindexing the relevant positions of the head with the external arguments of the head's
sisters. This relation is strictly local. The principle in (7b), External 6-role Assignment,
allows the external theta-role to be coindexed with the maximal projection of its head.
From there it is assigned by the principle in (7c), Predication, to a sister of the maximal
projection by coindexing.
Williams in his later work (cf Williams 1994; 1995) presents this theory within
an X1 phrase structure. There are a number of problems with adapting the theory,
however. First, internal 0-role assignment cannot require strict sisterhood because
binary branching will not allow ditransitive verbs to project all of their internal
arguments as a sister to V. As Williams (1994: 30; 1995: 104) notes, this difficulty is
overcome if the restriction is weakened to m-command, so that internal arguments of
the verb can be coindexed with any phrase inside the projection of the verb (this will
mean, however, that the verb will theta-govern any modifier phrase adjoined to V).
39
 Second, Williams (1994: 33; 1995: 107) argues that vertical binding is reducible to
properties of X' theory, given that X'-theory already requires information associated
with the head to be projected to its maximal phrase. If we see each node as a label, the
idea would work as in (9).

[9]
<V, Bar=max, External argument = i,....>
<V, Bar=l, External argument = i,....>
I
<V, Bar=0, External argument = i,....>

However, this proves to be too restrictive because X'-theory changes not only the
structure of the VP but also of the sentence: in the X'-version of S, the VP is not a
sister of the subject.

[10]
IP

Spec1 T

In fact Williams makes the reasonable assumption that the index associated with the
external argument can percolate to I1 as long as I is not filled with a lexeme with an
external argument of its own. Nevertheless if vertical binding is allowed to I' then it
must follow from the independent principle in (7b) not from general properties of X'-
theory. Predication will then proceed under c-command as before.
These are the three principles of argument projection formally defined by
Williams. However, he makes an important assumption concerning the indexing of
modifiers which might be seen as a fourth type of principle. Consider a sentence
containing a subject-oriented small clause.
40
[11] John ate his pizza nude.

Williams (1991: 304) observes that a realistic structure for (11) would be the structure
in (12).

Spec3/15

The main VP will externally bind the index of its external theta role in the usual fashion,
say the index a\ the modifier AP will also bind an index for its external theta role, say
the index b. If AP adjoins to VP (as distributional evidence suggests it does) then the
mother VP will have to bear the index of both the lower VP and the AP if the NP in IP-
Spec is to be the subject of both predicates. The binding of two indices is of course
disallowed. Williams (1994: 89) therefore assumes that the two indices can be equated,
so that the mother VP is effectively binding only one index. This is effectively a fourth
principle of argument discharge, especially given that argument projection in Williams
reduces to contexts of coindexation, variously interpreted. I will therefore assume that
there is a fourth implicit principle of argument projection, related to vertical binding.
Williams' theory is primarily a syntactic one: argument grids are visible to
syntactic rules; the binding principles are revised in terms of relations between theta-
roles, making implicit arguments available for binding; roles are related by coindexing,
where the indices are syntactic objects requiring explicit syntactic principles to regulate
their percolation and value assignments. The mapping between the syntactic
representations and the semantic translations which result are not treated systematically.
This is quite clear: all of the argument projection principles we have discussed involve
coindexing but the interpretation of coindexing is different in different contexts. The
41
effect of coindexing the verb's internal theta role with the referential role of the noun
phrase direct object is shown in (13).

[13] hit (x,y) & Bill(y)

The reference of the noun, Bill, acts as a condition on the interpretation of the variable
in the verb's structure in a translation often assigned to standard cases of modification.
When an index is vertically bound, the interpretation is of lambda abstraction over the
external argument.

[14] foe [hit (x,y) & Bill(y)]

However, this use of lambda abstraction is appropriate only to non-nouns. We do not

want the external index on a noun to be interpreted as a lambda abstract because the
VP/AP/PP will require a term of some sort to fill its external argument position, not a
one place function. (15) will not yield (16).

[15] * A,x [hit (x,y) & Bill(y)] (Xz John(z))

[16] hit(John,y) & Bill(y)

The effect of predication is then to make the subject the argument of this lambda
i1 i ( i

function, as in (17), which by lambda conversion will give the required (18).

[17] Xx [hit (x,y) & Bill(y)] (John)

[18] hit (John,y) & Bill(y)

There seems to be no significance to the fact that the object is interpreted by intersection
(y is a thing which is hit and a thing which is Bill) where the subject is interpreted by
substitution (hit is true of John if John hits something).
The effect of coindexing a subordinate predicate with the X' to which it adjoins,
as in (12), is presumably intersection and conjunction: however, coindexing in this case
42
appears to be merely a syntactic manoeuvre to allow the secondary predication to occur
in accordance with syntactic principles. As well as the principles of argument projection
discussed above we will therefore also require a set of explicit translation rules to spell
out exactly how coindexing is to be interpreted in each context.
Williams' focus is not however semantics but syntax and his theory is important
in showing how a sophisticated theory of argument projection can obviate the need for
certain transformational devices and empty categories. His theory will therefore provide
an important orientation point throughout this thesis.

43
Section 2.1.2: Higginbotham (1985: 1989)

Higginbotham in general discards the language of thematic roles and formulates

a theory explicitly in terms of straightforward argument positions. His approach is
motivated generally by a desire to "elucidate" the meaning of natural language
expressions and lexical items, and to do so he draws on the resources of predicate
logic. His theory is therefore intended to address in an explicit and balanced way both
the form and the interpretation of sentences and their constituent parts. Like Williams
and in keeping with standard analyses he assumes that every lexical argument comes
with an argument structure or thematic grid, consisting of bare positions (which may
however range over various types of object: individuals, events, propositions,
properties etc), as in (1). The grid is straightforwardly translatable into predicate logic,
as in (2).

[1]
[2] like(x,y)

The information in (1) is part of the syntactic information associated with an item. The
central conception of the problem of argument projection in Higginbotham is however
slightly different to that of Williams. For Higginbotham the question is how, given two
sister nodes in a tree and the argument information associated with them, we can work
out the argument structure of the mother of those nodes.

[3]
Y<z0...z n>

... n>

He argues that Universal Grammar provides a strictly limited way in which daughter
nodes may be related to their mother, i.e. functions which specify a particular type of
mapping. He suggests four main functions or "thematic transactions", as under (4).

44
[4] (a) Direct theta-marking
(b) Theta-binding
(c) Theta-identification
(d) Autonymous theta-marking

Direct theta-marking specifies a relation between a predicate with an open

position and a closed term, such that the mother has the argument structure of the
predicate with one less open position. The translation of this substitutes the reference of
the closed term into the open position of the predicate. Consider (5).

[5]
V <1,2*>

DPo

hit Bill

The verb hit is a transitive verb with two argument positions; its sister is a DP headed
by a proper name (a closed term). The two nodes project a mother with the argument
structure of hit, where one position has been saturated or closed (indicated by an
asterisk; one could also omit the "2"). The nodes in (5) would receive the logical
translation in (6).

[6] V: hit(x,y)
DP: Bill
V: hit(x,Bill)

Theta-binding involves the closing of an argument position of a predicate by an

operator. Consider (7).

f7]
D'<r>
Do NP<1>

the dog

45
In this view, the determiner, the, is an operator; dog is a classifier of individuals (i.e. it
defines the set of individuals of which dog is true). The operator binds the open
position, closing it. The translation for the nodes in (7) is as in (8).

[8] D: THE
NP: dog(x)
D1 : [THEx: dog(x)]

Theta-identification deals with simple cases of modification, as in (9), where the

open position in the argument grid of the modifier is identified with the open position in
the argument grid of the head node, the argument grid of the head then filtering up to
the mother unaltered.

9]

wall

As the translation of (9) in (10) shows, the semantic effect is literally one of
identification: the variable which the AP classifies is assigned the same value as the
variable which the lower N1 classifies.

[10] N1 : wall(x)
AP: white(y)
N1 : wall(x) & white(x)

Straightforward conjunction is, however, inappropriate for some adjectival

constructions, in one reading at least. For this final class, Higginbotham introduces
autonymous 0-marking. Consider the example in (11).

[Ill the big butterfly

46
 (11) could mean that something is both a butterfly and big (by some general standard);
it could also mean, however, that the butterfly is big in so far as it is a butterfly: a little
elephant is still going to be bigger than a big butterfly. Higginbotham therefore suggests
that the adjective big can be represented as a two-place relation between an entity and an
attribute, as in (12).

[12] big (y AAx F(x))

This means that y is big for an F. Consider the structure for modification in (13).

[13]

butterfly

As in the case of standard modification in (9), the first position of A 0-identifies

with the argument position of N\ Now, however, A also autonomously theta-marks AT
through its second position. Unlike the case of direct 0-marking where a closed term is
substituted into the argument position of a predicate (i.e. the argument of the predicate
is given the reference of the term), in this case, the second position is given the value of
the N' node, in this case the attribute of being a butterfly. Both positions of A are now
satisfied and the grid of the lower N1 passes up to the upper N1. The translation of the
nodes in (13) is as in (14).

[14] N': butterfly (x)

[15] A: big (y,AXz F(z))
[16] N1 : butterfly (x) & big (x,AXz butterfly (z))

"x is a butterfly and x is big with respect to being a butterfly"

47
 Higginbotham's theory of argument projection is therefore conceived in terms
of the systematic composition of meaning. He abandons the Montagovian restriction to
function-argument application and argues that Universal Grammar specifies a small
number of distinct principles of argument projection. There is an intimate trade-off in
this view between the syntax and the semantics. The mapping to the semantics forms an
integral part of the definition of the principles of projection themselves. Notice also that
it is not a syntactic theory in certain key respects. Where the indices in Williams are
syntactically real, the asterisks indicating argument saturation in Higginbotham are not:
they are a notational convenience for keeping track of the computation in progress.
Higginbotham must assume that the argument grid itself is a syntactic object because the
principles of semantic interpretation will need something to specify the addicity of the
predicate and relevant aspects of each position's interpretation. Higginbotham (p.c.)
$

assumes that the effect on the argument grid will be syntactically encoded as part of the
label information for each node, although this will be formally complex. This formal
function can be represented either in the Government and Binding approach of
Chomsky (1981) where the whole tree is presented for interpretation or in the bottom-
i
up Bare Phrase Structure account of Chomsky (1994). It is worth noting, however, that
i
there may be no need to represent any such information syntactically. Consider the case
of direct theta-marking. Assume that, except for the argument grid itself which specifies
the addicity of the predicate, the principles of argument projection do not have syntactic
reflexes. Consider then a VP with internal subject as in (17).

[17]

DPo

DPo

John

48
Here the transitive verb hit directly theta-marks the object Bill. Assume that this is not
marked in the syntax, so that the grid passes up in the expected manner. Now consider
the interpretation of V.

[18] hit(x,Bill)

Now assume that the verb directly theta-marks the subject, John, through the second
position again: this will be uninterpretable. These two nodes do not qualify for direct
theta-marking because the relevant position is not an open position but is already
referential (it refers to Bill). Direct theta-marking simply cannot apply to the same
position twice: the principles of argument projection are so closely integrated in syntax
and semantics that the uninterpretability of the structure will block the argument
projection. This approach will work best in a simultaneous computation analysis with a
bottom-up derivation. Whether this possibility is pursued or whether the thematic
transactions are defined in both syntactic and semantic terms, the relevant facts for our
analysis will follow.

49
Section 2.2: Some analyses of infinitives of result

Section 2.2.1: Faraci (1974)

Farad's doctoral thesis is the pioneering work in the study of infinitives of

purpose, establishing some of the major structural distinctions between the two
infinitives (Purpose and Rationale Clause) and isolating some key features of
interpretation. Farad's central claim is that the to-infinitive and the/or-NP construction
(and in fact the/or-gerund construction) are not simply related constructions, but are
identical at deep structure. Although his view would now be modified considerably, his
view that for is the head of this construction is still extremely influential. I take it,
however, that the syntactic characteristics he isolates are in fact true quite generally of
particular classes of adverbial modifiers, and that the semantic relations indicate lexical
classes that can be defined in terms of predicates occurring in the conditions on
reference of particular items. The key insight of the relation between the constructions,
however, comes from this thesis.
Faraci argues that the preposition for can take three types of complement at deep
structure.

1]
PP

for [NP]
[S NP to VP]
[S NP -ing VP

A late transformation then deletes for if it is adjacent to to. The parallels he offers are
striking and cover both distribution and interpretation.
He notes (Pge 48) that those verbs which subcategorise for /or-NP also
subcategorise for to-VP, as in (2-4).
50
[2] (a) John was waiting for a good movie.
(b) John was waiting to find a good movie.

[3] (a) John was hoping for a good movie.

(b) John was hoping to find a good movie.

[4] (a) John was anxious for a good movie.

(b) John was anxious to find a good movie.

More strikingly, he shows that the/or-NP construction exhibits exactly the same
alternation as the to-VP construction with regard to "purpose" and "rationale" readings
(Pge 49): i.e. the phrase may be interpreted as describing a function of the Theme in
which case it cannot be separated from the VP; or it may be interpreted as describing a
general intention of the Agent in which case it may optionally be separated from the VP.
The parallelism is shown in (5-7).

[5] (a) John built a robot to entertain his guests.

(b) John built a robot for the entertainment of his guests.

[6] (a) What John built to entertain his guests was a robot.
(b) What John built for the entertainment of his guests was a robot.

[7] (a) What John did to entertain his guests was build a robot.
(b) What John did for the entertainment of his guests was build a robot.

The sentences in (5) are ambiguous between purpose and rationale readings. (6)
favours a purpose reading and (7) requires a rationale reading. He also notes (Pge 50)
that if it is the infinitive which is the focus of the pseudo-cleft (i.e. the phrase which is
questioned by the wh-element) then the preposition for must appear (presumably as the
Case-marker of what).

[8] (a) What John built the robot for was to entertain his guests.
(b) What John built the robot for was the entertainment of his guests.

51
 The purpose reading and the rationale reading of each phrase may be
disambiguated by proposing (Pge 53), as in (9), where only the rationale reading is
possible.

[91 (a) To entertain his guests, John built a robot.

(b) For the entertainment of his guests, John built a robot.

When the two readings co-occur, parallel linear precedence restrictions also apply (Pge
57), as in (10-11).

[10] (a) John built a robot to do the housework to have fun.

(b) John built a robot for housework for fun.

[11] (a) *John built a robot to have fun to do the housework,

(b) *John built a robot for fun for the housework.

On the basis of such alternations, Faraci argues that the constructions are realised in the
configuration in (12). where the Purpose Clause is realised as the Complement of V in
VP, and the Rationale Clause is realised as the Complement of VP in VP.

[12]
VP'
T-
AUX VP PP
T
NP PP {NP}
{S}
{NP}
{S}

for for

There are a similar number of semantic parallelisms. As with the to-VP

construction, the property attributed to the object in afor-NP purpose construction must
be contingent on the main event; that is, the property attributed cannot exist prior to the
main event or independently of it.

52
[13] (a) We built the car to be durable at high speeds,
(b) We built the car for durability at high speeds.

In (13), the durability of the car must be result of the particular event of building in
which it is involved; it cannot be a pre-existing or inherent property (which would give
a reading of (13) in which the durability was the motivation for buying the car). The
property expressed by the NP in (13b) can be made an independently-existing property
of the object by using a possessive, as in (14).

[14] *We built the car for its durability at high speeds.

Its implies that durability is already a property of the car and that that property is a
motivation for buying. As expected, this sentence is ill-formed. A similar parallelism of
to-VP and/or-NP constructions holds for the rationale use of the /or-phrase. Both
constructions are incompatible with insentient subjects, volition-denying adverbs, and
statives where the state is not under the control of the subject.

[15] (a) *The wind knocked down the house to create some open spaces.
(b) The wind knocked down the house for the creation of some open
spaces.

[16] (a) *John accidentally entered the competition to win the big prize,
(b) *John accidentally entered the competition for the big prize.

[17] (a) *Mary has long legs to do modelling,

(b) *Mary has long legs for modelling.

There is interference in the interpretation of this sentence from two readings which we will not discuss here,
which may be paraphrased as in (a-b).

(a) Mary has legs which are rather (too) long for modelling.
(b) Mary has the legs for modelling.

I will assume that in (a) the infinitive modifies long as a degree-adjective and that in (b) it is an Infinitival
Relative (with a ideological reading). For the sake of Faraci's present point these two readings are irrelevant.
53
 As for the to-VP construction, the for-NP construction imposes heavy
selectional restrictions on the verb it modifies. In the terminology of Bach (1982), to be
discussed in the next section, only verbs with a "resultant state" of a "positive" sort are
compatible with the purpose reading of these constructions.

[ 18] (a) John built the table to put the flowers on.
(b) John built the table for the flowers.

[19] (a) * John destroyed the table to put on the bonfire,

(b) *John destroyed the table for the bonfire.

Both build and destroy are change of state verbs which concern the manipulation of the
integrity of an object, but only the positive verb, build, is allowed with to-VP andfor-
NP in the purpose construction.
Faraci also observes that both the/or-NP and the to-VP purpose construction
select the Theme, wherever it occurs.

[20] (a) John took the horse to the stable to get a rubdown.
(b) John took the horse to the stable for a rubdown.

[21] (a) John stopped by Henry's to get a massage,

(b) John stopped by Henry's for a massage.

[22] (a) John took Bill to Henry's to get a massage,

(b) John took Bill to Henry's for a massage.

In (20), it is the horse which is Theme and direct object of take, as is Bill in (22). In
(21), however, it is the subject, John, who is the Theme, and hence controls the
Purpose Clause.
Where the Theme is interpreted as being at the disposal of someone in its
attributed property, the person interpreted as being in charge is the Goal/Beneficiary
(the Agent being the default Beneficiary of an action).

[23] (a) Mary bought some rosary beads to do her penance on.
(b) Mary bought some rosary beads for penance.
54
[24] (a) Mary bought her daughter some rosary beads to do her penance on.
(b) Mary bought her daughter some rosary beads for penance.

In (24), Mary's daughter is the Beneficiary of the event of buying and hence controls
the Purpose Clause; in (23), Mary, as the Agent of the event, is taken as the default
Beneficiary.
Faraci thus establishes a striking parallelism between the to- VP construction and
the/or-NP construction. The parallelism falls in two domains: syntax and semantics. In
the syntax, both the to-VP construction and the/or-NP construction are strictly VP-
internal when they are used to modify the object of the verb (the purpose construction);
and both are VP-external when they are used to modify the subject (the rationale
construction). In the semantics, both the to-VP construction and the for-NP
construction select the Theme in their purpose reading and describe the properties that
the Theme comes to have or the purposes it comes to serve by virtue of its participation
in the event described by the modified verb; and both the to-VP construction and the
for-NP construction select the Agent in their Rationale reading and concern the general
purposes or intentions that motivate the Agent's participation in the event.
In Farad's analysis this parallelism in syntax and semantics is derived from
strict lexical and syntactic identity between the to-VP construction and the for-NP
construction. As can be seen in the diagram in (12), the only difference between the to-
VP construction and the/or-NP construction is the alternation between the nominal and
the sentential complement of underlying/or. As observed earlier, Faraci reinforces this
claim of strict identity by claiming that the two constructions are subcategorised for by
the same verbs and adjectives (e.g. wait, hope, be anxious).
While it is clear that the parallisms which Faraci discusses do exist, there is
strong evidence that they do not exist in the form argued for by Faraci. It will be argued
in this thesis that the parallelisms are not uniform and do not derive from strict
underlying lexical identity. Rather, the syntactic parallelisms fall out from a general
property of adverbial modification in the syntax, conditioned by constraints on
55
argument projection ?.nd c-command (and are not specific to the construction in
question); and the semantic parallelisms reflect shared units of meaning in the
conditions on interpretation for each lexical predicate, in much the way that verbs which
share particular thematic roles may be said to be similar but not identical in meaning. As
argued in Section 1.4, however, a satisfactory "elucidation" of the "units of meaning"
in question will not necessarily result in thematic relations of the standard type.
In essence, this thesis denies that the to-VP construction and the for-NP
construction are instances of the same construction but rather that they are part of a
broader pattern of syntactic and lexical generalisations. The best way to illustrate this is
to find a preposition v/hich parallels the to- and for- constructions in both syntax and
semantics in the relevant respects, but which is clearly not an instance of a covert for-
construction. Consider the behaviour of the preposition, as.
Like for and to, as has both a "purpose" and a "rationale" reading. The purpose
as-NP is forced to remain within VP,where the rationale oy-NP can be separated from
it, as in (25-28).

[25] John gave Bill a book as a study aid.

[26] John gave Bill a book as a joke.
[27] *What John did as study aid was give Bill a book.
[28] What John did as a joke was give Bill a book.

In (25), the phrase as a study aid describes the function that the book was intended to
serve having been given to Bill: it thus takes a "purpose" reading. It cannot therefore be
separated by pseudo-clefting from the VP containing its antecedent object, as shown by
the ill-formedness of (27). In (26) on the other hand, as a joke describes how the event
is intended to be taken (the general rationale behind the act of giving): as shown in (28),
this os-phrase can be separated from VP by pseudoclefting. The two phrase-types can
also co-occur. Where they do so the familiar linear precedence restrictions apply.

[29] John gave Bill a book as a study aid as a joke.

[30] *John gave Bill a book as a joke as a study aid.

56
Only the "rationale" phrase can be preposed.

[31] As a joke, John gave Bill a book.

[32] * As a study aid, John gave Bill a book.

Semantic restrictions pattern in a similar fashion. In the purpose reading, the

function which the object serves must be a consequence of the main event, not
antecedent to it; and the property must be intended rather than realised.

[33] John designed the car as an economy model.

In (33), the car has the property of being an economy model only in so far as it is
designed to be such (we cannot read (33) as meaning that being an economy model was
a property the car had to start with). Attribution of the property is only intended,
however, and can be denied without contradiction, as in (34).

[34] John designed the car as an economy model, but a number of elementary errors
in fact made it extremely inefficient.

Also as expected, the purpose as-phrase selects only "positive" resultant state verbs, as
in (35-36), always selects the Theme as controller, as in (37-38), and shows the
expected benefactive alternation, as in (39-40).

[35] John built the bench as a rest-place for the elderly.

[36] *John destroyed the bench as wood for the fire.
[37] John took Bill to the party as the local language expert.
[38] John stopped in on the party as the local languageexpert.
[39] Mary bought a computer game as a companion on the longflight.
[40] Mary bought her daughter a computer game as a companionon the long flight.

Similarly, in the rationale reading, the as-NP construction is incompatible with

insentient subjects, volition-denying adverbs, and states where the subject has no
control over that state.

57
[41] The river flooded as a threat to the local area.
[42] *John accidentally designed the car as an economy model.
[43] *Mary had long legs as a way of catching men's attention.

•i _

It seems that the parallels extend to the Telic Clause in certain important respects.

[46] John smashed the ball into the net, for a goal in the final minute.
[47] ...and inexplicably quickly the dinosaurs were gone, as a reminder to us all that
nothing will endure forever.

The use offor in (46) is clearly parallel to the Telic Clause in (48).

[48] John smashed the ball into the net, to score in the final minute.

Notice that the /or-phrase in (46) and the infinitive in (48) both receive a factive
reading, that is, the goal actually happens. The comma intonation suggests a similar
peripheral attachment for the/or-construction as has been argued for the Telic Clause.
The as-phrase in (47) is slightly different in interpretation from the other two
constructions but is similar in the relevant respects, namely it is factive, i.e. the passing
away of dinosaurs does remind us of mortal transience rather than being intended to do
so.
The preposition as clearly belongs to the same semantic family as to and for.
However, there is no evidence that the as-NP construction is the same as the/or-NP
construction; indeed any full specification of the meaning of as and for will show up
clear differences-- but in the relevant respects they are parallel. To account for this
family connection, we therefore need something other than syntactic or lexical identity,
a way of showing parallelism of interpretation without syntactic isomorphy. It was
suggested in Section 1.4 that it was precisely this role that thematic relations play and
that the relevant properties could be captured by taking these realtions as conditions on
the reference of classes of predicates. In this view, each of the related lexemes would
have an argument structure (lexically specified), whose conditions on reference would
include reference to the same predicate.
58
[49] (a) TO(......) iff P(x,e,AXz F(z)) & ... & ....
(b) FOR(......) iff P(x,e,AXz F(z)) & ... & ....
(c) AS(......) iff P(x,e,A?tz F(z)) & ... & ....

It is perfectly possible of course that, with the exception of the Purpose predicate, the
conditions on reference for each preposition will be entirely dissimilar. We thus gain
flexibility for lexical variation while giving a precise view of a common lexical element.
Such flexibility is necessary, even in Farad's own terms. As was observed
above, Faraci argues (Pge 48) that the/or-NP construction and the to-VP construction
are instances of the same construction because they are subcategorised by the same set
of verbs and adjectives (wait, hope, be anxious}, but this is not so in all cases. In
particular, there are verbs such as pay and send which subcategorise for for but not to.

[50] (a) John paid for the books.

(b) *John paid to own the books.
i <t f

[51] (a) John sent for Bill.

(b) ?*John sent to get Bill.

Further, we have already shown that as is parallel to for and to semantically in precisely
the way that Faraci argues that to is parallel to for, and yet as is not similarly
subcategorised.

[52] *John was waiting as a good movie.

[53] *John was hoping as a good movie.
[54] *John was anxious as a good movie.

Such lexical anomalies are predicted and allowed in the analysis adopted here because
parallelism of meaning points to shared restrictions in the conditions on interpretation,
not lexical identity.
Such an approach does not of course account for the striking syntactic
parallelisms (where both semantic and syntactic parallelisms in Farad's account flow

59
necessarily from constructional identity). However, it appears that the syntactic and
semantic parallelisms are not in fact the same; indeed the syntactic parallelism exhibited
by the three lexemes are a subpart of a much wider generalisation concerning adverbial
modification in the syntax. It appears that any object-oriented modifier is internal to VP
and any subject oriented modifier is external to (at least one) VP. Consider the case of
the preposition with.
With is clearly semantically distinct from to, for, and as with respect to the
predication of intentions and purposes, yet it shows exactly the same syntactic
restrictions as they do.

[55] John designed a car with an automatic sunroof.

[56] John built a car with great enthusiasm.

The w/r/i-modifier in (55) is object-oriented; the w/r/z-modifier in (56) is subject-

oriented. The object-oriented construction is incompatible with pseudoclefting and with
preposing, where the subject-oriented construction is compatible with both.

[57] *What John did with an automatic sunroof was design acar.
[58] *With an automatic sunroof, John designed a car.
[59] What John did with great enthusiasm was design a car.
[60] With great enthusiasm, John designed a car.

It appears that there is also a use of the with-construction which parallels the Telic
Clause in being strictly VP-external (a traditional S-adverb usage).

[61] John built the car, with any luck.

Like the Telic Clause, this construction is incompatible with the do so construction and
with t hough-movement (see Section 3.1 for a more detailed discussion of these
examples).

[62] ?Ruth gave David her article, with any luck, but Kerstin might have done so
too, unfortunately.
60
[63] ?*Give David the code, though Ruth may, with any luck,...

It therefore appears that there are a broad set of restrictions on the distribution of
adverbial modifiers, to the effect that object-modifiers must be VP-internal, subject-
modifiers must be VP-adjoined, and sentence-modifiers must be strictly VP-extemal.
This has nothing to do with the purpose and rationale constructions themselves,
whether with for, to, or as.
The ultimate evidence, however, that the syntactic parallelisms which Faraci sets
out derive, not from a shared constructional head, but from general properties of
adverbial modification in the syntax, comes from a set of constructions which Williams
(1983) takes to be instances of direct predication, and which are generally not taken to
be head-mediated- small clauses. It appears that, just like all the examples discussed so
far, where the small clause is object-oriented, as in (64), it is strictly VP-internal and
incompatible with pseudoclefting, as in (65), and preposing, as in (66); and where it is
subject-oriented, as in (67), it is VP-adjoined and compatible with pseudoclefting, as in
(68), and preposing, as in (69).

[64] Sven served the mutton hot.

[65] *What Sven did hot was serve the mutton.
[66] *Hot, Sven served the mutton.
[67] Sven served the mutton, trembling.
[68] What Sven did trembling was serve the mutton.
[69] Trembling, Sven served the mutton.

The explanation for the syntactic parallelism of for-NP and to-VP constructions
therefore lies in perfectly general principles of adverbial modification and not in any
construction-specific feature (such as a shared head).
Thus, although Faraci points to an important set of relations between a group of
purpose constructions in English, his view that the relations between them are syntactic
rather than lexical misses important generalisations concerning adverbial modification
and significant lexical idiosyncracies between the heads of the related constructions.

61
Section 2.2.2: The account of Bach (1982)

Bach (1982) works within the general framework of Montague grammar, using
a non-transformational syntax of the type developed in Bach & Partee (1980) and Partee
& Bach (1980), and a semantics like that in Montague (1974). The general aim of his
paper is to establish what type of information one might not want to include in a
syntactic or semantic account of the Purpose Clause (the focus of interest is the Purpose
Clause; the Rationale Clause is mentioned principally by way of contrast; the Telic
Clause is not mentioned at all).
The core of Bach's analysis lies in the notion of a "resultant state". He begins
his discussion by drawing out a number of parallels between the to-infinitive and the
/or-NP construction. We have looked at some of these parallels in our discussion of
Faraci (1974). Bach defines a resultant state, after Van Wright (1963) and Dowty
(1972), as involving an event of the form p/—ip or -»/?//?. The Purpose Clause is then
related to the second state of affairs. Consider (1-3) (Bach 1982: 39, his {34-36}).

[1] Mary got up for an hour.

[2] Mary got up for an hour but had to lie down again after fifteen minutes.
[3] For an hour, Mary got up.

(1) can have two interpretations. Firstly (and least plausibly) it could mean that Mary
got up over and over again during the course of one hour (the iterative reading).
Secondly it could mean that Mary got up, intending to be up for an hour (the Purpose
Clause meaning). The get up event will be structured as in (4), and, in the Purpose
Clause reading, the for an hour will apply to the second state of affairs (measuring the
duration of that state).

[4] Mary is not up I Mary is up

Further, in the Purpose Clause reading the/or-phrase denotes only an intended state of
affairs. Mary's intention to stay up for an hour could be thwarted, as in (2). As one
62
would expect, we cannot prepose the /0r-phrase without losing the Purpose Clause
reading: (3) is purely iterative. The notion of resultant state does indeed seem to be an
important one and I will return to it later.
It is worth noting in passing that Bach's discussion of the resultant state is odd
in two ways. First, he specifies in his list of compatible verb classes that the Purpose
Clause selects for transitive verbs; this is clearly an oversight as get up is an intransitive
verb. In fact, intransitive verbs, particularly the unaccusatives, seem perfectly
compatible with the Purpose Clause and indeed interact interestingly with it. Second,
(1) is a rather complex example because the iterative reading involves what seems to be
a quite different meaning offor. In his thesis, Faraci (cited by Bach) gives examples of
for phrases which clearly parallel the to-infinitive in having purpose readings alternating
with rationale readings. But the iterative for in (1) is in no sense a rationale or
intentional reading. This in fact points to the problem of insisting on some sort of
underlying lexical identity between to- and/<?r-constructions.
Bach uses the Purpose Clause to pose a general question concerning the nature
of linguistic semantics and in particular thematics. He observes that there are two
possible extremes in dealing with the problem of meaning. First, one could take the
view that the meaning of these clauses falls outside the domain of linguistics, requiring
general theories of human action and cognition. On this view there are no questions to
ask and the general issues which we have been discussing will be set aside as
linguistically uninteresting. Second, one could take the view that we need "a
decompositional treatment in which we actually give an explicit definition of the
constants we've used or the basis of more primitive components" (Pge 47). This has in
general been the approach adopted by linguists and Bach cites as examples "the
syntactic decompositions of generative semantics through the lexical decompositions of
some interpretivists to the decompositional translations into intensional logic of some
Montague grammarians". This thesis is also written in the latter tradition and Bach's
comments spell out the problems for such an approach.
He opens by discussing the sentence in (5) (Pge 48f, his (66)).
63
[5] John bought War and Peace to read.

He observes (Pge 49) that "we might give an explicit meaning rule for for' (or,
equivalently, give it a more elaborate translation into the intensional logic using either
constants corresponding to the English words we use in our meaning rule or semantic
primitives of some sort, which usually seem to boil down to the former unless we give
explicit meaning rules for them)". His attempt at such a rule is given under (6).

[6] I: for' denotes that function h from properties of individual concepts to

functions from TV-type intensions to TV denotations such that for all NP
intensions #1, #2 properties P, and intensions of TV meanings R,
/z(P)(R)(2>l)(!P2)=l at w, r, g (g an assignment of values to variables)...

I: (iff) R(fPO(22)=l at w, t, g, and the extension of &2 at w, t, g includes the

property of being an x such that x intends that the extension of T\ at w, r', g
include P at w, t\ g for some t' later than t.
for' might then be translated as:

Bach (Pge 49, his {67-68}) points out that this is inadequate, as it would make
both (7) and (8) acceptable, when they are in fact ill-formed.

[7] *John read War and Peace to start a fire with.

[8] *I hate War and Peace for John to destroy.

(7) is acceptable by (6) because (6) merely requires that John read War and Peace and
that he intend (after reading it) to start a fire with it. (8) is only well-formed by (6) in a
very curious reading. (6) requires in (8) that I hate War and Peace and intend (after
hating it) for John to destroy it. That temporal requirement, that John destroy the book
after the time of my hating it, is very odd indeed. (6) therefore seems to be quite
inadequate as a description of the semantics of the Purpose Clause. Bach does not,
however, suggest that there is any straightforward reason for this inadequacy. The
64
problem is presumably that the two conditions, (a) that John read the book and (b) that
he have the intention during the reading of burning it after the reading, are unrelated. It
is not enough that John perform the action denoted by the matrix verb and have an
intention, but that he intend that very action to be instrumental in achieving his
intention. But his reading of War and Peace is not for the purpose of destroying it, any
more than my hating of War and Peace is for John's destroying it. The predicate of
Purpose suggested in my basic analysis of the Purpose construction makes precisely
this link: the purpose predicate is a relation between an entity, an event, and a property,
such that the entity's involvement in the event is for the sake of its having the property.
John does not of course read the book for the sake of starting a fire with it; nor do I hate
i<
the book for the sake of John destroying it.
j :>,•
- 1 V

Bach (Pge 50) further objects to the idea of including reference to intentions and
agentivity in the translation (as I have done with the Rationale predicate, R) on the
grounds of examples such as (9).

[9] John bought The Golden Notebook for his children to read.

He comments,

Can't this be true in a situation where after months of clamoring, John

finally gave in? Here he is merely making it possible for his children to
read the book in question.

But he is surely not merely making the book available. An intention need not imply
active enthusiasm for the course of action taken, only that that course is pursued in the
knowledge of the nature and likely consequences of that action. John's lack of
enthusiasm for giving the book may remain constant before and after the event of
buying the book, what changes is his volitional state: he chooses to buy the book, and
does so knowing, indeed intending, that as a consequence his children will read the
book. We may concede that his choice is grudging, but a choice it remains.

65
 Ultimately, Bach seems dissatisfied with both extremes in addressing the
linguistic meaning of Purpose Clauses. He concludes (Pge 52) by suggesting a general
way forward which avoids both extremes, although his suggestions are (as he admits)
extremely programmatic.

I believe that there is another way of avoiding the uncomfortable

conclusions above: that is to recognize that there are at least two other
domains of linguistic inquiry beyond syntax and semantics. One is what I
have called elsewhere (Bach 1981) natural language metaphysics. In such
an endeavor we try to understand the conceptual underpinnings of natural
languages and natural language. This study has to deal with the place
where there is the most immediate and rich interplay between language,
the world, and our conceptualization of the world: lexical semantics. The
other domain is the theory of discourse, where we study how speakers and
hearers construct partial models based on texts in contexts (it thus
impinges on pragmatics, in several senses of the word).

Into the domain of natural language metaphysics would fall the theory of action,
volition, intention, and the attribution of properties; into the domain of discourse, the
way in which we work out who is most likely to have objects within their control in
different contexts. The issues that Bach raises here, especially with regard to natural
'i >
language metaphysics will be taken up throughout this thesis.

66
Section 2.2.3: The account of Chierchia (1989)

Chierchia (1939) offers an account of the Purpose Clause using structured

meanings, thematic relations, and semantic postulates. Meaning is structured in terms of
information units. Chierchia comments (Pge 137): "We will call atomic information
units "eventualities". If (3 is an information unit, E(p) is a proposition that can be read
as "it is the case that p"... We say that an information unit p is true in a world w iff w
[E(p)]." Thematic relations are then viewed as partial functions from eventualities into
one of their constituents (Pge 139). The definition for Theme is given under (1).

[1] Th(P) = xi, for that unique xiG p such that whenever p occurs, xi moves,
changes possession or is acted upon by Ag(p). If there is no such xi, Th(p) is
undefined.

Chierchia comments (Pge 139): "Thematic roles are therefore viewed not as primitives
but as partial functions defined in terms of the entailments associated with
eventualities".
Chierchia then defines how eventualities will be classified, as under (2) (Pge
144, his {5}). The formula in (3) means that a is a constituent of p; and the formula in
(4) means that there is an eventuality exactly like p except that y is uniformly substituted
for a.

[2] The type of an eventuality p is the relation r such that g(r) is the first member of
p. g is either an argument manipulating operation... or the identity map. We
represent an eventuality of type r as pr.

[3] p[cc]
[4] P[o/y]

Chierchia then offers his first version of a general control rule, as under (5) (Pge 144).

67
[5] (a) E(pr[P]) E(prfP/P*e(p))])
(b) Th > Go > Ag....

He paraphrases this as: "suppose that an r-eventuality that has property P as one of its
constituents is the case; then an r-eventuality obtained from p by replacing P with the
eventuality <P,x> (where x is the bearer of the role 9 with respect to p) is also the
case". The value of 0 is assigned by taking the highest theta-role in the hierarchy which
the eventuality type in question makes available.
(5) is a general rule for control. This means that it applies also to cases where
the "purpose clause" is subcategorised by the verb. The way the rule applies is shown
under (6) (Pge 144, his {7}).

[6] (a) John tries to swim

(b) E(<try',J,swim'>)
(c) E(ptrv[P]) E(ptry[P/P*(Ag(p))])
(d) E(<try! ,J,<swim ,J»)

(6b) is the prepositional meaning of (6a). (6c) represents the application of the
postulate in (5) to a fry-eventuality: try supplies only an Agent theta-role and therefore
6(P) is given the value Ag(p). Assuming (6c), (6b) now entails (6d).
The postulate in (5) captures three important generalisations: that control in these
structures is strictly local (the controller must be a lexically specified argument of the
main verb); that the controller must be unique; and that the controller must be overtly
present (accounting for resistance to detransitivisation where a Theme argument would
be selected and resistance to passivisation where an Agent argument would be selected).
This third generalisation, however, is problematic for a set of predicates which allow
control by an implicit argument, as in (7).

[7] It was decided to leave.

Chierchia therefore adopts a second control principle which effectively supplies the
required controller, as in (8) (Pge 149, his {3}).
68
[8] E(pr[P]) A 6(j5)=xi -> E(fr[P/Pe(p))])

The added conjunct in (8) allows reference to an implicit argument (0-role) of the verb.
This can now be selected as controller without being syntactically available. The
assumption is then that there is a parameter allowing predicates to select for either the
postulate under (5) or the expanded version in (8).
Chierchia now turns to the problem of the Purpose Clause, commenting that "I
will just assume that PC's are attached within the VP. This much seems to be fairly
clear, while it is not so clear whether they are VP-complements or VP-adjuncts".
Noting that Purpose Clauses are insensitive to the passive, and that they occur with
double object verbs v/ithout being sensitive to the double object alternation, he
concludes that the control principle for these clauses cannot be syntactic (governed by
Grammatical Functions such as Direct Object) but must be thematic. Considering
examples such as those in (9) (Pge 153, his {6}), he assumes that the accurate
generalisation concerning controller choice is that given in (10) (Pge 154, his {7}).

[9] (a) I gave thati to Johnj [__j to read _i].

(b) I gave Johnj thati [_j to read _i].
(c) Johnj was given thati [_j to read _i].

[ 1 0] (a) The obligatory gap of PC's is controlled by the matrix theme.

(b) The ndn obligatory gap of PC's is controlled by the matrix goal, if there
is one.

The control principle(s) which Chierchia has so far developed apply only to
cases where a predicate selects a particular set of thematic roles (or rather where an
eventuality is classified by a predicate as having certain sets of ordered entailments).
The Purpose Clause, however, seems to be an adjunct, at least syntactically. Chierchia
therefore adopts the theory of adverbial modification developed by Sally McConnell-
Ginet (1982), in which adverbs, previously classed as functions from TVP to TVP
intensions, are taken to be natural augmentations of a verb's meaning, i.e. to count as
69
extra arguments of the verb. Chierchia offers the postulate in (1 1) as his formalisation
of this approach, (Pge 155, his {11}).

[11] For every ad- verb £, £* is a function from properties into properties, such that
for any property P and any x, [£*(?)] (x) is the case iff x acts £-ly in doing P.

There is then another parameter allowing certain verb classes to select for the postulate
in (1 1). This analysis faces all the problems in accounting for the systematic entailments
of sentences containing Purpose Clause modification discussed in Section 1.3. The fact
that (11) is lexically selected allows Chierchia to account for the restraint on a "positive
outcome" for PC-verbs: verbs with a positive outcome, like build, select for (11); verbs
without, like destroy, do not. If a verb selects for (11), then the Purpose Clauses with
which it occurs will be converted into arguments of the verb, and consequently fall
under the rule for control already adopted. This solution to the positive outcome
problem seems extremely ad hoc and addresses none of the problems in defining
exactly what set of verbs really do allow Purpose Clause modification. A similar lexical
stipulation will be required to account for the restriction concerning resultant states.
In line with the general consensus, Chierchia treats the Purpose Clause as an
unsaturated structure. However, in his theory, there is no distinction between empty
pronominals (such as PRO) and empty operators. His version of the structure
suggested by Chomsky (1981), given under (12), is that given under (13) (Pge 156,
his {14}).

[ 1 2] <9i[PRO to read ti to the children]

[ 1 3] Xxi^xj [xjto read xi to the children]

Note that in this he departs not only for the position consistently adopted by Chomksy,
but also the position adopted by Bach, who treats all infinitival subjects as free
pronominals and all objects as obligatorily controlled gaps.

70
 ses (Pge
We can now see how Chierchia accounts for control in Purpose Clau
157ff). (14) gives his account of a single-gap Purpose Clause.

[14] (a) John designed a battery to operate at high temperatures.

(b) E(<design'toJohn,a battery ,Xx[x operate at high temperatures']>)
(c) E(<design'to,John,a battery,<Xx[x operate at high temperatures'],a
battery»)
(d) E(<design'to,John,a battery,[a battery operate at high temperatures']>)

postulate in (5)
(14b) gives the prepositional content of (14a). Application of the control
operator, giving
will select the Theme of design to supply the argument for the lambda
(14c). Lambda conversion will give (14d). The correct result.
(281) gives the first stage of his account of a double-gap Purpose Clause.

[15] (a) John gave Mary a book to read.

(b) E(<give'to>John,a book,Mary,XxXy[y read x']>)
(c) E(<give'toJohn,a book,Mary,<X,xXy[y read x'],a book»)
(d) E(<give'toJohn,a book,Mary,Xy[y read a book']>)

postulate in
(15b) gives the propositional content of (15a). Application of the control
da operator,
(5) will select the Theme of give to supply the argument for the first lamb
that P in (5)
giving (15c). Lambda conversion will give (15d). Chierchia assumes
rule (5) will
ranges over both propositional and eventuality functions, which means that
apply to (15d) again (Pge 158). However, the account now breaks down.

[16] (a) E(<give'to»Joh n>a book,Mary,<Xy[y read a book1],a book»)

(b) E(<give'toJohn,a book,Mary,[a book read a book']>)

Application of (5) to (15d) will of course select the Theme of give once
again, giving
rchia therefore
(16a). Lambda conversion will then give (16b): an incorrect result. Chie
of theta-identity
introduces a disjointness requirement (Pge 158f). He defines the notion
updates the
as in (17), where =0 may be read as "has the same theta-role as". He then

(parametrised) control principle to that given in (18).

71
[17] (a) p=0y=dfe(p)=e(Y)
(b) a=er=dfVx[a(x)=9y]

[18] (a) E(pr[cc]) (A 6(P)=x) -> E(pr[a/a*0(p))])

(b) where a is any function and 9 is the highest 0-role on the scale in c such
that 6(<r, xi,...., xn>) is defined and a*0P
(c) Th > Go > Ag....
(d) the parenthesized part may be selected by specific types of eventualities.

To return to (15d), we can now reapply the control postulate in (18). Because of
the condition of thematic disjointness in (18b), the next highest role in the thematic
hierarchy (i.e. the Goal) will be selected as the argument for the second lambda
abstractor, giving (19a). Lambda conversion will give (19b): the correct result.

[19] (a) E(<give'toJohn,a book,Mary,<Xy[y read a book'],Mary>)

(b) E(<give'to»John,a book,Mary,[Mary read a book']>)

One problem to note in Chierchia's account is his rejection of any distinction

between the subject- and object-gap of a two-gap infinitive. As has been pointed out
above, where the Purpose Clause contains two gaps, the object-gap is obligatorily
controlled by the Theme but the the subject gap seems to be freer, being controlled by
some sort of "beneficiary". This beneficiary can be an explicit Goal or Agent, but it can
be an implicit entity at whose disposal the Theme is placed (however generally), as in
(20).

[20] John brought some wine along to drink with dinner (which was very
considerate given that he is teetotal).

As the parenthetical extention suggests, (20) can be true where John does not intend to
drink the wine himself, but intends it to be generally available. How would Chierchia's
account extend to such examples?

72
 As a starting point, Chierchia would have to assume that the "beneficiary"
relation is a standard thematic relation, Ben(0); this is a possibility he allows for
straight-forwardly (Pge 139). Certain predicates would then be allowed to select for
Ben(6) as an implicit argument. Bring would need to be one of these predicates. Given
that bring allows control by an implicit Beneficiary, it would then have to select for the
extended version of the control rule in (18). However, if bring were to select for the
extended control rule it should allow implicit control in general. As (21) shows, it does
not.

[21] *John brought for Mary to drink with dinner.

The first problem Therefore is that the beneficiary relation would have to be
characterised as a thematic relation to play a role in the account at all. Its presence
would, however, require bring to select the extended control rule: but the extended
control rule is appropriate only for the Beneficiary argument of bring not for its other
arguments. We cannot separate the Beneficiary relation from the Theme relation,
however, because selection of the simple or the extended control rule is a property of
predicates not the roles for which they subcategorise.
The second problem is that the implicit Beneficiary reading of (20) is only one
reading (and the dispreferred one): the natural interpretation is that John brought the
wine along because he intended to drink it. This reading would follow straight-
•*'•'
forwardly if bring selected for the simple version of the control rule, because, after the
Theme had saturated the object gap, the Agent (the next available role on the thematic
hierarchy) would be selected to saturate the subject gap. This would mean, however,
that bring selects both for the simple and for the extended version of the control rule.
This would miss the intuition that the reason the Agent is selected is that an Agent is
viewed (by default) as a Beneficiary of his own actions. If this default equation were
possible, then only the extended version of the control rule would be required.
However, this would mean that an entity could bear two 0-roles with regard to one

73
predicate (this is a possiblity, but one not mentioned by Chierchia here). This problem
is made particularly clear in the overt contrast between (22) and (23) (discussed in
Section 4.3.3).

[22] e/ to explain e^ to him1.

David** brought RuthJ* some articles^1*1"
[23] David1 brought RuthJ some articles* el to explain ek to herJ.

In both (22) and (23) the object gap is controlled by the Theme as expected. However,
there is then a choice with regard to the control of the subject gap: it can be controlled
by the Goal as in (22), with the pronoun bound by the Agent; or it can be controlled by
the Agent as in (23), with the pronoun bound by the Goal. The alternation does seem to
be linked to some notion of beneficiarity but it is not the expected one: in (22) there is
an implication that David will benefit from the explanation, but of course the subject gap
is coreferential with Ruth; in (23) the implication and the control reverses. This
however is a pragmatic factor concerning our understanding of explainings, particularly
in an academic context. It seems that the notion of Beneficiary must be distinguished
from the general notion of "who benefits" or the explanation in terms of an implicit
beneficiary role must be abandoned. In any case, (22-23) constitute clear
counterevidence to the thematic hierarchy in (18c).
Chierchia's account is a particularly clear and well-argued defence of a thematic
view and this clarity also brings out sharply the problems which such an account faces.

74
Section 2.2.4: The account of Jones (1991)

The most comprehensive work on constructions of purpose is that of Jones

(1991) who offers a detailed and wide-ranging discussion of the Purpose Clause,
supplemented by an explicit account of the Rationale Clause. His account draws heavily
on the works of three scholars: Edwin Williams (1980; 1981; 1987a,b; 1989), Gennaro
Chierchia (1984; 1989), and Emmon Bach (1982).
He develops his account of the syntax of the Purpose Clause from Edwin
Williams' Theta Theory, which is intended to offer a non-transformational account of
the A-system of Universal Grammar and of certain types of modification. The general
thrust of Jones' analysis is to argue that Purpose Clauses are bare VPs which predicate
themselves of the direct object (i.e. directly assign their undischarged external theta-role
to the direct object); the Rationale Clause on the other hand is a fully clausal structure,
loosely related to the main sentence in both syntax and semantics. In Jones' view, the
Rationale Clause is a saturated structure (a CP); the external 6-role of the infinitival verb
is assigned in the usual fashion to PRO in IP-Spec. The infinitival CP then freely
adjoins either to VP or to IP, as in (1-2). It is interpreted by some general notion of
responsibility (after Farkas (1988)).

[1] [ipJohn [yptVPdesigned a battery] [cpPRO to win a prize]]]

[2] [[n>John [ypdesigned a battery]] [cpPRO to win a prize]]

The subject-gap Purpose Clause is a bare VP, vertically-binding an external

theta-role. It adjoins to VP only and assigns its role by predication to the Theme (direct
object) of the main verb, as in (3).

[3] John* [yptVPdesigned a batteryj ][yp to operate at high temperatures))]

[AgJ...]

75
The object-gap Purpose Clause is also a bare VP; it vertically binds its internal
argument. It also adjoins to VP and assigns its (now) external role to the Theme (direct
object) of the main verb, as in (4).

[4] John* [ VPt VPbought a book) ] [VP to read to the children^]

[...ThJ]

Let us look at this account in more detail.

The account of the Rationale Clause is relatively uninteresting in that it posits the
weakest relations possible between the infinitive and its containing sentence in both
syntax and semantics: free adjunction and discourse control. As is clear from my own
i.

account, I view the relation between the Rationale Clause and the sentence it modifies as
being more strictly controlled and richer in content. In particular, I have argued for a
single adjunction point for the infinitive and derived differences between the Purpose
Clause, the Rationale Clause, and the Telic Clause on the basis of systematically higher
adjunction points. Jones attempts to exploit free adjunction to VP and IP to explain the
Rationale Clause's ability to appear inside and outside of the VP with regard to do so
anaphora and negative scope. In such an account, however, he would presumably have
to treat the Telic Clause (which is never discussed) as a CP adjunct. Such a high
adjunction point would seem to be untenable. My discussion of the local nature of
Agent-selection specifically argues against Jones1 view that some general notion of
discourse control is sufficient for capturing the constraints on the interpretation of this
construction.
Although seemingly straightforward, his account of the subject-gap Purpose
Clause also faces a number of problems. First, consider his treatment of to. He argues
that to forms part of the VP, independently of INFL; indeed he treats to as some sort of
verbal head, giving the structures in (5) and (6).

[51 [VP to [VP-.]1

[6] [VP-..1

76
He argues that the subject-gap Purpose Clause and the normal matrix VP are essentially
parallel in that they predicate some argument of an NP subject. But of course the
distributions of a bare VP and the phrase containing to are not parallel, as shown in (7-
8).

[7] John designed a battery [yp to operate at high temperatures].

[8] *John designed a battery [yp operate at high temperatures].

However, there is no prima facie reason, if both constructions are simply VPs, why
they should not be distributionally equivalent. The only reason that the two
constructions behave differently is the presence of to. Jones will later use to as a way of
specifying an event type (which will trigger the relevant semantic postulate for the
interpretation of the construction). He must presumably use the presence of to here to
govern the syntactic distribution, although for reasons that will become apparent he
cannot allow to itself to have selectional properties. Nor can he allow to to be an
element of INFL (as in Bach) because this would require a clausal structure, which
Jones is arguing against.
Jones reinforces this move away from to as an element of INFL in his
discussion of event variables. He claims that only a VP governed by INFL has an event
varbiable which can be bound by aspectual and tense operators. The Rationale Clause
has such a variable, but, he argues, the Purpose Clause does not. Therefore the to
cannot be in INFL. Jones is therefore strongly committed to the view that to occurs as
the head of an upper VP (which does no work); in the Rationale Clause to raises to
INFL. Notice that to is playing an important role in conditioning the syntactic
distribution of the construction and its semantic interpretation, because the to- VP seems
to escape the ban on VPs adjoining to a matrix VP purely because of the presence of to
and the infinitive receives its interpretation because it counts as a ro-event type;
however, to has no content, it is placed in an extra VP-shell just to get it out of the way,

77
and that shell serves no other function than to host to. Given the necessity of using to to
distinguish the bare VP from the infinitive, we have focused properties of the
construciton in to itself.
Jones1 account of the subject-gap infinitive clearly cannot be extended straight-
forwardly to the object-gap infinitive because it is not the external argument of the
infinitival verb which is coindexed with the matrix Theme, but the internal argument.
The external argument is either realised by an NP case-marked by for or it is left
implicit. Jones thus needs a rule for supressing the external argument of the Object-gap
Purpose Clause and forcing the externalisation of the internal argument. He therefore
develops an analogy v/ith the passive.
Jones (Pge 99ff) cites Williams' (1981) suggestion that passive morphology
triggers a lexical rule which suppresses the external argument of the verbal head
(leaving it as an implicit argument on the verb's internal subcat list) and which forces
the externalisation of the main internal argument: i.e. a passive verb becomes the sort of
verb which vertically binds its internal argument and predicates it of its subject.
Jones suggests that there is a phrasal counterpart to this rule which operates in
the syntax. This rule takes a VP which is vertically binding its external 0-role and
internalises that external 0-role as an implicit argument. This of course does not force
externalisation of the internal argument, for which Jones seeks some more "general"
motivation. He argues that a verb is listed in the lexicon for the number of external
arguments it projects (i.e. { VQ, Vi}). This marking cannot be changed. Therefore, if a
syntactic rule removes a vertically bound external argument, then a verb marked V]
must project some argument externally. In fact, the phrasal rule for argument
internalisation is not exactly the same as its lexical equivalent.
In adopting this account of the OPC, Jones creates a discrepancy between the
subject-gap Purpose Clause and the object-gap Purpose Clause: the SPC is a
straightforward VP directly predicating itself of the matrix direct object; the OPC is the
output of a special syntactic rule. Jones therefore suggests that the OPC account be
generalised to cover both types of Purpose Clause. The Externalisation rule applies to a
78
 VP (presumably triggered by the presence of to although this is not clear). One of the
arguments must be externalised. If it is the usual external argument we get an SPC, if it
is the usual internal argument, we get an OPC. This Jones illustrates, as in (9), where
bold is used to mark the lexically-specified external argument (Pge 102, his {62}).

[9]

|E(0):XP|

, ThJ>

Verb is marked for one

external argument (V1):
anti-internalisation
forces projection of one
index.

External index Internal index

projected projected

Subject-gap Purpose Clause: Object-gap Purpose Clause:

John* brought BillJ along [to read to the childrenJJ fohn1 brought a book) along [to read to the childrenJJ
[AgL]______ ______________[AgThJ]

We now have an account of the Purpose Clause as a construction which

obligatorily undergoes an externalisation rule and may then randomly externalise any of
its roles. The most serious problem for this view is the fact that adjuncts are subject to
externalisation just as readily as arguments. He observes (Pge 112f) that the PP

79
argument of put allows an NP from inside its Locational PP to be externalised and
argues that this is because the Locational preposition is invisible to the thematic rule.
This is fine (although of course the passive does not apply to this role) but then the
Location PP is an argument and not an adjunct. He also mentions cases such as (10).

[10] They gave me New York to fly to e.

Here the PP is not strictly subcategorised. However, as Jones (Pge 68) points out,
citing Larson (1988), this type of adjunct is rather special given that it does not iterate.

[11] *John flew to Kennedy Airport to New York.

Larson takes these PPs to be implicit arguments of verbs of motion but Jones calls them
constrained adjuncts. Given their intimate connection with the verb Jones proposes that
.«Hs';
they can be reanalysed so that the NP is an object of a [Verb + Preposition] constituent.
However, both of these cases are irrelevant to the argument. The simple fact is that the
object-gap Purpose Clause can occur with any adjunct.

[12] John bakes cakes in the kitchen.

The PP in (12) is a classic adjunct subject to iteration (modulo certain set/subset

constraints).

[13] John bakes cakes in a corner in the kitchen in the house...

Any theory that treats the PP in the kitchen as an argument of some sort will be theory
which assumes that there is no distinction between arguments and adjuncts, a position
Jones is clearly keen to avoid. And yet just such adjuncts are compatible with OPC in a
straightforward way.

[14] John built a kitchen to bake cakes in e.

80
It is hard to see how implicit argument structure or reanalysis could capture this
convincingly.
There is however another, more subtle, problem with the analysis in (9). The
pattern in (9) is intended to account for the fact that Purpose Clauses with both subject
and object gaps occur which seem to behave similarly in the syntax. There is however a
problem which is obscured by the use of read, a verb which is ambiguous between a
transitive and an intransitive reading. In Williams' original proposal, adopted initially
by Jones for OPC, the external theta-role becomes an implicit argument (i.e. in the
verb's argument structure but unrealised) and the internal argument is projected. In this
account, the marking on the external argument is erased which is not equivalent to
internalising it as an implicit argument and then any argument in the argument grid can
be externalised. The question is then what happens to the arguments which are not
selected for externalisation. One might reply that they must be projected in VP, but
clearly the formerly external role does not need to be projected, as we can see by the
example with null marking in (9b). One the other hand, we cannot say that the
arguments which remain can be implicit as a matter of course because where the
originally external argument is itself selected as external argument, as in (9a), then all
of the internal arguments must be projected. Notice that although Jones lists a Theme
with an index in (9a) there is no object and in fact this is an occurrence of intransitive
read. Notice that object of read in (9a) has the classic type of lexical restriction on its
reference: the object must be stereotypical reading material, it cannot mean any object
which Bill thinks that he can "read" such as tealeaves. This covert asymmetry becomes
clear with a strictly transitive verb such as hit, as in (15-18).

[ 15] John brought Bill) along [QPCfor Tom1 to hitp.

. (Agi, Thi)
[ 16] John brought Bill) along [QPCto hit]l
(Ag.ThJ)
[17] John brought Bill1 along [SPC*o hit TomJp.
(Agi, ThJ)
81
 [ 18] * John brought Bill* along [SPCto hit]*.
(Agi, Th)

(15-18) represent the four possible outputs of (9). One role at least must be
externalised by the Anti-Internalisation Principle and therefore one role at least must
bear an index. It is presumably the case that the other role can also bear an index (i.e. it
is syntactically realised): this is Jones1 explicit position. It is clear from his treatment of
the Agent in (9b), however, that the Agent is allowed to be implicit (i.e. syntactically
unrealised but semantically interpreted): this is represented by a null index on a listed 6-
role. This must also presumably be a possibility for the Theme. In (15) we have a
realised Agent and a realised Theme with the Theme externalised (OPC); in (16) we
have an implicit Agent with a realised Theme with the Theme externalised (OPC); in
(17) we have a realised Agent and a realised Theme with the Agent externalised (SPC);
and in (18) we have a realised Agent and an implicit Theme with the Agent externalised
(SPC).
The fact that there must be an asymmtery between a subject gap in a double-gap
structure and an object-gap is shown by the fact that (16) is well-formed but (18) is ill-
formed6. The construction allows the Agent to be "implicit" in (16) but it does not
allow the Theme to be implicit in (18). Only the lexeme hit itself can allow an implicit
Theme and it does not do so. There is, however, no apparent reason for this asymmetry
in Jones1 account. Indeed the whole point of Jones1 syntactic rule is that it removes all
the distinctions between the internal and the external arguments: "with a flat argument
structure, all arguments are on an equal basis with respect to external projection" (Pge
102). If Jones were to recognise that this asymmetry is actually present then his

6 Shalom Lappin has pointed out to me that (18) is in fact fine where we are dealing with games such
as baseball, i.e. John brought Bill along so that Bill could act as batsman. Notice however that the
interpretation of the implicit object is then not free in the way that the implicit subject is free in (16).
In (16) the subject has the same interpretation as an overt subject, with free choice for controller; in
(18), the object must be interpreted under strict conditions, that is, the Theme is something which is
typically hit in the context of the game (i.e. a pitched ball). The Jones analysis blurs the distinction
between a free implicit argument and the occurrence of an implicit canonical or stereotypical object.

82
 analysis would cease to be "unified": his only reason for adopting this more complex
approach in the first place. This confusion between implicit arguments and arguments
which have simply been internalised makes Jones' discussion on this point extremely
problematic. We will now look at Jones approach to the semantics of the constructions.
The semantics Jones (Pge 216ff) offers for the Purpose Clause (both SPC and
OPC) is essentially that of Chierchia (1989), repeated here as (19), although he does
not include the paranthetical extention of (19a), presumably as he does not think it
relevant for the verbs under discussion.

[19] (a) E(pr[cc]) (A 6(p)=x) -> E(pr[a/oc*e(p))])

(b) where a is any function and 6 is the highest 6-role on the scale in c such
that 0(<r, xi,...., xn>) is defined and a*Qp
(c) Th > Go > Ag....
(d) the parenthesized part may be selected by specific types of eventualities.

Jones points out, however, that this rule is too general. Chierchia assumes that the sorts
of verbs which allow tlie Purpose Clause are those that take a Theme. However, there
are some verbs which take a Theme which do not allow the Purpose Clause, as in (20)
(cf Jones 1991: 209).

[20] *John slapped Bill to read to the children.

Jones (1991: 216ff) claims, taking up an idea in Bach (1982), that the problem
with sentences like (20) is that the Theme of the verb does not end up in a "resultant
state" with the right sort of "latent thematic properties". Consider (21).
i
i

[21] We brought John in to talk to.

The Theme of the verb bring ends up in a resultant state: the thing which is brought
ends up in the location it is brought to. In this case, John ends up "in" some discourse-
specified space. Jones' claim (Pge 220) is that the Purpose Clause is licensed only in so
far as we judge John to be more likely to fulfill the Patient-like role of being talked to
83
when he is "in" than when he was "out". Bring has a deictic element which implies, in
(21), that we are also "in"; if John ends up "in" then he will obviously be more easy for
us to talk to. Send has the opposite deictic element and, as we would expect, send
cannot be substituted for bring in (21).

[22] *We sent John out to talk to.

i

This is because if we are "in" and John ends up "out" then he will be harder not easier
to talk to.
Jones therefore rewrites Chierchia's rule to that in (23) (Pge 223, his {35}).

[23] PC Control as Thematic Matching

E(p[p0])

He paraphrases this as saying that "a proposition E(p) about an eventuality p that has as
one of its constituents a predicate P0 that projects an agent/patient 6-role entails an
interpretation in which that predicate P0 is predicated of the argument of p that has
matching latent agent/patient connotations in p', the resultant state of p...". He
therefore accounts for the ill-formedness of examples such as (20) because the resultant
state of a verb such as slap fails to have the appropriate agent/patient connotations.
This adaption of the Chierchia rule is, however, extremely problematic. First,
there is the problem of defining the content of 0(p') in (23). If we say that the value of
0 is Agent or Patient, are we using those terms in the standard way? The answer must
be yes if we are not to reopen all the problems of 0-role definition that Chierchia has
attempted to avoid in his systematic entailment analysis (an analysis which is already
problematic, cf the discussion of Dowty (1989; 1991)). Jones initially seems to equate
Q(P) with the established 0-roles; he comments (Pge 220): "we have arrived at this
point by considering particular cases in which Theme arguments are acted upon in ways
that enhance possible subsequent participation as agent or patient". Whatever "possible
subsequent participation" might mean, the notions Agent and Patient in this passage
84
seem to be the standard ones. However, Jones continues throughout his subsequent
discussion to refer to the roles, 6(p'), as "agent/patient connotations". The term
connotation implies that 6((?) defines a set of entailments which are very similar to the
set of entailments which one gets with Ag(6) or Pa(6) but which are not the same. This
slippage in Jones' treatment of the relation, 6(p'), flows from a second element in his
control rule in (23), the |i-operator.
The (Li-operator represents the second major problem with the rule in (23). Jones
states (Pge 223) that the ^-operator will "indicate the mediation of "latency" in the
agent/patient connotations", a definition which is essentially obscure. He attempts to
give some content to this definition by a discussion of the notion of latency. Consider
the sentence in (24) (Pge 219, his {28}).

[24] John put the glass on the edge of the table.

The putting of the glass on the edge of the table has the resultant state of the glass being
7
on the edge of the table, as shown in (26) .

[26] John put the glass on the edge of the table

The glass is on the edge of the table

It is in this resultant state that the relevant notion of latency occurs. Jones comments
(Pge 219) that "among the things we know about situations like the one in (24) is the
potential for the glass to fall... this kind of property is not a property of the glass itself,
nevertheless it seems to be something we know about the glass in (24)... the property
of being liable to fall is the kind of "latent" property I have in mind". Notice that if this

Note, incidentally, that this is not a resultant state as defined by Bach (1982), which would require the
definition in (25).

[25] The glass is not on the edge of the table

The glass is on the edge of the table

Jones does not discuss this technical problem and I set it aside here.
85
is the case then the notion of latency must concern some sort of modal relation between
some particular state and the events which might follow it. Consider (27).

[27] [The glass is on the edge of the table The glass falls to the floor]

We might paraphrase (27) as "it is possible that the glass being on the edge of the table
will lead to the glass falling to the floor". If this is what Jones intends, then we have a
further problem, as there are of course many possible outcomes of the state in question.
It is possible that the glass will be picked up and wrapped as a present for someone,
that it will be left there undisturbed for centuries, that it will be smashed by the singing
of the resident Russian soprano, etc. These possible outcomes are not the ones which
interest us. What seems to be needed is some notion of a "most probable outcome": the
glass is more likely to fall when it is on the edge of the table than it is if it is at the centre
(a fact on which Jones explicitly comments, Pge 219). But a notion of "modal
probability" is likely to be even more difficult to define than notion of "modal
t

possibility": are we, for instance, to define modal probability by asking whether more
" t

possible worlds have the glass falling off the table as a subsequent event than any other
outcome?
The difficulty is not helped by Jones1 ambivalence over the content of the \i-
operator. As we have discussed it here, it is the relation between a state (in this case a
resultant state of put) and an event which is the "outcome", in some sense, of that state.
Jones, however, continues to characterise the relation purely in terms of the resultant
state itself (and its relation to the main event whose resultant state it is). He comments
(Pge 219) that "this property of the glass, its potential to fall, arises directly as a result
of the put action; it is a property of the glass in the resultant state of this particular
instance of a put relation". But it is not of course a property of the pwr-relation itself;
rather it is a property of the resultant state which put specifies. By discussing the |i-
operator purely in terms of the resultant state and the verbal relation which specifies it,

86
Jones allows himself to introduce discussion of Agentive intentions, which are in fact
strictly irrelevant to the notion of "latency". Consider the sentence in (28).

[28] John put the glass on the edge of the table for the dog to knock e off.

As we have seen, the resultant state of put is the glass being on the edge of the
table. This resultant state is related by some notion of "modal probability" to a
subsequent event, namely, the glass falling off the table. John's intentions can be
described precisely in the light of the modal relation between the resultant state of put
and the "probable subsequent event": John acts as he does to achieve some aim (the dog
knocking the glass off the table), his choice of action determined by the nature of the
i

probable subsequent events of the resultant state of his action. However, Jones avoids
discussion of probable subsequent events (presumably because they will be rather
difficult to define); as we saw above, he attempts to claim that the properties of resultant
states which he discusses, and the intentions that are related to them in the Purpose
Clause, as simply properties of the main verbal relation itself.
The crucial point as discussed in Section 5.1.1 seems to be that the Purpose
Clause must apply to a state which is entailed to exist by the verb and that that state
must have causal efficacy. Attempting to formulate causal efficacy in terms of latent
properties of subsequent states to resultant states of events (where the properties are to
be understood as thenatic properties) seems to say too much. In this respect I follow
Bach in assuming that how we understand what properties the objects have in the states
and how they are causally efficacious falls outside of linguistic theory proper and
belongs to natural language metaphysics.
Jones' view leads to a peculiarity in his exposition of how the resultant state
analysis discounts examples such as (20) above, repeated here as (29), a peculiarity
which obscures a real problem with the notion of resultant state itself.

[291 *John slapped Bill to read to the children.

87
Jones comments (Pge 221) that "slapping [Bill] and ignoring him are not typically
actions that have as their goal an enhancement of [Bill's] subsequent possible
agentivity". Jones therefore explains the ill-formedness of (29) by saying that Agents
cannot have intentions of the appropriate type towards them, namely, they cannot intend
that by slapping the Theme entity they will increase the likelihood of Agent-like or
Patient-like properties in that entity. This simply isn't true: I might think it a very
efficient means of making Bill Agentive. "Verbs of violence" are in fact very likely to
have the effect on the Theme which the Agent desires, as in (30).

[30] John beat Bill into opening the safe.

Beating Bill might be considered a very effective means of getting him to open the safe.
Yet the verb, beat, does not allow the Purpose Clause construction.

[31] *John beat Bill to open the safe.

The confusion only arises if one focuses on the intentions of Agents in their
actions, rather than on the nature of the resultant states involved. The problem with (29)
and (31) in fact seems to be that the main verbs of these sentences do not have resultant
states of the appropriate type at all. It is clear that these events do have resultant states of
some sort: Bill will for one thing be very bruised and emotionally shaken as a result of
the slapping or beating he has taken. The problem lies in defining what a linguistically
relevant "resultant state" actually is: for instance, in neither (29) nor (31) does Bill end
up in a location lexically specified by the verb. The real problem with explaining the ill-
formedness of sentences containing "verbs of violence" lies in defining how those
verbs fail to have resultant states of a relevant type. This question is in fact the most
basic one that needs to be addressed in any analysis that uses the notion of a "resultant
state". By raising the problem of modal relations between states and subsequent events
and by then attempting to turn that into a discussion of Agent's intentions, Jones side-
steps what seems to be the central problem.
88
 CHAPTER 3: The External Syntax of Infinitives of
Result

Section 3.1: The tests themselves

The question that is addressed in the external syntax of a construction is at

which point the phrase or construction must be generated in the phrase structure tree;
the basic orientation point for answering such a question is the VP: is the constituent
VP-internal, VP-adjoined, VP-external, or entirely promiscuous in this respect? The
standard means of answering this question involves the use of various tests which
distinguish between VP-internal elements (all subcategorised material as well as object-
oriented modifiers), VP-adjoined elements (certain adverbials, usually
subject/Agent/event-oriented), and VP-external elements (other adverbials). These tests
fall into two groups: those involving substitution or anaphora (do so, pseudoclefting)
and those involving movement (VP-fronting, though-movement). In the section which
follows I will discuss the nature of these tests and point out particular features which
will be relevant in our discussion of infinitives of result.
It has standardly been observed that the phrase do so must replace not only the
verb but also the elements for which it subcategorises.

[1] *Ruth gave an article to David on Monday and Kerstin did so a monograph to
Jo on Tuesday.
[2] *Ruth gave ?::t article to David on Monday and Kerstin did so to Jo on
Tuesday.
[3] Ruth gave an article to David on Monday and Kerstin did so on Tuesday.
[4] Ruth gave ar article to David on Monday and Kerstin did so TOO.

Bold font is used to indicate the elements whose interpretation is included in the
interpretation of do so. In (1), do so is anaphoric only to the verb give, exluding the
direct object, the indirect object, and the adverb of time: the sentence is ill-formed. In
(2), do so is anaphoric to the verb and its direct object, but not to the indirect object or
89
the adverb of time: the sentence is again ill-formed. In (3), do so is anaphoric to the
verb and both its objects, but not to the adverb: this sentence is well-formed. Notice that
the adverb here takes a contrastive reading: all other aspects of the event were the same;
they differed solely with respect to time of occurrence. In (4), do so is anaphoric to the
verb, both its objects, and the adverb of time: the sentence is still well-formed. Notice
that the interpretation of the adverb really is present, i.e. Kerstin also gave David the
article on Monday. The strict identity of the event in which Ruth is Agent and the event
in which Kerstin is Agent is marked by the presence of too. Notice that including too in
(3), though marginally acceptable is extremely odd.

[5] ??Ruth gave an article to David on Monday and Kerstin did so TOO on
Tuesday.

The problem is that too implies identity of the events in which each individual was
Agent, but the inclusion of an adverb of time in the second conjunct asserts a difference
between them. I therefore take it that where inclusion of too is unnatural the adverb is in
the same class as adverbs of time, but that where inclusion of too improves the reading
that the modifier in the second conjunct is less closely integrated with the verb.
These facts are standardly accounted for, and follow straightforwardly, if it is
assumed (i) that the phrase do so is an anaphor which replaces VP; (ii) that all
subcategorised phrases are included within the projection of the verb (i.e. there is no
node, X, c-commanded by a subcategorised phrase where X=VP); (iii) that certain
adverbs, including adverbs of time, Chomsky-adjoin to VP (i.e. have a sister and
mother node which are both VPs); and (iv) that all other elements are excluded from VP
(i.e. there is no node, X, dominating these other elements, where X=VP). This would
give the skeletal phrase structure in (6).
[6]
X
«^

VP OTHER ADVERB

TIME-CLASS ADVERB

V OBJECT1 OBJECTn
90"
If do so replaces (or is anaphoric to) VP, then it will always replace the verb and its
objects (which are always included in the minimal VP); it will never replace the higher
adverb classes which are never included in a VP; and it will optionally replace the time-
class of adverbs because it can either replace the sister VP or the mother VP. Where the
sister VP is replaced, the time adverb takes a contrastive reading. Where the mother VP
is replaced, there is an identity reading. Notice that too is only included naturally where
all VP-material has been replaced.

[7] Mrs White attacked Rev. Green in the drawing-room with the candlestick
and Miss Scarlet did so in the kitchen with the dagger.
[8] Mrs White attacked Rev, Green in the drawing-room with the
candlestick and Miss Scarlet did so with the dagger.
[9] ??Mrs White attacked Rev. Green in the drawing-room with the
candlestick and Miss Scarlet did so TOO with the dagger.
[10] Mrs White attacked Rev. Green in the drawing-room with the
candlestick and Miss Scarlet did so TOO.

We can therefore assume that when too occurs naturally it indicates that the maximal VP
has been replaced. In a structure such as (11), tooinsertion will only occur in a natural
reading where VP^ h?s been replaced.

11]

VP4

VP3 ADV

Notice incidentally that do so really is replacing the lexical VP. It seems that it is not
enough simply to be pan of the extended projection of the verb. Consider the examples
in (12-16).

91
[12] John must have been shouting...
[13] ...and Bill must have been doing so too.
[14] *...and Bill must have done so too.
[15] *...and Bill must do so too.
[16] *...and Bill does so too.

In (13), do so has replaced the lexical verb shouting. In (14), we must try to interpret
do so as having replaced both the lexical verb and the Progressive. This is impossible.
The sentence is only well-formed on a reading where Bill's action is interpreted purely
as a Perfective: i.e. the Progressive is not included in the interpretation of do so, it has
simply dropped out. Similarly with (15), the sentence is only interpreted as well-formed
where the Perfective and Progressive have dropped out: do so does not include any
aspectual meaning at all. And similarly with (16), the modal meaning has dropped out
completely. This means that do so is purely an anaphor of the lexical VP: it cannot
replace whatever functional phrases mediate aspectual and modal interpretation of the
verb. In the crude phrase structure shown in (17), do so will replace only the VP.

[17]

M AuxP

Pert AuxP

Prog VP

The same can be said of the pseudocleft construction.

[ 18] What John must have been doing was reading.

[ 19] * What John must have was been reading.
[20] *What John must was have been reading.

VP-anaphora tests are therefore restricted to the lexical VP.

Within the lexical VP, the pseudo-cleft construction isolates the same elements
as do so. What in (21) is a vv/z-VP.
92
[21] What John did t was leave.

In this construction, a w/z-clause which is subject of copular be isolates some

constituent whose value is given by the Complement of the copular. The VP can be
questioned using what, triggering do-support, as in (21). The trace marks the position
of VP; and the value of VP is established by the copular Complement, [ypleave].
As we would expect, the w/z-element must replace not only the verb but also its
subcategorised Complements. The judgements are in fact much stronger than for do so.

[22] *What Ruth did t an article to David on Sunday was give.

[23] *What Ruth did t to David on Sunday was give an article.
[24] What Ruth did t on Sunday was give an article to David.
[25] What Ruth did t was give an article to David on Sunday.

In (22) what has replaced only the verb and the sentence is ill-formed. In (23) what has
replaced the verb and its direct object, but not the indirect object, and the sentence is ill-
formed. In (24), what has replaced the verb and both its objects, stranding the adverb,
and the sentence is well-formed. The interpretation of (25), however, requires caution.
As expected, association of the time adverb with the subcategorised VP constituents is
well-formed. It seems, however, that we cannot deduce from this that the adverb is
actually a part of VP itself.
There are adverbials which we clearly do not want to include in VP itself.
Consider the interpretation of with any luck in (26).

[26] Ruth gave David the article, with any luck.

The phrase with any luck is what is traditionally known as a disjunct and what is
sometimes known in the formal literature as a pragmatic adverb: it offers a personal
assessment, evaluation, or comment on the content of the sentence which precedes it.
Notice that it is marked off by comma intonation. As expected, the disjunct can appear

93
in the first part of a pseudocleft construction although heavy comma intonation is
required.

[27] What Ruth did ?, with any luck, was give David the article.
<

However, it can also appear in final position, as in (28).

[28] What Ruth did t was give David the article, with any luck.

This implies that any adverbial can appear in final position in the pseudocleft
construction. We cannot deduce from (28) that the disjunct is part of the VP, as in (29),
but must allow the possibility that it is an additional comment as in (30).

[29] What Ruth did t was give David the article, with any luck.
[30] What Ruth did t was give David the article, with any luck.

This implies that the only reliable contribution that the pseudocleft construction can
make to constituency testing is in what elements are allowed within the cleft itself,
rather than what can appear after the copular.
Consider also the behaviour of this disjunct in the do ^-construction.

[31] Ruth gave David the article, with any luck, and Kerstin did so too.

There are several points of interest to note here. First, the speaker implies that it is a
lucky thing that Ruth gave David the article; however, no evaluation is expressed
concerning Kerstin's action. The speaker may consider that also to be a lucky thing but
they may also have an entirely neutral attitude to it. That means that the VP-anaphor do
so does not carry over the interpretation of the disjunct, despite the fact that it is
stranded outside of the VP in the first conjunct; even the presence of too, the maximal
identity marker, does not force us to interpret the speaker as committing themselves to
an identical evaluation of the second conjunct.

94
 It is difficult to set up any sort of constrast between the disjunct in the first
conjunct and some other disjunct in the second conjunct. (32) is a possible example.

[32] ?Ruth gave David her article, with any luck, but Kerstin might have done so
too, unfortunately.

Notice, however, that too is still here, marking maximal identity between the two VPs:
with any luck and unfortunately cannot therefore represent a contrast between the two
VPs but between the tv/o sentences. Disjuncts therefore pose a general problem for the
VP-anaphora tests: they are more or less acceptable in the relevant distributions but they
have nothing to do with construction itself. These tests must therefore be treated with
caution. These issues will necessarily be raised again during a discussion of the external
syntax of the Telic Clause.
The do so and VP-pseudocleft constructions test for VP constituency by
substitution or co-reference. There are however other rules which are sensitive to VP
constituency: VP-fronting and though-movemeni. Both of these rules involve
topicalisation of a bare VP, in the case of VP-fronting, out of a root sentence, and in the
case of though-movement, out of a (concessive) subordinate clause governed by
• t '•

though. It is assumed that phrases which can topicalise with the VP form part of the VP
structure and those which do not, do not. It seems, however, that this test is only
consistent with VP-anaphora if it is assumed that the maximal VP alone can topicalise.
Consider first the cases of VP-fronting, applied to the base sentences in (33).

[33] (a) Ruth gave David the code at the conference.

(b) Ruth gave the code to David at the conference.

[34] *Give, Ruth may David the code at the conference, but...
[35] *Give David, Ruth may the code at the conference,but...
[36] *Give the code, Ruth may to David at the conference, but.
[37] ???Give David the code, Ruth may at the conference, but..
[38] ?Give David the code at the conference, Ruth may, but...

95
One of the problems with this test is that the preposed elements degrade quickly on the
basis of stylistic factors concerning weight (I have therefore used the double object
construction for topicalisation in (35) rather than the heavier prepositional object
version). It is clear, however, that it is quite ungrammatical to topicalise the verb
independently of its objects, as in (34-36). Topicalising the verb, its objects, and
time/place adverbs, seems to be acceptable, if stylistically heavy, as in (38).
Surprisingly, however, stranding of the VP-adverbs, as in (37) seems to me ill-formed.
Intuitively these topicalisation constructions place great weight on the action performed
by the Agent and it seems that all aspects of that event must be treated as important for
the construction to work. We cannot therefore use VP-topicalisation as a test for
Chomsky-adjunction, only for VP-internal constituency. Notice that, as we would
expect, disjuncts occur naturally in untopicalised position.

[39] ?Give David the code Ruth may, with any luck, but...

However, it can appear in pre-subject position also (if more marginally).

[40] ??Give David the code, with any luck, Ruth may, but...

Again, this is an ambivalent piece of evidence, though, as disjuncts can generally

appear in sentence initial position as it is.

[41] With any luck, Ruth may give David the code at the conference.

The general rule with disjuncts is therefore caution.

Though-movement is generally parallel to VP-fronting.

[42] *Give, though Ruth may David the code at the conference...
[43] *Give David, though Ruth may the code at the conference...
[44] *Give the code, though Ruth may to David at the conference...
[45] ??Give David the code, though Ruth may at the conference...
[46] Give David the code at the conference, though Ruth may...

96
As expected, the verb and its objects must be topicalised out of the r/zowg/z-clause,
otherwise the sentence will be ill-formed, as in (42-43). Where the verb, its object, and
the relevant adverbs are topicalised, the result is well-formed, as in (46). It seems
though that in this construction the adverbs can be stranded, as in (45), without being
too marginal. In this case though the adverb takes a sort of after-thought reading. It is
interesting to note that disjuncts to not seem to occur particularly naturally in these
contexts in any position.

[47] *Give David the code, with any luck, though Ruth may...
[48] ?*Give David the code, though Ruth may, with any luck,...

Perhaps evaluative modifiers do not occur naturally in concessive contexts.

One point should also be made here concerning object-oriented modifiers, such
as the AP, cold, in (49).

[49] The waiter served the soup cold.

Syntactic tests place this modifier unambiguously within VP: the modifier may not be
separated from VP by pseudoclefting, as shown in (50), or by preposing, as shown in
(51):

[50] *What the waiter did cold was serve the soup.
[51] *Serving the soup he may have been cold...

However, the AP is clearly acting as a modifier and other modifiers may be added
iteratively, just as we have seen for the VP-adjuncts.

[52] The waiter served the soup [cold] [with cream] [in a wooden bowl].

This suggests that there must be a position within VP available for recursive adjunction,
a position not offered by the traditional VP-rule in (53).
97
[53] VP -» V XP*

Within X'-theory there is of course the position V, and I take object-oriented modifiers
to be evidence for an X'-account of VP. Notice that this is strictly independent of the
issue of binary branching (cf Kayne 1983). I will therefore assume that the traditional
.-"?
i
account of VP is inadequate for an analysis of object-modifiers including the Purpose
Clause, prefering theories incorporating X'-structure.

98
Section 3.2: The Purpose Clause

One of the striking properties of the Purpose Clause is its incompatibility with
the w/i-clause of a pseudocleft construction (cf Faraci (1974), Kirkpatrick (1982),
Jones (1985; 1991), an incompatibility which holds for both the subject-gap Purpose
Clause and the object-gap Purpose Clause, as shown in (1-2). I have marked the phrase
which acts as antecedent to what in bold font and I take this to be a VP.

[1] *What Sven did to put in the fridge was take the ice cream into the
kitchen.
[2] *What Sven did to put the ice cream in the fridge was send Sigrun into the
kitchen.

The sentence in (1) is simply ill-formed: it has a gap in object position which must be
controlled by the ice cream and the ice cream is inaccessible as a controller. In (2), the
sentence will only be well-formed on a Rationale Clause reading where Sven is going to
put the ice cream in the fridge (in context an extremely odd reading); the subject-gap
Purpose Clause reading in which Sigrun is to put the ice cream in the fridge is
unavailable. This restriction will be accounted for straightforwardly if the Purpose
Clause must be a part of the VP. Notice that where the Purpose Clause forms part of the
copular Complement, as in (3-4), or where the object itself (independently of the VP) is
the focus of the pseudocleft, as in (5-6), the sentences are well-formed as expected.

[3] What Sven did was take the ice cream into the kitchen to put in the fridge.
[4] What Sven did was send Sigrun into the kitchen to put the ice cream in the
fridge.
[5] What Sven took into the kitchen to put in the fridge was the ice cream.
[6] (The person) who Sven sent into the kitchen to put the ice cream in the fridge
was Sigrun.

Judgements are the same, if less robust, with do so anaphora (cf data from
Jones (1991: 61) on VP-deletion).

[7] *Sven took his yoghurt into the kitchen to eat and Sigrun did so to put
in the fridge.
99
[8] *Sigrun sent Sven into the kitchen to find her a yoghurt and Anna did so
to cook her breakfast.

The sentences become well-formed when the Purpose Clause is dropped from the
second conjunct.

[9] Sven took his yoghurt into the kitchen to eat and Sigrun did so too.
[10] Sigrun sent Sven into the kitchen to find her a yoghurt and Anna did
so too.

Notice, however, that the interpretation of the Purpose Clause in the first conjunct is
included in the interpretation of do so, as indicated by the bold font: the identity of the
action performed by Sven and Sigrun in (9) and Sigrun and Anna in (10) (modulo the
sloppy identity of pronouns) is marked by the use of too clause-finally.
The tests used above to indicate the VP-internal status of the Purpose Clause
involve substitution, but it seems that movement rules are also sensitive to this,
although the test is less effective because the constructions which result are stylistically
marked anyway. First there is straightforward VP-preposing in which the bare VP is
fronted (cf Jones 1985). In this case it seems that the Purpose Clause must be fronted
with the VP.

[11] ...and [bring it along to drink] he did t.

[12] ...and [bring him along to drink the wine] he did r.
[13] *...and [bring it along] he did t to drink.
[14] *...and [bring him along] he did t to drink the wine.

In (11) we have an object-gap Purpose Clause which is fronted with the VP; the
sentence is well-formed. In (13) we have the same sentence, with the Purpose Clause
stranded outside the VP (whose base position is marked by a trace); the sentence is ill-
formed. In (12) we have a subject-gap Purpose Clause which is fronted with the VP;
the sentence is again well-formed. In (14) the Purpose Clause is stranded outside the
VP, and the sentence is ill-formed.

100
 Similar judgements occur when a VP is preposed outside of a subordinate
clause headed by though (though-movemeni).

[15] [Bring it along to drink] though he may...

[16] [Bring him along to drink the wine] though he may...
[17] *[Bring it along] though he may to drink...
[18] ?*[Bring him along] though he may to drink the wine...

In (15) we have a VP and object-gap Purpose Clause fronted out of a though-clause and
the result is well-formed. In (17), the Purpose Clause has been stranded inside the
though-clause and the result is ill-formed. In (16) we have a VP and subject-gap
Purpose Clause fronted out of a though-clause and the result is well-formed. In (18) the
Purpose Clause has again been stranded and the result is ill-formed. It does seem
however that the subject-gap Purpose Clauses are slightly more acceptable.
Evidence from VP-rules of various types therefore provides fairly strong
evidence that the Purpose Clause forms a part of the minimal VP. Another set of facts
support this view of intimate association with the verb. It seems that the Purpose Clause
always falls within the same scope as the verb in negative contexts (cf Jones 1991).
Consider a sentence such as (19).

[19] Nick didn't bake the cake to serve to Ute.

This should be ambiguous between two readings, represented schematically as in (20-

21).

[20] [Nick bake the cake] [NOT[to serve to Ute]]

[21] *[NOT[Nick bake the cake]] [to serve to Ute]

In (20), Nick does bake the cake but it was not his purpose in doing so to serve it to
Ute. In (21), Nick does not bake the cake and his intention in doing so is to serve it to
Ute. (21) is obviously not a possible reading of (19) which is unambiguous. These are
the standard scope facts. I take them to be generally accurate with two provisos. First,

101
the acceptable reading of (19) is not strictly as in (20). (20) asserts that Nick baked a
cake and denies that the purpose of the cake was to serve to Ute (i.e. it is focus
negation). In fact, (19) merely asserts that it was not the case that Nick baked a cake to
serve to Ute. Whether he did or did not bake the cake is indeterminate, i.e. negation
here is true wide-scope sentential negation, as in (22).

[22] [NOT [[Nick bake the cake] [to serve to Ute]]]

The excluded reading is however as shown in (21). Unfortunately this is a

potentially unhelpful example. The intuitive reason that (21) is ill-formed is that if Nick
did not bake a cake then he doesn't have a cake to serve to Ute, i.e. the reading is
excluded on semantic grounds which derive from notions of resultant states. Most
examples of Purpose Clauses in fact fall under this explanation and an analysis will
need to capture this fact. However, it is possible to come up with examples where verb-
negation is at least plausible, as in (23).

[23] Nick didn't cook the salmon to serve to Ute.

Here we have the standard reading, as in (24), namely that it was not the case that Nick
cooked the salmon to serve to Ute. However, we now have plausible alternative
reading, as in (25): imagine that Ute is a wild fan of sushi but that Nick's other friends
are not; he might cook the salmon for other guests at his dinner party, but leave it
uncooked for Ute.

[24] [NOT[[Nick cook the salmon] [to serve to Ute]]]

[25] ?*[NOT [Nick cook the salmon]] [to serve to Ute]]

I have to admit that I find this reading very strained; given the pragmatic plausibility of
the reading I therefore take it that what acceptability it has, it has by virtue of contextual
support, and that this scope is in fact grammatically impossible. This provides much

102
better evidence than the non-ambiguity of (19) derives from a scope restriction imposed
by the syntax rather than some more general semantic problem.
It seems that this pattern also holds for negated verbs of destruction, although
the judgements are rather delicate. As Bach (1982) observed, the Purpose Clause is
generally only possible with verbs that have a resultant state which is in some sense
positive, as in (26), rather than negative, as in (27) (a view we will discuss in some
detail in Section 5.1.1).

[26] Nick set up the files to show to Jo.

[27] *Nick destroyed the files to show to Jo.

Once again, the restriction in (27) has an obvious intuitive reason: if Nick destroys the
files then he doesn't have anything left to show to Jo. The negation of (27), on this line
of thought, has a perfectly plausible interpretation.

[28] ?*Nick didn't destroy the files to show to Jo.

Nick didn't destroy the files, i.e. he still has them, and that allows him to show them to
Jo. Once again I find this reading extremely strained (a sort of resumptive reading
where we supply a pronoun for the gap). This implies that the Purpose Clause cannot
fall outside of the scope of negation. It also implies that the Purpose Clause does not
modify any constituent which can include negation and in fact that it is sensitive to
lexically specified properties of the verb's event structure.
It is trivially obvious that the Purpose Clause also falls within the scope of
interrogation.

[29] Did Nick cook the salmon to serve to his guests?

The Purpose Clause in (29) forms a part of the question. Notice also, however, that it
can form the focus of the question quite naturally.

103
[30] Did Nick cook the salmon to serve to his guests or to have for himself?

Infinitives of result seem in general to allow this, although the Telic Clause is rather
marginal (see Section 3.4).
These tests provide strong evidence that the Purpose Clause must be contained
within the minimal VP (i.e. there must be no other VP within the c-domain of the
Purpose Clause). This suggests a structure as in (31).

31]
VPmin
.*-—•«*.
Vn Purpose Clause

As we will see in the following sections no other infinitive of result can occur within
this configuration. There is however another type of infinitival modifier which can
occur within VPmin, the Infinitival Relative (cf discussion in Faraci (1974) and Jones
(1985; 1991) amongst others). Where the Purpose Clause is a modifier of the verb, the
Infinitival Relative is a modifier of the noun. An infinitive which takes the nominal in
object position as its antecedent will usually be ambiguous between the two readings.
Consider the sentence in (32).

[32] Sven took the video e to show e to his mother.

In the Infinitival Relative reading the infinitive tells us something else about the video; it
is in fact a restrictive relative and therefore implies that there is some set of videos and
that this is the one which Sven intends to show to his mother. This infinitive therefore
occurs naturally with the N'-anaphor one in a contrastive reading.

[33] Sven took the video e to show e to his mother but not the one e to show e to his
father.

104
It does not occur with the pro-NP it which replaces the whole noun phrase containing
the infinitive.

[34] *Sven took it e to show e to his mother but not the one e to show e to his father.

Similar effects occur with the passive. The Infinitival Relative must move with the noun
it modifies, as in (35).

[35] The video e to show e to his mother was taken by Sven.

The pro-NP it must replace the whole [nominal + infinitive] string otherwise there is ill-
formedness, as in (36).

[36] *It e to show e to his mother was taken by Sven.

The Purpose Clause shows the inverse behaviour: it modifies the verb and is
entirely independent of the NP. In (37) we have two different purposes of the event of
not taking the video.

[37] Sven took the video e to show e to his mother but not e to show e to his father.

There is one video in (37) and two purposes, not two videos each with a distinct
purpose, as in (33). (37) is perfectly compatible with the pro-NP construction.

[38] Sven took it e to show e to his mother but not e to show e to his father.

In the passive, the Purpose Clause must be left clause-finally, as in (39).

[39] The video was taken by Sven e to show e to his mother.

This construction is then perfectly compatible with pro-NP as expected.

[40] It was taken by Sven e to show e to his mother.

105
 From these observations it follows that the Infinitival Relative occurs within
NP, as in (41).

41]

NP
1
Spec
r-
NT
N' infinitival Relative
1
N

We already know that the Purpose Clause must occur within the minimal VP. The
above observations show that it must occur outside of the direct object NP. One
possibility would be to suggest that the Purpose Clause adjoins to the NP rather than
being included within it, as in (42).

42]

NP Purpose Clause

Spec N1

N' Infinitival Relative

This would capture the fact that the Purpose Clause can occur outside of the NP (as it
has an NP sister). However, it also suggests that that Purpose Clause can ambiguously
appear inside and outside of pro-NP, which as we have seen is not the case. Notice in
particular that if (42) were the structure, we would expect (37) to imply the presence of

106
two videos. In fact it merely implies two purposes for the main event. This strongly
suggests that the Purpose Clause is not a nominal modifier of any sort. Given that the
Purpose Clause must occur within the minimal VP but that it cannot be associated with
a direct object NP, and given binary branching, this naturally suggests adjunction to V,
as in (43).

43]

VP
1
Spec V

V
r Purpose Clause
rV NP

107
Section 3.3: The Rationale Clause

The Rationale Clause shows the classic properties of a VP-adjunct, namely that
rules of substitution and movement which apply to the VP apply to the Rationale Clause
optionally, a fact which would follow under Chomsky-adjunction because the rules
could apply either to the VP to which the Rationale Clause adjoins or to the VP mother
of the Rationale Clause, created by the adjunction rule itself.
In cases of substitution this means that the Rationale Clause can either be
replaced by the VP-anaphor or left "outside" the substitution. For instance, consider the
cases of pseudo-clefting in (1-2).

[1] What Ute did t to please her friends was cook a salmon dinner.
[2] What Ute did t was cook a salmon dinner to please her friends.

In (1), the Rationale Clause is left outside of the wh-VP, whose antecedent is
Complement of the copular, a simple VP; I follow Kirkpatrick (1982) and Jones (1991)
in assuming that (1) is well-formed, pace Faraci (1974). In (2), the Rationale Clause
forms part of the antecedent VP itself.
Similar judgements occur with do so-anaphora (again compare with data in
Jones (1991: 61) on VP-deletion).

[3] Ute cooked dinner to please her friends and Sigrun did so too.
[4] Ute cooked dinner to please her friends and Sven did so to get out of the
washing-up.

In (3), do so takes as its antecedent the entire [VP + Rationale Clause] string: notice that
(3) asserts that Sigrun also cooked dinner with the intention of pleasing her friends
(again, sloppy identity). Unless do so is ambiguous in the antecedents it allows, this
implies that the [VP + Rationale Clause] string form a VP constituent. As expected,
however, the modified VP can itself be replaced by do so, excluding the Rationale

108
Clause, to give the well-formed (4). It therefore appears that the Rationale Clause is
Chomsky-adjoined to VP, forming a split VP.
Movement rules which apply to VP also seem to respect this segmentation
(although as suggested above, these tests are less effective given the stylistic weight).
Consider the case of straightforward VP-preposing (cf Jones 1985).

[5] ...and [get wildly drunk to offend the guests] John did.
[6] ...and [get wildly drunk] John did, to offend the guests.

It seems to be possible to prepose the [VP + Rationale Clause] as in (5) or to leave the
Rationale Clause in situ and prepose just the minimal VP segment, as in (6). There is a
problem here, however. The force of the Rationale Clause is different in (5-6). In (5)
the Rationale Clause is taken to be part and parcel of what John did: the whole point
about what happened was that he acted as he did with the express intention of offending
the guests. In (6) on the other hand, all that John did was get wildly drunk: it happens
also to be true that in doing so he intended to offend the guests, but that piece of
information is an afterthought. This is not inconsistent with the analysis. The syntactic
rule simply makes two forms of topicalisation available; the truth-conditional semantics
guarantees an intentional reading of the infinitive; and then it may be that the pragmatics
of topicalisation and weight of information units takes over. However, I suggest that
his data should be treated with some care; a point which will become more important in
our discussion of the behaviour of the Telic Clause in these constructions (in Section
3.4). The same intuitions (with the same proviso) occur in the r/zowg/z-movement
examples, as in (7-8).

[7] [Get wildly drunk to offend the guests] though John may...
[8] [Get wildly drunk] though John may, to offend the guests,...

The Rationale Clause may be preposed with the VP, as in (7), or left in situ as in (8),
although (8) has the flavour of an aftercomment.

109
 Interestingly, exactly the same ambiguity occurs with negation, which is strictly
a scopal phenomenon (cf Jones (1991) for an alternative view of these facts). Consider
(9).

[9] Jo didn't make a pass at the guy to impress us.

•I-
This sentence is ambiguous, as in (10-11).

[10] [NOT [[Jo make a pass at the guy] [to impress us]]
[11] [NOT [Jo make a pass at the guy]] [to impress us]

In (10) we have the reading where it simply isn't the case that Jo made a pass at the guy
with the intention of impressing us. In (11) we have the reading where Jo does not
make a pass at the guy and not doing so is intended to impress us. There are two
possible solutions to this problem. It might be the case that the Rationale Clause moves
in and out of the scope of negation, as proposed by Jones (1991). However, this will
not fit in well with our discussion of the distribution of the Telic Clause which must
clearly be structurally superior to the Rationale Clause: if the Rationale Clause adjoins
ambiguously to VP or IP then the Telic Clause will be forced to adjoin too high in the
tree. The other solution takes up the parallel that clearly emerges between negation and
modification as discussed so far. It may in fact be the case that negation adjoins to VP,
hence taking ambiguous scope over the Rationale Clause, and is then raised to NegP at
LF for checking.
As with the Purpose Clause, the Rationale Clause is unambiguously in the
scope of interrogation, as in (12).

[12] Did Jo make a pass at the guy to impress us?

The nature of Jo's intention is a part of what is questioned in (12). The Rationale
Clause also naturally forms part of the focus of interrogation, as in (13).

110
[13] Did Jo make a ;oass at the guy to impress us or to offend them?

It therefore seems that the Rationale Clause is a standard VP-adjunct which is

ambiguous with respect to VP-rules but unambiguously within the scope of sentential
operators. This suggests the structure in (14) for the infinitives of result considered so
far.

[14]

Rationale Clause

Spec

V Purpose Clause

1 1 1
Section 3.4: The Telic Clause

As was suggested in Section 3.1, the evidence from VP-anaphora and

movement for the distribution of higher level modifiers is often ambivalent. Some of the
best evidence in fact comes from scope effects, though the judgements are still delicate.
Consider the interaction of the Telic Clause with negative scope. It seems that the Telic
Clause is marginal within the scope of sentential negation.

[1] ???A sitting Prime Minister hasn't ever resigned as leader of the party before, to
be re-elected with an even bigger majority.

As before we would expect two possible scopes, with pragmatic factors in (1)
favouring the scope in (2).

[2] [NOT [[a PM resign] [to be re-elected]]]

[3] [NOT [a PM resign]] [to be re-elected]

(2) simply denies that it is the case that a sitting Prime Minister has resigned and then
been re-elected; (3) asserts that a sitting Prime Minister did not resign and was re-
elected. The natural interpretation is the one in (2), yet this reading does not seem to be
available, i.e. the Telic Clause lies outside of the scope of negation.
Notice that the interpretation is a perfectly plausible one which becomes
available under negative topicalisation.

[4] Never before has a sitting Prime Minister resigned as leader of the party, to be
re-elected with an even bigger majority.

The sentence in (4) is perfectly well-formed with the interpretation intended for (1).
Notice that the well-formedness of (4) has two important implications. First, the Telic
Clause must form part of the syntactic computation which generates the sentence it
modifies: it is tempting, given the comma intonation and intuitive sense that this
infinitive is a sort of comment on the whole sentence, to hypothesise that it is a sort of
1 12
non-restrictive relative clause whose value is computed independently of the matrix;
given the interpretation of (4), which is effectively as in (2), this simply cannot be the
case. The Telic Clause is integrated into the sentence in the normal way and its
interpretation is integrated with the interpretation of the sentence it modifies.
The second implication of the well-formedness of (4) is that the Telic Clause can
be adjoined no higher than C, assuming that in the construction of negative
topicalisation, the negative raises to CP-Spec and the auxiliary to Cft. The question is
then whether the adjunction is as high as the projection of Cft. There is evidence to
think that it is not. It seems that the Telic Clause also falls within the scope of subject
negation, as in (5).

[5] No sitting Prime Minister has ever resigned as leader of the party before, to be
re-elected with an even bigger majority.

Again, this sentence is perfectly well-formed, effectively with the interpretation in (2).
This means, however, that the Telic Clause must be within the scope of the subject in
IP-Spec. The natural assumption, therefore, is that the Telic Clause is adjoined to I1.
The data, however, is not entirely consistent. Although it seems that the Telic
Clause can be interpreted in the scope of a negative subject and of topicalised negation,
it seems that negative polarity items are not licensed naturally in these positions.

[6]
(a) No Prime Minister has ever demanded the support of his party, to be challenged
by someone in his own cabinet.
(b) *No Prime Minister has ever demanded the support of his party, to be
challenged by anyone in his own cabinet.
[7]
(a) Never has a Prime Minister demanded the support of his party, to be challenged
by someone in his own cabinet.
(b) *Never has a Prime Minister demanded the support of his party, to be
challenged by anyone in his own cabinet.

The examples degrade considerably when only is included.

113
[8] *No Prime Minister has ever demanded the support of his party, only to be
challenged by anyone in his own cabinet.
[9] *Never has a Prime Minister demanded the support of his party, only to be
challenged by anyone in his own cabinet.

Given the clear well-formedness of examples such as (4-5), however, it is possible that
the restriction on negative polarity items is unrelated to scope and linked to some other
aspect of the Telic Clause's interpretation. As Shalom Lappin has pointed out to me, all
weak NPs block the licensing of negative polarity items, even within the immediate
scope of a negative, as in (10-11).

[10] Never have books with any inappropriate pictures been given to the children.
[11] *Never have the/five books with any inappropriate pictures been given to the
children.

In (10) the complex NP, [Npbooks with any inapproapriate pictures], falls within the
scope of a topicalised negative item; any is licensed in a negative polarity reading. The
simple addition to the NP of the definite determiner the or of the numeral five, as in
(11), blocks this reading. It is therefore clear that negative polarity licensing is sensitive
to factors other than scope and that the interpretation of the Telic Clause may interfere
with this licensing just as weak NPs do. It will, however, become apparent as the
discussion progresses that consideration of the distribution of the Telic Clause provides
evidence pulling strongly in both directions: towards closer integration with the verb
and towards extreme peripheral attachment. The negative polarity evidence suggests a
peripheral attachment, although negative scope in general suggests that that attachment
cannot be too high.
Similar data comes from interrogative scope. It seems that the Telic Clause must
be adjoined below the complementiser position for the correct interpretation of certain
types of subordinate clause.

[12] John wondered whether the Prime Minister would resign, to be challenged by a
member of his own Cabinet.

1 14
Here the "object of wonderment" is not just whether or not the Prime Minister would
resign (John might be quite certain of that in advance); the question is whether that
resignation will lead to a challenge from within the Cabinet. For this interpretation to
follow the Telic Clause must be integrated into the interpretation of IP, as would be the
case for an I'-adjunction. The situation in root sentences is however less clear. It is
clear that the Telic Clause can appear as part of a question.

[13] Will the Prime Minister resign, to be challenged by a member of his own
Cabinet?

(13) is well-formed as a standard question. As in the subordinate case

discussed above, what is questioned is not just the Prime Minister's resignation, but the
resignation and consequence sequence. However, it seems that (13) is less acceptable
where the Telic Clause is intended as the focus of the question, as in (14).

[14] ??Will the Prime Minister resign, to be challenged by a member of his own
Cabinet or to sail through the first round unopposed?

Nevertheless, (14) is still more acceptable than traditional content disjuncts in these
contexts.

[ 15] * Will the Prime Minister resign, with any luck?

[ 16] *Will the Prime Minister resign, with any luck or unfortunately?

The evaluative disjunct, with any luck, is bad even in a simple question, as in (15); but
as the focus of a question, as in (16), it is ludicrous. Once again therefore, it is clear
that the Telic Clause is more peripheral than the Purpose and Rationale Clauses with
regard to question-focus. However, we cannot assume that it is maximally peripheral,
as it does not behave -i'^e a true disjunct.
The evidence v/e have discussed so far has suggested a relatively high point of
adjunction for the Telic Clause. It is worth noting a couple of constructions which again
suggest that this infinitive cannot be adjoined too high in the tree: reduced clause
115
structures. Consider first the construction in (17) which Williams, after Jesperson, calls
a "nexus of deprecation".

[17] Me get the chance to visit Hollywood, only to turn it down because I'm
working you must be kidding!!!

This is a classic small clause construction which is assumed to lack the properties of a
full clause. It is nevertheless possible for a Telic Clause to occur here. This suggests
that the Telic Clause does not require adjunction to a fully sentential projection. In our
discussion of the semantics of Telic Clause in Section 5.3, I will argue that this is
i
exactly what we would expect of an I1 adjunct, as I will take I' to denote an event. A
similar observation can be made of perceptual sentences.

[18] Jody saw Maverick win $1 million on the first hand, only to lose it on the
second.

Syntactically, the complement of see is some sort of small clause structure, denoting an
event The Telic Clause is not only syntactically acceptable here, it actually constitutes a
part of the event perceived (see Section 5.3 for a more detailed discussion).
Constructions such as (17) and (18) indicate that caution is required in suggesting a
peripheral point of adjunction for the Telic Clause.
The delicacy of the judgements which the Telic Clause produces are nowhere
clearer than in constructions involving VP-sensitive rules (a problem anticipated in
Section 3.1). We will now turn to these constructions. First, consider the case of
pseudoclefting in (19).

[19] What Ute did t, only to discover that Reuben was vegetarian, was cook squid
pasta.

In (19), the Telic Clause is excluded from the raised wh-VP which takes the bare VP,
cook squid pasta, as its antecedent. The sentence is well-formed (although stylistically
heavy, requiring a strong intonation break around the infinitive). This suggests that the
1 16
Telic Clause can occur outside of a VP segment. The Telic Clause can also appear
sentence-finally in a pseudo-cleft construction, as in (20).

[20] What Ute did t was cook squid pasta, only to discover that Reuben was
vegetarian.

However, as was observed in Section 3.1, this does not give any reliable evidence that
the Telic Clause can appear within the VP. In fact, sentence final position seems
available for any type of adverbial, however wide its scope. (19) suggests that the Telic
Clause can appear external to VP; there is no evidence however that it appears within
VP.
The interaction with do so is also rather problematic, in part because the
examples are so stylistically awkward.

[21] ??Ute cooked beef for the dinner party, only to realise that her guests
were vegetarian, and Nick did so, only to realise that that they were Hindu.

The sentence in (21) is odd but perfectly interpretable: Ute and Nick both performed the
same actions but the their actions had different unfortunate outcomes. It seems
intuitively clear that the reason for the oddness is that the Telic Clause is an
aftercomment on the event rather than an integral part of it, which suggests that despite
its oddness (21) can still be taken as evidence that the Telic Clause falls outside of VP.
Strictly however this is rather weak evidence. There is however one interesting fact
which supports this view. I find (21) more acceptable where the identity of Ute and
Nick's actions is stressed as well as the contrast in the outcome of those actions. This is
achieved by the inclusion of too, as in (22).
'«
[22] ?Ute cooked beef for the dinner party, only to realise that her guests were
vegetarian, and Nick did so TOO, only to realise that that they were Hindu.

1 17
The inclusion of too stresses identity of action. But of course it is only inserted where
i
the reference of the two maximal VPs is identical. This will only be the case if the
Telic Clause does not form part of the VP itself.
Now consider the case where do so putatively replaces the whole [VP + Telic
Clause] string.

[23] ??Ute cooked beef for the dinner party, only to realise that her guests were
vegetarian, and Nick did so too.

Again, this is an odd construction but perfectly interpretible. Notice that the crucial
point here is that the interpretation of the Telic Clause be included in the reference of do
so, i.e. it is not enough for Nick simply to have cooked beef for his dinner party; that
must also have had the unfortunate outcome that the guests turned out to be vegetarian.
It is clear that we can understand this to be the point: but do we get to that interpretation
by charitable interpretation of a disjointed construction or by the semantic rules making
it a part of the truth conditions? My suspicion is that here we have an instance of
pragmatic compensation salvaging an anomalous construction. VP-external elements are
only allowed in the first conjunct if they are balanced by some other VP-external
element in the second conjunct. Given that there is no balancing element we assume
identity. The problem is that a Telic Clause is not in any sense related to the action
performed (by an Agent) but to the event which occurs as a whole. Notice that the Telic
Clause is compatible with the happen construction which is anomalous with the
Rationale and Purpose Clauses.

[24] Ute cooked beef for the dinner parry, only to realise that her guests were
vegetarian, and that happened to Nick too.
[25] *Ute cooked salmon to please her friends and that happened to Nick too.
[26] *Sven took the ice cream into the kitchen to put in the fridge and that happened
to Nick too.

I will set the questions concerning VP-anaphora aside here.

118
 Fortunately movement rules produce clearer judgements. Consider a context
such as given in (27a) and a continuation as in (27b).

[27]
(a) Chrissie told Ute that she would fly to Oxford if Ute prepared salmon for her,...
(b) ...so, Ute cooked her some salmon, only to realise that she wanted it raw.

Now consider the effects of VP-preposing, as in (28-29), where the Telic Clause is
unaffected by the preposing rule in (28) but is carried along by it in (29).

[28] ...and cook her some salmon Ute did, only to realise that she wanted it raw.
[29] *...and cook her some salmon only to realise that she wanted it raw, Ute did.

The example in (28) is perfectly well-formed; indeed unlike the examples of stranded
Rationale Clauses given above, the Telic Clause is quite natural in this position.
However, the sentence in (29) is ill-formed, in stark contrast to the Rationale and
Purpose Clauses: notice that if (29) is to be interpreted at all we must assume that the
realisation is in some sense a part of what Ute did and perhaps therefore that she
intended the realisation. This is of course quite impossible.
The Telic Ciause seems entirely incompatible with though-movement,
presumably because it does not naturally form part of a concession.

[30] *Cook Chrissie salmon though Ute may, only to realise that she wants it raw,
she will not come to Oxford.
[31] *Cook Chrissie salmon only to realise that she wants it raw though Ute may,
she will not come to Oxford.

Both (30) and (31) are bad. Notice, however, that in (31) the Telic Clause is read as a
part of Ute's intented action, which is anomalous.
There is one further fact which suggests that the Telic Clause is in the class of
disjuncts, distinct from Purpose and Rationale Clauses, although the reason for the
restriction is unclear. It appears that the Telic Clause, unlike the Rationale and Purpose
Clauses, cannot appear in a simple cleft construction.

1 19
[32] *It was only to realise that Chrissie wanted it raw that Ute cooked the salmon.
[33] It was to please Chrissie that Ute cooked the salmon.
[34] It was to enjoy the excellent salmon that Ute took Chrissie to the sushi bar.

Although the data is rather complex it seems that there is good evidence,
particularly from negative scope, that the Telic Clause is, on the one hand, external to
VP, and on the other, internal to the sentence. I therefore suggest that it adjoins to F.
This gives us a basic structure for the constructions as in (35).

[35]
IP
1
Spec r
rr Telic Clause

Rationale Clause

Spec

V Purpose Clause

NP

120
Section 3.5: Some other observations

So, the evidence discussed in Sections 3.1-4 suggests an external syntax for the
infinitives along the line of (1).

Telic Clause

Rationale Clause

Spec

Purpose Clause

NP

Notice that this constituency straightforwardly predicts another striking fact about the
external syntax of infinitives of result: the constructions can co-occur but where they do
so, strict linear precedence restrictions apply (both Faraci (1974) and Jones (1991) use
linear precedence as an argument for relative hierarchical position; Jones attributes the
argument to Williams (1974)). The Purpose Clause must be closest to the verb,
followed by the Rationale Clause, followed by the Telic Clause.

[2] Sven brought Armistead along to the meeting [pc to discuss his books] [RatC
to impress his friends] [TC only to discover that they had already met him].

Of the six logically possible orderings for the three infinitives, this is the only one

which is acceptable.

121
 The analysis might also allow an explanation for another curious fact about the
constructions. Of the three infinitives, only the Rationale Clause can be preposed.

[3] *[pc To discuss his books], Sven brought Armistead along to the meeting.

[4] [RatC To impress his friends], Sven brought Armistead along to the meeting.
[5] *[JC Only to discover that his friends had already met him], Sven brought
Armistead along to the meeting.

Faraci (1974) follows Chomsky (1965) in assuming that complements of the verb do
not prepose naturally (although Higginbotham (p.c.) informs me that the general
assumption is that definites do not prepose).

[6] ???The book, John bought.

This for me is literary/rhetorical rather than ungrammatical; however, it is clearly

marked in comparison with the preposing of an adjunct.

[7] On Monday, John bought a book.

One might argue therefore that it is not just arguments of the verb, but VP-internal
elements (internal arguments plus a restricted set of object-oriented modifiers) which
cannot prepose. This argument is rather weak given the contrast between the
unacceptability of (6) and (3). However, it clearly cannot be more than an ancilliary
factor here because the Telic Clause exhibits exactly the same restriction and the Telic
Clause is clearly not a VP internal element. Either therefore we have two separate
explanations (one of which is already weak) or we find a more general explanation. In
the phrase structure in (1) the possibility of more general explanation arises if an
assumption of Chomsky (1994) is adopted. Chomsky proposes that only heads and
maximal projections are visible to the computational component: a phrase can be
adjoined to an intermediate projection, but once there is becomes invisible. I am

122
assuming that proposing is a topicalisation rule of some sort8. If Chomsky's hypothesis
is correct then the facts on preposing follow directly: the Purpose and Telic Clauses are
X' adjuncts which are invisible to the computational component and cannot therefore be
moved; the Rationale Clause is a VP-adjunct and can therefore be affected by movement
o
and topicalisation rules .
It is tempting to see this distinction at work in another distinction between the
Rationale Clause and the Purpose Clause: recursion. As Jones (1991) observes, the
Rationale Clause allows recursive adjunction where the Purpose Clause does not. The
point, however, requires clarification as I believe that Jones' discussion is misleading
on this point.
Jones (1991: 52ff) claims that the Rationale Clause, which is simply a CP, is
subject to free adjunction to either VP or IP; where the Purpose Clause, which is a bare
VP, must, because of its special character, adjoin within the matrix VP, presumably to
V. The evidence which Jones uses to support this claim is the contrast in acceptability
between the Rationale Clause in (8) and the Purpose Clause in (9).

[8] Fred started a food co-op in order to save on his grocery bills in order to pay his
phone bills in order to be able to call his girlfriend from his home.
[9] *They brought John to talk to them to show them the way.

" As Shalom Lappin has pointed out to me, this view relies on the assumption that topicalisation of this sort
is a rule of core grammar, as movement for stylistic purposes is taken to fall outside of the constraints on core
movement One piece of evidence favours this view. Contrast (a) and (b).

(a) John can't have been buying drugs to impress anyone.

(b) *Buying himself drugs to impress anyone, John can't have been.

The illformedness here seems to follow because the negative polarity item, anyone, is moved outside of the
scope of negation. This is only possible, however, if movement of the VP to the front of the sentence is an
instance of core movement: if the VP were moved later under "stylistic movement" we would not expect the
negative polarity violation.

9
Jones (1991: 56) mentions some examples pointed out to him by Barbara Partee which suggest that
topicalisation of the Rationale Clause is not subject to reconstruction.

(a) *To impress anyone, John didn't buy a car.

In cases where topicalised Purpose Clauses are allowed they show reconstruction effects.

(b) ???and to impress anyone with I wouldn't bring that.

123
Now it seems immediately apparent to me that these examples are quite different from
each other. In (8), as Jones himself points out, the Rationale Clauses do not represent a
string of random reasons for Fred's starting a food co-op: Fred did not start his co-op
to save on grocery bills and to pay his phone bills and to be able to call his girlfriend
from his home. Rather, he started his food co-op to save on grocery bills. He wanted to
save on grocery bills to pay his phone bill. He wanted to pay his phone bill to be able to
call his girlfriend from home. Admittedly, his ultimate aim in starting the food co-op is
to be able to call his girlfriend from home, but (8) requires us to think of this ultimate
aim in terms of a chain of motivations, each dependent on the preceding one.
This, of course, is quite unlike (9), which really does represent a list of random
motivations. John was brought along (i) that he might talk to them, and (ii) that he
might show them the way. As soon as one constructs genuinely parallel examples for
the two types of clause the contrast disappears. Consider (10-15).

[10] *They brought John to talk to them to show them the way to help them with the
language.

(10) is ill-formed because it contains a string of Purpose Clauses which are unrelated to
each other. (12), containing Rationale Clauses, is also ill-formed, however, for the
same reason.

[12] *Fred went home to have a rest to see his mother to record some music to visit
his grandparents to look after the garden.

(12), which contains Rationale Clauses, is, however, well-formed for the reasons given
above; but so is (15), which contains a sequence of Purpose Clauses.

[14] Fred started a food co-op [to save on his grocery bills [to pay his phone bills [to
be able to call his girlfriend from his home]]].
[15] They brought the general along [e to send the soldier into town [e to find a priest
[e to marry the couple]]].

124
The Purpose Clauses in (15) are all related to each other in the same way as the
Rationale Clauses in (13): they brought the general along because they wanted him to
send a soldier into town; they wanted him to send a soldier into town to find a priest;
they wanted him to find a priest to marry the couple. The marriage is the ultimate aim,
but is mediated by a sequence of subsidiary aims. Notice that the Telic Clause is also
iterable, as in (16).

[16] John began typing his final chapter, only to be disturbed by a knock at the door,
only to discover that it was his girlfriend, only to get into a fight with her, only
to be reconciled and spend the evening by the fire.

Notice once again that each Telic Clause represents an unexpected outcome to the
immediately preceding event.
So the to-infinitive is trivially iterable in each of the constructions, in what Jones
(1991: 54) calls a "cascade of dependencies" construction. With regard to the examples
he gives there is no reason to posit any fundamental difference between the
constructions. Thus, although I agree with his distinction between the Rationale and
Purpose Clauses (which can be extended to the Telic Clause), I find his evidence
spurious. The Rationale Clause differs from the other infinitives not in the examples
that Jones gives (which I will call internal adjunction) but in allowing a rather special
type of recursion which I will call external adjunction.
External adjunction involves recursive adjunction to a single node, where the
antecedent of each Rationale Clause is the entire complex of elements which precedes
it. Consider the Rationale Clauses in (17).

[17] Catholics {[have sex] [solely) to procreate]} [to conform with the Church's
teaching]

Here we are not saying that Cathoics procreate to conform with the Church's teaching.
Rather we are saying that Catholics have sex solely for the purpose of procreation and
the reason for THAT complex event is that they might conform with the Church's

125
teaching. The construction is marked and relies on our understanding of complex
actions being the object of an Agent's intention. This construction does not occur
naturally with the other infinitives.
It is tempting to search for a way of deriving this restriction from the invisibility
of X'-nodes to the computational component. However, this seems inadvisable. It
seems that recursive adjunction to X'-nodes is allowed as long as the modifiers are of
different types.

[18] John served beef [raw] [with a tomato sauce] [in a large dish] [for the guests to
try as an appetizer].

This is of course a possibility predicted by X'-theory as standardly understood (the

adjunct rule is recursive). If there is a restriction on multiple Purpose and Telic Clauses
it seems that that must have something to do with their meaning. This view is supported
by the intuition that Rationale Clauses are allowed to undergo external recursion
because they involve intentions: the Rationale Clauses build up increasingly complex,
internally-dependent intentions for particular actions.
However, although the other infinitives of result do not allow interpretations of
quite the sort which occur with Rationale Clauses which externally adjoin, it does seem
that it is possible to construct examples at least of Purpose Clauses which allow external
adjunction of some sort. Consider (18).

[19] I set the computer up [e to operate at 100MHz] [for my son to enjoy e\.

Notice in this case that we must have an external adjunction of Purpose Clauses: the
first infinitive is a subject-gap Purpose Clause (it is the computer which is operating at
100MHz not me) and the verb operate does not take an object. The second infinitive,
which is an object-gap Purpose Clause, cannot therefore be modifying the subject-gap
Purpose Clause, because there is no direct object or Theme to control the gap. Notice
that (20) is ill-formed.

126
[20] *The computer operated at 100MHz for my son to enjoy e.

The second infinitive is also therefore modifying the Theme of set up, requiring
recursive adjunction to the same node (the V projected by the matrix verb). Notice
further that it is not enough in (19) for me to set up the computer for my son to enjoy: it
is implied that my son's enjoyment of the computer is reliant on the fact that it is set up
to operate at 100MHz. It therefore seems that external recursion of some sort is
available for the Purpose Clause.
The link with Agentive intention remains a tempting one: it seems that Telic
Clauses simply do not allow external recursion. I found it impossible to construct ill-
formed examples, simply because it is unclear what they would mean; in both the
Purpose and the Telic Clauses it is the nature of complex intention which seems to
'/ '

support the construction. I leave this point here.

127
Section 3.6: Further observations: external syntax and conditions on
argument projection

Given the questions that will be raised during the course of this thesis, it is
important to clarify a number of assumptions concerning phrase structure
representations. I take it that there are three basic types of phrase structure which will be
of interest.
[1]

Subject

[2]

Derived Subject Internal Subject

[3]

Spec

Derived Subject Internal Subject Direct Object

128
 (1) is a traditional phrase structure in which the internal structure of all phrases is
individually specified by rewrite rule, allowing, for instance, phrases without heads
(there is no head for S). (2) is the canonical structure for the sentence within the
Government and Binding tradition after the adoption of the internal subject hypothesis
(cf Kuroda (1988)). Each phrase (lexical and functional) conforms to a general
structural schema which places strict constraints on permissible phrasal projections. A
phrase structure along the lines of (3) is proposed by Larson (1988; 1991) and by
Bowers (1993) amongst others. Bowers argues that the direct object is a sort of
secondary subject, dominating all other internal arguments. For our purposes, (2-3) are
t
t

clearly preferable to (1) as they provide a position for the recursive adjunction of object-
i

oriented modifiers (see the end of Section 3.1). They also offer a considerably more
restrictive theory of phrase structure. Given the more articulated structure offered by
X'-theory and following Minimalist assumptions, I will also assume binary branching
(cf Kayne 1983).
What are important for our purposes, however, are the implications for theories
of argument projection of adopting these structures. Consider the case of the Purpose
Clause. It is standardly assumed the modifiers like the Purpose Clause act as one place
predicates in some sense (although what is meant by that term differs widely, see
Section 4.1.2.2). It is:assumed that they discharge their argument by predication and
j
that predication only occurs under the strictly local conditions, specifically, mutual c-
command. As Williams (1994: 29) puts it:

The structural locality of the "argument of relation is the tightest of all

grammatical relations. It is essentially as tight as it can get- that is,
immediate sister nodes may enter in the relation, and nothing else.

It is important to note that the term "sisters" refers to phrases which c-command each
other. This is important as we do not wish predicates to assign their theta-roles to
phrases considerably further up or further down the tree than they are (say across clause

129
boundaries in either direction). This requirement is met straightforwardly in (4), where
the modifier simply acts as another sister of the verb.

[4]

fSubject- 1 irect Object Modifier

\ I
[Derived Subject)

However, the structure in (4) does not allow for iteration of object-oriented modifiers
and fails to distinguish them from true arguments of the verb.
X'-theory provides V as a node for recusive adjunction within VP, giving the
modifier structure in (5).

[5]

Spec

Derived Subject Internal Subject Direct Object Modifier

Here the modifier c-commands the direct object, but the direct object of which the
modifier is predicated does not c-command the predicate. If we assume Williams' view
that predication is merely a case of coindexing, the structure in (5) can be maintained if

130
the conditions under which predication is allowed are weakened. Williams (1994; 1995)
in fact assumes a definition of m-command, in which all phrases are sisters which are
dominated by the same maximal projection: the modifier and the direct object in (5) are
of course dominated by the same maximal projection (VP) and therefore m-command
holds. Coindexation/predication can occur. Notice that this is a considerable weakening
of our definition of "strict locality". It will mean that all of the phrases in VP down to
the complement of the verb itself will be mutually c-commanding: yet predication
requires only that the complement of the verb c-command a V'-adjunct.
Even assuming this definition of "c-command", it is not weak enough. Consider
the problem posed by VP-adjuncts. There is good evidence that subject-oriented
modifiers such as the small clause in (6) are in fact adjoined to VP, given that they can
be preposed with the VP as in (7) and stranded outside of the VP in the pseudocleft
construction as in (8).

[6] Sven is serving our dinner nude.

[7] Serving our dinner nude, Sven may be, but...
[8] What Sven was doing nude was serving our dinner.

In the structure offered by (2), this would give (9).

[9]

Derived Subject Internal Subject Direct Object Modifier

131
Assume for a moment, following Williams, that the subject of the sentence is actually
base-generated in IP-Spec. We then require mutual "c-command" between the modifier,
ZP, and the subject in IP-Spec. Under our present definition this is impossible, because
ZP is dominated by Y13 and IP-Spec is dominated only by IP. They are therefore not
contained in the same maximal projection: IP-Spec (strictly) c-commands ZP, but ZP
does not even m-command IP-Spec. Shalom Lappin has pointed out to me that the
definition of c-command offered in May (1985) will in fact allow the mutual c-
command required.
May (Pge 34, his <9>) begins with a standard definition of c-command as m-
command:

[10] a c-commands (3 =df every maximal projection dominating a dominates (3, and
a does not dominate p.
!.

The key to his theory is his definition (Pge 57) of the notion of maximal projection
itself.

Let us... assume instead that [the nodes arising as a structural effect of
Chomsky-adjunction] do not constitute distinct categorial projections, now
understanding the occurrence of a projection (at a given bar level) to be
made up of a set of occurrences of nodes that are featurally non-distinct.
It is these nodes, taken collectively, that constitute the membership of a
projection.

What this means is that to be dominated by a maximal projection, one must be

dominated by every segment of that projection. Consider the tree diagram in (9)
again. Notice that V, NP and VP-Spec are contained within every segment of VP: VP is
therefore the first maximal projection dominating all of them. IP-Spec, I, and VP are
contained within every segment of IP, and so IP is the first maximal projection
dominating all of them. Now consider the VP-adjunct, ZP. ZP is dominated only by
one segment of VP, not by every segment; therefore ZP is not contained in the maximal
projection, VP, which is the collectivity of VP-segments. The lowest full phrase, every

132
segment of which dominates ZP, is IP. Therefore ZP c-commands everything in IP.
Mutual c-command therefore holds between IP-Spec and ZP in May's view.
Notice, however, that our definition of "strictly local" c-command is now even
weaker. Under strict c-command mutual c-command only holds between a and P if a
and p are immediately dominated by the same node (whatever it is); under standard m-
command mutual c-command holds between a and P if a and P are immediately
dominated by the same maximal projection, so that a head c-commands and is c-
commanded by its complement, its specifier, and all X'-adjuncts. Under May's
definition of c-command, mutual c-command holds between a and p if a and P are
immediately dominated by the same complete set of maximal projection segments: this
means that a head c-commands and is c-commanded by, not only its complement, its
specifier, and all X'-adjuncts, but also all phrases adjoined to the head's complement
(and presumably those adjoined to its specifier or adjoined to its X'-adjuncts).
Despite this considerable weakening of c-command, May's definition does not
allow predication of a 6-role by a VP-adjunct to a VP-internal subject. Recall that by his
definition, the c-domain of VP-Spec is VP, because VP-Spec is contained in every
segment of VP. VP-Spec does not therefore c-command ZP in (9), as required for
predication. Mutual c-command will hold in this case only if we take the standard
definition of m-command and assume that the term maximal projection refers only to the
maximal maximal projection.
It appears that this apprach to argument projection, operating even over the
structure in (2), is insufficiently restrictive. Consider for a moment the highly restrictive
theory of argument projection developed in Higginbotham (1985; 1989).
Higginbotham's theory is forumalated in terms of the systematic projection of argument
structure from daughters to mother, as in (11).

[11]

133
The question that is addressed is how we can derive the argument grid of y, given the
argument grids of a and p. Higginbotham defines four possible "thematic transactions"
which specify a particular type of mapping from {a,PJ to {y}. One of those
"transactions" or principles of argument discharge is theta identification, where an open
position in p is assigned the value of an open position in a and then the grid of a filters
up to y. Notice that the argument relation here really is local and cannot be otherwise. In
fact, this theory forms the basis for a view of argument projection in which syntax and
semantics are built up in parallel during the course of computation, in keeping with the
main thrust of the Minimalist program (cf Higginbotham, forthcoming).
As one would expect given the restrictiveness of this theory, the phrase
sturcutre in (4) is problematic because it offers a multiple branching structure for
interpretation. If we have a multiple branching structure, such as that in (12), then
defining a set of projection principles will be extremely difficult.

[12]

The mapping of argument structures of any open-ended set of nodes to a mother will be
complex and unconstrained. To maintain this view of the syntax-semantics interface a
more restricted input is required.
Now consider the theory of phrase structure in (5). The modifier adjoins to V;
we assume that the modifier discharges its argument by theta identification. Theta
identification requires that there be an open argument position in a which will discharge
the open position in P by identification. However, in (5), the open position associated
with the direct object has already been closed because the direct object was directly
theta-marked by V in V. Theta-identification will fail. In this case we cannot naturally

134
resort to m-command as an explanation because m-command will make all phrases in
VP available, including the internal subject: this will effectively reduplicate the multiple
t

projection problem in (12).

The problem is solved straightforwardly in the theory outlined in (3), which
would have a modifier structure such as that in (13).

[13]

Derived Subject In ernal Subject Direct Object odifier

Here the Purpose Clause would adjoin to V as in (2). Now, however, the modifier is
in the c-domain of the direct object rather than vice versa; the verb has not directly theta-
marked the object within V and therefore V has an open argument position for the
direct object which will subsequently be closed by it. Identification will proceed as
expected. The restrictive theory of phrase structure in (13) therefore accomodates
naturally the theory of argument projection developped by Higginbotham, maintaining
strict locality and a simple mapping from daughters to mother. The structure allows for
object-oriented modification naturally, unlike (4), and requires no weakening of
definitions of c-command, as required for (5).
Given the structure in (13) one can continue to assume that the Telic Clause
adjoins to I1. The question is then where the Rationale Clause adjoins. If the arguments
in Section 3.5 are correct then we must assume adjunction to a maximal projection and
135
therefore presumably adjunction to VP. However, it would be possible simply to
reinterpret all the data from VP-tests as tests for X' in (13) and take the adjunction point
of the Rationale Clause as X1 (Pred1 in a Bower's phrase structure). This would allow
us to state that infinitives of result are modifiers of X1 and to see them as maximally
symmetrical in their distribution. However, this would require a rejection of the
suggestions concerning restrictions on proposing in Section 3.5. I will leave the
question open, but will continue to refer to the Rationale Clause as a VP-adjunct for
expository convenience.

136
 CHAPTER 4: The Internal Syntax of Infinitives of
Result

Section 4.1: The distribution of gaps

Section 4.1.1: The Rationale and Telic Clauses

In the Rationale Clause the distribution of gaps is straightforward. A gap may

only occur in subject position, as shown in (1-3); and that subject gap alternates
straightforwardly with a lexical NP, as in (4) (I will assume that for appears purely for
Case-marking purposes).

[1] Mary slapped Bill (in order) e to hurt him.

[2] *Mary slapped Bill (in order) e to hurt e.
[3] *Mary hit Bill with a hammer (in order) e to hurt him with e.
[4] Mary slapped Bill (in order) for Tom to get annoyed.

The distribution of gaps in the Telic Clause is exactly parallel to that in the Rationale
Clause, namely, a gap may only appear in subject position, as in (5), and a subject gap
alternates straightforwardly with a lexical NP, as in (6).

[5] Mary bought the house on Monday only e to notice on Tuesday that it had rising
damp.
[6] Mary bought the house on Monday only for her partner to notice on Tuesday
that it had rising damp.

The natural assumption, therefore, is that the Rationale Clause and Telic Clause may
have an empty pronominal in subject position (PRO). Subject position is the only
position in English which licenses PRO. As a pronominal, PRO alternates with normal
referential lexical expressions (except that it does not require the presence of a Case-
assigner). The internal structure of the Rationale and Telic Clauses is therefore as in (7-
8).

137
[71 Telic Clause: [CPtlPPRO to VP]]
[CP for[ip NP to VP]]
[8] Rationale Clause: [CPtlPPRO to VP]]
[CP forfip NP to VP]]

The question is then how the reference of PRO is determined. It seems that in
the case of the Telic Clause the assignment of reference is passive sensitive (but see
Section 5.3).

[9] Goliath* defeated an entire armyj, only PRO* to be overcome by a shepherd boy.
[10] !! An entire armyJ was defeated by Goliath*, only PRO) to be overcome by a
shepherd boy.

(10) is well-formed but only in a different reading to (9). The active and passive differ
with respect to which argument of the verb takes the privileged position of subject of
the sentence and in theories which accept some version of the Internal Subject
Hypothesis (cf Kuroda (1988) or, more recently, Bowers (1993)) this is represented
structurally: the arguments of the verb are projected within the c-domain of 1^ and the
syntactic subject raises to IP-Spec. A natural suggestion, in the spirit of Rosenbaum
(1970), is then that FRO is controlled under a Principle of Minimal Distance, as in (11).

[11] PRO is controlled by a minimally distant c-commanding NP.

The Rationale Clause is however controlled by the subject only if it is Agent-like (but
see Section 5.2 for qualifications). We have already suggested, however, that it is more
deeply embedded in the tree. We might therefore take it that the minimally c-
commanding NP for the Rationale Clause is the internal subject. Control would then
follow by the principle in (11) applying to the tree in (12).

138
[12]

Telic Clause

Rationale Clause

Purpose Clause

Syntactic In anal Direct

Subject Subject Object

A problem is then posbd by the passive: the Agent is not realised lexically as the internal
i
subject and therefore the minimally distant NP will be in IP-Spec. Control by the raised
NP will be excluded on interpretational grounds (see Section 5.2) and so the sentence
will be ill-formed. But of course the Rationale Clause standardly allows passivisation.

[13] The lights were turned off to conserve electricity.

However, as I argue in some detail, following a suggestion by Williams (1974; 1987b),

it seems that passivisation is restricted to those Rationale Clauses allowing event
control. Following Hegarty (1989) it would then be possible to argue that the Rationale
Clause is in fact controlled by the event, through A We might then suggest that the
i
principle of minimal distance be formulated along the lines of (14).

139
[14] PRO is controlled by a minimally distant nominal, where

(i) nominals = NP or
(ii) nominals in the same projection are equidistant with respect to control.

The fact that the Rationale Clause and Telic Clause must still be hierarchically
distinct follows fromi the fact that the Rationale Clause seems to be sensitive to the
insertion of an Agentive fry-phrase even when event control is unavailable, where the
Telic Clause is not.

[15]
(a) John printed out the file to give himself a better idea of the layout.
(b) *The file was printed out to give himself a better idea of the layout.
(c) The file was printed out by John to give himself a better idea of the layout.
[16]
(a) John printed out the file only to realise that the layout was wrong.
(b) *The file was printed out only to realise that the layout was wrong.
(c) ?*The file was printed out by John only to realise that the layout was wrong.

Given that the Telic Clause imposes no interpretational constraints on its antecedent, we
would expect it readily to take the fry-phrase as an antecedent if that phrase were
structurally accessible. However, the fry-phrase appears to be inaccessible to the Telic
Clause. On the other hand, the Rationale Clause, which is ill-formed in the passive,
becomes well-formed when the agentive fry-phrase is inserted, implying that it is
structurally accessible. Although it is not clear how this sensitivity is to be captured, it
is clear that some sort of structural distinction will be important.

140
Section 4.1.2: The internal syntax of the Purpose Clause
Section 4.1.2.1: Gaps in the Purpose Clause

Questions which are relevant to the internal syntax of the Purpose Clause might
be the following.

[1] What gaps can occur in the Purpose Clause and why?
[2] Given that the gap is obligatorily controlled by an argument of the matrix verb,
how is the link between the gap and the controller mediated?
[3] How does the form of the infinitive relate to its interpretation?
[4] In a w/i-movement analysis, why must the w/z-element be an empty operator
rather than an overt w/i-word?

In the section which follows I will suggest following Chomsky (1977), Browning
(1987), and Wilder (1989) that the Purpose Clause contains w/z-movement and that the
distribution of the gaps is what one would expect of w/z-extraction. I will argue that the
"link" between the gap and the matrix "controller" is established by an independent
predicate which heads the Purpose Clause; in fact this is an example of true predication
and does not involve control-style coindexing. The infinitive takes the form that it does
because it is required by the Purpose predicate to denote a property. The v^-element
i

must be an empty operator and not an overt w/z-element because only the empty
operator is interpreted as a pure lambda abstractor creating a one-place property rather
than as some sort of operator.
The Purpose Clause will allow a gap to appear in any of three positions: subject
position, as in (5); object position, as in (6); or adjunct position, as in (7).

[5] John brought the tiger along e to frighten Bill.

[6] John brought the tiger along for Tom to frighten e.
[7] John brought the tiger along for Tom to frighten Bill with e.

141
In each of these cases the gap is obligatory and obligatorily controlled by the matrix
Theme. Where the gaps are filled, the result is a Rationale Clause allowing in order
insertion (the result is sometimes marginal).

[8] John brought the tiger along (in order) for Tom/it to frighten Bill.
[9] John brought the tiger along (in order) for Tom to frighten Bill/it.
[10] John brought the tiger along (in order) for Tom to frighten Bill with it.

The Purpose Clause may contain two gaps, with the strict requirement that one
of the gaps be in subject position.

[11] John brought the tiger along e to frighten e.

[12] *John brought the tiger along for Tom to frighten e with e.

The Purpose Clause, however, may never contain three gaps, a fact which is
presumably reducible to a restriction that only one non-subject gap is allowed.

[13] *John brought the tiger along e to frighten e with e.

Notice also that in (11) there is an asymmetry between the subject and the object gap.
The object gap acts exactly like the single gaps in (5-7), i.e. its reference is strictly
controlled by the matrix Theme. The subject-gap is freer. In (11) it seems to be
controlled by the Agent. Where we have a benefactive construction, as in (14) , it
seems to be controlled by the Goal/Beneficiary.

[14] John brought Mary a tiger e to frighten e.

Some readers might find the context here peculiar. Compare with examples such as those in (a-b).

(a) John brought a cake along e to eat e.

(b) John brought a video home e to watch e.
Once again, the reader might find the context here peculiar. Compare with the examples in (a-b).

(a) John brought Mary a cake along e \aezle.

(b) John brought Mary a video home e to watch e.
142
It seems, however, that the reference of a subject-gap in a double-gap infinitive can be
entirely free of the matrix, as in (15).

[15] John brought some wine e to drink e with dinner, which was very good of him
given that he's teetotal.

In (15), John has made a general contribution: whoever wants the wine may drink it.
I will assume following Chomsky (1977) that the object-gap Purpose Clause
contains w/z-movement of an empty operator and following Wilder (1989) that the
subject-gap Purpose Clause also contains covert w/j-movement. Jones (1991) has
suggested that gapping in the Purpose Clauses is an instance of free projection of an
external argument from an undifferentiated list of verbal arguments. I favour the Af-
analysis of Chomsky (1977) over the A-analysis of Jones because the distribution of
gaps patterns straightforwardly with the distribution of w/z-trace rather than NP-trace.
As Jones himself points out, gaps can occur in the Purpose Clause anywhere that wh-
extraction is allowed where A-movement such as the passive (reanalysed as external
projection of an internal argument after Williams (1981)) is more restrictive.

[16]
(a) Sven brought the CD along to play e.
(b) What did Sven play el
(c) The CD was played e.
[17]
(a) Sven brought the bag along to hide the money in e.
(b) What did Sven hide the money in el
(c) *The bag was hidden the money in e.
[18]
(a) Sven brought Sigrun along e to sing some traditional songs.
(b) Who e sang some traditional songs?
(c) *E was sung some traditional songs.

Applying the diagnostics for w/z-movement suggested by Chomsky (1977: 86)

is rather difficult with this data, but some evidence supports this analysis. We have
already seen that the construction contains an obligatory gap. In (19), we have an
apparent violation of the Propositional Island Constraint in the presence of a bridge
143
verb; in (20), we have a violation of the Wh-ls\and Constraint; and in (21), we have a
violation of the Complex NP Constraint.

[19] John brought some medicine along for us to tell/order Bill to take.
[20] * John brought some medicine along for us to tell Bill why to take.
[21] *John brought some medicine along for us to give Bill the order to take.

Notice also that the object-gap constructions license parasitic gaps where the subject-
gap constructions do not: an asymmetry parallelled in overt w/z-movement.

[22]
(a) Sven put the letter in the tray Op e to file t without reading t.
(b) What did Sven file t without reading tl
(c) *The letter was filed without reading r.
[23]
(a) ?Sven opened the draw Op e to put the letter in t without reading r.
(b) ?What did Sven put the letter in t without reading r?
(c) *The letter was put t in the draw without reading r?
[24]
(a) *The manager brought in a secretary Op t to do the work without rewarding r.
(b) *Who did the work without rewarding r?
(c) *The work was done t without rewarding r.

The internal structure for the infinitives is then as in (25-27).

[25] Purpose Clause: [cpO [jpt to [VPtV'tV V NP][pp P NP]]]

[26] Purpose Clause: [cpO [jpPRO to [VPtV'tV V t][pp P NP]]]
[27] Purpose Clause: [cpO [ipPRO to [VPtV'tV V NP][pp P t]]]

I am thus adopting an analysis in which the subject-gap Purpose Clause is

formed, like the object-gap Purpose Clause, by w/i-movement of an operator to CP-
'•\

Spec. This symmetrical approach to the derivation of the subject- and object-gap
Purpose Clauses goes back to the original analysis in Chomsky and Lasnik (1977),
which is developed in more detail in Wilder (1989). Specifically, Wilder argues against
the departure from the symmetrical analysis introduced by Williams (1980), in which

144
the subject-gap Purpose Clause is assumed (like the Rationale Clause) to have a PRO
subject. This split analysis is pursued by Browning (1987) for the subject-gap Purpose
Clause and Jones (1985) for the related subject-gap Infinitival Relative (although of
course Jones (1991) argues for a symmetrical analysis, if radically different from
Wilder's).
As Wilder observes (1989: 214), Williams' (1980) departure from the
symmetrical analysis is based on problems posed for the version of Case Theory he
adopts (as in Chomsky (1980)), problems arising because extraction from the subject
position of an infinitive will require a caseless w/z-trace. Wilder (1989) argues that this
position is in fact not a Caseless position, but rather that it is Case-marked by the
Complementiser, for, which deletes on the way to Phonological Form. This analysis
therefore parallels the standard analysis of the infinitival complement of want, which
Chomsky (1981: 69) assumes to be a CP whose Complementiser, for, deletes when
next to the verb.

[28] John wants very much for Bill to leave as soon as possible.
[29] *John wants very much Bill to leave as soon as possible.
[30] John wants (??for) Bill to leave as soon as possible.

In some dialects, for is standard in sentences such as (30). Wilder observes that when
the subject of the infinitive is extracted, deletion of for is obligatory, even in contexts
where it is usually required.

[31] *Who does John want very much for t to leave as soon as possible?
[32] Who does John want very much t to leave as soon as possible?
[33] Who does John want t to leave as soon as possible?

Wilder goes on to argue that the reason that English licenses subject-gap Purpose
Clauses and Infinitival Relatives is precisely that the Complementiser is a Case-
assigner; Wilder claims that languages such as Dutch, French, and Swedish which do
not have Case-assigning Complementisers also do not license w/z-extraction from the

145
subject of the infinitive. I will assume that an analysis along the lines of Wilder (1989)
is in fact correct.
The symmtrical analysis of these constructions also allows us to follow up a
possible link between the argument structure of the Purpose predicate and the internal
syntax of the Purpose Clause. Recall that in Section 1.2 I argued that the Purpose
predicate which heads this construction denoted a relation between an entity, an event,
and a property, such that the entity's participation in the event was for the sake of the
entity having the property. Assume then that the translation of the empty operator is as a
lambda abstractor, which creates a one-place property. We can now explain the
occurrence of w/i-movement because the predicate requires it if the infinitive is to be
theta-marked as the property argument. The reason we understand a particular argument
of the main verb as being the subject of predication is because the Purpose predicate is
itself a relation between an entity and a property, where the infinitival CP denotes the
property (and is directly theta-marked) and the direct object of the main verb is
coreferential with the entity argument by theta-identification. I therefore differ from
i

previous views of the empty operator in not seeing it as an operator whose value must
be fixed by an antecedent, requiring a theory of wh-predication such as that of
Browning (1987) (see Section 4.1.2.2): there is no theory of predication and
coindexing. Rather the operator receives a translation as a lambda abstractor creating a
one-place property; the purpose predicate theta-marks this property and then attributes it
to its external entity argument, according to general principles of argument discharge.
This then suggests an answer to the problem raised by Hegarty concerning
why, if the Purpose Clause contains w/z-movement, there is no alternation with overt
wh-words. Overt wh words are excluded from Purpose Clause contexts, as in (28-30);
and the empty operator is excluded from overt w/z-contexts, as in (37-38), although this
cannot be shown for the subject-gap Purpose Clause because of homophony with a
sentence containing PRO.

[34] *Sven brought Sigrun along who t to perform some traditional music.
[35] *Sven brought Sigrun along who e to introduce t to his friends.
146
[36] *Sven brought Sigrun along who e to sit next to t at the dinner.
[37]
(a) *Sven asked Sigrun Op e to do t.
(b) Sven asked Sigrun what e to do t.
[38]
(a) *Sven asked Sigrun Op e to give it to t.
(b) Sven asked Sigrun who e to give it to t.

The answer may be that overt w/z-words are not interpreted as lambda abstractors, but
rather as quantifiers of some sort. We might paraphrase the interpretation for (37b) by
the intuitive (39).

[39] Sven asked Sigrun [[for which x] e did x].

The interpretation of (37a) containing the empty operator on the other hand would
receive the paraphrase in (40).

[40] Sven asked Sigrun [the property of being an x such that e did x].

In this view you can ask someone to identify an answer from a set of propositions; you
cannot ask them a property.
Thus the questions under (1-4) are answered as in (41-44).

[41] Gaps can occur in any position from which w/z-extraction is allowed because
the Purpose Clause contains w/i-movement; the presence of an additional
subject gap is optional because PRO can always be present as subject.

[42] There is no direct link via coindexing between the w/z-trace and the NP
argument of the modified verb with which it is coreferential: the two are linked
because of the relation specified by the purpose predicate, namely that the
individual (argument of main verb) have the property denoted by the infinitive.

[43] The infinitive has the form which it has because it must denote a property.

[44] The infinitive contains movement of an empty operator because the empty
operator receives a translation as a lambda operator; overt w/z-words are
excluded because they are not translated as lambda operators but as quantifiers
of some sort.

147
Section 4.1.2.2: The Purpose Clause and predication

In general, theories of syntactic predication have addressed the problem of co-

reference between an overt argument of a matrix verb and a non-overt (usually non-
subject) argument of a subordinate verb, as in (1).

[1] Sigrun brought a jumper1 along for Sven to wear ei at the firework display.

The verb wear has an object argument which is not overtly realised. We must interpret
this object as referring to some individual denoted by a jumper. A jumper is the object
of the matrix verb bring. Some theories conceive of the problem of "predication" in
purely these terms: the linking of an argument position of one verb with the argument of
another verb, where the dependent position is clearly embedded at some distance from
its antecedent. Solutions to this problem have taken two forms. There is the A'-solution
(cf Browning (1987) and Wilder (1989)) in which a non-overt syntactic phrase
mediates the formation of a syntactic chain by which the gapped position is coindexed
with the antecedent position; and there is the the A-solution, adopted by Jones (1991)
following Williams (1980; 1981; 1987a,b; 1989), in which the principles of argument
projection (the Theta Calculus) allow the index associated with a particular argument
position to be assigned directly by a series of local steps over the required distance. The
important point to notice in these analyses is that the theory of predication is set up
solely to mediate coreference between two argument structures: it is not clear what sort
of independent contribution to the interpretation of the sentence the modifier is
supposed to make. Hegarty (1989) views this as problematic and argues that the
purpose modifier has the interpretation which it does because it is headed by a predicate
in the traditional sense, i.e. a relation between individuals. In his view the
complementiser,/0r, denotes a relation between an individual and an irrealis event and
it is this predicate which mediates the coindexing relation in (1), while contributing its

148
own independent element of interpretation. In the analysis I will propose, the relation
between the infinitive in (1) and the clause it modifies will be seen as a relation of
predication purely in the traditional sense, where the infinitive acts as an argument of
the purpose predicate and the interpretation of the infinitive with regard to the modified
clause, as well as the "coindexing" relations, follow as reflexes of the interpretation of
the predicate itself.
Consider first the analysis of Browning (1987). Browning assumes, following
Chomsky (1977), that the gap in the Purpose Clause is created by w/z-movement of an
empty operator to the Specifier position of Cfl. Under standard assumptions the trace is
coindexed with its antecedent as in (2).

[2]

Sven J to wear t l

Assuming that the infinitive adjoins to a position which makes the relevant NP an
accessible, c-commanding antecedent we must then explain how pro mediates
coindexation between the wh-trace and the NP as in (3).

[3]

a jumper1 pro1 or SvenJ to wear t1

149
Browning notes that pro is standardly assumed to require identification, usually by the
person features of 1^, under certain morphological conditions. This is clearly not
possible in (3). She argues that pro is identified in this construction by coindexing with
a referential NP antecedent: hence the predication relation is motivated by a requirement
on the empty category itself. Assume that predication itself involves the coindexing of
two phrases under the condition of c-command (or m-command). This will allow the
structure in (4).

[4]

NP 1

The assumption that heads agree with their specifiers has been standard since Chomsky
(1986b), giving the coindexation in (5),

[5]

Spec1

as is the assumption that maximal projections agree with their heads, giving the co-
indexation in (6).

[6]

150
Between these three agreement relations (predication, Head-Spec, XU-X111^), a chain
is formed which links pro in CP-Spec, and hence the w/z-trace, with its antecedent, as
in (7): {NP,CP,C<Vo,r}.

[7]

Sped C'i

O IP

John' pro 1 for Bill to meet t

Jones (1991) argues that viewing predication in this context as involving A'-
movement introduces an asymmetry into the theory of predication given that it is
standardly used to mediate the assignment of external theta-roles, as in (8).

[8]

Sigrun brought a jumper J

(Ag *, Th J)

Here the external theta-role of the verb bring is the Agent role. Following the theory of
Williams, the index associated with this external role is transmitted up the tree to I1
which is then predicated of the subject by coindexation. The principles which allow for
the transmission and assignment of theta roles Williams calls the Theta Calculus. Jones

151
suggests that the theory of predication naturally falls within the domain of A-rules rather
than A'-rules. He assumes that there is a syntactic rule which allows the VP to
projection any argument externally: if the Agent is not projected it simply becomes an
implicit argument. In the case of (1) therefore, the infinitival verb could project the
Theme argument as its external argument, as in (9).

[9]
VPJ

to wear
(Ag, ThJ)

The infinitive is taken to be a bare VP: this bare VP predicate will then adjoin and
i

directly assign its external theta-role to its antecedent, as in (10).

[10]

a jumper to wear
J) (Ag, ThJ)

As noted above, the accounts of both Browning and Jones share the assumption
that the problem is merely one of assigning co-reference between an non-overt,
dependent argument of a subordinate verb and the overt, antecedent argument of the
matrix verb under appropriately local conditions. The question of why the construction
receives the interpretation that it does is left open. Hegarty assumes that this is
inadequate and argues that the interpretation is introduced by an independent predicate,
realised by complementiser/or. He assumes that for is a two-place relation between an

152
 an
entity and an irrealis event (for the Purpose Clause, as in (11)) or an event and
irrealis event (for the Rationale Clause, as in (12)).

[11] for(x,Ae)
[12] for (e/^e)

d
He assumes that for is the head of a CP (although his arguments are intended to exten
straightforwardly to the heads of PPs). Consider the sentence in (13).

[13] John brought some chicken home for Bill to eat.

Part of the internal structure for the Purpose Clause in (13) is given in (14).

[14]

for Bill eat

He assumes that an IP headed by to denotes an "irrealis event"; for can therefore

directly theta-mark its IP complement through its second argument position, closing that
position off. The projection of argument structure for (14) will therefore be as in (15).

[15]

o VP<E l:

153
As is clear from sentence (13), however, for clearly also has a purely syntactic relation
to its IP complement in that overt NPs are allowed to occur in IP-Spec and must
therefore presumably be Case-marked. Hegarty follows the standard assumption that
for Case-marks the NP in IP-Spec by exceptional case marking. However, he takes this
Case-marking property offor to be exceptionless (and of course must assume thai for is
present in all infinitives because it is the construction's head). This means that IP-Spec
in all infinitives is a governed position and hence that PRO cannot occur in that
position. He therefore also assumes that PRO is base-generated in VP-Spec and does
not raise. This means that in his view infinitives contain an empty IP-Spec (in violation
of the extended projection principle) which is redundantly assigned case.
For therefore both theta- and Case-governs its complement. Following a
suggestion by Harry Leder, based on Burzio's Generalisation, Hegarty suggests that if
a head assigns Case to an internal argument then it must assign a theta-role to an NP
subject. This means that for must directly theta-mark an NP in CP-Spec through its first
argument position, as in (16).

[16]
CPo

NP <>

The question now arises as to the nature of the category to which the theta-role is being
assigned.
Hegarty introduces his discussion on this topic by commenting (Page 22):

This external argument refers to the entity (object or event) of which the
adjunct is predicated. Since this entity is introduced in an argument
position within the matrix clause, it shouldn't be introduced again with
an independently referring R-expression in the external argument
position of [for]. Therefore, the external argument of [for] should not be
an overt lexical NP, but rather a null NP, PRO or pro, which gets its
reference from the matrix NP.
154
Hegarty's assumption that "an independently referring R-expression" excludes overt
NPs begs a number of important questions (why for instance will an overt w/z-word not
serve the necessary purpose?). This is further complicated by his straightforward
assumption that NP in CP-Spec must be PRO or pro, given that he later rejects the view
that PRO and pro are the only base-generated empty categories and that both NP-trace
and w/i-trace are derived from applications of Move Alpha, by introducing a base-
generated empty anaphor. It seems in fact entirely possible that the category in CP-Spec
could be an anaphor if an appropriate definition of government is assumed. The NP in
CP-Spec is referentially dependent on a c-commanding NP. That c-commanding NP
will always be within the XP to which the CP has adjoined (the object in VP, the
subject in DP); given VP-internal subjects, that XP will always also contain a SUBJECT
(the external argument of the verb in VP or the syntactic subject/AGR in IP). The
anaphor will therefore be bound in its governing category.
The choice between pro and PRO is not central to Hegarty's basic analysis. As
he observes (Page 22):

Either empty category can be chosen as the basis of a theory of

secondary predication: which we choose depends on background
assumptions about the directionality of government.

In the end he argues that the choice of pro poses less difficulties for the binding theory
and proposes that Rizzi's (1986) licencing schema for pro be extended to
complementiser/or, as under (17), (Page 26, his {30}).

[17]
(a) pro is governed by Pu
(b) pro is not assigned case by P^

Pro in CP-Spec will then be bound by a minimally c-commanding nominal (which may
include AGR in INFL): this coindexing Hegarty calls "secondary predication". The

155
 semantic effect of such coindexing seems to have essentially the same semantic
properties as direct theta-marking in that all variables to which pro assigns a value are
given the value of the controller. Hegarty does not discuss the coindexing in these
terms, nor does he comment on the way this type of argument discharge interacts with
the theory of argument discharge developed in Higginbotham (1985), which he adopts
as his theory of argument projection.
We now have a syntax and semantics for the predicate/or, as in (18).

[18]

C<1,E2> IP<E3*>

VP<E3>

bring a chicken pro for o

The main verb is bring which directly theta-marks its direct object through its second
argument position, producing a V with an argument position for the Agent of bring and
an event position. The infinitival verb is eat. The VP containing eat has an open event
position. This is theta-bound by to in INFL, producing an "irrealis event". IP is directly
ir

theta-marked by for through its event argument position, producing a C with one open
position, for directly theta-marks pro in CP-Spec through its first argument position.
Therefore CP is a closed structure, which adjoins to V, producing an identical node.
The relation between the infinitive and the main clause is therefore exclusively mediated
by the anaphoric binding relation between pro and the direct object: there is some x
which is both a chicken which John brings and a thing which is for Bill's eating. One
of the aims of Hegarty's paper is to reduce complex facts concerning the interpretation
of these infinitives to particular features of the representation. In a simple coindexing
156
account, no semantic relation of any sort is specified and semantic relations must be
stipulated on a construction by construction basis. By assuming a predicate as the head
of the construction, the richness of the interpretation can be reduced to the interpretation
of that head. It seems however that Hegarty's view of the argument structure of the
head is insufficiently rich: it is not merely the case that John brings along a chicken and
that the chicken is for him to eat, but that it is precisely by virtue of his bringing it along
that the chicken is to become available for eating. It seems that the interpretation of the
Purpose Clause requires us to understand that the matrix event plays a significant role in
the agent's intention to bring about the "irrealis event". This is not clear in the
representation, not even in the coindexing relation, and it is not clear how it could be
made clear except by a construction specific rule. In the basic account developed in
Section 1.2, this follows straightfowardly (for the very reasons that Hegarty has
argued) because the Purpose predicate relates not only an entity and a property, but an
entity, an event, and a property such that the entity's involvement in that very event is
efficacious with respect to it having the property. The principle (theta identification)
which tells us that the entity is the same entity as the matrix Theme will also tell us that
the event in question is in fact the matrix event. Hegarty cannot however simply include
an event position because it would then have to be projected as an empty category but
there is no position to which it could be projected. I therefore follow Williams (1983) in
assuming that argument projection is not necessarily mediated by empty categories and
that general principles of argument projection can link argument positions of predicates
independently of phrasal realisation. However, in this case, I also run contrary to
Williams' general assumption that there can be only one external argument: it preceisly
this restriction, in syntactic form, which makes Hegarty's analysis of the link between
the purpose predicate and the matrix predicate too weak.
There remains, of course, one of the key problems in accounting for the internal
syntax of purpose constructions: the distribution of gaps in the infinitival VP itself. The
structure offered in (18) is in fact a misleading abbreviation of the structure of the
infinitival VP. The eat which heads the VP is in fact transitive: there is something which
157
Bill is intended to eat (in fact, the chicken). This means that there must be an empty
category in object position which is interpreted as co-referential with both pro and the
NP, the chicken. In fact, this chain, {NP, pro, e}, would mimick the chain in a
standard "empty operator" analysis (Chomsky (1977)), {NP, Op, wh-t}. Consider
Hegarty's analysis of the VP-intemal gap and the way in which it is bound by pro.
In the empty operator analysis of the Purpose Clause, the empty operator is base
generated in the relevant argument position of the infinitival verb and is then raised to
CP-Spec, leaving a wh-t in the base generated position. Wh-i is theta-marked and case-
marked and is free in its governing category (IP); it is bound by the operator which is in
an A'-position. The relationship between the trace and the operator obeys island
constraints. Hegarty cannot however adopt this analysis because CP-Spec is an
argument position for the predicate for which is independent of the argument structure
of the infinitival verb. The gap in the VP must therefore be independently base
generated and bound due to some other mechanism than chain formation (and Move
Alpha).
Hegarty argues (Page 51) that e cannot be a variable (i.e. w/z-t) because it is A-
bound by pro which he counts as a violation of Binding Condition C. Nor can it be
PRO as it is governed by the verb. It cannot be pro as V is not a licenser in English.
Therefore he argues that it must be an empty anaphor, [+anaphor,
-pronominal]. Remember, however, that in discussing the nature of the empty category
in CP-Spec, Hegarty assumed that the category could only be pro or PRO because it
was base generated in an argument position. He must either revise this assumption and
consider the possiblity that the empty category in CP-Spec is an empty anaphor, or he
must dismiss the possiblity of e in VP being [+a,-p]. Let us assume for a moment that
such a base-generated empty anaphor is available. Another question then arises. The
only empty anaphor standardly available is NP-t, but Hegarty never discusses e as an
instance of base-generated NP-t (a possibility which Chomsky recognises but
dismisses). If he really is suggesting that e is NP-t, then he must account for why

158
sentences such as the one in (19) are ill-formed, given the well-formedness of parallel
sentences with lexical anaphors, as in (20).

[19] John killed NP-t.

[20] John killed himself.

If he argues that e is an empty anaphor but not NP-t (and hence exempt from principles
restricting the distribution of NP-t, although falling under Binding Condition A), then
he must introduce a new empty category into the typology (exclusively for these
constructions) which the same features as NP-t but is distinguished from it (i.e. the
label NP-t would then be more than a label for the features [+a,-p], having real
content).
Even assuming that e is an empty anaphor, distinct for NP-t, which is base-
generated, pro is still outside of its governing category (which is standardly considered
to be IP). Hegarty therefore assumes that an e base generated in object position can
raise to adjoin to VP, where it is governed by for (I being a defective governor), and
hence has CP as its governing category, as in (21).

159
In this case of course, the empty anaphor will leave a real NP-t which will be bound by
it, so that the argument chain for a sentence like (21) will be, {NP,/?ro, e, NP-t}. It is
not clear why the empty anaphor will not take John as its antecedent. It is also unclear
how such a chain could be made to mimick island constraints (as Hegarty assumes it
does, Page 5If).

[22] John brought some medicine along pro for us to tell Bill PRO to tell Mary PRO
to take e.

In (22) e must be bound by pro: this would involve it adjoining to the lowest VP,
raising over the lowest CP to adjoing to the lowest VP headed by tell, raising over the
next CP to adjoin to the next VP headed by tell, where it will be governed by pro. If
such a chain is allowed, then it must be allowed independently of principles of
Subjacency which would allow w/z-words to move up successive cyclically in (22). In
the empty operator analysis, the operator is assimilated to the w/z-class and therefore
falls under the same explanations. Hegarty's analysis of the internal gap therefore
seems to run into problems at each turn of the argument.
One of the principle reasons Hegarty gives for adopting this analysis is to
explain the non-interchangibility of overt and covert w/z-elements in the Purpose
Clause, which the empty operator analysis, by assimilating the two categories, implies
should occur.

[23] I brought it along to eat.

[24] *I brought it along what to eat.

Hegarty argues (Page 53) that his theory accounts for this asymmetry because CP-Spec
\

is filled by pro and cannot therefore be filled by a lexical w/z-word. The implication is
that w/z-elements are base-generated as arguments of the infinitival verb. It is not clear,
however, that that need by the case (Hegarty has made free use of base-generated traces
already). If a w/z-word was base-generated in CP-Spec it could presumably bind an
empty anaphor of the type described by Hegarty in the same way as pro without any
160
problem. The only conceivable restriction is that an overt lexical w/i-word requires
case, where pro (under the licencing scheme above) does not (in the empty operator
analysis the w/z-words are licenced by heading a case-marked chain).
Although extremely different in many details, Hegarty's account is similar in
spirit to the position defended in this thesis; I assume, however, that a number of
problems can be avoided with slightly different approach. In particular, argument
structure is seen as not necessarily mediated by empty categories but may be organised
by general principles of argument projection, leaving CP-Spec free; the covert wh-
movement can therefore be movement of a standard sort to CP-Spec; the particular
interpretation of the Purpose Clause and the non-alternation with an infinitive containing
overt w/j-movement follows from translating the operator not as some sort of
quantifier, but as a lambda abstractor. These assumptions do however raise some
interesting questions concerning the most appropriate realisations for the predicates
heading infinitives of result, a question to which I now turn.

161
Section 4.2: Internal syntax and conditions on argument projection

what
Notice that if one takes each predicate individually, the question of
lexical item if
category would mosi naturally occur as complement (and hence which
The Purpose
any most naturally lexicalises the predicate) receives a different answer.
The natural
Clause takes a property argument derived by abstract w/i-movement.
complement is therefore CP.

[1]

sation. The
The purpose predicate would be an abstract preposition without overt reali
l closure of
Rationale Clause takes a prepositional argument, resulting from existentia
the event variable. The natural complement is therefore IP.

[2]

Tor Si| run enjoy berseir

R

162
The Rationale Clause might therefore be lexicalised by for. The Telic Clause takes an
event argument, resulting from theta-binding of the event position. The natural
complement is therefore VP.
[3]

discover that..

The Telic Clause might therefore be lexicalised by to. This multiple approach is rather
unattractive. Apart from anything else, if the three predicates are not in complementary
distribution one would expect them to co-occur which does not appear to be the case
(although of course cur understanding of this is linked to interpretation). I assume that
the relevant predicate is to for two reasons. To is obligatorily present with this
infinitive: when it does not occur we have a bare infinitive which has a different
distribution and intepretation; for never occurs with a bare infinitive independently of
to. Further, the construction is derived historically from a preposition + verbal noun
construction: the preposition which governed the case of the verbal noun was to. It is
perfectly true that to has the distribution of a modal auxiliary and it may therefore be the
case that the standard position (cf Bach (1982) and Hegarty (1989)) is correct that to is
merely a modal operator and that the predicates are lexicalised by
Complementiser/Prepositional/or. It may also be the case that all of the predicates are
abstract prepositions taking CP complements: the occurrence of for will then follow
from Complement selection, and the occurence of to will follow from that.
Whatever the case, the arguments must be motivated by facts concerning the
argument structure of the predicates and the most natural way of discharging those
arguments. I do not find convincing the traditional appeal (following Faraci (1974)) to

163
the fact that a class of for + NP constructions share similar properties to these
predicates. This follows in part from perfectly general facts concerning modifier
distribution. Where there really does seem to be a lexical correspondence it is wider
than the traditional position suggests.
The decision to associate each of the result predicates with the lexeme, to,
brings out clearly an important feature of this analysis, namely that it is what I will call a
radical polysemy analysis. The assumption is that the lexeme, to, will be associated in
its lexical entry with e, number of alternative predicates. In this thesis I have offered a
detailed discussion of three of these predicates. There will, however, presumably be
others, for instance the predicate which heads the Infinitival Relative construction.
Making such a move raises two important questions: (i) is the assumption of radical
polysemy in itself valid?; (ii) what does this imply for the study of modifier typology?
In answer to the first question, a comment in Larson & Segal (1995 :487) on the
significance of polysemous uses of thematic roles such as Agent seems relevant.

Worries like these have led some researchers to be skeptical about

thematic roles altogether. For our part, we do not find their arguments
decisive. For example, it seems to us that the fragmentation problem noted
with the notion of Agent arises with equal force in the definiton of many
predicates of natural language. Any detailed dictionary will give entries
for prepositions like to, by, and with that show equal fragmentation into
subentries, with divergence in meaning at least as great...

They then offer an entry from Webster's Nineth New Collegiate Dictionary for to
(ironically enough). This entry has eight major headings and upto eighteen
subheadings. One might object that the polysemy which is being discussed here is
genuinely lexical, that is, that it is variation in the interpretation of the predicate which is
marked in the conditions on the reference of the predicate, not that it is syntactic or
formally semantic, involving variation in the argument structure itself. But this simply
is not true. The variation in lexical meaning will be reflected in variation in the
complementation of th? predicate in several cases (subcategorisation for NP and VP in
different uses, at the least) and sometimes for specific word classes, as in its time use,

164
five to ten.. The lexicon will by its nature be fragmentary and a central problem in
organising the lexicon is to limit the range of this fragmentation. To turn the argument
for rejecting a polysemy analysis on its head, it seems that where ambiguity is
necessary it is best isolated in the lexicon, with the rules of syntactic and semantic
composition being maximally
i' general. This is the case in the analysis which I have
adopted. The Inflectional head, to, is polysemous in that it has (at least) three lexical
subentries; but those subentries represent well-motivated independent aspects of to's
meaning; and the principles by which those argument structures are projected into the
syntax are maximally general. Worries over polysemy do not therefore seem to me to be
a strong objection to this approach.
More interesting is the objection that a polysemy analysis obscures
generalisations in the behaviour of different classes of modifier. This objection seems to
me also, however, to 'be
J'
backwards, for the assignment of coherent and well-motivated
argument structures to modifiers is surely the best way to begin the investigation of
modifier typology in the first place. It is presumably of little value to assign modifiers to
particular semantic classes on the basis of their syntactic distribution without some clear
notion of their semantic interpretation. It may prove to be the case, for instance, that
VP-intemal modifiers often relate properties to individuals in (substates of) an event.
But this must surely flow from an investigation of the argument structure of the
modifier in question: indeed it may be that specific features of argument structure by
which groups of modifiers are classed only become apparent when a proper elucidation
of the meaning of those modifiers (in the sense of Higginbotham (1989)) has been
i
offered. Should such generalisations emerge they will presumably offer real insights
into the nature of the relation between syntactic and semantic representations as
mediated by argument structure. I therefore take it that this polysemy approach, far
from obscuring an investigation of modifier typology, in fact facilitates it and provides
the basis for the relevant generalisations.

165
Section 4.3: Gaps, pronouns, and binding

One important question which is not systematically addressed in the literature

concerns the types of binding relations that are licensed in the infinitives of result and in
particular the nature of the alternation between pronouns and gaps. It is an important
question given that sentences such as (1) are standardly cited as examples of Rationale
Clauses and statements concerning the distribution of Rationale Clauses are made on the
basis of such examples.

[1] John* brought SillJ along PRO* to talk to him!

The problem is that such sentences are clearly minimally distinct from Purpose Clauses:
the gap in the Purpose Clause is replaced by a pronoun in the Rationale Clause. I will
therefore call the infinitive in (1) a "resumptive infinitive".

[2] John1 brought BillJ along PRO1 to talk to d.

The resumptive infinitive in (1) does not behave like a standard Rationale Clause, for
instance, resisting preposing.

[3] *PRQi to talk to himJ, John* brought BillJ along.

The question is then whether this unusual behaviour occurs because of some unnoticed
restriction on Rationale Clauses or whether the pronoun in (1) is in fact a resumptive
pronoun of some sort, giving the derived infinitive mixed properties. In the following
sections I will discuss some of the properties of each infinitive of result with regard to
binding in an attempt to define the parameters of the argument. I will begin with the
Rationale and Telic Clauses whose behaviour in this respect is more straightforward

than that of the Purpose Clause.

166
Section 4.3.1: Pronominal binding in the Rationale Clause

I will use the verb introduce as the matrix verb in the examples which follow to
cut down interference from the subject-gap Purpose Clause in relevant examples. I
assume that the verb introduce does not license Purpose Clauses as shown in (1).

[1] ???Jose introduced Bill to Gudrun e to show e around.

(1) is extremely marginal and it is not clear what the control relations are supposed to
be (who is showing around whom). To make the Purpose Clause reading even more
inaccessible I will preface the examples with an optional in order.

[2] Jose* introduced SverJ to Gudrun^ in order PRO* to show her^ that students at
Oxford
•
could be interesting.
* 1 * "
[3] Jose1 introduced SvenJ to GudrunK in order PRO1 to show him) that students at
Oxford could be interesting.

The infinitives in (2-3) are both Rationale Clauses and therefore PRO is obligatorily
controlled by the Agent subject (Jose). Either the direct object, as in (3) or the indirect
object, as in (2), may bind a pronoun in the infinitive. Consider the effect of wh-
extracting the direct object.

[4] WhoJ did Jose1 introduce tl to Gudrun^ in order PRO1 to show her^ that
students at Oxford could be interesting?
[5] *WhoJ did Jose1 introduce tl to Gudrun^ in order PRO1 to show himJ that
students at Oxford could be interesting?

(4), where the pronoun is bound by the indirect object, is well-formed when the direct
object is w/z-extracted, but (5), where the pronoun is bound by the extracted direct
object itself, is not. Now consider extraction of the indirect object.

[6] *Whok did Jose1 introduce SvenJ to ** in order PRO1 to show herk that students
at Oxford could be interesting?
[7] Who did Jose1 introduce SvenJ to fi in order PRO1 to show him) that students at
Oxford could be interesting?
167
 (7), where the pronoun is bound by the direct object, is well-formed when the indirect
object is w/i-extracted, but (6), where the pronoun is bound by the extracted indirect
object itself, is not. Both (2-3) are fine when the syntactic subject is extracted, as
shown in (8-9).

[8] Who1 /* intrc^ned SvenJ to Gudrun^ in order PRO* to show her^ that students
at Oxford could DC interesting?
[9] Who1 r* introduced SverJ to Gudrun^ in order PRO1 to show him) that students
at Oxford could be interesting?

This suggests the perfectly straightforward restriction in (10).

[10] Condition on pronominal binding (Rationale Clause)

A pronoun in a Rationale Clause cannot be bound by a w/z-extracted element in
the matrix.

It also suggests that PRO in a Rationale Clause can be so bound (i.e. w/z-extraction
does not interfere with control).
One further interesting fact which supports (10) (and which offers further
interesting data for the semantics of "event" control, see Section 5.2) comes from
sentences such as (11).

[11] Jose1 introduced SvenJ to Gudrun^ in order PRO1 to give him1/) the chance to
meet her friend, Ute.

Notice that the Rationale Clause in (11) will allows "event control" as suggested in the
paraphrase in (12).

[12] Jose's introducing Sven to Gudrun gave him the chance to meet her friend, Ute.

This seems to allow the Agent subject (Jose) to bind the pronoun (contrary to
expectation): the introduction might allow Sven to meet Ute, or it could be an excuse to

168
allow Jose to meet Ute while Sven and Gudrun chat. Notice now that if the subject is
w/i-extracted it may not act as an antecedent, in accordance with the restriction in (10).

Who1 11 introduced SvenJ to Gudrunk in order PRO1 to give him ^J the chance
* * * 1 * 2fc " /*

[13]
to meet her friend, Ute?

Now consider the effects of passivising (2-3), as shown in (14-15).

[14] SvenJ was introduced to Gudrun^ in order PRO1 to show her^ that students at
Oxford
•
could be interesting. 1 * "
[15] SvenJ was introduced to Gudrun* in order PRO1 to show himJ that students at
Oxford could be interesting.

It seems that passivisation does not effect the binding of a pronominal in a Rationale
Clause. In fact, the condition on passivisation does seem to be based exclusively on an
"event paraphrase" constraint (see Section 5.2). Where the infinitival verb is a verb
such as explain which does not allow an event subject, passivisation is excluded.

[16] Jose1 showed GudrunJ the diagrams^ in order PRO1 to explain them^ to herJ
before the interview.
[17] *GudrurJ was shown the diagrams^ in order PRO1 to explain them^ to herJ
before the interview.
[18] *The diagrams^ were shown to GudrunJ in order PRO1 to explain them^ to herJ
before the interview.

The question is then whether an implicit Agent can bind a pronominal in a Rationale
Clause which does allow the event paraphrase; consider the passivisation of (1 1) in
(19).

[19] SvenJ was introduced to Gudrun^ (by Jose) in order PRO6 to give him?1/] the
chance to meet her friend, Ute.

It is clear that Sven can bind the pronoun him straightforwardly; judgements on binding
by the implicit Agent are more delicate. It seems however that the Agent realised in a

169
by-phrase can binding the pronoun marginally and the reading improves considerably
when stress is placed on the pronoun.

[20] SvenJ was introduced to Gudrun^ (by Jose) in order PRO6 to give him1 the
chance to meet her friend, Ute.

In this case it appears that pronominal binding in the Rationale Clause is entirely
insensitive to the passive. The sole restriction is therefore the one in (10).

170
Section 4.3.2: Pronominal binding in the Telic Clause

The situation in the Telic Clause would appear to be even more straightforward.
Consider the case of a standard Telic Clause as in (1).

[1] John1 won MaryJ a car^ on the first spin of the wheel, only PRO* to lose it* for
herJ on the second.

Now consider the effect of w/i-extraction.

[2] What^ did John* win for MaryJ ft on the first spin of the wheel, only PRO1 to
lose it^
•
for herJ' •on the second?
I • •
[3] WhoJ did John1 win a car* for tl on the first spin of the wheel, only PRO1 to
lose it^ for herJ on the second?
[4] Who1 fl won MaryJ a car^ on the first spin of the wheel, only PRO1 to lose it^
for herJ on the second.

All of the sentences in (2-4) are well-formed. This suggests that w/z-extraction does not
affect the binding of pronouns whose antecedents are internal arguments (as in (2-3)). It
also suggests that the binding of PRO is unaffected by w/i-extraction. The question
remains whether the subject of the matrix can bind a pronoun in the Telic Clause and if
so, whether this binding relation is affected by w/z-extraction. Fortunately the Telic
Clause, like the Rationale Clause, allows "event control", as in (5).

[5] Derby1 beat BoltonJ decisively PRO6 to give them1 a shot at the championship.

Event control is suggested by the well-formed paraphrase in (6).

[6] [Derby's beating Bolton decisively] gave them a shot at the championship.

Now consider the effect of w/i-extracting the object, as in (7), and the subject, as in (8).

[7] WhoJ did Derby1 beat ti decisively PRO6 to give them1 a shot at the
championship?
171
 • • • •

[8] *Who1 f* beat BoltonJ decisively PRO6 to give them1 a shot at the championship?

(7) is well-formed as we would expect, suggesting that the base sentence in (6) is
perfectly compatible with w/z-extraction. However, the extraction of a subject, which
binds the pronominal in the Telic Clause, here produces ill-formedness. It therefore
appears that pronouns in the Telic Clause are sensitive to w/z-extraction of the subject
alone (where pronouns in the Rationale Clause were sensitive to all w/z-extraction). The
relevant condition is stated under (9).

[9] Condition on pronominal binding (Telic Claused

A pronoun in a Telic Clause cannot be bound by a w/z-extracted subject
in the matrix.

This restriction clearly covers a subset of the cases covered by the condition on
pronominal binding in the Rationale Clause in Section 4.3.1. It also adds to the weight
of evidence that the Telic Clause is less integrated with the modified verb than the
Rationale Clause: pronouns in the Rationale Clause are sensitive to w/z-extraction of all
arguments; pronouns in the Telic Clause are sensitive only to w/z-extraction of the
subject.

172
Section 4.3.3: Pronominal binding in the Purpose Clause

The situation in the Purpose Clause is considerably more complicated,

particularly given that the relatively free binding possibilities for the PRO subject seem
to affect well-formedness.

_ • « * • * •
[1] David1 brought RuthJ some articlesK PRO) to explain & to him1.
David1 brought Ruth] some articlesK for the professor to explain fl to him1.
* \r I '
[2]
[3] David1 brought RuthJ some articles^ PRO1 to explain fl to herJ.
* 1 1 '
[4] David1 brought RuthJ some articles* for the professor to explain fl to herJ.

The infinitives in (1-4) are all Purpose Clauses and the gap in object position is
obligatorily controlled by the matrix Theme (the articles}. This aside, the referential
possibilities are fairly, Sree. (2) and (4) show the expected alternation between PRO and
a lexical subject; I will assume that ill-formedness which occurs because of a control
violation will disappear where a lexical subject is used and that ill-formedness which
results irrespective of the alternation between PRO and a lexical subject is an instance of
illicit binding of the pronominal.
In (1), the pronominal is bound by the matrix subject and PRO by the matrix
Goal: David brings some articles along to Ruth so that she can explain these articles to
him. In (3) we have the inverse, the pronominal is bound by the matrix Goal and PRO
is bound by the matrix subject: David brings some articles along to Ruth so that he can
explain the articles tc her. The Purpose Clause in (1) is of the type standardly cited,
with the PRO subject controlled by the Goal when present. The well-formedness of (3)
is however significant. It is a counterexample to the control rule of Chierchia which
already has difficulties with implicit Goal/Beneficiaries (see Section 2.2.3); it also
suggests that fixing the control of PRO on any sort of thematic hierarchy will not work
unless the Agent and Goal are ranked equally (in which case the notion of hierarchy is
weakened). It does appear to be the case, however, that the control exhibited in (3) is

173
non-standard and the infinitive is much more restricted in its binding possibilities than
the standard (1), as we will see.
Consider now what happens when the Goal is w/i-extracted in (5-6) (I will
include a stranded preposition as this seems easier to parse).

[5] WhoJ did David1 bring some articles^ to */, PROJ to explain & to him1?
[6] WhoJ did David1 bring some articles^ to */, for the professor to explain fl to
him1?

Both sentences are well-formed: the pronoun is bound by David and PRO takes the wh-
trace as its antecedent. Now consider w/z-extraction of the Goal in (7-8).

[7] *Who) did David1 bring some articles^ to (/', PRO1 to explain ft to herJ?
[8] *WhoJ did David1 bring some articles^ to */, for the professor to explain ft to
herJ?

12
Both sentences here are ill-formed . Notice that replacing the PRO (controlled by the
syntactic subject, David) with an independently referential phrase does not affect the
well-formedness of (7); as stated above, I take this to indicate that the problem here is
with the binding of the pronominal by the w/z-extracted element, rather than with the
control of PRO.
Now consider what happens when the subject is w/z-extracted from (1-2).

[9] Whoj 1* brought RuthJ some articlesk PROJ to explain fl to him1?

[10] Who1 /* brought RuthJ some articles^ for the professor to explain fi to him1?

Both sentences are well-formed with PRO controlled by Ruth and the pronoun bound
by the w/z-extracted subject. This suggests that there cannot be a general restriction on

12
Given that parsing strategies are not as politically correct as one might wish, it could be the case that
these sentences are bad because the w/i-element is binding the "marked" feminine pronoun. This cannot,
however, be the case, because reversing the matrix participants makes no difference.

(a) *WhJ did Ruth1 bring some articlesk to </. PRO1 to explain f* to himJ?

(b) *WhoJ did Ruth1 bring some articlesk to «/'. for the professor to explain tk to himJ?
174
the binding of pronouns in the Purpose Clause by w/z-extracted elements. The situation
becomes particularly intriguing when the subject in (3-4) is w/z-extracted.

[11] *Whoi f* brought Ruth! some articlesk PRO1 to explain ft to her)?

[12] Who1 r* brought Ruth) some articles^ for the professor to explain ft to herJ?

There is an asymmetry here. It seems that the ill-formedness of (1 1) cannot be attributed

to the binding of the pronoun, because (12) is well-formed. The minimal difference
between these two setnences is the replacement in (12) of PRO by a full referential NP,
i.e. it seems that (11) is ill-formed because of a control violation: PRO cannot be
controlled in (11) by the extracted w/z-phrase. All of the Purpose Clauses, however,
seem to perfectly compatible with extraction of the matrix Theme.

[ 1 3] Whatk did Davidj bring ft to Ruth) PRO) to explain ft to him1?

[14] Whatk did David* bring fa to RuthJ for the professor to explain ft to him*?
[15] Whatk did David1 bring ft to RuthJ PRO1 to explain ft to herJ?
WhatK did David1 bring /* to RuthJ for the professor to explain tfc to her)?
'Ir* I*
1

[16]

The behaviour in the passive is more intriguing.

[17] Ruth) was brought some articles^ PRO) to explain ft to him1.

[18] Ruth) was brought some articles^ for the professor to explain ft to him1.
[19] *Ruth) was brought some articles^ PRO1 to explain ft to herJ.
[20] RuthJ was brought some articles^ for the professor to explain ft to her).

Where Ruth is not binding a pronoun in the infinitive (examples (17-18)), passivisation
is fine. Where Ruth is binding a pronoun, extraction is only allowed where there is a
full lexical NP subject for the infinitive. This suggests that in this case the Purpose
Clause is in fact passive-sensitive as far as the control of a PRO subject is concerned
and hence will select Ruth as controller in (19), giving a Binding Condition B violation
in the infinitive.
At first glance it appears that passivisation of the Theme itself interferes with

pronominal binding.
175
[21] *Some articles^ were brought to Ruth) PRQJ to explain & to him*.
[22] *Some articles* were brought to RuthJ for the professor to explain ft to him*.
[23] *Some articles* were brought to Ruth) PRO* to explain ft to herJ.
[24] Some articles* were brought to RuthJ for the professor to explain ft to her).

Notice, however, that the only example which is well-formed contains no binding by
the implicit Agent. (21-22) become well-formed when the pronoun is understood as
non-coreferential with the Agent.

[25] Some articles^ were brought to RuthJ PROJ to explain ft to the class/them.
[26] Some articles^ were brought to RuthJ for the professor to explain ft to the
class/them.

(23) cannot be saved in this way because the controller is naturally understood to be the
Agent. The ill-formedness of (21-23) therefore seems to be linked to binding by an
implicit Agent. Why this restriction should hold for passivisation of the Theme but not
of the Goal is unclear.
The data on Purpose Clauses is therefore extremely complicated, particularly in
comparison to the other infinitives of result: this is perhaps what we would expect given
that the Purpose Clause contains abstract w/Mnovement where the other two infinitives
do not. In general the standard Purpose Clause in (1) seems to be relatively free, being
insensitive to w/z-extractions of all types. The only restriction seems to occur with
passivisation and that seems to be linked to a restriction concerning control by an
implicit Agent.

176
Section 4.3.4: Classifying the "resumptive infinitive"
?

Let us now return to the general problem of establishing the correct

classification for what I have called "resumptive infinitives" which are identical to
Purpose Clauses except that they have a pronoun in place of the obligatory gap.

[1] David| bought RuthJ some articlesk PRO1 to explain themk to her).
[2] David1 brought RuthJ some articlesk for the professor to explain thern^ to herJ.

Consider the effect of w/z-extracting the Goal phrase in (1-2), as shown in (3-4).

[3] *WhoJ* did David** bring^"^ some articleski to tl, •PRO* to explain themk to herJ? 1
[4] *WhoJ did David1 bring some articles* to */, for the professor to explain themk
to herJ?

Both (3-4) are ill-formed, with the w/z-extracted NP binding the pronoun, her. This is
would be expected if these were Rationale Clauses because of the principle in Section
4.3.2 which excludes binding of a pronominal in the Rationale Clause by a vv/z-
extracted element.
Now consider the effect of extracting the Agent subject, David, as shown in (5-
6).

[5] *Whoi /* brought Ruth) some articles^ PRO1 to explain themk to herJ?
[6] Who1 tf brought RuthJ some articlesk for the professor to explain themk to herJ?
t

(5-6) is unexpected if these are Rationale Clauses. In our discussion above we saw that
Rationale Clauses allow PRO to be controlled by a wh -extracted element

straightforwardly, as in (7-8).

[7] Who1 r1' introduced SvenJ to Gemmak PRO1 to show herk that students at
Oxford could be interesting?
[8] Who1 r1' introduced SvenJ to Gemmak PRO1 to show him) that students at
Oxford could be interesting?

177
This problem really does seem to be related to the nature of these "derived Rationale
Clauses" which are directly related to Purpose Clauses. Notice that true Rationale
Clauses will allow the binding of pronominal in the infinitive by a matrix Theme and
control of PRO by a w/i-extracted subject, as in (9-10).

• • i • • i
[9] Who1 11 introduced the twins* to GemmaJ before dinner PRO1 to discuss themK
with herJ
* •
afterwards? I •
[10] Who1 tl introduced
1
the
*
twins* to GemmaJ before dinner for her partner to
discuss themK with herJ afterwards?

Notice that the infinitives in (9-10) must be Rationale Clauses because the matrix verb,
introduce is incompatible with Purpose Clauses and therefore there is no pronoun-gap
alternation, as shown in (11).

[11] *Hugh introduced the twins to Gemma before dinner PRO to discuss e with her
afterwards.

In this respect then, the sentences in (5-6) behave exactly like their Purpose Clause
equivalents in (14-13).

[12] *Who| r( brought Ruth) some articlesk PRO| to explain them* to herJ?
[ 13] *Who1
• •
f* brought RuthJ
•
some articlesk
I
PRO1 to explain if*- to herJ ? I •
[14] Who| t1 brought RuthJ some articlesK for the professor to explain them* to herJ?
[15] Who1 if brought RuthJ some articles* for the professor to explain ft to herJ?

However, this seems to be a fairly limited exception to the general pattern.

Notice the striking effect of w/z-extracting the Theme phrase (the articles).

[ 1 6] *Whatk did David1 bring tk to RuthJ PRO1 to explain themk to herJ?

[ 1 7] *Whatk did David1 bring ^ to RuthJ for the professor to explain themk to her)?

This is exactly the behaviour one finds in related Rationale Clauses, as in (18-19).

178
[ 18] *Whok did Hugh* introduce tk to Gemma) before dinner PROi to discuss themk
with herJ afterwards?
[19] *Whok did Hugh* introduce tk to GemmaJ before dinner for her partner to
discuss themk with her) afterwards?

The related Purpose Clauses (with a gap instead of a pronoun) are all perfectly well-
formed with the matrix Theme w/z-extracted. Once again this follows the prediction that
Rationale Clauses resist binding of pronouns by matrix phrases which have been wh-
extracted where Purpose Clauses do not.
Assume for a moment that the "resumptive infinitive" really is a Rationale
Clause. We would then expect to find examples of event control which license
passivisation and preposing. The example we have been discussing above does not
passivise when PRO is present.

[20] *Some articles^ were brought to Ruth) PRO* to explain themk to herJ.
[21] Some articles^ were brought to RuthJ for the professor to explain themk to herJ.

This parallels the case with the Rationale Clause.

[22] The twinsk

•
were introduced to GemmaJ before dinner PRO* to discuss themk
with her) afterwards.
[23] The twinsk were introduced to Gemma) before dinner for her partner to discuss
with herl afterwards.

However, we saw that event control allows passivisation. It appears that the infinitive
under discussion also allows event control and that that event control licenses
passivisation and preposing.

[24] Sigruni showed SvenJ the snakes^ PRO* to make themk seem less frightening to
him).
[25] The * snakes^ were shown to SvenJ PRO* to make themk seem less frightening to
him).
[26] PRO* to make themk seem less frightening to him), SigiW showed SvenJ the
snakes^.
[27] Who* I* showed SvenJ the snakesk PRO1 to make themk seem less frightening to
him)?

179
Given the cluster of facts in (24-27) it seems reasonable to assume that the "resumptive
infinitive" really is a Rationale Clause: the unusual behaviour of some such infinitives in
passivisation and preposing contexts seems to be related to more general types of
i;
restriction on the Rationale Clause, correlated to event control.

180
 CHAPTER 5: Argument Structure and Control

Section 5.1: The Purpose Clause: objects. Themes, and resultant states

Section 5.1.1: Conditions on Purpose Clause antecedence

I have already argued in some detail that there are syntactic conditions on
Purpose Clause antecedence. It was argued in Section 3.2 that there is strong evidence
that the Purpose Clause adjoins within VP and is subject to all rules which affect the
immediate phrasal projection of the verb, and this seems to be linked to conditions on
the theta-marking of the direct object by the modified verb. Adopting the view that the
Purpose Clause has an argument structure which includes a position ultimately
associated with the direct object and that that argument position must be discharged
within a binary branching structure under strictly local conditions, as specified in the
theory of Higginbotham (1985), has already forced us to adopt a particular view of the
internal structure of the VP and the projection of arguments within it, namely that the
object is a "secondary subject" in the sense of Bowers, realised in the Specifier position
of the verb's immediate projection (see Section 3.6). It appears, however, that some
sort of semantic restriction is also necessary, independently of these syntactic
:' r

constraints. *
First, current transformational analyses of the dative alternation are incompatible
with data from the Purpose Clause construction. Antecedence for the Purpose Clause is
unaffected by the dative alternation, just as it is unaffected by the passive.

[1] Sven gave a corkscrew to Jose e to open the wine with e.

[2] Sven gave Jose a corkscrew e to open the wine with e.
[3] Sven gave an executive assistant to Jose e to type his letters for him.
[4] Sven gave Jose an executive assistant e to type his letters for him.

Neither the account of Larson (1988) nor the account of Bowers (1993) predicts that the
Purpose Clause will be insensitive to dative movement.
181
 Consider first the two structures offered by Larson (1988).

[5]

Svengave v tv toSigmn

[6]
VP

Sven gave 1 Sigrun j

In Larson's view the dative alternation is analagous the passive alternation, in that a
subject gets demoted (in the passive the primary subject, in the dative alternation the
secondary subject or object) and Case is absorbed (structural Accusative Case in the
passive, inherent dative Case, realised by prepositional to, in the dative alternation): the
Caseless internal argument then raises to the Case-marked Specifier position to receive
Case (Nominative Case in the passive; Accusative Case in the dative alternation). Of
course in the passive, the demoted argument is realised only optionally and when it is
realised is realised in a prepositional phrase (headed by by). In the dative alternation the
demoted argument is realised obligatorily and is realised as a bare NP. It therefore

182
 may than of course
requires Case. Larson argues that the V is reanalysed as V, which
assign Accusative Case.
take as an
The structure in (5) correctly predicts that the Purpose Clause will
in VP-Spec. The
antecedent the Theme argument which is directly theta-marked
with anything in
predictions of (6) are less clear. The Purpose Clause cannot identify
Goal is not selected
VP-Spec which is now a S'-position. This is correct in so far as the
se adjoins to V
as the subject of the Purpose Clause's predication. If the Purpose Clau
or outside of the c-
then one might assume that it will be realised ambiguously inside
domain of the demoted direct object, as in (7-8).

Purpose Clause

demoted Direct Object

demoted Direct Object

Purpose Clause

object, as in (7),
If the Purpose Clause adjoins outside the c-domain of the demoted
for the Purpose
then there will be no open position in the verb's argument structure
m-command (with
Clause to theta-identify. If we decide to weaken the theory and admit
i
bly the Purpose
all the concomitant problems set out in Section 3.6), then presuma
is m-commanded.
Clause could also predicate itself of the Goal argument by which it
Purpose Clause can
The structure in (8) gets out of this problem straightforwardly: the
structure before the
theta-identify with the direct object position in the verb's argument
more problematic
direct object is theta-marked. The structure in (8) however is even
ose Clause and
than the structure in (7). First there is the simple fact that the Purp
183
demoted object are in the wrong linear order and therefore, in addition to the
assumptions already made, it would need to be assumed that the Purpose Clause was
obligatorily extraposed. This still will not save the structure, however. Notice that the
object requires Casel Larson has argued that it gains Case by its sister V being
reanalysed as V. However, this would require the Purpose Clause to become a part of
the verb. Not only is this strongly counterintuitive but there seem to be no cases of
structures such as (9) which would be predicted by the reanalysis.

[9] *Sven [gave to listen to] Sigrun a CD.

The structure in (6) is therefore entirely incompatible with the data from Purpose Clause
modification. The problem though is the general one noted at the outset: dative
alternation has no effect whatsoever on Purpose Clause modification, so any theory
i
which requires quite different sets of assumptions to get the right antecedence in each
i
case is suspect. A similar problem faces Bowers.
Consider the two structures suggested by Bowers (1993) for the dative
alternation.

Svenl gave 1 Si run J tv

184
 V PredP
[+cause] >1
Spec
r*^
Fred'

Fred
^>
Sfc ' f

^ j pi>
1, I /^\
Sveni gave v

Bowers views these constructions as causative constructions in which the upper VP is

headed by a causative feature. The alternation is then supposed to correlate with the
paraphrase in (12-13) for (10-11) respectively.

[12] Sven caused [Sigrun have a CD]

[13] Sven caused [a CD belong to Sigrun]

Bowers argues for these structures from a range of evidence, in particular data on
quantifier stranding. Consider how this analysis applies to infinitives of result.
Assume again that the Purpose Clause adjoins to V. We then have a choice. If
we adopt a predication analysis of these constructions of the sort already discussed then
we will only be concerned with instances of V which are "thematically active", i.e.
those whose Specifier position is a 6-position. This will be the case for the lower, but
not for the upper, VP in both cases. If we adopt the view that the Purpose Clause's
antecedence is fixed by minimal distance, or if we mimick derivation by the vertical
binding of argument indices, then antecedence by an argument in the Specifier of the
upper VP will become available. This is clearly undesirable because the dative shifted
185
version in (26) has the Goal argument in the Spec of the upper VP. I will therefore take
it that this is further evidence of the advantage of a theory of argument structure as
necessarily local as that of Higginbotham (1985).
This analysis fares better than that of Larson (1988). If the Purpose Clause
adjoins to the lower V, then it will in both structures be able to theta-identify with the
position of the Theme argument which is base-generated in (lower) VP-Spec.
However, in the shifted structure in (10), the Goal argument is directly theta-marked
external to the lower VP: its position in the argument structure will therefore also be
available in this construction (although not in the unshifted version, in (11)). This
problem would be removed if the Purpose predicate imposed thematic restrictions on its
antecedent: of the Theme-NP and the Goal-NP made available in the structure in (10),
only the Theme-NP would be compatible with Purpose Clause interpretation. We must
already, in fact, assume such a restriction to prevent the Purpose predicate identifying
i>
with the Agent position of the verb in standard structures.
When we consider the distribution of the Rationale Clause, the problems
become more serious.We must assume that the Rationale Clause adjoins to either VP or
Pred' to account for the "VP-adjunction" data discussed in Section 3.3. However, if it
can adjoin to any VP or Pred' then it will be able to adjoin to the lower VP or the lower
Pred': this will allow it to theta-identify with the Goal-NP in (10). Even worse, just as
the Rationale Clause may adjoin to the lower VP or Pred', so the Purpose Clause may
adjoin to the upper V; if both occur simultaneously then the Purpose Clause will
hierarchically dominate the Rationale Clause, contrary to all evidence. Thematic
restrictions will of course force the Purpose Clause to adjoin within the lower VP and
the Rationale Clause to adjoin within the upper PredP, as required.
The analysis of Larson (1988) is therefore incompatible with facts concerning
the Purpose Clause and the analysis of Bowers (1993) is consistent only in so far as
thematic restrictions are also available. Some sort of "argument" restriction will
therefore be necessary under current syntactic accounts.

186
 A second reason for assuming argument constraints of some sort comes from
the Purpose Clause's insensitivity to alternations such as the locative alternation: it
appears that the Purpose Clause will always take the Theme argument of a predicate as
its antecedent, regardless of its realisation as direct object or oblique (PP) argument.

[14] Jose had a surplus of turnips, so he loaded his cart with them e to sell e at the
market.

The third reason for assuming that the Purpose Clause imposes some sort of
argument restriction on the VP in which it occurs is that only verbs denoting certain
types of event are compatible with Purpose Clause modification (a restriction on the
event-type of a verb is clearly an argument restriction, entirely unrelated to the
grammatical functions of particular NPs). As Bach (1982) has pointed out, there is a
class of verbs excluded from modification by the Purpose Clause which are robustly
transitive (i.e. they have direct objects) and which have often been classified with
Theme internal arguments.

[15] *Ruth slapped Katherine e to open the safe.

[16] *Ruth tickled Katherine e to tell the secret.

The ill-formedness of (15-16) argues against a purely syntactic analysis because verbs
of this class (slap, tickle, hit, beat) strictly subcategorise for a direct object, as shown
by the ill-formedness of (17), and yet are almost uninterpretable with the relevant
reading in (15-16).

[17] *Ruth slapped.

*Ruth tickled.
*Ruth hit.
*Ruth beat.

Nevertheless the reading required is a perfectly reasonable one; (16) might be

paraphrased by the gerundive construction in (18).

187
[ 1 8] Ruth tickled Katherine into telling the secret.

The problem, as Bach observed, is to do with the lexical entailments of these verbs:
these verbs do not entail a change of state in their object, unlike the verbs bring or give.
If I bring something to a place then that object must end up in that location having been
elsewhere; if I give something to someone, then that object must come to be in that
person's possession or control having not been in their possession or control before.
For the verbs slap, tickle, hit, beat this is not so. If Ruth slaps Katherine she might be
upset, hurt, or offended, but the verb does not entail that any of these things be the
case; it does not even entail that Katherine notice that she has been slapped (a change of
mental state). Similarly, if Ruth tickles Katherine, she might be angry, amused, or
enthusiastic but the verb itself does not entail that there be any reaction at all. We may
feel that slapping, hitting, or tickling are extremely effective tortures for making people
serve the purposes which we intend for them; they may be assessed in terms of our
world knowledge as being particularly effective ways of making people do things; but it
seems that linguistically that simply is not enough. If the verb does not entail a change
of state in the object then it will not allow modification by a Purpose Clause because the
Purpose Clause tells us what purpose an object serves by virtue of its change of

One note of caution should be sounded here. It was pointed out to me by James
Higginbotham that there is a state with respect to which all predicates entail a change:
the perfective state. That is, if some verb V is true of an object NP, then it will be true
in consequence that the NP is in the state of having been V-ed: if I bend a coathanger at
some point in time, t^, then it will be true of the coathanger at every subsequent point in
time, f^ +/I, that the coathanger will be in the state of having been bent. Clearly, we
want to rule out the perfective state from the set of states that are relevant to Purpose
Clause modification (otherwise all verbs would be compatible with modification by the

1 ^ I set aside here the problem of unrealised or incomplete objects and assume that for both build and
design there is an "object" to which we can refer.
188
Purpose Clause). It seems that what is required is a notion of causal efficacy: the state
in question must be causally efficacious with respect to the object having the property in
question. If I repeatedly bend a coathanger it may well end up in a brittle state; that
present state will be causally efficacious with respect to a breaking of the coathanger.
The event of bending may itself be causally efficacious with respect to the state of
brittleness. But the perfective state of having been bent is not causally efficacious in the
sense that there is no possible causal consequence of the coathanger's having been bent
which is not reducible to the causal efficacy of the initial event of bending and the states
to which it leads. James Higginbotham has observed to me that this is reminiscent of a
debate in the natural sciences concerning the formulation of natural laws. Can we
formulate laws in terms of perfective states? Can we say for instance that if I have bent
a coathanger repeatedly, it will break? It seems that the answer is no. The law must be
formulated in terms of present events and their causal consequences: if I bend a
coathanger repeatedly then it may come to be in a certain state; that state may be such
that some other event may occur.
We can therefore maintain our position that the Purpose Clause is related to a
t

change of state in the object where the object comes to have a property by virtue of
ending up in a particular state. Notice however that the Purpose Clause seems to have
access only to information lexically provided by the verb concerning the nature of the
modified event, i.e. the Purpose Clause can make reference only to the verb's lexical
entailments in evaluating whether a change of state of the relevant sort has occurred.
Notice that (19) ill-formed.

[191 *Nick bent the coathanger e to break e.

Although an event of bending a coathanger might in actual fact be causally efficacious

with respect to a state of brittleness which is itself causally efficicious with respect to an
event of breaking, that is not entailed by the verb bend: say that the coathanger is made
of an imperishable alloy, then the bending will have no consequence. Notice that where

189
the purpose served by the coathanger relates to a change of shape then the Purpose
Clause becomes acceptable.

[20] Nick bent the eoathanger e to fit around the doorknob.

An event of bending entails a change of shape in the object: an event of bending is

causally efficacious with respect to the state of the object being in some (different)
shape and the verb bend cannot be true unless the bending does in fact have that
consequence. The conditions imposed by the Purpose Clause are therefore met: the
modified verb lexically entails a change of state in its object and it is in that very state
that the object has the property. The reason that verbs such as slap, tickle, beat and hit
may not be modified by a Purpose Clause is that they entail no change of state in their
object with respect to which the object comes to have a property, just as the verb bend
is incompatible with a Purpose Clause when the property in question is not one which
the object has in the lexically entailed resultant state.
This analysis also allows us to explain a rather curious restriction on the internal
structure of the Purpose Clause. Jones (1991) argues that the Purpose Clause cannot
occur in the perfective, a fact which follows necessarily from his view that the Purpose
Clause is a bare VP. It seems, however, that Purpose Clauses can occur in the
perfective in certain contexts.

[21] He set up the security system to have closed the gate before electrifying the
fence.

Two points should be noticed about such examples. First, the infinitival verb does
occur in the perfective and therefore a syntactic position must be available for it: this
favours the view of Purpose Clauses as fully clausal structures. Second, the perfectivity
of the infinitival verb is internally oriented: the closing of the gate is assessed as being
in the past with respect to the electrifying of the fence not with regard to the setting up
of the security system. We might link this restriction to our observation concerning

190
causal efficacy. The resultant state of an event cannot by definition be causally
efficacious with respect to an event which has already occurred. In terms of the
attribution of properties, a resultant state can only be causally efficacious in attributing
properties which are direct consequences of its own properties. Thus in (7) the setting
up could not be causally efficacious with respect to a having closed the gate. It could
however be efficacious with respect to a closing of the gate which is such that it will
always precede an electrifying of the fence.
Our account therefore predicts the two crucial properties of the Purpose Clause
in (21). The Purpose Clause is a fully clausal structure and therefore allows auxiliaries
such as the perfective. It will only however allow its complement to denote properties
with respect to which the resultant state of the main verb can be causally efficacious and
therefore the interpretation of perfectivity in the Purpose Clause must be bound within
the scope of the lambda abstractor itself.
We have therefore linked the Purpose Clause to the notion of a change of state,
constrained by the notion of causal efficacy. It seems however that the Purpose Clause
is compatible with any verb which involves, as a necessary part of its meaning, a state
of some sort (as long as it is causally efficacious). There are in fact four types of event
with which the Puipose Clause is compatible.

[22] a telic event

[23] an atelic event
[24] a continuance of state
[25] a state

Examples of these, with the Purpose Clause modification, as given in (26-29).

[26] Sven pushed the cart into the arena for the crowds to see e.
[27] Sven pushed the cart along the road for the crowds to see e.
[28] Sven kept the cart in the arena for the crowds to see e.
[29] The cart was in the arena for the crowds to see e.

Bach's (1982) notion of resultant state is clearly directed at cases such as (26); it does
not however take in (27-28), although Bach does mention change and continuance of
191
state as suitable for Purpose Clause modification. (29) is not generally mentioned as an
instance of Purpose Clause modification, although its status as such an infinitive,
parallel to (26-28) is I think clear. I assume that what (26-28) all have in common is that
each is understood with respect to states. If something is pushed into the arena then it
comes to be in the state of being in the arena; if something is pushed along the road it
comes to be in a state of motion (and in fact to pass through a series of states of being at
points along the trajectory); if something is kept in the arena then, through the agency of
some individual, the object is maintained in a particular state, namely being in the arena;
and of course if something is in the arena then it is by definition in the state of being in
the arena. Let us then say that there is a function 5 from events to the states which are
lexically specified as parts of those events. S applied to a telic pushing will give the
final state; S applied to an atelic pushing will give the state of motion; S applied to verbs
such as keep will give the state of enduring in some state; and S applied to states is
identity. We then have the argument grid for the Purpose predicate in (30),

[30] P(x,e,AXz (3el) O(z,el))

and the conditions on reference in (31).

i ..
[31] e is a state of x and e is to be causally efficacious in bringing it about that x has
the property cf being a z such that (Be1) O(z,e!)

The question is then how these two are to be related. Notice that under the present
analysis the e-position in the argument grid of P is theta-identified with the event
position of the modified verb: but if e is a pushing into the arena then e is not itself a
state of x in which the crowds can see x; rather it is the state of being in the arena which
gives x that property, i.e. it is S(e) not e itself.
We then have two options. The first is to say that the event position in (30) is in
fact identified with the matrix event but that the conditions on the predicate themselves
make use of S(e), as in (32).
192
[32] P(x,e,AXz (3ei) O(z,el)) <-» [3ef: e'=S(e)] e' is a state of x and e1 is to be
causally efficacious in bringing it about that x has the
property of being a z such that (3el) <X>(z,e!)

This formulation will bring out the relevant features of the Purpose Clause's
interpretation, namely that it concerns the properties which objects are to have in
particular states. However, there is a sense in which the result is achieved by sleight of
hand. Our intuitions tell us strongly that x has the relevant property because it is in that
state, i.e. P is a relation between an entity, a state, and property such that x is in that
state for the sake of it having the relevant property; this is not what (18) says. (18) says
that x is involved in an event for the sake of having a property and that interpretation is
licensed by the fact that the event is related in the appropriate way to a state.
Higginbotham has suggested to me that this provides evidence that theta identification
might in fact be relativised to functions on events. When the Purpose Clause adjoins to
V we have a situation as in (33).

[33]

<X,E> CP<Y,E'>

It may be that theta-identification of the event variables in (33) can be translated directly
in terms of a function on e, as in (34).

[34] push(x,e) & P(x,S(e),AXz (3el) O(z, el))

Now the Purpose predicate directly relates an entity to a state, derived from the matrix
event. The conditions on the reference of the predicate will then relate directly to the
derived state, as in (35).

193
[35] P(x,S(e),AXz (3el) O(z, el)) <-» S(e) is a state of x and S(e) is to be
causally efficacious in bringing it about that x has
the property of being a z such that (3el) <£(z,el)

Such a revision of the principle of theta identification would need to be justified

on more general grounds. It appears however that reference to states which are lexically
available parts of events is indeed generally available. Higginbotham has observed
examples such as (36-37).

[36] Sven wrapped the present nicely.

[37] Sven tossed the caber nicely.

Notice in (36) that it is not the wrapping which is nice or the present itself, but the
present in the wrapped state. The verb wrap lexically specifies a resultant state of the
object and it is the object in that state which is nice. Even more strikingly in (37), it is
not the tossing of the caber as a whole event or the caber itself which is nice but
i
presumably the trajectory which it follows (cf a nice shot in football is one which
traverses an impressive trajectory even if it does not end up in the right place). It seems
that S(e) is therefore generally available to this group of object modifiers.
This account predicts that where two appropriate states are available it will be
ambiguous, as appears to be the case.

[38] Sven pushed the cart to the end of the street for the crowds to see e.

It may be that there is a crowd congregated at the end of the street and that the cart has
the property that the crowds can see it by virtue of push having the resultant state of the
cart being at the end of the street. Or it may be that the crowd is lining the entire street
(although there is no one congregated at the end) and hence the cart has the property that
the crowds can see it by virtue of push specifying a state of motion, namely it passing
along the street. Notice that this view is compatible with either of two accounts of telic-
atelic alternation which occurs with push. We might say that the verb push is

194
ambiguous between a verb push which takes an ordered pair of events, where one is a
process and the other a telos, and a verb push which takes only a process; or we might
say that the PP itself introduces either a process (as for towards) or a process and telos
(as for to). In either case, V will have associated with it argument grids which lexically
specify the relevant types of event.
The view adopted here also suggests a solution to a complex set of data
involving verbs of creation. Bach (1982) notes that only verbs with a "positive resultant
state" are available for Purpose Clause modification, thus we have (39) but not (40).

[39] John built the swing in the garden to annoy his neighbours with.
[40] *John destroyed the swing in the garden to annoy his neighbours with.

This contrast follows straightforwardly on this view: the result of building is that there
is some object; the Purpose Clause gives us the purpose of the object's existence. The
result of destroying, however, is that there is no object and hence the Purpose Clause
cannot apply because it tells us the purpose for which objects are in particular states, but
here there is no object to be in a state. The restriction here is rather more specific than
the notion of "positive" or "negative" resultant states. Notice that there are some
"negative" result state verbs which are perfectly compatible with Purpose Clause
modification.

[41] John broke up the wooden stool to put on the fire.

[42] John shredded the newspaper to stuff into the mattress.

Verbs such as shred and break up lexically specify a material result (and one which is
clearly related to the disassembled object). The Purpose Clause tells us the purpose
which these objects serve by being in this disassembled state. There is, however, one
further class of verbs. A verb like burn will have a material result, namely ash, but it
seems to pattern with destroy.

[43] ?*John burnt the bulky newspapers to put in the special ash containers.

195
 (43) might mean that the newspapers were too bulky to store, so John burned them so
that, as ash, they could be put in the ash containers. Remember that the Purpose
predicate must theta-identify with the object of the verb: it may be that certain types of
material result are not considered to be instances of the original object. We might
willingly describe a pile of shredded paper or a pile of wood as "the newspaper" or "the
stool" of which they are parts. We may not willingly describe a pile of ash as an
instance of the object from which it is derived. We must be careful here however, as
examples such as (44) seem more acceptable.

[44] ?John burnt the newspapers to blacken his face with.

It may be that the purpose in (43) is not so much a purpose of the burnt newspapers as
the purpose of the event of burning itself (we therefore do not have the right sort of
antecedence for the property); in (44) it really is the burnt newspaper which is serving a
purpose.
Whatever the solution to this problem, it appears that acceptability of the
Purpose Clause is linked to the particular state of the object, which is provided by S(e).
The crucial point is that S(e) makes reference only to lexical properties represented in
the argument grids of the relevant predicates. Thus the "impact verbs" slap, tickle etc
will be excluded because they are not lexically associated with a state of any sort
(except the perfective state which is not causally efficacious); the "negative resultant
state verbs" such as destroy will also be excluded because there state is not one in
which the object has the relevant property as the Purpose Clause requires. This complex
set of selectional restrictions fall out striaghtforwardly from the meaning of the Purpose
predicate itself, integrated into the structure of the modified sentence by a generalised
theta theory. Notice that the restriction to Theme antecedents, in so far as it is valid, is
parasitic on the notion of event structure because Themes are usually taken to be the
subject of states.

196
 This will obviously raise the question of the nature of the function S(e) itself.
Thus far we have restricted the function, S(e), to events and substates listed in the
verb's lexical entry; and we have defined the function itself merely in terms of a list of
possible outputs. Ultimately, we will need a precise definition of the semantic criteria
which identify an event, e, as having a substate of the sort required by the function S(e)
(just as we need a precise definition of what it is that makes a Theme a Theme or an
Agent an Agent). I take this to be a problem for further reasearch.
Assuming this analysis of the semantic restriction imposed by the Purpose
Clause, this then raises the interesting question of whether the Purpose Clause reading
is available with unaccusative verbs such as go or leave which are standardly taken to
have a Theme subject. According to the definition set out above they should be
compatible with a Purpose Clause reading because they lexically specify a state (a state
of motion). For the prediction to have weight the reading must be distinguishable from
a possible Rationale Clause reading which will presumably appear the same in the linear
string. Consider (45).

[45] John left to frighten Bill.

There do in fact seem to be two readings and those readings are distinguishable in the
terms already adopted and seem to have the expected syntactic ramifications. Consider
first a situation in which John and Bill are in a dark room and John sneaks out in the
hope that Bill, discovering his absence will panic. (45) might reasonably be asserted of
such a situation. This reading is compatible with the insertion of in order.

[46] John left in order to frighten Bill.

It is also compatible with an infinitive that does not contain a gap. Let us say that Bill is
extremely confident when John is around; Bill is frightened of Fred in general but John
makes him feel safe. We might then imagine a situation where John maliciously

197
abandons Bill to the company of Fred, knowing that this will give Fred the chance to
threaten Bill. (47) might be asserted of this rather bizarre scenario.

[47] John left (in order) for Fred to frighten Bill.

In both cases, the infinitive specifies some general intention that the Agent hopes to
have fulfilled as a result of performing the action. The intention is held with respect to
leaving in general: i.e. it has some property or other that the Agent believes will
facilitate the fulfilment of his intention.
(45) is however compatible with another, more specific reading. Consider a
situation in which John has been hired to threaten Bill; he is setting off to fulfil his
contract. That is, he is at some location (at home maybe) and he leaves so as to get to
some other location where he will perform the relevant task. He needs to leave the place
he is at because he cannot perform the task there; it is by being at some other place that
he is enabled to act in the required way. This is of course precisely the reading we
would expect of a Purpose Clause. Some object changes state and because of that
change of state it acquires some property. The Purpose reading is licensed by the
change of state and the object is taken only to acquire the relevant property because its
state has changed in the relevant manner. Notice that in this reading we cannot have in
order insertion as in (46). Nor does it make sense to have a lexical subject (as in (47))
because the whole point of this reading is that the Agent intends to perform some action
by getting him to a particular location. Further, if one has a sentence containing two
infinitives, one having the general intention reading and one having the resultant state
reading, the ordering restrictions are as expected.

[48] John left to tall- to Bill to irritate Tom.

Take (48) to mean that John left his present location to go to some other location at
which he intended to talk to Bill (Purpose Clause reading) and that he did so with the
general intention that this "leaving to talk to Bill" irritate Tom (Rationale Clause
198
reading). If we construe the two infinitives in this way then we cannot reverse their
order, as expected.

[49] *John left to irritate Tom to talk to Bill.

Note also that if we take both infinitives to be Purpose Clauses (i.e. to denote what
John intends to do when he gets to his destination) then the sentence is ill-formed any
way: we cannot take the two infinitives simply to be two purposes of the same sort.
This is of course again what we would expect because we know that Purpose Clauses
do not naturally iterate in "external adjunction". Preposing has the effect of entirely
disambiguating the sentence.

[50] To frighten Bill, John left.

This can only mean that John intended his leaving to frighten Bill, it cannot mean that
he went somewhere where he intended to frighten Bill.
It therefore appears that the selectional restriction suggested for the Purpose
Clause correctly predicts its compatibility with a range of constructions including telic
predicates involving a resultant state, atelic predicates involving a change of state, verbs
such as keep and remain involving a continuance of state, states proper, and
unaccusatives, while excluding the "impact verbs".

199
Section 5.1.2: Some problems

There are a number of cases which stand as clear counterexamples to the

restriction of Purpose Clause control to the Theme of the sentence and which are
incompatible with the specific theory suggested so far, although some notion of
"resultant state" is clearly necessary. It appears that the subject gap Purpose Clause (and
only the subject-gap Purpose Clause) can be controlled by non-Theme internal
arguments of the verb. Consider for instance the case in (1).

[1] Lobbyists gave MPs1 £1,000 each e^ to ask questions in Parliament.

Like a standard Purpose Clause, this sentence can be passivised, as in (2), and like a
standard Purpose Clause it cannot be preposed, as in (3).

[2] MPs were given £1,000 each to ask questions in Parliament.

[3] *To ask questions in Parliament, lobbyists gave MPs £1,000 each.

Notice that this cannot be linked to the fact that in the double object construction, the
Goal acts as the direct object of the verb, because the PP version of (2) is also
acceptable.

• •
[4] Lobbyists gave £1,000 to each MP1 el to ask questions in Parliament.

It therefore seems possible to have control of a subject-gap Purpose Clause by the Goal
or Location argument, irrespective of its realisation. Notice that the event analysis
cannot be used as a means of avoiding this conclusion in this case.

[5] *Giving MPs £ 1,000 each asked questions in Parliament.

Notice that this construction is indeed only compatible with the subject-gap Purpose
Clauses.
200
 • •

[6] *Lobbyists gave MPs1 £1,000 each for their backers to be able to ask el
questions.
[7] *Lobbyists gave MPs1 £1,000 each for their backers to be able to talk to e*
privately.

The ill-formedness here is purely syntactic: replacement of the gaps with pronouns
produces well-formed sentences.

_ •
[8] The lobbyist gave MPs1 £1,000 each in order for his backer to be able to ask
them questions.
[9] The lobbyist gave MPs1 £1,000 each in order for his backer to be able to talk to
them privately.

We therefore have a robust of example where the Goal of the main verb controls
the obligatory gap in the Purpose Clause, not the Theme. However, it would be
possible to argue that this is not precisely the case: £1,000 is in fact what Jackendoff
calls a secondary Theme. Notice that secondary Themes are often distinguished from
Themes proper in the syntax, for instance in (10) which is the same as (1) except that
the infinitive here is an implicit argument.

[10] Lobbyists offered the MPs £1,000 to ask questions in Parliament.

)

Here again it is the Goal which controls the subject gap, and again the control relation is
insensitive to the dative alternation, as in (11).

[11] Lobbyists offered £ 1,000 to the MPs to ask questions in Parliament.

There is also however a version of (10) with a Goal subject, as in (12).

[ 12] The MPs offered to ask questions in Parliament for £ 1,000.

Notice that the/or-p^rase belongs with the main verb not the infinitive.

201
[13] The MPs offered, for £1,000, to ask questions in Parliament.

The preposition for is a standard marker of a secondary Theme with transactional verbs.
We might therefore ssek to argue that secondary Themes are quite distinct from Themes
proper when it comes to control. However, it appears that secondary Themes behave
exactly like Themes proper in other examples.

[14] Lobbyists gave MPs £1,000 to spend on holiday.

Thus, no theory which tries to explain this anomaly by taking the secondary Theme to
be inaccessible as a controller in general (or placing it low on the Thematic Hierarchy)
will work: for any object-gap Purpose Clause it behaves exactly like an ordinary
Theme. In fact the assumption that the subject-gap Purpose Clause is controlled by the
Goal irrespective of syntactic realisation captures the broader distribution of the
construction, as shown in (15-18).

[15] The MPs got £1,000 (*in order) to ask questions in Parliament.
[ 16] The MPs got £ 1,000 (*in order) to spend on holiday.
[17] £1,000 went to each MP (*in order) to ask questions in Parliament.
[18] £1,000 went to each MP (*in order) to spend on holiday.

Another counterexample to the Theme-only analysis of subject-gap Purpose

Clauses was first pointed out to me by James Higginbotham.

• * I • •

[19] John1 put the photosJ under the spotlightK e to illuminate them) for him1.

Notice that the gap in (19) cannot be controlled by John which already binds the
.,: •?.

pronoun, him\ this is confirmed by the ill-formedness of (20).

[20] "John1 illuminated them for him1.

202
Nor can it be controlled by the photos which control the pronoun, them', this is
confirmed by the ill-formedness of (21).

[21] *The photos) illuminated them) for him.

Control by the spotlight would however be possible, as shown by (22).

i

[22] The spotlight1' illuminated them) for him1.

This however runs against standard analyses in two ways. First the c-command
constraint is violated because the NP is governed by a pronoun. In a theory such as that
of Williams which makes a distinction between subcategorised, semantically empty
prepositions, and contentful prepositions this problem is resolved relatively
straightforwardly, as the external index of the NP will be realised on the PP. Second,
spotlight bears the role of Location not of Theme. It clearly forms part of the "resultant
state" in some way, but not in the way expected.
Notice two points further, however. First it is hard to construct similar
i%
examples where the controlled gap is VP-internal.

[23] *Johni put the photosJ under the spotlight^ e* to illuminate them in A

Further, the Location cannot act as the subject-controller in an object-gap Purpose

Clause.

[24] * John* put the photos) under the spotlight^ e^ to illuminate el for him.

The sentence in (19) therefore seems to be strictly limited to a Location-controlled

subject-gap infinitive. The question is then whether (19) is in fact a Purpose Clause at
all. It is possible that it is in fact a Rationale Clause with event control (see Section 5.2).
This would give the paraphrase in (25).

203
[25] Putting the phctos under the spotlight illuminated them for him.

Unfortunately assessing the plausibility of (25) is complicated by the fact that Rationale
Clauses containing bound pronouns do not exhibit the full range of behaviour (see
Section 4.3.1). They do not prepose, as (19) does not.

[26] *To illuminate them for him, John put the photos in the spotlight.

They do not allow passivisation, as (19) does not.

[27] *The photos were put in the spotlight to illuminate them for him.
, ><

The only diagnostic which can be interestingly applied is in order insertion which seems
to come out marginally against the interpretation of (19) as a Rationale Clause.

[28] *John put the photos in the spotlight in order to illuminate them for him.

This evidence becomes stronger when one considers that the same sentence in which
the pronoun is omitted or in which it is replaced with a reflexive is perfectly acceptable.

[29] John put the photos in the spotlight in order to illuminate them (for himself).

This suggests that the event control analysis, though plausible, is in fact untenable. It
therefore seems that (19) really is a (subject-gap) Purpose Clause with Location control.
It is clear that both the examples with give and the examples with put obey the
general constraint that the verb must lexically specify a state in which some individual
has a property. It is not clear however why control by the non-Theme is acceptable in
these cases but not in other standard cases of dative alternation, such as those discussed
in Section 5.1.1. It is also obscure why the possibility is available only to subject-gap
Purpose Clauses. I leave the question here.

204
Section 5.2: The Rationale Clause: intentions and control

The Rationale Clause cannot be controlled by an internal argument of the matrix

verb: sentences ambiguous between a subject-gap Purpose Clause and a Rationale
Clause are easily distinguished by inserting the phrase in order.

[1]
(a) Sven brought Armistead to the party to impress the guests.
(b) Sven brought Armistead to the party in order to impress the guests.

In (la) we have two readings: either it is Sven who is impressing the guests by bringing
along Armistead; or it is Armistead impressing the guests (which is why Sven brought
him along). The first is a Rationale reading and the second is a Purpose reading. In (Ib)
we have only the Rationale reading (i.e. Sven is doing the impressing).
In general, Rationale Clauses are only compatible with the Agent of the sentence
(i.e. the Rationale Clause will not tolerate any subject of any role). If we assume some
sort of "internal subject", then the Rationale Clause selects an internal subject which is
an Agent.

[2] The police blocked the road to stop the convoy from coming through.
[3] The road was blocked by the police to stop the convoy from coming through.

In both (2) and (3), the police bears the role Agent with respect to the verb stop (it is the
police who are doing the stopping). In (2), the police is the Subject; in (3), it is an
Adjunct, governed by by. Nevertheless, the police remains the controller of the
infinitive, regardless of this change of Grammatical Function. However, it appears that
restricting controller selection to the role of Agent is too strong. First, it seems that the
Rationale Clause is compatible with other roles as long as they are susceptible to an
intentional reading. The verb receive has a Goal as subject and need not receive an

intentional reading, as in (4).

[4] John received a death threat.

205
However, it can occur with a Rationale Clause, but in this case the subject is construed
as acting with active intention, as in (5).

[5] The Queen received the ambassador to placate the offended prince.

Further, it seems that the Rationale Clause is perfectly compatible with copular
predication where the attribution of the property to the subject is seen to be one which is
under the subject's control.

[6] John is a member of CND to offend his father (not out of any sense of
commitment).

Where the property is conceived of as inalienable the Rationale reading fails.

[7] *John is tall to get onto the basketball team.

The facts follow straightforwardly in the analysis suggested because the

predicate is a relation between an individual, an event, and an intention such that the
individual participates in the event with the intention in mind: in this case the event must
be the sort of event that an individual can manipulate with a particular intention (hence
being a member of an organisation is fine but being a certain height is not); the subject
need not be an Agent but whatever role it plays in the event it must play with active
intention (as in (5)); and of course for verbs which do have an Agent it is the Agent
which is necessarily read as the intender of action, so the restriction to Agent in
standard cases is predicted. We might therefore have the conditions on reference in (8).

[8] R(x,e,Ap) <-> x acts in e with the intention that p

One problem with this analysis is posed by the passive as it appears that the
Rationale Clause can appear without any lexically realised Agent.
206
[9] The lights were turned off to conserve electricity.

On the basis of such examples Jones suggests that the Rationale Clause be taken as an
instance of discourse control. However, this seems far too weak (as Jones himself
implies by recognising that where the Agent is lexically realised, it must control). If the
Rationale Clause simply supplied an intention which someone had with respect to the
matrix event, and the attribution of that intention was left to general principles of
discourse, it would presumably be possible for a salient "discourse Agent" to control
the Rationale Clause. This seems not to be the case. Consider (10).

[10] John and Bill trashed their table to attract the attention of the police.

Consider (10) asserted in the following situation: a barman wishes to summon the
police because of a troublemaker in the bar, but he does not want the troublemaker to
realise that he is intentionally summoning the police; he therefore asks two friends,
John and Bill, to cause a disturbance which will attract a passing policeman's attention.
In this situation, the intention to attract the attention of the police is clearly the barman's
and the initiation of the table-trashing is at the barman's instigation. The use of a
Rationale Clause is perfectly fine in (10), despite the fact that the barman is not himself
refered to lexically (or lexically implicit) in the sentence.
This argument seems to me equally spurious. The predicate I have suggested
requires the syntactic Agent to intend the proposition, it does not prevent anyone else
having intentions of the same sort, even, perhaps, stronger intentions. And notice that
the syntactic Agent must be interpreted as being at least complicitous in the intention to
summon the police. This seems to me intuitively clear, but to bring it out, notice that
one cannot attribute to John and Bill a lack of knowledge about what they are doing.

[11) *John and Bill (unwittingly) trashed their table (unwittingly), to attract the
attention of the police.

207
So the syntactic Agent must be at least complicitous in the relevant intention; other
entities may or may not have such intentions and to any degree of intensity.
It is clear then that a simple theory of discourse control is unacceptable because
the relation involved in Rationale Clause modification is strictly local. Of course, Jones
himself observes tha' where there is an Agent lexically present it will be the controller.
If however discourse control is too weak in general then motivating it using the passive
seems rather weak evidence. Two quite separate sets of facts argue against using the
passive as a motivation for discourse control. First there is the extremely robust contrast
with the middle construction, as in (12-13).

[12] The ship was sunk to collect the insurance.

[ 13] *The ship sank to collect the insurance.

It seems that the Agent is syntactically available in (12) even though it is not realised by
r
an overt lexical phrase: it may be that the theta grid of passive sunk contains a position
for the Agent which is in some way saturated by the passive morphology itself, where
the theta grid for middle sink contains no position for the Agent at all. Clearly the
Rationale Clause in (12) describes the Agent's intentions (the sinker intends the
collection). Returning to the discussion of (10) above, it seems to me that a
passivisation of (10) makes the same point: whatever x trashes the tables also intends to
alert the police.

[ 14] The table was trashed to attract the attention of the police.

Notice again that the inclusion of unwittingly is anomalous.

[ 15] *The table was trashed unwittingly to attract the attention of the police.

If the passive (unlike the active) really did allow discourse control then one might
expect that the property of unwittingness might be predicated of the table-trashers who
were being manipulated by the discourse Agent whose intentions are being described by
208
the Rationale Clause. This seems to me to be quite impossible: we may not know who
trashed the tables but we know that Agency in the trashing of the tables and intention
with respect to that trashing are linked. Notice that this still allows a certain vagueness
in the interpretation of Agency which might be either lexically specified or quite general
in the passive. Consider (16).

[16] The Mafia sank the ship to collect the insurance.

In (16) we do not assume that every member of the Mafia was physically involved in
the sinking of the ship; nevertheless (16) is true even where one individual sinks the
ship as long as that individual is appropriately related to the Mafia and acts in the
sinking of the ship purely as part of that relationship (rather than through private
motives). Similarly we may assume that the individual who sinks the ship does not
have an intention to collect the insurance but we must assume that "the Mafia" (however
that institution is to be understood) does have that intention. The same point may be
made when the Rationale Clause modifies a passive sentence: the Agent of the action
and the intender of the proposition must be "the same individual" in some relevant
sense.
The above examples therefore suggest that the attribution of intention by the
Rationale Clause is strictly local (modulo vague readings as discussed). Another set of
examples also argues against using the passive to support a discourse control analysis:
what Williams (1974; 1987b) calls event control.

[17] John turned off the lights to save money.

[ 18] Turning off the lights saved money.

Jones (1991: 37, fn 1) mentions this observation and cites an anonymous reviewer as
arguing against Williams' analysis with examples such as the one in (19).

• • • •
[ 19] Mary1 brought JohnJ along in order el to talk to him).

209
It seems however that the Rationale Clause should not be taken to have only event
control or only individual control, but (as Hegarty (1989) suggests in a similar case) to
allow both. This seems to be related to an important constraint on passivisation with the
Rationale Clause, which I believe has not been mentioned before. Consider the effect of
inserting a reflexive into (17) and passivising as in (20-21).

[20] The lights were turned off to save money.

[21] *The fights were turned off to save himself some money.

Suppose that Rationale Clauses are controlled in two ways: either by the subject of the
matrix clause, if the subject bears the relevant role with respect to the matrix event; or
by the matrix event itself if there is an x which bears that role with respect to it.
Consider how this would work with regard to the examples discussed so far.

[22]
(a) John turned off the lights to save money.
(b) John* eJ turned off the lights^ [PROW*k to save money].
(c) John saved money.
(d) Turning off the lights saved money.
[23]
(a) The lights were turned off to save money.
(b) The lights* eJ were turned off [PRO**/J to save money].
(c) *The lights saved money.
(d) Turning off the lights saved money.
[24]
(a) John • turned
•
off the lights to 1save himself
* /sk" /ski
some money. *
(b) John1 eJ turned off the lights* [PRO1/ J' * to save himself1 some money].
(c) John saved himself some money.
(d) Turning off the lights saved himself money.
[25]
(a) *The lights were turned off to save himself some money.
^^ I • ^^1 /"Jr * *

(b) *The lights* eJ were turned off [PRO */ J to save himself1 some money].
(c) *The lights saved himself some money.
(d) Turning off the lights saved himself some money.

A standard example of a Rationale Clause is given in (22a). The relevant co-

indexing relations are shown in (22b). Notice that according to the generalisation stated
210
above, PRO can be controlled by the subject, which is the Agent of the matrix event; or
by the event itself (which has John as an Agent); but not by the object, which fulfils
neither criterion. Control by John will give the paraphrase (22c) for the interpretation of
the Rationale Clause. Control by the event will give the paraphrase in (22d). Both
paraphrases are well-formed and the sentence is also well-formed.
The passive equivalent of (22a) is given in (23a). The coindexing possibilities
are given in (23b). As before, PRO can be controlled by the matrix event (which still
has John as an Agenc). We therefore have the well-formed paraphrase in (23d).
However, PRO cannot be controlled by the subject because the subject is the Theme of
turn off not the Agent: we hence have the ill-formed paraphrase in (23c). Nor can PRO
be controlled by John, who is no longer the subject.
The situation becomes particularly interesting when a reflexive is introduced as
in (24a). Consider the coindexing possibilities in (24b). As for (22b), PRO can be
controlled by the subject which is the Agent of the matrix event, but not by the object.
However, PRO cannot be controlled by the event, as it was in (22b). This because the
PRO in (24b) binds the anaphor, himself, and must therefore denote a human male.
John is therefore the only possible antecedent of PRO in (24b). This is shown by the
well-formedness of the paraphrase in (24c) and the ill-formedness of the paraphrase in
(24d).
The passive equivalent of (24a) is given in (25a), which is the equivalent of
(23a), except that it includes a reflexive. Notice in (25b) that there are no possible
antecedents for PRO. As in (23b), the subject in (25b) is not a possible antecedent
because it is the Theme and not the Agent of the matrix event. Now, however, the event
cannot control PRO for the same reasons that it could not in the active equivalent in
(24b), namely, that reflexive, himself, requires a human antecedent.
This analysis does, however, face a number of problems. On the one hand,
there seem to be examples in which there is no event paraphrase but passivisation is
possible. This is shown in (26).

211
[26]
(a) John sank the ship to collect the insurance.
(b) The ship was sunk to collect the insurance.
(c) * John's sinking the ship collected the insurance.

(26a) is a well-formed active sentence; (26b) is a well-formed passive sentence; but

passivisation cannot here be licensed by event control because the event paraphrase in
(26c) is ill-formed.
On the other hand, there are sentences in which there is a plausible event
paraphrase, but passivisation is not allowed, as in (27).

[27]
(a) John saved the cat to get into the local newspapers.
(b) *The cat was saved to get into the local newspapers.
(c) John's saving the cat got into the local newspapers.

(27a) is a well-formed active sentence; its passive equivalent in (27b) is ill-formed,

despite having a plausible event paraphrase, as in (27c). The examples in (27),
however, suggest that we should handle the notion of an "event paraphrase" with care.
What we mean when we say that a sentence has an event paraphrase is that event
control accurately captures the interpretation of the infinitive. In (27c) this is not the
case. If John saved the cat to get into the local newspapers, then it is John's intention
that he get into the local newspapers; we naturally assume that he manages to do so by
the story of the rescue itself appearing, but it may be that you have a chance of getting
mentioned if you have done several good deeds. That is, John may want to be
mentioned in some context or other himself and doing a good deed is one means en
route to this end; the actual saving of the cat may not itself appear in the paper at all.
(27c) assumes that the object of intention is that the saving itself be reported. It may
therefore be that passivisation is only allowed where there is the right sort of relation
between the matrix event and the infinitival event. This would mean that there is a
constraint that the Rationale Clause places on the events with which it occurs, in terms
of which the constraints on the entity arguments themselves are understood. The
situation would directly parallel the case of the Purpose Clause where the constraint
212
requiring a "Theme" argument was reduced to a constraint involving objects related in
particular ways to states.
However thess problems are to be addressed, it seems that the notion of event
control is extremely ussful in accounting for the behaviour of both the Rationale and
Telic Clauses. Although this makes it more difficult to use controller selection as a
further piece of evidence for adjunction point in the syntax, linear precedence, VP-
anaphor, and negation facts still argue strongly for the distributional non-equivalence of
these infinitives. Further evidence for this is provided when we address a distinction
between the two constructions which has been raised earlier: the Rationale Clauses
which do not allow event control will in fact allow passivisation of the matrix if the
Agent is lexically realised but similar Telic Clauses will not.

[28]
(a) John printed out the file to give himself a better idea of the layout.
(b) *The file was r rimed out to give himself a better idea of the layout.
(c) The file was printed out by John to give himself a better idea of the layout.
[29]
(a) John printed out the file only to realise that the layout was wrong.
(b) *The file was printed out only to realise that the layout was wrong.
(c) ?*The file was printed out by John only to realise that the layout was wrong.

However the contrast in (28-29) is to be represented, it is clear that the Telic Clause has
access only to the syntactic subject. We know already that the Telic Clause will happily
take an antecedent with any thematic role: if the Agent fry-phrase was structurally
accessible to it it would be ambiguous between control by the syntactic subject and
control by the Agent. However, the syntactic subject is the only possible controller. The
i ;"
Rationale Clause on the other hand must have the fry-phrase structurally available,
otherwise it would not matter whether there was an overt Agent or not. Although it is
not clear how these facts are to be accounted for, the implication that the Telic Clause is
structurally less integrated with the verb than the Rationale Clause comes through.

213
 The above discussion however stresses the distinction between control of PRO
(establishing the reference of the empty subject of the infinitive) and attribution of
intention via predication. Consider the sentence in (30).

[30] John drove his familly to the church [for his wife to join the ladies' group] [for
their children to gain access to the children's facilities].

The sentence in (30) describes three basic events: John drives to church; his wife joins a
group; their children gain access to some facilities. These events are linked in terms of
intended consequence: If John drives to church, his wife can join the group; if his wife
joins the group, his children will gain access to the facilities. If the intention described
by the Rationale Clause is attributed to the local Agent, then we would expect it to be
necessary for the wife to intend that the children gain access to the facilities as a
consequence of her joining the group, as well as it being necessary for John to intend
his wife to join the ladies' group. In other words a Rationale Clause would be evaluated
with regard only to its local Agent. This seems not to be the case. Imagine a case in
which John decides that the children using the facilities is a good thing and knows that
the only way for this to happen is for his wife to join the ladies' group. He does not
consult his wife or children but drives them to the church where he makes his wife join
the group. (30) could truly describe such a situation because there is no reason for the
wife or the children to act with any intentions whatsoever (certainly not with the
intentions expressed in the Rationale Clause). Admittedly they must perform the actions
in question and can therefore be seen as Agentive in those individual actions. But it
seems spurious to attribute active intention to them of the sort required. The point is all
the more important as it excludes an iteration in the interpretation which is not present.
This becomes clear when you consider the way we are forced to understand
John's intentions. We can imagine a scenario in which the wife decides that the children
using the facilities is a good idea, but knows that her husband hates church activities.
She asks to be driven to church so that she can sort out some business, and only there
does she make it clear what she intends. (30) cannot be truthfully asserted of such a
214
scene, because John at least must intend his wife to join and intend the children to use
the facilities. As with the table-trashing example in (10-11), other Agents in the
sentence and in the discourse can have whatever sort of intentions they like but the
Agent of the root event must have the relevant intentions.
Notice then that we have an asymmetry between the control of PRO where it is
present and the control of the infinitive as a whole by an intending agent. Consider
(31).

[31] John drove to town [for Mary to register at the club] [PRO to get
*himself/him/herself a pass to the squash courts].

Here PRO must be controlled by Mary but it is John that must intend the whole series
of events.
This suggests that contrary to our original suggestion, the Rationale Clause
cannot have the intending individual as an argument in its argument grid because the
strict locality of this relation will force selection of the immediate Agent, which we have
already argued would be an undesirable result.

[32]
VP <X,E>

VP <X,E> CP <Y,E>

We could therefore remove the individual argument from the argument grid of R and
take the Agent to be introduced as a condition on the reference of the predicate, as in
(33).

[33] R(e,Ap) <-» ... & intend(x,Ap)

However, this gives us the original problem, namely that the selection of an intender is
clearly not discourse free but is necessarily linked to the syntax. This will not follow

215
from (33) because the attribution of intention is a condition on the interpretation of the
predicate which is entirely unrelated to its syntactic realisation. The fact of the matter is
that the intention is attributed to the "agent" of the root event (i.e. the "agent" of the first
event which assigned deictic temporal reference): with iterated Rationale Clauses, PRO
is controlled strictly locally but intention is assigned to this "root agent". I have no
solution to this problem.

216
 Section 5.3: The Telic Clause: evaluating the consequences

It was said in Section 1.2 that the Telic Clause was headed by a predicate
denoting a relation between two events, where the second was the inevitable conclusion
of the first. As is clear from the phrase structure suggested in Section 3.4, the Telic
Clause adjoins outside of the domain of argument projection. It is therefore no surprise
i
to find that the Telic Clause places no selectional restrictions on the argument structure
of the verb it modifies (it does not select for Theme-like or Agent-like antecedents); nor
does it select for particular aspectual or event-internal structure (it does not select for
resultant states or events which are "manipulable"). We have seen already that the Telic
Clause is passive sensitive in that it will select whatever NP is syntactic subject as its
controller, regardless of thematic role or (base) grammatical function.

[1] Sven1 gave SigmnJ the CD^, only e* to realise that she already had it.
[2] SigrunJ was given the CDk by Sven*, only el to realise that he wanted it for
himself.
[3] The CD^ was given to SigrunJ by Sven^, only e^ to get damaged before she
had a chance to play it.

In (1) the Agent subject of give is understood as the subject of realise; in (2) the Goal
subject of passive give is understood as the subject of realise; and in (3) the Theme
subject of passive give is understood as the subject of get.
Similarly, it seems to be insensitive to standard aspectual categories and
whatever event substructure those categories might be associated with. It is quite clear
that the Telic Clause marks some sort of transition between events: it can mark the
culmination of a process, as in (4).

[4] Leaves rolled along the pavement to settle in the gutter at the end of the street.

In Vendlerean terms, roll along denotes an Activity (an open-ended process); settle
denotes an Achievement: in (4) the settling is taken to be the culmination of the rolling.

217
When the Telic Clause occurs with only it marks a strong sense of disruption between
events, as in (5).
:i)
[5] The ball sailed towards the goal only to be blown off course by the wind.

Here the trajectory described by sail towards the goal is disrupted by another event, the
blowing of the wind. We might therefore expect that the Telic Clause would select for
events which mark transitions of some sort (Achievements or Accomplishments) or that
it might select for pairs of events which build up a transition (Activity + Achievement).
However, this is not the case. It appears that the Telic Clause places no aspectual
restrictions on the events with which it occurs; in fact, it is perfectly compatible the least
dynamic of all event-types: the State. It can be used straightforwardly to relate two
states, as in (6).

[6] Sven was out in town for most of the day, only to be home when Sigrun
arrived.

The Telic Clause in (6) requires us to assume that Sven's being at home when Sigrun
arrives is undesirable: the State of Sven being in town would have avoided this but the
culminating event is (in contrast) the State of Sven being at home. It therefore seems
that what is necessary for the occurrence of the Telic Clause is that the external
relation between events be of an appropriate type, not that they have aspect or argument
structures (which I takes to be "internal" in some sense) of a particular class.
The external relation is obviously the one introduced by the Telic Clause itself
(rather than a property of the modified verbs which is selected); the question is then
what the nature of this relation is. A crucial element is that it is in some sense evaluative:
that is, the Telic Clause predicates relations between events which we judge to be
natural or appropriate, or contrastive and adversative. The discussion of (4-5) above
implied that the Telic Clause merely built up larger aspectual structures, so that the
Activity of rolling culminated (naturally) in the Achievement of settling or the Activity

218
of sailing culminated (unexpectedly) in the Achievement of blowing off course, a
pattern which suggests the substructure of the Accomplishment: process + telos.
However, the relation predicated by the Telic Clause is much broader than this. Notice
that in (6) it is not the case that the State of Sven's being in town "culminates" in his
being at home, rather there is a generally adversative reading: Sven's being at home is
what ends up happening despite his being in town most of the day, and that culmination
is unexpected/ unplanned/ undesirable etc. The two events must still be related (the
being at home is what ends up happening in spite of the being in town), but the notion
of telos here is a fairly broad one. The notion of "ending up" must bear a lot of weight
here, as it is presumably that element which triggers the evaluative reading which I take
to involve some general form of pragmatic inference. It is not enough for one event
simply to happen after another or for the two events simply to be in some coherent
sequence, but the second event must be seen to terminate the general situation in which
the first event is involved. It seems that we must therefore judge that termination to be
particularly appropriate and fitting or particularly contrary to expectation for the relation
to hold. The reading which one gets when the two events are opposed as inconsistent or
contrary to expection I will call an adversative reading, and it appears to be the easiest to
construct. The reading where one event is judged as an appropriate and fitting outcome
I take to be a resolutional reading. These are more difficult to construct because they
rely on a sense of stereotypical sequence: relations between events which are judged to
be ideologically natural in some sense (here perhaps is another source of analogy
between the Telic Clause and Purpose Clause, which also allows non-intended
attribution of properties as long as they are "natural"). We might therefore have
conditions on reference along the lines of (7).

[7] T(<el,e2>) <-» el finds its conclusion in e^

I have assumed throughout this discussion that the Telic Clause denotes a
relation between events rather than between propositions. The question is clearly a
219
delicate one given the fact that the Telic Clause does not seem to relate to event-internal
structure. However, I believe that striking evidence for this view is provided by the
interpretation of the Telic Clause in perception contexts. Consider (8) and (9).

[8] John saw Brutus stab Caesar.

[9] John saw that Brutus stabbed Caesar.

There are a number of important distinctions between the interpretations of (8) and (9)
(see Higginbotham (1983a) for a detailed discussion). Notice in particular that the
complement of see in (8) is referentially transparent whereas the complement of see in
(9) is referentially opaque. (8) must be asserted where John actually witnesses the
stabbing of Caesar by Brutus, although he may not realise that this is what he saw (he
might think that Brutus merely stumbled and bumped into Caesar; or he might believe
that the incident he has witnessed actually involved two entirely different individuals).
(9) carries no implication that John actually witnessed the stabbing of Caesar by Brutus,
merely that he perceives it to be true that that is what has happened: it is necessary to the
content of his perception that he understand the event is a stabbing and that Brutus is
Brutus and Caesar is Caesar and so on. Higginbotham (1983a) argues in detail that
these facts amongst others follow straightforwardly in a theory which includes both
events and propositions as ontological primitives. In (8), John witnesses an event
which is described in a particular way although we know nothing of his perception of
that event; in (9) John perceives a proposition to be true, in which case the content of
the proposition forms the content of his belief. In Higginbotham1 s theory, sentences
such as (8-9) receive translations as in (10-11).

[10] (Be 1 ) [3e2: kill (Brutus, Caesar, e2)] see (John, e2, e 1 )
[11] (Be 1 ) see (John, A[(3e2) kill (Brutus, Caesar, e2)], e 1 )

Now consider the interpretation of a Telic Clause occuring in a perception

context, as in (12).

220
[12] John saw Bill win one million pounds on the first spin of the wheel only to lose
it all on the second.

It is clear that (12) has the properties associated with the event reading rather than the
propositional reading. If John saw Bill win one million pounds on the first spin of the
wheel only to lose it all on the second, then he actually witnesses those two events; he
may be under the misapprehension that he is actually seeing Fred rather than Bill, but
as long as it is actually Bill who is involved in the event, then (12) can be truthfully
asserted. Significantly, Bill's losing the money does not constitute a part of John's
beliefs about the situation. It therefore appears that the Telic Clause in (12) denotes an
event (referentially transparent) which John can perceive in the way he can perceive any
other event; it is not a proposition. It seems therefore that the correct translation for (12)
is the one given in (13).

[13] (Be 1 ) [(3e2e3 : T(<e2,e3>) & win(e2) & lose(e3)] see(John,<e2,e3>,e 1 )

This leaves us with the question of the control of PRO. It has been observed
already that the empty category in the Telic Clause shows exactly the same properties as
the empty category in the Rationale Clause, modulo a difference in controller selection
which has been attributed to structural factors. However, PRO in the Rationale Clause
allows event control as well as individual control and this alternation has been linked to
restrictions on the passive. The question then arises whether the Telic Clause also
allows event control and with what properties. It was observed above that the Telic
Clause is passive-sensitive in that passivisation forces a shift in controller, as in (1-3).
However, as for the Rationale Clause, there do seem to be instances of Telic Clauses
which are able to passivise without changing the interpretation, as in (14-15).

[14] Derby won the match 6-2, to continue their phenomenal recovery this season.
[15] The match was won by Derby 6-2, to continue their phenomenal recovery this
season.

221
 In both (14) and (15) it is Derby who are continuing their phenomenal recovery. As for
the examples with the Rationale Clause we could posit an ambiguity with respect to
subject-vs-event control, as in (16-19).

[16]
(a) John bought Bill a CD only to realise that he already had it.
(b) John* ei bought Billk a CD only PROi/*J/*k to realise that he already had it.
(c) John realised that he already had it
(d) *Buying Bill a CD realised that he already had it
[17]
(a) Bill was bought a CD only to realise that he already had it.
(b) Billk eJ was bought a CD only PRO^J to realise that he already had it.
(c) Bill realised that he already had it.
(d) *Buying a CD realised that he already had it.
[18]
(a) Derby won the match 6-2, to continue their phenomenal recovery this season.
(b) Derby* eJ won the matchk 6-2, PROVJ/*k to continue their phenomenal recovery
this season.
(c) Derby continued their phenomenal recovery this season.
(d) Winning the match contintued their phenomenal recovery this season.
[19]
(a) The match was won by Derby 6-2, to continue their phenomenal recovery this
season.
(b) The matchk eJ was won by Derby* 6-2, PRO* W^ to continue their phenomenal
recovery this season.
(c) The match continued their phenomenal recovery this season.
(d) Winning the match continued their phenomenal recovery this season.

In (16b), we can see that the Telic Clause can be controlled by the matrix subject, John,
giving the well-formed paraphrase in (16c); it cannot, however, be controlled by the
event because of seiectional restrictions imposed by the verb realise as illustrated in
(16d); the first object, Bill is not accessible. With the passive, Bill becomes accessible
and is a possible controller as shown by the well-formed paraphrase in (17c); however,
the event cannot control the infinitive for the same reasons as in (16), as illustrated in
(17d). In (18) the situation is the same as in (16) except that event control is now
allowed because the verb continue allows event subjects, as illustrated in (18d). Under
passive this control possibility remains constant, as shown in the co-indexing

222
possibilities in (19b) and the paraphrase in (19d). Subject control is available in the
expected fashion, as in (19c).
The situation becomes particularly interesting when a reflexive/pronoun
alternation is introduced

[20] Derby won the match 6-2, to give themselves a shot at the championship.
[21] Derby won the match 6-2, to give them a shot at the championship.

Like the Rationale Clause, the Telic Clause allows the use of reflexives which are
(indirectly) co-referential with the controlling subject. This means, incidentally, that
(20) is ambiguous between a Rationale Clause reading in which Derby won the match
(by that particular margin) with the intention that they have a shot at the championship;
and a Telic Clause reading in which winning the match by that margin does give them a
shot a championship (as a matter of course rather than intention): we are only interested
in the Telic Clause reading here. Unlike the Rationale Clause, the Telic Clause also
allows the use of a pronoun co-referential with the subject (making (21) unambiguous).
As for the Rationale Clause, the Telic Clause with a reflexive will not allow the matrix
to passivise, where the Telic without reflexive will.

[22] *The match was won 6-2, to give themselves a shot at the championship.
[23] The match was won 6-2, to give them a shot at the championship.

Notice that, unlike with the Rationale Clause, the inclusion of a lexical Agent makes no
difference.

[24] The match was won by Derby 6-2, to give themselves a shot at the
championship.
[25] The match was won by Derby 6-2, to give them a shot at the championship.

Once again these facts will fall out of an analysis like the one offered above for
the Rationale Clause.

223
[26]
(a) Derby won the match 6-2, to give them a shot at the championship.
(b) Derby1 ei won the matchk 6-2, PRO*W*^ to give them a shot at the
championship.
(c) *Derby gave them a shot at the championship.
(d) Winning the match gave them a shot at the championship.
[27]
(a) The match was won 6-2, to give them a shot at the championship.
(b) The matchk eJ was won 6-2, PRO^/J to give them a shot at the championship.
(c) The match gave them a shot at the championship.
(d) Winning the match gave them a shot at the championship.
[28]
(a) Derby won the match 6-2, to give themselves a shot at the championship.
(b) Derby* eJ won the matchk 6-2, PRQV*j/*k to give themselves a shot at the
championship.
(c) Derby gave themselves a shot at the championship.
(d) *Winning the match gave themselves a shot at the championship.
[29]
(a) *The match was won 6-2, to give themselves a shot at the championship.
(b) *The matchk d was won 6-2, PRO*^ to give themselves a shot at the
championship.
(c) *The match gave themselves a shot at the championship.
(d) *Winning the match gave themselves a shot at the championship.

In (26b), we see that the event can control PRO, as shown by the well-formed
paraphrase in (26d); the match is structurally unavailable; Derby is structurally available
but the presence of the pronoun in the infinitive prevents it from controlling PRO
(which would give a Binding Condition B violation), as shown in the ill-formed
paraphrase in (26c). When (26a) passivises we have the coindexing relations in (27b).
The match is now structurally available and can control PRO as expected. The event
control is also available, as shown in the paraphrase in (27d) and the sentence is well-
formed.
Replacement of the pronoun in (26a) with a reflexive, as in (28a), gives the
binding relations in (28b). As before, the object, the match, is structurally unavailable.
The reflexive however reverses the remaining binding possibilities: the event is not a
>. i
proper antecedent for the reflexive, as shown in the ill-formed paraphrase in (28d), but
the subject is now available without any Binding Condition violations, as in (28c). As
before, the reflexive forces subject control. When (28a) is passivised, giving (29a), we

224
therefore have the binding relations in (29b). Neither the subject nor the event are
possible antecedents for the reflexive, as shown in (29c-d) and therefore the sentence is
ill-formed.
The Telic Clause thus resembles the Rationale Clause in a number of respects. It
is tempting to see the parallel alternation between subject and event control as indicating
a more fundamental parallel. Evidence has already been offered, however, in Section
3.4 that the Telic Clause is in fact less deeply embedded than the Rationale Clause.
Although we cannot take the orientation to Agent to be evidence that the Rationale
Clause is minimally c-commanded by the "internal subject", there does seem to be
evidence of some sort, not just that the Rationale Clause is semantically interpreted with
regard to Agents, but that it is syntactically sensitive to the presence an Agent in a
way that a Telic Clause is not. We have already seen that there is a difference with
respect to pronominal binding. The Rationale Clauses which do not allow event control
will in fact allow passivisation of the matrix if the Agent is lexically realised but similar
Telic Clauses will not. Recall the contrast discussed in Section 5.2, repeated here.
[30]
(a) John printed out the file to give himself a better idea of the layout.
(b) *The file was printed out to give himself a better idea of the layout.
(c) The file was printed out by John to give himself a better idea of the layout.
[31]
(a) John printed out the file only to realise that the layout was wrong.
(b) *The file was printed out only to realise that the layout was wrong.
(c) ?*The file was printed out by John only to realise that the layout was wrong.

Given the Telic Clause is insensitive to thematic relation we would expect it readily to
take the fry-phrase as an antecedent if that phrase is structurally accessible, but it
appears not to be. The Telic Clause does therefore seem to be structurally less
embedded than the other infinitives of result and to modify the event irrespective of
thematic or aspectual relations which it specifies: this is just what we expect from the
general outcome reading.
The Telic Clause thus relates to the other infinitives of result in an extremely
complex and interesting way: it is distributionally extremely loosely tied to the sentence
225
when one considers scope and its interpetation is the most abstract and the hardest to
define. And yet it bears a clear relation to the Rationale Clause with respect to the event
control alternation. It can clearly be closely integrated with verbs of perception where it
is interpreted as straightforward event and where it seems to adjoin to at most a reduced
clausal structure. I take it that as we understand the nature of the constraints on control
of the Rationale Clause better in the syntax and gain a clearer idea of the sort of
'• :. *

causation which are important in linguistic interpretation, the place of the Telic Clause
will become clearer. I will leave the discussion here. Before concluding, however, I
will offer an analysis of a sentence including each of the infinitives of result to refocus
the general position taken in this paper, namely that the infinitives of result are headed
constructions which discharge thier arguments by general principles of argument
discharge.

226
Section 5.4: The typology of Result Predicates and in order insertion

The typology suggested in this thesis for the infinitives of result implies both the
independence of each construction from the other and a close link between them in
syntax and semantics. Each infinitive has its own distinct internal syntax, argument
composition, point of adjunction, and interpretation. On the other hand, each predicate
is lexicalised by the Inflectional element, to, leading to the projection of a full CP clause
structure, and each construction is a instance of simple event modification, integrated
into the modified structure by the same general principles of argument discharge.
Further, it appears that the meaning of the predicates is closely related and that this
relation between the predicates can be viewed as a sort of metaphorical extension,
where metaphorical extension is understood as the systematic weakening of constraints
on the interpretation of a predicate.
I have argued that the Purpose Predicate is ideological in a general sense, that
is, that it concerns the attribution of properties to objects, where the attribution may take
the form of a natural function, a designed feature, or a general intention, as in (1-3)
[repeated from (26-28a), Section 1.2].

[1] Grass is green to photosynthesise more efficiently.

[2] This car is contoured to accelerate more quickly.
[3] John bought some cream cakes at the shop to eat with his grandmother.

The Rationale Clause focuses on the notion of intended outcome and excludes reference
to objects and their properties. The Telic Clause focuses on the notion of outcome and
excludes reference to intentions. The conceptual relations between the constructions
might then be viewed as in (4).

[4]
- PURPOSE PREDICATE r RATIONALE PREDICATE r TELIC PREDICATE
* Natural Function ^-Intended Outcome pr Final Outcome

l * Designed Feature
I ^General Intention
227
 Specific features of each construction are then related to the way these general
notional relations are given formal realisation in the syntax and the semantics. The
Purpose Predicate deals with the attribution of properties in a general sense and
therefore the infinitive is required to be interpreted as a property in the semantics and
hence to contain w/z-movement of an empty operator in the syntax; as it concerns
properties which objects gain by virtue of their participation in the event, the purpose
predicate applies to stative subevents of the modified event, S(e), in which the object
has the property in question, i.e. it selects for "resultant state" verbs.
The Rationale Clause on the other hand deals with intended outcomes of events.
The infinitive must therefore be interpreted propositionally in the semantics, requiring a
CP which contains no movement in the syntax. Further, the predicate now applies to
the modified event itself, not its substates. The Telic Clause deals with actual outcomes
of events and the infinitive must therefore be interpreted directly as an event in a
transparent context. Once again the modified event is the event itself, not one of its
subevents.
This typology of the constructions also allows us to raise the probelm of in
order insertion in a more fruitful context. As mentioned in Section 1.2, Jones (1991)
treats the preface, in order, as a straightforward marker of the Rationale Clause, that is,
all contexts in which in order insertion is possible, a simple Rationale Clause is
possible; and all infinitives prefaced by in order are instances of Rationale Clauses.
Neither of these assumptions appears to be the case.
First there is a construction in which Rationale Clauses appear to be at best
marginal, but in which In Order infinitives are perfectly acceptable: the passive. It
seems that Rationale Clauses resist passivisation: compare the passive (5a) with the
active (5b). The contrast is minimal with the In Order infinitive.

[5] (a) ?*Jo flattered the bouncer to be admitted to the club,

(b) Jo flattered the bouncer to gain admittance to the club.
[6] (a) Jo flattered the bouncer in order to be admitted to the club.
228
 (b) Jo flattered the bouncer in order to gain admittance to the club.
i

\ j

Second, the preface, in order, seems to be compatible with the Purpose Clause
under certain extremely strict (but rather unsurprising) conditions. Recall that it was
argued in Section 5.1.1 that there are four event types which are compatible with the
Purpose Predicate.

[7] John sent the troops into the hall [e to pacify the crowds].
[8] John sent troops along the hall [e to pacify the crowds].
[9] John left the troops in the hall [e to pacify the crowds].
[10] The troops were in the hall [e to pacify the crowds].

In (7), we have a telic event, in (8) an atelic event, in (9) a continuance of state, and in
(10) a state. Developing Bach's (1982) resultant state analysis, I argued that the
Purpose Predicate is only compatible with those verbs which contain a state in which
the object has the property in question. The telic use of send includes a final or resultant
state (by being in the hall the troops are able to pacify the crowds); the atelic use of send
includes a state of motion (the troops are able to pacify the crowds by virtue of moving
along the hall); the verb leave includes (once again) the state of the troops being in the
hall); and of course in (10), the main sentential predication refers directly the state of the
troops being in the hall. Significantly, (10) is taken to contain a Purpose Clause because
the main clause refers to the very state which acts as a resultant or continuant state in (7)
T;
and (9) respectively, and the interpretation of the troops with respect to the infinitive in
(10) is presumably identical to the interpretation of the troops with respect to the
infinitive in (7) and (9).
Despite the fact that (10) contains a Purpose Clause, however, in order insertion
is possible, where it is not in (7-9).

[11] *John sent the troops into the hall [in order e to pacify the crowds].
[ 12] * John sent troops along the hah1 [in order e to pacify the crowds].
[ 13] *John left the troops in the hall [in order e to pacify the crowds].
[ 14] The troops were in the hall [in order e to pacify the crowds].

229
This implies that the preface in order is not a marker for Rationale Clauses but rather
that there are conditions on its insertion which makes in order always compatible with
Rationale Clauses, though also potentially compatible with other infinitives. The
relevant conditions seem to be three.
First, as the contrast between (11-13) and (14) shows, in order can only be
inserted in front of an infinitive which modifies the main event itself (rather than a
subevent). This is always true of Rationale Clauses. It is also true of the Purpose
Clause in (14) because the function S(e) will in this case simply provide the state, e,
itself (no substate is required or available). It is not, however, true of (11-13), where
the state in question is some substate of the main event. This effectively restricts the
occurrence of in order with the Purpose Clause to cases where the modified sentence is
an instance of copular predication.
Second, in order insertion is never compatible with object-gap Purpose Clauses,
even when they are applied to sentences containing copular predication, as shwon in
(15-18).

[15] *Sven pushed the cart into the arena in order for the crowds to see e.
[16] *Sven pushed the cart along the road for the crowds to see e.
[17] *Sven kept the cart in the arena in order for the crowds to see e.
[ 18] *The cart was in the arena in order for the crowds to see e.

We must therefore assume that in order insertion is only available where both gaps
occur in subject-position (being thus compatible with Rationale Clauses and subject-gap
Purpose Clauses, but not object-gap Purpose Clauses).
Thirdly, in order insertion appears only to be allowed (unsurprisingly) where
there is an intention reading. Once again, this is always the case with the Rationale
Clause which has the notion of "intended outcome" built formally into its meaning.
However, as shown in (4), the Purpose Clause, being ideological in a general sense,
has an intentional reading as only one of the types of reading available. The example
given above, (14), is a case of "general intention", i.e. John, by sending the troops into
the hall has the general intention that they be able to pacify the crowds. It seems that
230
Purpose Clauses with a "designed feature" reading are also compatible with in order
insertion, because the property, though in some sense inherent to the object, is present
by an act of intentional creation.

[19] The car was designed with special contours in order to accelerate more quickly.

However, the generic use of the Purpose Clause I find incompatible with the use of in
order even where other conditions are met.

[20] #Grass is green in order to photosynthesis more efficiently.

I have marked the sentence as semantically ill-formed because it is unacceptable on a

particular reading. According to my intuitions, the insertion of in order in (20) forces us
to assume that someone intentionally made grass green so that it would function as it
does (presumably God); the statement is not merely a general statement about the
Ideological properties of grass itself. I therefore have a contrast between (20) and (21).

"•t
[21] Grass at driving ranges is red in order to show up golfballs more clearly.

I find (21) perfectly acceptable on the assumption that the grass has been designed in a
particular way. It seems therefore that in order insertion is incompatible with the true
generic reading of a Purpose Clause and that this is so because it requires an intention
reading. Needless to say, the restriction to intentional readings also excludes its
occurence with the Telic Clause.
To summarise, we can assume that in order insertion is generally available
under three conditions, shown in (22-24).

(22) The infinitive must apply to the modified event itself, not to a subevent. This
excludes all Purpose Clauses except those modifying copular predication
structures.

231
(23) A dependency relation must hold only between subject gaps. This excludes
object-gap Purpose Clauses.
(24) The interpretation of the infinitive must include reference to intention. This
restricts insertion as follows:

Ilie Intention Restriction on in order insertion

TELIC PREDICATE
Final Outcome

RATIONALE PREDICATE
IN ORDER INSERTION Intended Outcome

PURPOSE PREDICATE
Natural Function
Designed Feature
General Intention

232
Section 5.5: A sample analysis

[1] John designed a battery [pc to operate at high temperatures] [RatC to win a
prize] [TC only to realise that he was not eligible to enter]

I will assume that the main predicates in this sentence are those in (2-8); I have
distinguished all entity and event variables so that it is clear how the process of
argument projection proceeds; for the sake of space, I have represented a property by TI
and a proposition by p.
j;

[2] design (x.y.e 1 )

[3] to 1 (z,e2,7i)
[4] operate (a,e3)
[5] to2 (b,e4,Ap)
[6] win (c,d,e5)
[7] to3 (<e6,e7>)
[8] realise (g,Ap,e8)

As should be clear from this list, the projection of the arguments of all of these
predicates into a unified sentence is no trivial matter: for instance, we have eight event
variables and eight entity variables all differently associated. However, given the
specific properties of each of these predicates, and the general principles of argument
discharge suggested in Higginbotham (1985), the projection of argument structure in
the complex sentence in (1) will proceed in a general and systematic fashion.
Let us begin with the internal argument structure of the three infinitives. In (1),
the infinitival verb in the Purpose Clause is the verb operate. Operate will project the
structure in (9).

[9]
VP [WH] <E3>

S{*c"[WH]o '<A,E3>

1 °<A,E3>

0[WH]

233
 Being intransitive, operate has no complement and its argument grid filters up to V. It
directly 0-marks an NP (a closed expression) in VP-Spec, which will in fact be an
empty operator (or its trace), as suggested. I will assume that the [WH] feature on the
specifier will be visible on the VP by Agreement. This will give a translation as in (10).

[10] operate (a,e3)

The VP headed by operate will now appear as the complement to 1^, which
dominates to1 . We will then have the structure in (1 1).

CP(Z,E2)

I will now make two assumptions. First, to1 triggers existential closure in its
complement VP, giving the translation in (12).

[12] A(3e3) operate (a,e3)

Second, the presence of a WH-feature is interpreted as lambda abstraction over a

variable represented by the operator (or rather its trace at LF).

[ 1 3] AXa (3e3) operate (a,e3)

We now have a property. This fulfils the selectional restrictions imposed by to1 on its
third argument; to1 therefore directly 9-marks VP. The first two arguments of to 1 now

234
filter up to the CP node, awaiting discharge. This gives the translation for the Purpose
Clause infinitive in (14).

[14] tol(z, e2, Ata (3e3) operate (a,e3))

Now consider the Rationale Clause in (1). The infinitival verb in this infinitive
is win, a transitive verb. It will project the structure in (15).

[15]
VP<E5>

Speco V <C,E5>

\0<C,D,E5> o

PRO win a prize

>

Win takes an NP complement (a closed expression) which it directly 6-marks, closing

its second argument position. Its argument grid filters up to V. It directly 6-marks an
NP subject, closing its first argument position. The event position filters up to VP. This
gives the translation in (16).

[16] win (PRO, a prize, e5)

This VP appears as a complement of to?, giving the structure in (17).

[17] V.ft
(B.E4) '*
CP(B,I

' (B.E4)

'(B.E4)

lO(B,E4,P) VP<E5>

PRO win a prize

235
Once again I will assume that to2 triggers existential closure in its complement, giving
the translation in (18).

[ 18] *(3e5) win (PRO, a prize, c5)

: •>

Now, however, there is no WH-feature to trigger lambda abstraction; but of course a

proposition is precisely what to? requires for its second argument. It therefore directly
6-marks VP through its third argument position (closing it) and the first two argument
positions filter up to CP, awaiting discharge. The Rationale Clause will therefore have
the translation in (19).

[19] to2(b, e4, A(3e5) win (PRO, a prize, e5))

Finally, we have the Telic Clause. The verb which heads this infinitive is
realise, a transitive verb which will project the structure in (20).

[20]
VP <E8>

PRO realise that be was not eligible to enter

Realise takes a CP complement (a proposition, i.e. a closed expression) which it

directly 9-marks, closing its second argument position. Its argument grid filters up to
V. It then directly 6-marks an NP subject, closing its first argument position. The
event position filters up to VP. This gives the translation in (21).

[21 ] realise (PRO, Ahe was not eligible to enter, e8)

236
This VP appears as a complement of tcJ, giving the structure in (22).

[22]
CP (<E6,E7>)

PRO realise that be is not eligible

Unlike to^ and to?, to? does not trigger existential closure in its complement VP (it is a
pure relation between events). Rather its second argument position 0-identifies with the
VP's event variable. Notice that this means that e$=e~/\ I will therefore now relabel the
event variable of realise in (21) as e 7. The whole argument grid of the head,
therefore filters up to C?. The translation for the Telic Clause is therefore as in (23).

[23] to3 (<e6,e7>) '& realise (PRO, Ahe was not eligible to enter, e7)

We can now turn to the problem of integrating the infinitives into the argument
structure of the main clause. The main verb in (1) is design, a transitive verb. The
internal structure of the lower VP, headed by design, will be as in (24).

[24]

Sjfeco V2 '<X,Y,E1>

a battery design 0 to operate at high temperatures

237
 Cli.

The direct object of design is base-generated in V?2-Spec; there is nothing in

"Complement" position, and therefore the argument grid of design filters up to V21 .
The Purpose Clause now adjoins to V21; it has an entity argument and an event
argument to assign (as seen in (11)). The entity argument of the Purpose Clause 6-
identifies with the second entity argument of design, i.e. z=v. The event argument of
the Purpose Clause 0-identifies with the event argument of design, i.e. e^-e^. This will
give the translation in (25) for the upper ¥2'.
. \ ';! '

[25] design (x.y.e 1 ) & to 1 (y, e 1 , AXa (3e3) operate (a,e3))

Now design directly 9-marks its direct object in VP2-Spec, substituting the reference of
a battery for all instances of the variable y. The translation for VP2 will therefore be that
given in (26).

[26] design (x, a battery, e 1 ) & to 1 (a battery, e 1 , AXa (3e^) operate

VP2 is the Complement of Vi. The internal structure of VPi is that given in
(27).

[27]

Spec o VI' <X,E1>

CP<B,E4>

John design a battery PRO to win a prize

The argument grid of VP2 filters up to Vl', to which the Rationale Clause adjoins, with
'.?'
two arguments to discharge (as shown in (17)). The entity argument of the Rationale
Clause 0-identifies with the Agent argument of design, i.e. b=x', the event argument of
238
the Rationale Clause 9-identifies with the event argument of design, i.e. e4=e1 . This
will give the translation in (28) for the upper Vi'.

[28] design (x, a battery, e 1 ) & to2 (x, e 1 , A(3e5) win (PRO, a prize, e5))

Design now directly 0-marks its subject in VPl-Spec, substituting John for all
occurrences of the variable, x. This gives the translation in (29) for VPi .

[29] design (John, a battery, e 1 ) &...& to2 (John, e 1 , A(3e5 ) win (PRO, a prize,
e5))

VPi is the Complement of lO. The internal structure of DP is given in (30).

[30]
IP<E1,E7>

<E6,E7>

[PAST] John design a battery only PRO to realise...

The grid of VPi filters up to I1, to which the Telic Clause adjoins. The first event
i I1

variable of the Telic Clause 0-identifies with the event variable of design, i.e. e^=e^ .
All free event variables then filter up to IP (i.e. the first event variable of to? which has
been 6-identified with the event variable of the matrix verb as well as the second event
variable of to? which has been 9-identified with the event variable of the infinitival
verb: e-e\ e-e}. I will assume that at IP the remaining event variables receive
default existential closure. This will give a translation for IP as in (31).

[31] (3C 1 ) (Be?) design (John, a battery, e 1 ) &...& to3 (<e!,e7>) & realise (PRO,
Ap, e?)
*'_'.
*" 239
This will give a final full translation for the sentence in (1) as in (32).

[32] (3C 1 ) (3e7) design (John, a battery, e 1 ) & to 1 (a battery, e 1 , AXa (3e3) operate
(a,e3)) & to2 (John, e 1 , A(3e5) win (PRO, a prize, e5 )) & to3 (<e 1 ,e7 >) &
realise (PRO, Ap, e7)

240
 CONCLUSION
i,

In this thesis I have introduced an infinitival modifier which has not been
recognised in the literature as an independent construction. I have argued on the basis of
a range of different syntactic and semantic evidence that the Telic Clause is a distinct
construction from the more familiar infinitival modifiers in the literature. Its interaction
with VP-anaphora rules and scope indicate that it is adjoined at the sentence level where
the Rationale and Purpose Clauses are associated with the VP. Its compatibility with
perception verbs and its interpretation in those contexts suggests that it denotes an
event, where the Rationale and Purpose Clauses denote propositions or prepositional
functions. I have attempted in this thesis to set out a general theory of infinitives of
result which draws out the similarities and differences between them in a systematic
fashion: in this respect my analysis approaches the task at a different angle from others
in the literature, which have tended to focus on the "horizontal" relation between the
Purpose Clause and other obligatorily gapped infinitives.
My principle focus of interest in these constructions has been the problems that
they raise for a theory of modification: the infinitives of result clearly make distinctive
contributions to the meaning of the sentences in which they occur and yet this is usually
not expressed, particularly in theories of the Purpose Clause. These constructions
provide good evidence for the view that the syntax and semantics of modification
should be integrated into the general domain of argument projection. Each infinitive of
result has a head which acts as a predicate, with an argument structure like any other
predicate. The problem is then how that argument structure is integrated with the
argument structure of the modified verb. Addressing this problem requires a flexible
view of argument structure and its relation to the syntax, one which is not tied to the

notion of thematic roles and privileged relation types.

I show that where a flexible view of argument structure is adopted, along with
general principles of argument discharge, certain problems of interpretation are posed
particularly sharply (for instance the problem of the attribution of intention by the
241
Rationale Clause) but unexpected types of solution to syntactic problems also become
available: for instance the linking of gapping in the Purpose Clause to lambda
abstraction in the semantics, or the striking compatibility of the Telic Clause with
perception predicates relating directly to the fact that it denotes a relation between events
rather than propositions. I take this rich interaction between the syntax and semantics to
be one of the most interesting possibilities which is opened up when a flexible theory of
argument structure is applied to constructions of the complexity of these infinitives of
result.
From the syntax proper and the interface level of argument structure, the
analysis has also pointed out to broader problems of interpretation, concerning relations
between events and states, actors and intentions, and relations of causal efficacy of a
general type. This I take to be part of the domain of inquiry which Bach (1982) calls
natural language metaphysics. It seems that an adequate understanding of the
"conceptual underpinnings" of language must go beyond the restricted notion of a
thematic relation to a more general study of the properties of individuals and the
relations between them.
The focus of interest in argument structure thus allows one to look both ways,
at strict questions of form and more general questions of interpretation. The "vertical"
approach to the study of infinitives of result encourages this, moving from the close
integration of the Purpose Clause with the event structure of the verb to the more
general intepretation of the Telic Clause. I therefore hope that, in offering a systematic
account of this group of infinitives, this thesis will have contributed to an understanding
of the complex relation between syntax and semantics in natural language.

242
 Bibliography
Authier, J.-Marc P., (1986), Two types of empty operator, Linguistic Inquiry, 1 17-

Bach, Emmon, (1979), Control in Montague Grammar, Linguistic Inquiry 10, 515-
*J J A

Bach, Emmon, (1982), Purpose Clauses and Control. In Jacobson, P. & Pullum,
O.K. (eds.), The Nature of Syntactic Representation (35-57). Reidel,
Dordrecht.

Bach, Emmon & Partee, Barbara H., (1980), Anaphora and Semantic Structure.
Proceedings of the Chicago Linguistic Society, Vol. 16.
Bowers, John, (1993), The syntax of predication, Linguistic Inquiry 24, 591-656.
Browning, M., (1987), Null operator constructions. Unpublished Ph.D. Dissertation,
MIT.
Carroll, Susanne, (1983), Remarks on FOR-TO Infinitives, Linguistic Analysis 12,
415-451. ;'

Chierchia, Gennaro, (1984), Topics in the Syntax and Semantics of Infinitives and
Gerunds. Unpublished Ph.D. Dissertation, University of Massachusetts,
Amherst.
Chierchia, Gennaro, (1985), Formal Semantics and the Grammar of Predication,
Linguistic Inquiry 16, 417-443.
Chierchia, Gennaro, (1989), Structured Meanings, Thematic Roles, and Control. In
Chierchia, Gennaro & Partee, Barbara & Turner, Raymond (eds.), Properties,
Types and Meanings (Vol. 2, 131-166). Kluwer, Dordrecht.
Chomsky, Noam, (1965), Aspects of the Theory of Syntax. MIT Press, Cambridge,
MA.
Chomsky, Noam, (1976), Conditions on Rules of Grammar, Linguistic Analysis 2.
Chomsky, Noam, (1977), On wh-movement. In Culicover, Peter & Wasow, Thomas
& Akmajian, Adrian (eds.), Formal Syntax (71-132). Academic Press, New
York.
Chomsky, Noam, (1980), On Binding, Linguistic Inquiry 11, 1-46.
Chomsky, Noam, (1981), Lectures on Government and Binding. Foris, Dordrecht.
Chomsky, Noam, (1982), Some concepts and consequences of the theory of
government and binding. MIT Press, Cambridge, MA.
Chomsky, Noam, (1986a), Barriers. MIT Press, Cambridge, MA.
Chomsky, Noam, (1986b), Knowledge of Language: Its Nature, Origin, and Use.
Praeger, New York.

243
Chomsky, Noam, (1991), Some Notes on Economy of Derivation and Representation.
In Freidin, R. (ed.), Principles and Parameters in Comparative Grammar (4IT-
454). MIT Press, Cambridge, MA.
Chomsky, Noam, (1993), A Minimalist Program for Linguistic Theory. In Hale, K. &
Keyser, S.J. (eds.), The View from Building 20. MIT Press, Cambridge, MA.
Chomsky, Noam, (1994a), Bare Phrase Structure. MIT Occasional Papers in
Linguistics, Number 5.
Chomsky, Noam, (1994b), Language and Thought. Mover Bell, Wakefield (RI) and
London.
Chomsky, N. & Lasnik, H., (1977), Filters and Control, Linguistic Inquiry 8, 425-
504.
Cormack, Annabel & Smith, Neil, (1994), Serial Verbs. UCL Working Papers in
Linguistics, Vol. 6, Pages 63-88.
Cresswell, M.J., (1985), Adverbial Modification: Interval Semantics and its Rivals. D.
Reidel Publishing Company, Dordrecht.
Culicover, Peter W. & Wilkins, Wendy, (1986), Control, PRO, and the Projection
Principle, Language 62, 120-153.
Curry, H.B., (1963), Some logical aspects of grammatical structure. In Jakobson
(ed.), Structure of Language and Its Mathematical Aspects: Proceedings of the
Twelfth Symposium in Applied Mathematics (56-68). American Mathematical
Society,.
Davidson, Donald, (1966), The Logical Form of Action Sentences. In Davidson,
Donald (ed.), Essays on Actions and Events ((1980), 105-122). Oxford
University Press, Oxford.
Davidson, Donald, (1980), Essays on Actions and Events. Oxford University Press,
Oxford.
Dowty, David R., (1972), Studies in the Logic of Verb Aspect and Time Reference in
English. Unpublished Ph.D. Dissertation, University of Texas, Austin.
Dowty, David, (1982), Grammatical Relations and Montague Grammar. In Jacobson,
Pauline & Pullum, Geoffrey K. (eds.), The Nature of Syntactic Representation
(79-130). Reidei, Dordrecht.
Dowty, David R., (1989), On the Semantic Content of the Notion "Thematic Role". In
Chierchia, Genriaro & Partee, Barbara & Turner, Richard (eds.), Properties,
Types and Meanings (Vol. 2, 69-130). Kluwer, Dordrecht.
Dowty, David R., (1991), Thematic Proto-Roles and Argument Selection, Language
67, 547-619.
Duffley, Patrick J., (1992), English Language Series, The English Infinitive.
Longman, London and New York.
Epstein, Samuel David, (1989), Quantifiction in Null Operator Constructions,
Linguistic Inquiry 20, 647-658.

244
 Phrases.
Faraci, Robert, (1974), Aspects of the grammar of Infinitives and For
Unpublished Ph.D. Dissertation, MIT.
Farkas, Donka F., (1985), Obligatorily controlled subjects in Rom anian. Papers from
the Twenty-First Regional Meeting of the Chicago Linguistic Soci ety, Vol.
21/1, Pages 90-100. Department of Linguistics, University of Chicago,
Chicago IL.
osophy 11, 27-58.
Farkas, Donka, (1988), On obligatory control, Linguistics and Phil
Fillmore, Charles, (1966), The case for case. In Bach, Emmon & Harms, R.T. (eds.),
.
Universals in Linguistic Theory. Holt, Rinehart, and Winston, New York
ons.
Fukui, Naoki, (1986), A theory of category projections and its applicati
Unpublished Ph.D. Dissertation, MIT.
Greenbaum, Sidney, (1969), Studies in English Adverbial Usage. Long
mans, London
and Harlow.
Grimshaw, Jane, (1988), Adjuncts and Argument Structure. MIT Lexicon Project
Working Paper, No. 21; Occasional Paper, No. 36.
Grimshaw, Jane, (1990), Argument Structure. MIT Press, Cambridge,
MA.
. Dissertation,
Gruber, J.S., (1965), Studies in Lexical Relations. Unpublished Ph.D
MIT.
Hegarty, Michael, (1989), An Analysis of Null Operators in English.
Unpublished
paper (Syntax and Semantics Generals, MIT).
s. Unpublished
Hegarty, Michael, (1992), Adjunct Extraction and Chain Configuration
Ph.D. Dissertation, MIT.
Higginbotham, James, (1983a), The Logic of Perceptual Repo rts: An Extensional
Alternative tc Situation Semantics, Journal of Philosophy 80,100-127.
, Linguistic
i

Higginbotham, James, (1983b), Logical Form, Binding, and Nominals

Inquiry 14, 395-420.
Higginbotham, James, (1985), On Semantics, Linguistic Inquiry 16,
547-593.
Philosophy
Higginbotham, James, (1989), Elucidations of Meaning, Linguistics and
12, 465-517.
uage 6, 344-
Higginbotham, James, (1991), Belief and Logical Form, Mind and Lang
369.
Higginbotham, James, (1992), Reference and Control. In Lars on, Richard & latridou,
mar
Sabine & Lahiri, Utpal & Higginbotham, James (eds.), Control and Gram
(79-108). Kluwer Academic Publishers, Dordrecht.
Higginbotham, James, (1995), Sense and Syntax. An Inauguraby l Lecture delivered
r of
before the University of Oxford on 20th October 1994 the Professo
General Linguistics. Clarendon Press, Oxford.
Language
Higginbotham, James, (forthcoming), Semantic Computation, Natural
Semantics.
245
Higgins, Roger, (1979), The Pseudo-cleft Construction in English. Unpublished Ph.D.
Dissertation, MIT.
Hudson, Richard, (1992), So-Called "Double Objects" and Grammatical Relations,
Language 68,251-276.
Jackendoff, Ray, (1972), Semantic Interpretation in Generative Grammar. MIT Press,
Cambridge, MA.
Jackendoff, Ray, (1977), X-Bar Syntax: A Study of Phrase Structure. MIT Press,
Cambridge, MA.
Jackendoff, Ray, (1983), Semantics and Cognition. MIT Press, Cambridge, MA.
Jackendoff, Ray, (1990a), On Larson's treatment of the double object construction,
Linguistic Inquiry 21, 427-456.
Jackendoff, Ray, (1990b), Semantic Structures. MIT Press, Cambridge, MA.
Jackendoff, Ray, (1992), Languages of the Mind: Essays on Mental Representation.
MIT Press, Cambridge, MA.
Jones, Charles, (1985), Syntax and Thematics of Infinitival Adjuncts. Unpublished
Ph.D. Dissertation, University of Massachusetts, Amherst.
Jones, Charles, (1988), Thematic Relations in Control. In Wilkins, Wendy (ed.),
Syntax and Semantics: Thematic Relations (Vol. 21,75-89). Academic Press,
New York.
Jones, Charles, (1991), Purpose Clauses: Syntax, Thematics, and Semantics of
English Purpose Constructions. Kluwer Academic Publishers,
Dordrecht/Boston/London.
Kayne, R.S., (1983), Connectedness and Binary Branching. Foris Publications,
Dordrecht.
Kirkpatrick, Charles, (1982), The transitive Purpose Clause in English, Texas
Linguistic Forum 19, 99-123.
Kitagawa, Yoshihisa, (1986), Subjects in Japanese and English. Unpublished Ph.D.
Dissertation, University of Massachusetts, Amherst.
Koopman, Hilda & Sportiche, Dominique, (1991), The position of subjects, Lingua
85,211-258.
Kopcke, Klaus-Michael & Panther, Klaus-Uwe, (1991), Kontrolle und
Kontrollwechsel im Deutschen, Zeitschriftfur Phonetik, Sprachwissenschaft,
und Kommunikationsforschung 44, 143-166.
Krifka, Manfred, (1989), Nominal Reference, Temporal Constitution and
Quantification in Event Semantics. In Semantics and Contextual Expressions.
Foris, Dordrecht.
I,'

Kuroda, S.-Y., (198C), Whether we agree or not: Rough ideas about the comparative
syntax of English and Japanese, Lingvisticae Investigations 12, 1-47.

246
Ladusaw, William A. & Dowty, David, (1988), Toward a non-grammatical account of
theta roles. In Wilkins, Wendy (ed.), Syntax and Semantics: Thematic Relations
(Vol. 21, 62-73). Academic Press, New York.
Lappin, Shalom, (1982), The Theta Criterion and Pronominal Binding, ms (paper
delivered to NELS).
Larson, Richard, (1988), On the double object construction, Linguistic Inquiry 19,
335-391.
Larson, Richard, (1990;, Double objects revisited: reply to Jackendoff, Linguistic
Inquiry 21, 589-632.
Larson, Richard, (1991), Promise and the Theory of Control, Linguistic Inquiry 22,
103-139.
Larson, Richard & Segal, Gabriel, (1995), Knowledge of Meaning: An Introduction to
Semantic Theory. MIT Press, Cambridge, MA.
Lewis, D., (1975), Adverbs of Quantification. In Keenan, E. (ed.), Formal Semantics
of Natural Language (3-15). CUP, Cambridge. (N)
Leys, Odo, (1988), Prospektives Urn, Deutsche Sprache, 97-102.
Leys, Odo, (1991), Skizze einer Kognitiv-Semantischen Typologie derDeutschen Um-
Infinitive. (ms).
Manzini, Rita, (1983), On Control and Control Theory, Linguistic Inquiry 14, 421-
446.
May, Robert, (1985), Logical Form: Its Structure and Derivation. MIT Press,
Cambridge, MA. ' ' < •:

McConnel-Ginet, Sally, (1982), Adverbs and Logical Form: A Linguistically Realistic

Theory, Language 58, 144-184.
Montague, Richard, (1974), Formal Philosophy. Yale University Press, New Haven.
Nakayama, Mineharu & Tajima, Kazuhiko, (1993), Purpose Clauses in Japanese,
Lingua 91, 1-32.
Nishigauchi, Taisuke, (1984), Control and the Thematic Domain, Language 60, 215-
250.
Panther, Klaus-Uwe & Kopcke, Klaus-Michael, (1993), A Cognitive Approach to
Obligatory Control Phenomena in English and German, Folia Linguistica 27,
57-105.
Parsons, Terry, (199C), Events in the Semantics of English: A Study in Subatomic
Semantics. MIT Press, Cambridge, MA.
Partee, Barbara H. A Bach, Emmon, (1980), Quantification, Pronouns, and VP
Anaphora. IL! Groenendijk, J. & Janssen, T. & Stokhof, M. (eds.), Formal
Methods in the Study of Language: Proceedings of the Third Amsterdam
Colloquium (Vol. 2, 445-481). Mathematisch Centrum, Amsterdam.
Pustejovsky, James, (1992), The Syntax of Event Structure. In Levin, Beth & Pinker,
Steven (eds.), Lexical and Conceptual Semantics (A Cognition Special Edition'
247
 ed., First published in 1991 by Elsevier Science Publishers, B.V.Amsterdam,
The Netherlands. This edition reprinted from Cognition: International Journal of
Cognitive Science (1991), Vol 41, No. 1-3. Blackwell., 47-81). Blackwell,
Oxford.
Quirk, Randolph & Greenbaum, Sidney & Leech, Geoffrey & Svartvik, Jan, (1985), A
Comprehensive Grammar of the English Language. Longman, London and
New York.
Ramchand, Gillian Catriona, (1993), Aspect and Predication: The Semantics of
Argument Structure, ms.
i

Raposo, Eduardo, (1987), Case Theory and Infl-to-Comp: The Inflected Infinitive in
European Portuguese, Linguistic Inquiry 18, 85-109.
Rizzi, Luigi, (1986),-Null Objects in Italian and the Theory of pro, Linguistic Inquiry
17, 501-557.
Roberts, lan, (1987), The Representation of Implicit and Dethematized Subjects. Foris,
Dordrecht.
Roca, I.M. (ed.), (1992), Thematic Structure: Its Role in Grammar. Foris, Berlin.
Rosenbaum, P.S., (1970), A principle governing deletion in English sentential
complementation. In Jacobs, R. & Rosenbaum, P.S. (eds.), Readings in
English Transformational Grammar. Ginn, Waltham, MA.
Ruzicka, Rudolf, (1983), Remarks on Control, Linguistic Inquiry 14, 309-324.
Sag, Ivan A. & Pollard, Carl, (1991), An integrated theory of complement control,
Language 67, 63-113.
«V
i
Schein, Barry, (1992), Plurals and Events, ms.
»\'
Schoenfinkel, Moses, -t( 1924), Uber die Bausteine der mathematischen Logik,
Mathematische Annalen 92, 305-316.
Stowell, T., (1981), Origins of Phrase Structure. Unpublished Ph.D. Dissertation,
MIT.
Stowell, T., (1982), The tense of infinitives, Linguistic Inquiry 13, 561-570.
Tenny, Carol, (1987), Grammaticalising Aspect and Affectedness. Unpublished Ph.D.
Dissertation, MIT.
Vendler, Z., (1967), Linguistics in Philosophy. Cornell University, Ithaca, New York.
Wilder, Christopher, (1989), Predication, null operator infinitives and fox-deletion.
UCL Working Papers in Linguistics, Vol 1,1989.
i

Wilkins, Wendy (ed.Y, (1988), Syntax and Semantics: Thematic Relations (Vol. 21).
Academic Pr^rs, New York.
Williams, Edwin, (1574), Rule Ordering in Syntax. Unpublished Ph.D. Dissertation
MIT.
Williams, Edwin, (1980), Predication, Linguistic Inquiry 11, 203-238.

248
Williams, Edwin, (1981), Argument Structure and Morphology, The Linguistic Review
1,81-114. v
Williams, Edwin, (!?£;), Against Small Clauses, Linguistic Inquiry 14, 287-308.
Williams, Edwin, (1985), PRO and Subject of NP, Natural Language and Linguistic
Theory 3, 297-315.
Williams, Edwin, (1987a), Implicit Arguments, the Binding Theory, and Control,
Natural Language and Linguistic Theory 5,151-180.
Williams, Edwin, (1987b), NP Trace in Theta Theory, Linguistics and Philosophy 10,
433-447.
Williams, Edwin, (1989), The Anaphoric Nature of Theta-Roles, Linguistic Inquiry 20,
425-456.
Williams, Edwin, (1991), The Argument-Bound Empty Categories. In Freidin, Robert
(ed.), Principles and Parameters in Comparative Grammar (77-98). MTT Press,
Cambridge, MA.
Williams, Edwin, (1992), Adjunct Control. In Larson, Richard & latridou, Sabine &
Lahiri, Utpal ''& Higginbotham, James (eds.), Control and Grammar (297-321).
Kluwer Academic Publishers, Dordrecht.
Williams, Edwin, (1994), Thematic Structure in Syntax. Linguistic Inquiry Monograph
23. (N)
Williams, Edwin, (1995), Theta Theory. In Webelhuth, Gert (ed.), Government and
Binding Theory and the Minimalist Program (97-124). Blackwell, Oxford.
Van Wright, George, (1980), Norm and Action. Humanities Press, New York.
Zucchi, Alessandro, (1993), The Language of Propositions and Events: Issues in the
Syntax and Semantics ofNominalization. Kluwer, Dordrecht.
Zwarts, Joost, (1992), X'-Syntax - X'-Semantics: On the Interpretation of Functional
and Lexical Heads. LEd. (OTS dissertation series), Utrecht.

249