Cann

This book provides a clear and accessible introduction to formal semantics within a
linguistic framework. It presupposes no previous background in logic, but takes

students step-by-step from simple predicate/argument structures and their
interpretation through to Montague's intensional logic, bringing them to a level where
they can tackle the primary technical literature. It assumes some basic knowledge of
linguistics, but aims to be as non-technical as possible within a technical subject.
Formal semantics will be welcomed by students and researchers in linguistics,
artificial intelligence and cognitive science alike.
CAMBRIDGE TEXTBOOKS IN LINGUISTICS
General Editors: J.BRESNAN, B. COMRIE, W. DRESSLER, R. HUDDLESTON,
R. LASS, D. LIGHTFOOT, J. LYONS, P. H. MATTHEWS, R. POSNER,
S. ROMAINE, N. V. SMITH, N. VINCENT
FORMAL SEMANTICS
In this series:
P. H. MATTHEWS Morphology Second edition

B. COMRIEAspect
R. M. K EM PS ON Semantic Theory
T. B Y N O N Historical Linguistics
J. ALL WOOD, L.-G. ANDERSON, O. D AHL Logic in Linguistics
D. B. F R Y The Psysics of Speech
R. A. HUDSON Sociolinguistics
J. K. CHAMBERS and P. T R U D GIL L Dialectology
A. J. ELLIOT Child Language
P. H. MATT HEW S Syntax
A. RADFORD Transformational Syntax
L. BAUER English Word-formation
S. C. LEVIN SON Pragmatics
G. BROWN and G. YULE Discourse Analysis
R. HU DD LES TO N Introduction to the Grammar of English
R. LASS Phonology
B. C O M R I E T ^
W. KL EIN Second Language Acquisition
A. CRUT TEN DEN Intonation
A. J. WOODS, P. FLETCHERandA. HUGHES Statistics in Language
Studies
D. A. C R U S E Lexical Semantics
F. R. PALMER Mood and Modality
A. RADFORD Transformational Grammar
M. G ARM AN Psycholinguistics
G. C O R B E T T G ^ r
H. GIEGERICHEnglish Phonology
R. C ANN Formal Semantics
FORMAL SEMANTICS
An introduction
RONNIE CANN
Department of Linguistics, University of Edinburgh
CAMBRIDGE
UNIVERSITY PRESS
Published by the Press Syndicate of the University of Cambridge
The Pitt Building, Trumptington Street, Cambridge CB2 1RP
40 West 20th Street, New York, NY 10011-4211, USA
10 Stamford Road, Oakleigh, Victoria 3166, Australia
© Cambridge University Press 1993
First published 1993
A catalogue record for this book is available from the British Library
Library of Congress cataloguing in publication data applied for
ISBN 0 521 37463 4 hardback

ISBN 0 521 37610 6 paperback
Transferred to digital printing 2003
AU
For Bertie
CONTENTS
List of figures page xn

Preface xiii
List of symbols xv
1 INTRODUCTION 1
1.1 Semantics and semantic theory 1
1.1.1 Compositionality 2
1.1.2 Meaning relations 5
1.1.3 Ambiguity 8
1.1.4 Denotation 10
1.2 Interpretation and representation 13
1.2.1 Truth-conditions 15
1.2.2 Formalisation, models and sets 17
1.3 Beyond truth-conditions 19
1.3.1 Knowledge, behaviour and use 19
1.3.2 Context 22
1.4 A note on method 23
1.5 Further reading 25
2 P R E D I C A T E S AND ARGUMENTS 27
2.1 Translating English into a logical language 27
2.1.1 The syntax of Lp 27
2.1.2 A grammar fragment for English 28
2.1.3 The translation procedure 32
2.2 Interpreting LP 38
2.2.1 Individuals and identity 41
2.2.2 A little light set theory 44
2.2.3 Interpreting predicates 47
2.2.4 Finishing up 52
3 NEGATION AND C O - O R D I N A T I O N 54
3.1 Compound sentences 54
3.2 Complex formulae 55
3.3 Interpretation 60
3.3.1 Negation 60
3.3.2 Conjunction 63
3.3.3 Disjunction 67
3.3.4 Implication 70
3.3.5 Equivalence 73
3.4 Formal interpretation 74
3.4.1 Formal models 75
ix
3.4.2 Model theory 78
4 TYPE THEORY 82
4.1 Verb phrases and other constituents 82
4.2 A typed logical language 83
4.2.1 Semantic types 83
4.2.2 Translating verb phrases 88
4.3 More set theory 93
4.3.1 Relations and functions 93
4.3.2 Sets again 96
4.4 Interpreting L ^ 98
4.4.1 Denotation 98
4.4.2 Revising the theory 103
4.5 Adverbs 107
5 THE LAMBDA OPERATOR 112

5.1 The passive 112
5.2 Introducing the lambda operator 115
5.2.1 Extending L ^ 116
5.2.2 Interpreting lambda expressions 119
5.2.3 The passive again 126
5.3 Generalising lambda expressions 127
5.4 Reviewing co-ordination 136
5.4.1 Sentential co-ordination 136
5.4.2 Co-ordinating other categories 143
6 QUANTIFICATION 150
6.1 The variety of noun phrases 150
6.2 Introducing the logical quantifiers 151
6.2.1 The quantifiers 152
6.2.2 Interpreting LQ 154
6.2.3 Quantification and negation 158
6.3 A compositional approach 159
6.3.1 Translating quantifier pronouns 159
6.3.2 Complex NPs 162
6.3.3 Nominal modifiers 169
6.4 Proper names and definite descriptions 172
6.5 Two problems 177
6.5.1 Type raising 177
6.5.2 Scope ambiguities 180
6.6 Generalised quantifiers 187
6.7 Further Reading 195
7 INFERENCE 197
7.1 Making inferences 197
7.2 Logical deduction 200
7.2.1 Using the connectives 204
7.2.2 Reasoning with quantifiers 210
7.3 Lexical meaning 215
7.4 Non-truth-conditional aspects of the connectives 224
7.4.1 And 224
7.4.2 Or 276
7.4.3 If 229
7.5 Further Reading 231
8 TIME, TENSE AND ASPECT 233

8.1 Temporal contingency 233
8.2 Time 233
8.2.1 Intervals of time 234
8.2.2 Temporal models 236
8.3 Tense 241
8.3.1 Past, present and future 241
8.3.2 Complex tenses 246
8.4 Simple aspect 251
8.4.1 Perfective and imperfective 251
8.4.2 States and actions 256
8.5 Scope ambiguities 259
9 P O S S I B L E WORLDS 263
9.1 Where entailments fail 263
9.2 Intension and extension 267
9.3 Introducing other worlds 269
9.3.1 Simple modality 270
9.3.2 Accessible worlds 276
10 I N T E N S I O N A L SEMANTICS 282
10.1 Modelling intensions 282
10.2 The intensional language LIL 288
10.2.1 Intensional expressions... 289
10.2.2 ... and their interpretation 293
10.3 Interpreting opaque contexts 299
10.3.1 Oblique transitive verbs 299
10.3.2 Control verbs 306
10.3.3 Propositional attitudes 308
10.4 Two problems 315
10.4.1 Intensional equivalence 316
10.4.2 Cross-world reference 319
10.5 Postscript 320
Answers to selected exercises 323

References 333
Index 339
xi
FIGURES
The book is on the table page 11

The book is not on the table 14
The entities in Mi 41
If Jo is over six foot tall, he is over four foot tall 71
A set and its complement 97
Every student liked Jo 163
A lecturer screamed 166
Universal sublimation of [student']M 168
Existential sublimation of [lecturer']M 169
Individual sublimation of [fiona']M 174
Contingent and non-contingent formulae 271
Necessary and possible formulae 272
Negative sublimation of [dog']M 328
xn
PREFACE
Although billed as an introduction to formal semantics in general, this textbook is

concerned primarily with what has come to be called Montague Semantics and is
therefore based primarily on Montague (1970a; 1970b; 1973). A good deal of research
within Montague's general framework has been carried out since the 1970s and this
has led to many changes in, and many variations of, the original theory. Other
research has also led to reactions to Montague's programme and the development of
rival theories. Only a few of these revisions and extensions to Montague's theory
have, however, found their way into the text of the book. This may seem retrogressive,
but it is my conviction that many of the questions being asked in formal semantics and
the directions of research are best understood by learning about the more radical
elements of Montague's original approach, particularly the semantic analysis of noun
phrases and the theory of intensionality. Once these have been grasped, later
developments can be understood more easily. For this reason, the exposition develops
an account of the now classical version of Montague's theory, but references are given
for the major revisions and extensions at the end of each chapter for readers to pursue
as their interests dictate. Furthermore, there is no attempt in this book to give more of
the logical and mathematical background than is necessary to understand how such
things can help in the analysis of the semantics of natural languages. This is not an
introduction to formal semantics for logicians, but for linguists. No discussion of
completeness, consistency, formal proofs, or axiomatisation will therefore be found
in these pages. This book thus provides an introduction to certain techniques in formal
semantics as applied to natural languages (exemplified by English). Although readers
may find that the domain of meaning that appears to be amenable to the sort of
semantic analysis discussed in this book is somewhat small, the fundamental ideas of
truth-conditional semantics and the possibility of applying the techniques of formal
semantics to natural languages remain important for the development of linguistic
semantics.
As an introduction, no logical or mathematical background is assumed for the
reader, but some knowledge of Linguistics is taken for granted, particularly with
regard to syntax, but acquaintance with an introductory book in general linguistic
semantics would be helpful. An explanation of all the relevant concepts from logic
and set theory is, however, provided within the text, and references to other textbooks
that deal with these concepts are given in the further reading sections of each chapter.
Readers who have a knowledge of basic predicate logic may skip Chapters 2 and 3,
but those who do not are advised to take each chapter in turn. It is, moreover,
advisable to attempt the exercises as they come along, provided that the reader does
not get stuck on them. It is better to move on to the rest of the text than to spend many
fruitless hours struggling over a single problem (in the exercises or in the reading).
The answers to some of these exercises appear at the end of the book, but none are
given for those that have an asterisk in front of them. These latter are in the nature of
research exercises designed to make readers consider some of the sorts of problems
and difficulties that are encountered in providing formal semantic analyses of the
xiii
sentences of a natural language. References to particular solutions to some of these
exercises may, however, be found in the further reading.
This project began as a joint venture between myself and Lesley Stirling.
Unfortunately, circumstances conspired against our co-authorship and the final text
was written entirely by myself. Lesley wrote an early draft of Chapter 1 and some
notes for Chapter 8 and, while none of her original text appears in this final version,
her ideas in our early discussions have obviously influenced its final shape.
Co-teaching with her in 1986/7 also influenced the way that my later courses in formal
semantics developed and thus ultimately how certain of the topics in this book are
presented. For these, alas now unidentifiable, benefits, she has my heartfelt thanks. My
thanks also go to John Lyons who acted as a very careful editor. This book has
benefited greatly from his meticulous notes on the manuscript and his numerous
criticisms of substance and style and his positive suggestions have helped to improve
the text considerably. I also wish to thank my colleagues Jim Hurford and Jim Miller,
for their many comments on earlier versions of the text. Chapter 8 benefited greatly
from the latter's careful comments. Thanks also to Ruth Kempson for her comments
on the earliest text written by myself and Lesley Stirling and to Marguerite Nesling
for her insightful comments on the philosophical aspects of the text. The book has
grown out of a series of courses in formal semantics for undergraduate and first year
postgraduate students and I have to thank all the students at Edinburgh University, too
numerous to mention by name, who took the Advanced Semantics class between the
academic years 1986/7 and 1991/2. Their vociferous suffering over the numerous
typos of the various versions of the text and their confusion at wrong diagrams and
impossibly difficult exercises inspired me to change things well after the time that I
thought I had a finished version. The text is the better for their complaints. None of
those mentioned here are, of course, responsible for any remaining mistakes or
infelicities for which the computer is entirely to blame.
Judith Ayling and her staff at Cambridge University Press have my gratitude
for their careful editing of the final text and their considerable help with the
preparation of the camara ready copy from which this book is printed. Robert
Bradford helped a good deal with the latter stages of the production of the text, the
final version of which was prepared on a NeXTstation.
Many thanks too to Fiona Chapman who heroically produced the illustrations
while simultaneously building a house and looking after three children. Thanks also
to Tom and Betty Watson for endless gin and sympathy and to those people in the
village who were continually surprised that I had still not finished the book. Finally,
my thanks to Robert Bradford for all his help and support, emotional, financial and
computational. His help in solving the repetitive tasks in word-processing and his
patience in listening to endless rantings about logic, time and the meaning of meaning
went far beyond the realms of duty. I dedicate this book to him with my love.
Pool o' Muckhart
xiv
SYMBOLS
Logical symbols
& conjunction (and)
v inclusive disjunction (or)
ve exclusive disjunction
-» material implication (if...theri)
<-» equivalence (if and only if)
= identity operator
negation (not)
X lambda operator
» X-converts into
V universal quantifier (all/every)
3 existential quantifier (a/some)
Pres present tense operator
Past past tense operator
Fut future tense operator
Perf perfective operator
Impf imperfective operator
o possibility operator
• necessity operator
f(a) functor f applied to argument a
Predn n-place predicate
t type of a formula
e type of an individual
<a,b> complex type
<s,a> intensional type
h entailment
Hh paraphrase
*= contradiction
Variables
x,y,z individual (type e)
r individual concept (type <s,e>)
P,Q set (type <e,t>)
A property (type < s , < e , t » )
R binary relation (type < e , < e , t » )
SR relation in intension (type <s,<e,<e,t>»)
p,q,r propositional (type t or (ch. 10) <s,t>)
P generalised quantifier (type «e,t>,t>)
p intensional generalised quantifier (type <s,«e,t>,t»)
xv
Set theory
{a,..,b} set consisting of elements a,...,b
{x I f(x)} set of elements x satisfying property f
<a,b> ordered pair of elements
<a,b,c> ordered triple of elements
0 the null (or empty) set
G set membership (is a member of)
£ set non-membership (is not a member of)
n set intersection
u set union
c proper subset relation
z> proper superset relation
c subset relation
3 superset relation
< is less than or equal to
> is greater than or equal to
> is greater than
|A | cardinality of set A
AxB cartesian product of sets A and B
({<a,b>la€ A&be B})
AB set of functions from domain B to range A
M o d e l theory
[(X]M,g,w,i interpretation of a with respect to M, g, w and i
1 truth
0 falsity
M Model
A set of entities
F denotation assignment function
T set of times
I set of temporal intervals defined on T
< precedence relation
W set of possible worlds
R accessibility relation on W
g assignment of values to variables
gaA, assignment of values to variables where element g(u) = a
denotation type of expressions of type a
[ti,tj] temporal interval between times tj and tj
ft] moment of time
o overlap relation (of temporal intervals)
C subinterval relation
3[(|)]M,8 intension of <>| with respect to M and g
l[<j)] M -8 sense of c>
| with respect to M and g
XVI
Grammatical symbols
s sentence
NP noun phrase
Npr proper noun
Pro pronoun
N common noun
CNP common noun phrase
Det determiner
A adjective
A[-PRD] attributive adjective
VP verb phrase
VP[PAS] passive verb phrase
Vi intransitive verb
vt transitive verb
vVodt ditransitive verb
impersonal verb
[+FIN] finite/non-finite (verb)
Adv adverb
CONJ conjunction
NEG negative
PPby by prepositional phrase
PP«o to prepositional phrase
[iPAST] past/non-past
—» rewrite arrow
translates as
XVll
1 Introduction
1.1 Semantics and semantic theory

In its broadest sense, semantics is the study of meaning and linguistic semantics is the
study of meaning as expressed by the words, phrases and sentences of human
languages. It is, however, more usual within linguistics to interpret the term more
narrowly, as concerning the study of those aspects of meaning encoded in linguistic
expressions that are independent of their use on particular occasions by particular
individuals within a particular speech community. In other words, semantics is the
study of meaning abstracted away from those aspects that are derived from the
intentions of speakers, their psychological states and the socio-cultural aspects of the
context in which their utterances are made. A further narrowing of the term is also
commonly made in separating the study of semantics from that of pragmatics.
Unfortunately, the nature of the object of inquiry of the discipline (what constitutes
semantic meaning, as opposed to pragmatic meaning) and the domain of the inquiry
(what aspects of meaning should be addressed by the discipline) remain difficult and
controversial questions. There are, however, three central aspects of the meaning of
linguistic expressions that are currently accepted by most semanticists as forming the
core concern of linguistic semantics. These central concerns of semantic theory,
adapted from Kempson (1977:4), are stated in (1) and may be adopted as criteria for
ascertaining the adequacy of semantic theories which apply in addition to the general
conditions on scientific theories of falsifiability and rigour.
(1) A semantic theory must:
1. capture for any language the nature of the meaning of words, phrases and
sentences and explain the nature of the relation between them;
2. be able to predict the ambiguities in the expressions of a language;
3. characterise and explain the systematic meaning relations between the
words, the phrases and the sentences of a language.
One may add to these the condition that a semantic theory should provide an
account of the relation between linguistic expressions and what may be called 'things
in the world'. In other words, it is a primary concern of a semantic theory to explain
how human beings can use their language to convey information about the external
world. We may thus require a semantic theory to conform also to the criterion of
adequacy in (2).
(2) A semantic theory must provide an account of the relation between
linguistic expressions and the things that they can be used to talk about.
There are many other aspects of meaning that can be included in the domain of
linguistic semantics, but a theory conforming to the four criteria in (1) and (2) will
cover the main ground of the discipline and provide a firm basis for further research.
In this book, we will be looking at a particular theory of semantics that goes a long
way towards satisfying these criteria and that has been very influential in linguistic
1
1 Introduction
semantics over the last two decades. This theory is a formal theory of semantics and
is distinguished from general linguistic semantics by its greater use of mathematical
techniques and reliance on logical precision. This is not to say that formal semantics
and general linguistic semantics are completely separate disciplines. It sometimes
appears that these two approaches to the semantics of natural languages are mutually
incompatible, but this is not obviously true. The former draws heavily on the long
tradition of research in the latter which in turn benefits from the greater precision of
the former. Both approaches enable us to understand more about meaning and greater
integration between them would doubtless bring greater benefits to the discipline.
Formal semantics itself was devised as a means of providing a precise
interpretation for formal languages, i.e. the logical and mathematical languages that
are opposed to natural languages that are spoken or written as the native languages
of human beings. Many logicians considered it to be impossible to apply the same
rigour to the semantics of human languages, because of their supposedly inexact
syntax, their vagueness and their ambiguity. In the late nineteen-sixties, however, the
philosopher Richard Montague asserted that it was possible to use the same techniques
in analysing the meanings of sentences in English. In three articles, English as a
formal language, Universal grammar and The proper treatment of quantification in
English, all published or presented in 1970, Montague gave arguments for his
hypothesis that:
There is in my opinion no important theoretical difference between natural languages and the
artificial languages of logicians; indeed, I consider it possible to comprehend the syntax and
semantics of both kinds of language within a single, natural and mathematically precise
theory. Montague (1974: 222)
Throughout the nineteen-seventies, after his tragic death in 1971, Montague's work
had a radical effect on the study of semantics in linguistics. Indeed, his ideas on the
semantics of human languages have become central to the understanding of many of
the questions and theories being discussed in linguistic semantics today. Owing to the
relatively recent application of the tools of formal semantics to the analysis of natural
languages, however, there are many topics in linguistic semantics that have not yet
been formally analysed, but it is hoped that ultimately a good deal of linguistic
meaning will be amenable to the sort of rigorous treatment envisaged by Montague.
It is the exposition of Montague's theory in its now classical form that constitutes the
subject matter of this book, but, before the main points of his semantic theory are
introduced, the four criteria of adequacy in (1) and (2) above will be discussed in more
detail in order to provide a clearer idea of the fundamental issues that underlie the
development of the theory in later chapters.
1.1.1 Compositionality
A fundamental property that any semantic theory must have is the ability to pair the
syntactic expressions of a language with their meanings. In the first condition of
adequacy in (l.a), above, this property is characterised as a requirement that a
semantic theory account for the nature of the meaning of linguistic expressions and
be able to pair every expression in a language (words, phrases and sentences) with an
appropriate meaning. As already mentioned, the characterisation of meaning is a
Semantics and semantic theory
controversial matter, but whatever meanings are taken to be within a theory, it is

obvious that there must be some way of associating them with appropriate linguistic
expressions. This is not a trivial matter, however, and there are a number of important
points that need to be discussed with respect to this property.
In the first place, let us consider more closely what it is that is to be assigned
meanings by a semantic theory. Condition (l.a) refers to words and sentences as the
carriers of meaning. The term sentence here is being used in its abstract sense,
common in linguistics, as the largest unit of syntactic description, independently of
its realisation in spoken or written texts. Like the term sentence, word is also
ambiguous in everyday English. Within semantics, the notion of word that is most
useful is that of the lexeme which is an abstract grammatical construct that underlies
a set of word forms which are recognised as representatives of 'the same word' in
different syntactic environments. For example, the word forms sing, sings, singing,
sang and sung are particular inflectional variants of a lexeme which we may represent
for the time being as SING. It is to lexemes and not to word forms that meanings
should be assigned, because while the inflectional properties of the verb SING may
vary in different syntactic environments, the sort of action described by the verb
remains the same. For this reason, the meaning of words is referred to in this book as
lexical meaning, rather than word meaning.
Although it is possible for the meanings assigned to lexemes and sentences to be
very different from each other, it is reasonable to expect the meanings of sentences to
be related to the meanings of the lexemes underlying the word-forms they contain. It
is intuitively implausible for there to be a language where the relation between the
meaning of a sentence and the meanings of its component lexemes is entirely random.
While languages do contain idiomatic phrases and sentences where lexical and
sentential meaning are not transparently related (e.g. kick the bucket meaning the same
as DIE in English), this is never the general situation. If there were no direct relation
between lexical and sentential meaning, of course, the meaning of each sentence in a
language would have to be listed. Since the number of sentences that make up a
language is infinite, this would mean that no human being would be able to determine
the meanings of all the sentences of any language owing to the finite resources of the
brain. This is absurd, of course, and just as sentences are defined recursively by
syntactic rules, taking words (or morphemes) as their basis, so their meanings should
also be defined recursively from the meanings ascribed to the lexemes they contain.
Thus, in addition to associating each expression in a language with a meaning,
an adequate semantic theory must also be able to explain how the meanings of smaller
expressions contribute to the meanings of larger ones that contain them. A theory that
derives the meaning of larger expressions from those of smaller ones is said to be
compositional. The Principle of Compositionaiity, given an initial definition in (3),
is generally attributed to the German philosopher Gottlob Frege, and is thus
sometimes referred to as the Fregean Principle of Compositionaiity or just the
Fregean Principle, although it is unlikely that he ever stated the principle in precisely
this way.
(3) The meaning of an expression is a function of the meaning of its parts.
The notion of a function will be discussed in more detail in Chapter 4, but
essentially it is an operation that derives a single result given a specified input. Thus,
1 Introduction
the principle (3) minimally requires that the meaning of a larger expression be
uniquely determined from the meanings of its component parts. This cannot be all
there is to compositionality, however, since, otherwise, we would expect that
sentences containing the same words mean the same thing. This is, of course, not true.
The sentence Jo kicked Chester does not mean the same as Chester kicked Jo. It must
also be the case, therefore, that the syntactic structure of an expression is relevant to
the derivation of its meaning. Indeed, we may strengthen the principle of
compositionality so that, in deriving the meaning of a composite expression, the
meaning of its component expressions are combined in some way that refers to the
way they are combined by the syntax. This implies that wherever meanings are
combined in a particular way to derive the meaning of a composite expression, all
other composite expressions of the same sort have their meanings determined in the
same way. In other words, the construction of meanings is rule-governed, in the same
way that the construction of the well-formed syntactic expressions of a language is
rule-governed. For example, whatever rule derives the meaning of the sentence Jo
sang from the subject Jo and intransitive verb sang applies to all declarative sentences
derived by combining a subject noun phrase with the appropriate form of an
intransitive verb.
Furthermore, it is a general property of human languages that all the
sub-expressions of a grammatically well-formed phrase have a role to play in the
interpretation of a sentence, even if, on occasions, this role is predictably redundant
(as, for example, in double negative constructions in certain dialects of English like
/ never did nothing where the second negative expression merely reinforces the idea
of negation introduced by the first). Semantic rules should, therefore, not be allowed
to delete meanings during the derivation of the meaning of a composite expression.
The effect of this restriction is to make the creation of the meanings of larger
expressions monotonic with respect to their component parts where a derivation is
said to be monotonic if all properties of previous parts of a derivation are maintained
throughout. In other words, once information is introduced into a monotonic
derivation, it is not lost thereafter. The initial definition of compositionality in (3) may
thus be strengthened to give the statement in (4).
(4) The principle of compositionality: The meaning of an expression is a
monotonic function of the meaning of its parts and the way they are put
together.
The implications of this interpretation of the principle of compositionality is that
meanings should be ascribed not only to lexemes and sentences but also to other
syntactic constituents. It is thus generally assumed that meanings should be assigned
to all the well-formed constituents of a language, not just to its words (lexemes) and
sentences. Indeed, the concept of syntax as a bridge between phonology and
semantics, current in many grammatical theories, would seem to require that all
constituents be assigned a meaning by the semantics and, furthermore, that (surface)
syntactic structure should directly determine how the meanings of sentences are
derived. It is common to assume that semantic constituency parallels syntactic
constituency and hence that an adequate semantic theory must be able to ascribe
appropriate meanings to noun phrases like the old cat, Jo's mother, Chester and verb
phrases like sang, kicked the cat, ran slowly, etc, according to their syntactic structure.
One way in which this may be achieved is to adopt the hypothesis that for each
syntactic rule of the grammar (or syntactic structure admitted by the grammar) there
is a corresponding semantic rule that derives the meaning of the resultant expression
(or structure). For example, assuming that there is a rule that defines a sentence in
English as consisting of a noun phrase plus a verb phrase, then the adoption of the
rule-to-rule hypothesis in (5), together with the principle of compositionality in (4),
requires that there be a corresponding semantic rule deriving the meaning of the
sentence from the meanings of its immediate constituents, NP and VP.
(5) Rule-to-rule hypothesis: For each syntactic rule there is a corresponding
semantic rule.
The principle of compositionality in (4) is assumed to be a constraint on semantic
theories and, indeed, will be seen to be the primary motivator behind much of the
discussion in later chapters. The rule-to-rule hypothesis, on the other hand, is not a
necessary requirement of a semantic theory, but a means of achieving
compositionality. We will have cause to question the validity of this hypothesis in the
later chapters of this book, but it is used in the earlier chapters to maintain a
transparent relation between syntactic structures and semantic representations.
1.1.2 Meaning relations

Another aspect of meaning that must be accounted for by any semantic theory is the
systematic relations that hold between different expressions in a language. According
to the condition of adequacy in (1 .b), a theory must satisfactorily analyse the intuitions
speakers of English have about the semantic relations between lexemes and between
sentences. This assumes that expressions in a language which may not be syntactically
related may be related semantically and, indeed, such is the case. Consider the
sentences in (6). Assuming that the reference of the name Jo and the discourse context
are held constant for all the sentences in (6), then the sentences in (6.b) to (6.i) are
semantically related to that in (6.a), even though it is not always the case that there is
a direct syntactic relation between them.
(6) a. Jo stroked a cat.
b. A cat was stroked.
c. There was a cat.
d. No-one stroked a cat.
e. There are no such things as cats.
f. A cat was stroked by Jo.
g. It was Jo who stroked the cat.
h. Jo touched a cat.
i. Jo stroked an animal.
The relation between the sentences in (6.a) and those in (6.b) and (6.c) is one of
entailment, as it is intuitively impossible for it to be true that Jo stroked a cat on
some occasion without it also being true that a cat was stroked or that there existed a
cat to be stroked on that occasion. We may thus define a sentence S, as entailing a
sentence S2 if the latter truly describes a situation whenever the former also does. The
I Introduction
negation of an entailment always derives a contradiction and a sentence Si may be

said to contradict a sentence S2 if the former must be false when the latter is true (or
vice versa). For example, the assertion that Jo stroked a cat is contradicted by the
non-existence of cats, making (6.e) a contradiction of (6.a). The relation between (6.a)
and the sentences in (6.f) and (6.g) is also primarily semantic, although most syntactic
theories recognise a syntactic relation between the sentences as well. Using a common
term in a technical way, we may say these sentences are paraphrases of each other,
since they all have the same core meaning. Another way of putting this is to say that
they mutually entail each other. Hence, we may say that a sentence Si is a paraphrase
of a sentence S2 if Si entails S2 and S2 entails S,. An adequate theory of semantics
must, therefore, provide an account of entailment, contradiction and paraphrase that
allows one to identify which sentences are entailed by, or contradict or paraphrase,
another in a language. Indeed, this concern, along with compositionality, is a major
motivation for the theoretical programme developed in this book.
Other sorts of implication between sentences are also recognised in general
linguistic semantics. Some of these derive from lexical meaning as in (6.h) and (6.i)
which are related to (6.a) by virtue of the meanings of the lexemes STROKE and CAT,
respectively. It is part of the meaning of the lexeme STROKE thai an action of stroking
also involves an action of touching, so that (6.a) implies (6.h). Furthermore, it is part
of the meaning of CAT that anything that is a cat is also an animal and thus (6.a)
implies (6.i). The meaning relations that hold between the lexemes of a language (or
between lexemes and larger expressions) are called sense relations and include
hyponymy, which holds if the sense of one lexeme includes that of another (e.g.
between CAT and ANIMAL); synonymy, where two lexemes have the same sense (e.g.
betv/eenMERCURYand QUICKSILVER); and oppositeness, where two lexemes have
opposing senses (e.g. BIG and SMALL and DEAD and ALIVE). Hence, an adequate
theory of semantics must give some account of lexical meaning and, in particular, of
the sense relations that hold between lexemes in a particular language.
Other implicational meaning relations hold between sentences in addition to
those that have been noted above. One of these is so-called conversational
implicature, which is derived not from the conventional meanings of words or
sentences, but from principles of discourse and context. For example, in the exchange
in (7), the apparent irrelevance of Ethel's reply in (7.b) to Bertie's question in (7.a)
leads the latter to infer (7.c). The reason behind inference has to do with Bertie's
expectations about Ethel's co-operativeness in providing him with the information he
needs. The fact that she has not given a straight answer leads Bertie, not to assume that
she is being deliberately unhelpful, but to look for some piece of information that is
relevant to his question that is indirectly implied by Ethel's response. This example
is discussed in more detail in Section 1.3.1 below.
(7) a. Bertie: Is Fiona a good lecturer?
b. Ethel: She has a good line in sweaters.
c. Bertie (thinks): Fiona is not a good lecturer.
Another sort of implication between sentences is called presupposition. A
sentence is said to presuppose another if its truth and that of its negation both imply
that the presupposed sentence is also true. In other words, presupposition deals with
aspects of meaning that are assumed to hold of a situation when a sentence is uttered
to describe that situation. For example, the use of the definite article in a noun phrase
is said to presuppose the existence of something that has the property described by the
common noun in the same NP. This is illustrated in (8) where the truth of the
sentences in (8.c) and (8.d) is presupposed by that of (8.a) and its negation in (8.b),
because of the use of the in the subject noun phrase. (8.c) is not implied by (8.e) which
replaces the in the subject NP by every (as shown by the bracketed causal clause
which denies the truth of (8.c), apparently without contradiction) and, while (8.f)
implies (indeed, according to the discussion above, entails) (8.c), its negation in (8.g)
does not.
(8) a. The Duchess of Muckhart terrorised the village.
b. The Duchess of Muckhart didn't terrorise the village.
c. There is a Duchess of Muckhart.
d. There is a village.
e. Every Duchess of Muckhart terrorised the village (because there is no
Duchess of Muckhart).
f. A Duchess of Muckhart terrorised the village.
g. A Duchess of Muckhart didn't terrorise the village.
It is usually assumed that implicatures such as that in (7) result from principles of
conversation and thus form part of the domain of pragmatics rather than semantics
(see Chapter 7 for some further discussion). More controversial, however, is the status
of presupposition. Whether it should be included in semantic or pragmatic theory is
an extremely vexed question, as indeed is the definition and status of the phenomenon
itself. As the inclusion of this topic would require considerable discussion, it is
omitted from consideration in this book. This is for convenience only and should not
be taken to reflect on the importance of the topic, only on the controversiality of its
analysis. The reader is referred for further information on this vexed topic to the books
and articles mentioned at the end of this chapter.
In addition to accounting for these semantic relations, a semantic theory may
also be required to provide some account of anomaly in the meaning of expressions
in some language. It should, therefore, be able to explain why certain expressions
which are syntactically well-formed are unacceptable or deviant from the semantic
point of view. For example, the sentence in (9.a) is syntactically well-formed and
semantically coherent in English. Those in (9.b) and (9.c), however, are semantically
anomalous despite the fact that they have the same syntactic structure as (9.a). Such
sentences can, however, be given some sort of non-literal interpretation (although (9.c)
is harder to find an interpretation for than (9.b)), unlike the completely ill-formed
expression in (9.d) which is simply not English. This decline in acceptability from
(9.a) to (9.c) and the incoherence of (9.d) should thus be explained by an adequate
theory of semantics.
(9) a. Green Wellington boots are very popular now.
b. Green ideas are very popular now.
c. Green corollaries are very popular now.
d. *very Wellington are boots popular now green.
1 Introduction
1.1.3 Ambiguity
The third area of meaning that Kempson (1977) suggests must be explained by a
semantic theory is ambiguity. A sentence is said to be ambiguous whenever it can
be associated with two or more different meanings. Ambiguity can arise in a sentence
for a number of reasons: through the ascription of multiple meanings to single words
(e.g. (lO.a)); through the assignment of different syntactic structures to a sentence
(e.g. (lO.b)); or through the use of certain expressions that may have different
semantic scope (e.g. (lO.c)).
(10) a. Ethel's punch was impressive.
b. The strike was called by radical lecturers and students.
c. Every good politician loves a cause.
The first sort of ambiguity occurs where an expression is associated with two or more
unrelated meanings, as in (10.a) where the word punch may be interpreted as a drink
or as an action. Lexemes whose word forms have this property are called homonyms
and can be subdivided into homophones, where the forms of the lexeme sound the
same but may be written differently, e.g. draft and draught which can both be
represented phonemically as /draft/, and homographs, e.g. lead, which are written the
same, but which are pronounced differently. Some lexemes are both homophones and
homographs, like PUNCH. Homonyms can be divided into full homonyms (like
BANK, PUNCH), where all of the lexeme's associated word forms are phonetically
or orthographically identical, and partial homonyms (like FIND, FOUND), where just
some of its word forms are identical.
Homonymy is often contrasted with polysemy. A polysemous lexeme is one that
is interpreted as having multiple senses that are not entirely distinct, as is the case in
the standard examples of homonyms. The classic example of a poly seme in English
is the lexeme MOUTH which has different interpretations depending on what sort of
entity is described as having a mouth. There are, for example, human mouths, mouths
of caves, mouths of bottles, mouths of rivers, and so on. In each of these cases, the
properties of the entity described by MOUTH are different, but not absolutely
different, as each one refers to an opening of some sort. The difference between
homonymy and polysemy is one of degree, and precise definitions of these terms are
difficult and controversial. As this book is not primarily concerned with lexical
meaning, no attempt will be made to differentiate the two notions or to incorporate
polysemy within the theory at all. As will be seen in Chapter 2, the approach to
homonymy taken here is very simplistic: the senses of homonymous lexemes are
simply differentiated formally by the use of superscripts, where necessary. Although
an account of polysemy and a better approach to homonymy may be possible within
the theory of formal semantics presented in later chapters, these matters are not central
to the concerns of this book and an adequate discussion of the issues involved would
only serve to increase the size of the book without serving any great purpose. The
decision to exclude polysemy from consideration and to take a simplistic view of
homonymy is taken on the grounds of expository convenience and readers are again
referred to the further reading noted at the end of the chapter.
A more interesting source of ambiguity from the point of view of the formal
semanticist is illustrated in (lO.b). Here the ambiguity results from the possibility of
assigning two or more syntactic structures to a single grammatical string of words.
To ascertain the meaning of (lO.b), for example, it is necessary to know whether the
adjective radical modifies the nominal phrase, lecturers and students, in which case
both the lecturers and the students who called the strike are all radical, or whether it
modifies just the noun lecturers, in which case the lecturers who called the strike are
said to be radical but the political attitude of the students who did so is not specified.
These two readings are illustrated in (11) where the labelled bracketings of the
agentive noun phrase in (ll.b) and (ll.d) correspond to the readings indicated in
(1 l.a) and (1 l.c), respectively.
(11) a. The strike was called by lecturers who are radical and by students.
b. [Np [NI [NI radical lecturers] and students]].
c. The strike was called by lecturers who are radical and by students who are
radical.
d. [NP [Ni radical [N1 lecturers and students]]].
In the above example, what is at issue is the scope of the adjective, radical. In
(ll.a), it modifies, and thus has scope over, the noun lecturers, while in (ll.b) its
scope is the nominal phrase lecturers and students. Scope is an important concept in
semantics and a primary source of ambiguity which involves not only adjectives, but
also conjunctions, like and, or, etc and quantifiers, like every, all, and some in
English. Structural ambiguity of this sort thus has its source in the syntax of a
language, but there are other scope ambiguities that do not directly depend on the
syntactic structure of a sentence. Such ambiguity usually involves negation (not),
quantification (every, some) and other elements like tense, which do not vary their
syntactic position according to the reading of the sentence. For example, the two
readings of the sentence in (lO.c) can be made clear by those in (12). In (12.a), there
is only one cause that every good politician loves, while in (12.b) each politician may
love a different cause. The sentence in (lO.c), however, is usually only assigned a
single surface constituent structure, so that this ambiguity cannot be directly
attributed to a syntactic source and is referred to as a semantic scope ambiguity.
(12) a. Every politician loves a cause and that is their own career.
b. Every good politician loves a cause and each one loves a cause that
everyone else loathes.
An adequate semantic theory must thus be able to predict where structural
ambiguity is likely to arise in a language and provide a means of differentiating the
interpretations of the different structures to an ambiguous sentence by the grammar,
where this is relevant. It should also ensure that sentences that have two (or more)
syntactic derivations, but only one semantic interpretation, are not assigned more than
one meaning (see Chapter 3 for examples involving the conjunctions and and or). The
theory should also provide an account of scope ambiguities where these are not
directly reflected in syntactic derivations, and be able to differentiate the scopes of
particular expressions independently of the syntax.
1 Introduction
1.1.4 Denotation
The final criterion of adequacy that is considered here is stated in (2), above, and is
the most important for our purposes, since it forms the basis of the semantic theory to
be proposed in the rest of this book. This criterion requires a semantic theory to give
an account of the relation between linguistic expressions and what they can be used
to talk about. Since language can be used to talk about what is outside the linguistic
system, it is essential that a semantic theory should be able to associate linguistic
expressions with extra-linguistic objects. Language is not used solely to talk about
itself, but rather it is most commonly used to convey information about the situations
in which human beings find themselves. Since a listener can in general understand the
meaning of what is being said by a speaker, meanings must be publicly accessible in
some sense. One way that this public accessibility must be realised is in the
association of linguistic expressions with publicly identifiable entities and situations.
For example, the utterance of a sentence like The book is on the table conveys
information about two entities, one of which is conventionally called a book in
English and one of which is conventionally called a table, and the relation between
them. Someone who hears an utterance of this sentence associates it with the situation
pictorially represented in (13). Although (13) is itself a representation of an actual (or
possible) situation, it is nonetheless a non-linguistic representation and a theory of
semantics should be capable of relating the meaning of the sentence to the picture and,
indeed, to concrete, non-representational situations where there is a (single) book on
the table.
The association between the sentence The book is on the table and the situation
represented in (13) depends in part on there being, in the situation described, an
instance of a thing that is conventionally called a book and one that is conventionally
called a table in English. In other words, part of the meaning of the sentence depends
on the sorts of extra-linguistic entities that can be referred to by the lexemes BOOK
and TABLE. The aspect of the meaning of an expression that concerns its relation to
such objects is called its denotation and an expression is said to denote particular
sorts of extra-linguistic objects. Although this relation has often been called the
reference of an expression, this book follows the usage of Lyons (1977) and reserves
this latter term for the act of picking out a particular entity denoted by the expression
through the utterance of that expression on some occasion. For example, in uttering
the sentence The book is on the table, a speaker is said to be referring to two
particular, contextually unique, entities. The entities being referred to by the use of the
definite noun phrases, the book and the table, are single elements in the class of
entities denoted by the lexemes BOOK and TABLE.
Thus, a speaker may use linguistic expressions to refer, but linguistic expressions
themselves denote. No more will be said here about the act of reference, and the
differences between denotation and reference, but for more details the reader is urged
to consult the further reading at the end of this chapter.
Informally, we may think of the denotation as the relation between an expression
and a class of various sorts of individuals, events, properties and relations that may be
referred to by the use of the expression on some particular occasion. The lexeme
BOOK may, therefore, be thought of as denoting the set of all books, TABLE as
denoting the set of all tables, while the preposition ON may be thought of as denoting
10
The book is on the table
the set of all the pairs of entities of which one is on the other. It is easy to grasp the
notion of denotation with respect to lexemes that denote concrete entities like books
and tables, but the question arises about whether abstract lexemes like LOVE,
KNOWLEDGE or THEOREM or ones denoting fictitious entities like UNICORN or
HOBBIThzve denotations in the same way. The answer is, as might be expected,
controversial, but here the position is taken that there is no essential difference
between such expressions and those that denote concrete entities. Thus, the noun
LOVE is taken to denote a set of entities just like BOOK. The difference between
them is that the entities denoted by the former are abstract while those denoted by the
latter are concrete. Although the postulation of abstract entities of this sort may cause
problems from a philosophical point of view, it does have the advantage of reflecting
the fact that the same sorts of linguistic expressions (e.g. nouns) are used in many, if
not all, languages to refer to both abstract and concrete things.
In a similar fashion, lexemes describing fictitious entities like hobbits, or entities
that are no longer extant like dodos, are also assumed to have a denotation. It is,
however, useful to distinguish between the denotations of lexemes that may be used
to refer to entities that exist in the real world (including abstract ones) and those that
do not. We can do this by making a distinction between two aspects of denotation.
Nouns like BOOK may be used to refer to entities in the world, but the entities of
which one can truthfully say That is a book all share a certain common property, their
'bookness', so to speak. In the same way, the set of entities that are red all have the
property of redness and the set of entities that run all have the property of running. In
other words, we distinguish between the property denoted by a common noun like
BOOK, adjective like RED or intransitive verb like RUN and the entities it can be used
to refer to. The former part of the meaning of the lexeme is often referred to as its
sense and is opposed to the idea of its reference. However, just as the latter term is
used here for a different notion, so too is that of sense which is used solely with
respect to the sense relations that hold between the lexemes of a language (see Section
1.2 and Chapter 7). The distinction between the different aspects of the meaning of
BOOK noted above are treated as a distinction between the intension of an expression
11
I Introduction
and its extension. The former corresponds to the property aspect of common nouns,
whereas the latter corresponds to the entities that they may be used to refer to in the
world. Thus, the extension of the lexeme BOOK is the set of all books whilst its
intension is the property of being a book. Both properties and sets of entities are
external to the linguistic system and thus constitute aspects of denotation. This
distinction allows a differentiation between entities like books which have existence
in the real world and those like unicorns that, presumably, do not. In the latter case,
the lexeme UNICORN has no extension in the real world, but it does have an
intension, the property of being a unicorn. Thus, we may speak about unicorns and
other entities without them needing to exist in the real world. This distinction is
discussed in more detail in Chapters 9 and 10, where a specific theory of intensionality
is put forward. Until then, we will be concerned primarily with the notion of extension
and thus be concentrating on the relation between linguistic expressions and existing
entities.
It is not necessary to restrict the notion of denotation to lexemes, but it may be
extended to all well-formed linguistic expressions. For example, the verb phrase
kicked a cat may be taken to denote the class of actions involving the kicking of a cat,
which is distinct from the class of all actions involving kicking a dog which would be
denoted by the expression kicked a dog, and so on. The denotations of other
expressions, like quantified noun phrases, are less easy to specify informally, but we
will see in Chapter 6 how they can be defined. Those expressions that are not used to
pick out external entities in any way may also have denotations. Such expressions are
typically described as functional or grammatical expressions, like determiners and
conjunctions, as opposed to the content expressions, like nouns, verbs and adjectives.
This distinction amongst the syntactic categories of a language is a traditional (and
very useful) one and can be reflected in semantics by assigning rather different sorts
of denotation to the two sorts of expression. Grammatical expressions are taken to
denote logical relations between groups of entities that are denoted by content
expressions. This subject is dealt with in proper detail in later chapters of the book,
where the differences between the denotations of content and functional elements will
become clear.
A theory of denotation is thus not a trivial one and any semantic theory that
provides an account of this important relation has already achieved a great deal.
Throughout this book the terms denotation, denotes, intension and extension are used
a great deal. Although denotation most properly refers to the relation between an
expression and some entity, event, property or relation, it will also be used below to
refer to what an expression denotes (what Lyons (1977) calls the denotatum of an
expression). The term is used in this way in discussions of general importance that
include both extensional and intensional denotation. Where reference is being made
to what a specific expression denotes, the terms extension or intension are used,
depending on which aspect of denotation is relevant. Thus, the extension of an
expression is taken below to refer to what the expression extensionally denotes, and
similarly for the term intension. This terminology should cause no confusion, but the
reader is asked to pay special attention to their use in the earlier chapters of this book.
12
Interpretation and representation
1.2 Interpretation and representation

In the last section, we looked at those aspects of meaning that constitute the minimum
domain of an adequate theory of linguistic semantics. For the rest of this book, we will
be exploring a theory that attempts to provide an account of such phenomena and, in
so doing, to define what semantic meaning is. There are a number of competing
theories of meaning and the one that is adopted in this book starts from the premiss
that one of the most important aspects of language is to talk about entities in the
world. The central core of meaning in this theory is thus determined by the relation
between linguistic expressions and the entities that they can be used to refer to. In
other words, denotation, and specifically extensional denotation, forms the basis of the
theory of meaning developed here. Although this theory has many critics and gives
rise to a number of difficult problems, it nevertheless continues to have a large amount
of support and has proved extremely fruitful in the study of linguistic semantics. In
this section, a general overview is presented of the major aspects of the theory that are
presented in more detail in the following chapters.
The main question that arises at this point is where the analysis should start.
Many semantic theories have taken the meaning of words (lexemes) as basic and
derived sentential (and phrasal) meanings from it. Whatever the successes or failures
of such theories, for a theory whose basis resides in the informational content of
language, what languages may be used to talk about, it is more appropriate to begin
with sentence meaning and derive the meanings of its constituent parts from this.
Declarative sentences are typically used to make statements which convey
information directly about some state of affairs. Expressions smaller than a sentence,
however, require more information from context for their interpretation than sentences
do. For example, in a situation in which there are two people, one of whom is
searching for a particular book, the second person knows that the book that the first
is looking for is on the table and may utter any of the four expressions in (14) to
convey this information to his or her partner. Only the first, a full declarative sentence,
is complete and unambiguous, requiring the minimum amount of extra information
to be provided by the context. Thus, while (14.a) may be used to describe a situation
like that depicted in (13), it cannot be used to describe truthfully the one depicted in
(15) in which the book is not on the table, but under it. The utterance of the
expressions in (14.a) to (14.d) could, however, be used to describe this situation (i.e.
where the speaker is referring to the vase being on the table, not the book) and a good
many more, depending on the context.
(14) a. The book is on the table.
b. On the table.
c. The table.
d. Table.
Declarative sentences, when uttered to make statements, thus form the basis of
our investigation into semantic meaning. Such expressions, uttered to such a purpose
(and not, for example, indirectly as a polite command or a question), are said by
philosophers to express propositions and this usage has become commonplace in
linguistics. Like most useful and interesting ideas, however, propositions have had
a controversial and contentious life and been variously defined. Lyons (1977)
13
1 Introduction
The book is not on the table
provides the definition in (16.a), while Hurford and Heasley (1983) give that in
(16.b). Other definitions ascribe them the further properties in (16.c) and (16.d).
(16) A proposition:
a. is what is expressed by a declarative sentence when that sentence is uttered
to make a statement. Lyons (1977:141)
b. is that part of the meaning of a sentence that describes a state of affairs.
Hurford and Heasley (1983:19)
c. may be true or false.
d. may be known, believed or doubted.
We will sidestep the controversy surrounding definitions like those in (16) and
adopt all four statements as descriptions of the properties that propositions have and
make the assumption that declarative sentences (which we will henceforth assume to
be uttered only as statements, thus ignoring other uses of such sentence forms) express
propositions on particular occasions of utterance. According to (16.c), the proposition
expressed by a sentence may be true or false and this truth or falsity may vary with
respect to different occasions on which the sentence is uttered. For example, the
proposition expressed by the sentence The book is on the table uttered with respect to
the situation depicted in (13) is true, but it is false, if uttered with respect to the
situation pictorially illustrated in (15). Propositions may thus vary in their truth value
from utterance occasion to utterance occasion. In terms of denotation, we may,
therefore, think of sentences as extensionally denoting truth values on particular
occasions of utterance (and, in Chapter 10, it will be suggested that sentences
intensionally denote propositions). This rather abstract idea of a sentence
extensionally denoting truth or falsity forms the starting point of the semantic theory
to be pursued through the rest of this book and it is to a consideration of what it means
for something to be true that we now turn.
14
1.2.1 Truth-conditions
The definition of what it means for a statement to be true that is adopted in this book
is the correspondence theory of truth given in (17) which is so called because a
statement is defined as being true if, and only if, the state of affairs described by the
statement holds.
(17) The correspondence theory of truth: A statement in some language is true
if, and only if, it corresponds to some state-of-affairs.
The definition in (17) applies to statements in any language, whether a natural
language learnt and spoken by human beings or an artificial language devised for
specific purposes. Since declarative sentences are used to make statements in English
(and other languages) and since, according to (16.a) above, declarative sentences are
taken to express propositions when uttered as statements, we can revise (17) to apply
to the particular case of human languages as in (18).
(18) The proposition expressed by a declarative sentence uttered as a statement
is true on some particular occasion if, and only if, that proposition
corresponds to some state-of-affairs that obtains on that occasion.
This definition makes specific reference to a state-of-affairs that holds on the occasion
of the utterance of the sentence, and this is an important point. As we have seen, the
proposition expressed by an utterance of The book is on the table is true with respect
to the situation depicted in (13), but not to that depicted in (15), because in the latter
case there is no correspondence between the state-of-affairs described by the sentence
and that portrayed in the picture. The situation in which, or with respect to which, a
sentence is uttered constitutes an important part of the interpretation of sentences, and
other expressions, to be discussed in later chapters of the book.
Just knowing whether a statement is true or false does not, of course, itself tell
us what the statement means. Speakers of a language know the meaning of a sentence
in their language even if they do not know the truth or falsity of the proposition
expressed by that sentence. For example, Ronnie has just dropped a cup of tea on the
floor is a perfectly comprehensible sentence in English whose meaning is clear,
irrespective of whether readers know whether it describes a true state-of-affairs.
Although readers may not know whether the proposition expressed by the sentence is
true or false, they do know the conditions that must obtain for it to be true, i.e. they
know what the world must be like for the statement to correspond to some actual
state-of-affairs. This idea forms the basis of truth-conditional semantics, whose
central hypothesis is that the core meaning of a sentence is its truth-conditions.
(19) To know the core meaning of a sentence uttered as a statement is to
understand the conditions under which it could be true.
The theory thus equates core meaning with knowledge of how the truth of a
declarative sentence can be ascertained without requiring the truth or falsity of a
sentence to be known or know able in any particular situation. Every sentence in the
language whose semantics is being defined must, therefore, be associated by the
theory with at least one set of truth-conditions (more if the sentence is ambiguous).
Hence, for our semantic theory to be adequate, it must be able to derive an infinite set
15
1 Introduction
of statements of the form of (2O.a) where 5 is a sentence in the object language and
p is a set of truth-conditions. By hypothesis, the statement in (2O.a) is equivalent to
that in (2O.b). (Here, and elsewhere, the abbreviation //f stands for the phrase if and
only if.)
(20) a. S is true iff p.
b. S centrally means that p.
Under this interpretation of meaning, the statement in (21) gives the core meaning of
the sentence The book is on the table.
(21) The book is on the table is true iff the book is on the table.
Of course, (21) is not very informative and it appears that the meaning of the
sentence is being defined in terms of itself, making the theory hopelessly circular.
However, this apparent circularity is caused by the fact that the language that is being
used to explicate the truth-conditions of the sentence is the same as the language of
the sentence itself. A language that is being analysed is called an object language and
that used to talk about an object language is called a metalanguage. The apparent
circularity on (21) is lessened if the object language and the metalanguage differ. For
example, if we change the object language to Modern German, as in (22.a), or adopt
Modern French as the metalanguage, as in (22 .b), the statements become more
informative and less apparently circular.
(22) a. Das Buch ist aufdem Tisch is true iff the book is on the table.
b. The book is on the table est vrai si, et seulement si, le livre est sur la table.
Even using English as both the object language and the metalanguage, we can state
the truth-conditions for our example sentence in a way that makes the import of (2O.a)
clearer. The statements in (23) specify the truth-conditions of the object sentence The
book is on the table and must all hold if the proposition it expresses is true on some
particular occasion of utterance. According to the equivalence hypothesised between
(2O.a) and (2O.b), the specification of the truth-conditions shown in (23) thus provides
a theory of the core meaning of the sentence The book is on the table.
(23) a. There is a contextually unique entity which is in the (extensional) denotation
of the lexeme BOOK.
b. There is a contextually unique entity which is in the (extensional) denotation
of the lexeme TABLE.
c. The entity in (a) stands in the relation of being on the entity in (b).
Much of what is done in later chapters concerns the development of a
metalanguage, still based on English but with a rigorous interpretation, that may be
used to specify the truth-conditions of a range of English sentence types (and by
hypothesis equivalent types in other human languages). This metalanguage is used
to connect the proposition(s) expressed by a sentence to states-of-affairs by detailing
the relationship between the syntactic structure of the sentence and the denotations of
the lexemes and other expressions that make this up. In this way, the core meaning of
the constituent expressions of a sentence is defined as the contribution they (or more
strictly their denotations) make to the truth-conditions associated with that sentence.
16
1.2.2 Formalisation, models and sets

The notion of truth introduced in the previous subsection is based on the work of the
logician Alfred Tarski who intended that the correspondence theory of truth would be
used to provide a semantics for formal or logical languages quite unlike the languages
that are spoken by human beings. Hence, the application of the truth-conditional
method has often been referred to as formal semantics. Its formal nature also resides
in the fact that it utilises very precise notations and concepts to define as accurately
and as rigorously as possible how expressions in the object language are to be
interpreted. All aspects of the metalanguage in which the semantics is defined are
precisely determined. There is no room for ambiguity, vagueness or inaccuracy and
hence much of the specification of the metalanguage is mathematical or logical in
form. One of the advantages of constructing a formal metalanguage with a precise
interpretation is that it is thereby possible to check the consequences of one's theory
and prove, or disprove, that such-and-such a consequence results from the way the
semantics is set up. Within logic, the construction of formal semantic systems is
bound up with the idea of proving that certain forms of argument within some formal
language are valid whilst others are not. For example, within propositional logic the
pattern of inference in (24) is valid, whilst that in (25) is not and these patterns are
reflected in the English sentences that accompany the logical expressions below.
There is nothing in the syntax of the logical statements in (24) that determines that the
inference in (24.c) is valid and it is the purpose of the semantic theory to show that
whatever p and q stand for, the truth of the expression p —> q and of p always
guarantees the truth of q while the truth ofp—>q and q does not always guarantee the
truth of p.
(24) a. p —> q If Ethel is singing, the cat is unhappy.
b. p Ethel is singing.
c. q Therefore, the cat is unhappy.
(25) a. p —> q If Ethel is singing, the cat is unhappy.
b. q The cat is unhappy.
c. p Therefore, Ethel is singing.
It is interesting to note here that while formalisation within syntactic theory has
been accepted as a useful, if not a necessary, part of syntactic analysis, its adoption
within semantics remains a subject of considerable debate. However, the adoption of
a formal interpretation procedure for a particular object language makes it possible to
ascertain accurately whether or not one's theory is correct as an account of the
meaning of particular object languages, just as a formalisation of syntax into formal
rules or principles enables the syntactician to test whether they generate
ungrammatical expressions or fail to generate grammatical ones. It is furthermore
possible with such systems to prove that a semantics is consistent (i.e. contains no
internal contradictions) and complete (i.e. it derives the set of all valid arguments of
the language as theorems). Although these last two aspects of formal semantics will
not be discussed in this book, the fact that these things can be more easily be proved
with respect to semantic theories that are formalised than to those that are not will be
taken as positive, if implicit, arguments in favour of formalisation as a tool of the
semantic analysis of human languages.
17
1 Introduction
In the discussion above, much has been made of the fact that the propositions
expressed by declarative sentences may differ in truth value with respect to different
situations. In other words, their truth is not absolute but depends on when the sentence
expressing the proposition is uttered, on the location and other aspects of the context
of utterance (including the modality of the sentence itself)- This idea may be
formalised by constructing mathematical models of a state-of-affairs and defining
truth with respect to particular models. For example, a mathematical model can be
constructed of the situation depicted in (13) with respect to which the propositions
expressed by the sentences in (26) are all true. A model of the situation depicted in
(15) does not, however, support the truth of (26.a), (26.b) or (26.d) but does support
that of the other two sentences. We thus speak of truth with respect to a particular
model which represents the general context with respect to which a sentence is uttered.
(26) a. The book is on the table.

b. There is a book on the table.
c. The book is open.
d. A table is supporting a book.
e. There is no cup of coffee.
As will be seen in more detail from Chapter 3 onwards, model-theoretic
interpretation has two parts. In the first place, there is a model that represents
precisely what events, properties and relations make up the situation being modelled.
The model provides a description of the denotations of all the basic expressions
(content lexemes) in the object language. This reflects the fact that human beings
make statements in order to convey information about some state-of-affairs. Although
one can know the meaning of a sentence without knowing what specific situation is
being talked about (by knowing its truth-conditions), the sentence conveys no
information unless it is associated with particular individuals and the relations that
hold between them.
The second part of the theory provides the rules for interpreting expressions in
the object language with respect to any arbitrary model. It does this by providing a
recursive definition that specifies how the denotations of composite expressions are
constructed from those of their component parts. In other words, the model theory
provides a specification of the truth-conditions of the sentences in the object language.
The truth-conditions as specified by the model theory hold independently of particular
models, but the interpretation of particular sentences may be carried out only with
respect to some model or other. One may also think of model theory as constraining
the sorts of model that are admissible for the interpretation of some language. Not all
possible models are interpretable according to a particular model theory and there is
thus a two-way relation between the theory and the model. On the one hand, models
are constructed to represent a given set of phenomena and, on the other, the way that
phenomena can be represented is constrained by the theory of interpretation. From the
linguistic perspective, the theory of meaning thus influences the way states-of-affairs
can be represented and talked about.
It has been said that the models used to represent situations with respect to which
sentences are to be interpreted are mathematical. In fact, the models in this book are
based on set theory. The denotations of lexemes are thus modelled as sets containing
18
Beyond truth-conditions
various sorts of entity or as mathematical functions over these. The model theory uses
fundamental notions from set theory, like set membership, union and intersection, etc.,
to specify how the denotations of more composite expressions can be constructed from
these. Ultimately the notion of truth is characterised according to the axioms of set
theory, thus providing the rigorous and formal metalanguage for interpretation that
was discussed above. No previous knowledge of set theory is, however, assumed in
the discussion of later chapters and all relevant concepts are introduced as and when
they are needed.
13 Beyond truth-conditions
In the discussion of the last section, it was suggested that the central meaning of a
sentence is the set of conditions that guarantee its truth with respect to any occasion
of utterance. This does not mean that truth-conditions exhaust the whole of sentence
meaning, let alone all aspects of linguistic meaning. A complete explication of the
latter requires considerably more than just truth-conditions. However, the hypothesis
that truth-conditional meaning is central predicts that other aspects of meaning
interact with this to derive the meaning of an utterance. In other words, we may think
of linguistic meaning being determined by the interaction of a number of theories
dealing with particular domains of meaning. This section takes a brief look at some
of the other theories of meaning that are needed to interact with truth-conditional
interpretation to provide a more adequate account of meaning than is possible using
truth-conditions on their own.
1.3.1 Knowledge, behaviour and use

In the discussion of truth-conditions, we have referred only to the interpretation of
declarative sentences uttered as statements. But, of course, declarative sentences may
be used to ask questions and issue commands, as well as make statements, and there
are other sentence types that are typically used in the latter ways. For example, the
sentence in (27 .a) is a declarative that is typically used to make a statement. The
addition of a question mark in the orthography as in (27.b), orrisingintonation at the
end of the sentence in spoken English, turns the statement into a question that has the
same meaning as the interrogative in (27.c). Finally, the imperative sentence in (27.d)
is typically used in issuing a command.
(27) a. Jo kicked the cat again.
b. Jo kicked the cat again?
c. Did Jo kick the cat again?
d. Kick the cat again, Jo!
Although the sentences in (27) sentences are all typically used to do different
things (make statements, ask questions and issue commands), they contain a shared
component of meaning: they all make reference to a situation in which someone called
Jo engages in the activity of kicking a (contextually unique) cat and that this is not the
first time he or she has done so. These are, of course, the truth-conditions associated
with the declarative sentence in (27.a), and the truth-conditions of a declarative
sentence may be equated with its propositional content, which is distinct from the
19
1 Introduction
proposition expressed by a sentence and from the truth or falsity of that proposition
on some occasion of utterance. It is propositional content that may be considered to
be the common core of meaning that syntactically related sentences share, whatever
their surface form. Thus, each of the sentences in (27) share the same propositional
content but in their different uses bear a different relation to that content. (27 .a) states
that the situation described by the propositional content actually occurred. On the
other hand, (27.b) and (27.c) ask whether the situation described by the propositional
content has occurred or not, i.e. whether the proposition expressed by (27.a) is true
or false with respect to some context of utterance, and (27.d) is typically uttered to
bring about a situation in which the proposition expressed by (27.a) is true.
The sentences in (27) thus possess the same propositional content but differ in
their illocutionary force. The way content relates to use forms the study of speech act
theory which concentrates, not so much on the abstract meaning of a sentence, but on
what people are doing when they utter it. It, therefore, takes into account
extralinguistic information like speaker intentions (what perlocutionary effect a
speaker intends to have by the utterance of an expression), socio-cultural knowledge
(whether the utterance of some expression is felicitous or not) and so on. Such things
do not form part of truth-conditional meaning, but can form the basis of the relation
between abstract sentence meaning and the meaning that an utterance of that sentence
has in particular situations.
Another aspect of meaning that does not come within the realm of
truth-conditional semantics has its basis in discourse behaviour. This concerns the
sorts of inferences that are drawn by participants in a discourse. In Section 1.2, we
looked at the sorts of meaning relations between sentences that are within the possible
domain of truth-conditional semantics. These relations allow inferences to be drawn
from utterances because the truth of a paraphrase or an entailment, or the falsity of a
contradiction, are guaranteed by the truth of the proposition expressed by the original
sentence. The assumption that participants in a discourse are being co-operative and,
in general, telling the truth is one of the conversational maxims put forward by the
philosopher Paul Grice in the nineteen-sixties. This assumption allows inferences to
be made from entailments even if the participants have no way of checking the truth
or falsity of any assertion.
As mentioned in Section 1.1.2, inferences can be drawn from discourse that do
not follow directly from the truth-conditions of a sentence. We have already seen an
example in (7) above which is repeated below.
(7) a. Bertie: Is Fiona a good lecturer?
b. Ethel: She has a good line in sweaters.
c. Bertie (thinks): Fiona is not a good lecturer.
The inference in (7.c) from the sentences in (7.a) and (7.b) is not guaranteed by the
truth-conditional meaning of the latter, but by the apparent flouting of a conversational
maxim. Ethel, the speaker of (7.b), replies to Bertie's question in (7.a) in an apparently
irrelevant manner. In order for Bertie to maintain the assumption that the former is
sticking to the co-operative principle in (28), he must look for some relevant piece
of information that Ethel has not given that answers the original question. Since Ethel
has been obviously irrelevant and not answered the question in (7.a) directly, thus
flouting the maxim of relation in (29.c), the most consistent inference to make is that
20
Beyond truth-conditions
she is intending to convey that Fiona is not a good lecturer, but does not wish to say
so directly.
(28) The co-operative principle: make your contribution such as is required, at
the stage at which it is required, by the accepted purpose or direction of the
talk exchange in which you are engaged.
Grice suggested a number of conversational maxims, which are given in (29).
The interaction of these maxims with the truth-conditional import of the full sentences
that underlie the expressions uttered in any discourse gives rise to conversational
implicatures that allow a speaker to convey information over and above what is
actually said (i.e. in terms of truth-conditions) in any exchange. Implicatures of the
sort illustrated by (7) are highly context-specific and often rely on assumptions being
made by the hearer that are cmpletely dependent on the context. For example, a
similar exchange to that in (7.a) and (7.b) where the context contains information to
the effect that there is a good lecturer who has a good line in sweaters will lead Bertie
to infer the opposite of (7.c), i.e. that Fiona is a good lecturer. Since entailments result
from truth-conditional meaning, which is constant from context to context,
implicatures must form part of a theory of meaning that is non-truth-conditional.
Grice's theory of conversation has been very successful in accounting for some of the
non-truth-conditional meaning of certain expressions, giving further support to the
idea that truth-conditions form a central core of meaning that interacts with other
modular theories to account for a greater range of phenomena.
(29) a. Maxim of quantity:
i. make your contribution as informative as is required (for the current
purposes of the exchange),
ii. do not make your contribution more informative than is required.
b. Maxim of quality:
i. do not say what you believe to be false.
ii. do not say that for which you lack adequate evidence.
c. Maxim of relation:
make your contribution relevant.
d. Maxim of manner:
i. avoid obscurity of expression,
ii. avoid ambiguity,
iii. be brief.
Grice's work has not gone unchallenged or unrevised in the literature and, in
recent years, an attempt has been made to reduce the maxims in (29) to a single
explanatory principle that guides all inference making. Relevance theory reduces all
the maxims to one, essentially the maxim of relation, and defines relevance in terms
of cognitive effect (the amount and importance of information contained in an
inferred proposition for a particular participant in a particular situation) balanced
against processing effort (the number of steps that need to be taken to get to the
inferred proposition). A definition of these ideas is given in (30).
21
1 Introduction
(30) Theory of relevance:

a. Other things being equal, the greater the cognitive effect achieved by the
processing of a given piece of information, the greater its relevance for the
individual who processes it.
b. Other things being equal, the greater the effort involved in the processing
of a given piece of information, the smaller its relevance for the individual
who processes it.
Wilson and Sperber (1988:140)
Relevance theory has currently developed into a theory of cognition, rather than a
theory of linguistic meaning specifically, and, although it has some advantages over
the Gricean approach, it is less well known and there is, as yet, no basic introduction
to the subject. For these reasons, it will not be used in Chapter 7 where inference is
discussed in more detail.
1.3.2 Context
One of the most important influences on meaning is that of the context of utterance.
The context plays a vital role in determining how a particular utterance is to be
interpreted on any occasion. In particular, it is needed to restore ellipses, resolve
ambiguity, provide referents for deictic elements and resolve anaphoric dependencies.
The dialogue in (31) illustrates how important context is in enabling
interpretation to be carried out.
(31) a. Fiona: Hello?
b. Jo: In here!
c. Fiona: Any chance of a coffee?
d. Jo: There's no milk in the fridge.
e. Fiona: Oh hell! I bought some yesterday. You must have used it all in that
disgusting punch you made.
As (31) shows, sentences in discourse (or written texts) may not be complete
realisations of full sentences and so not sufficient for the purposes of truth-conditional
interpretation. To interpret (31.b) and (31.c), it is necessary to recover the system
sentences that underlie them: to I'm in here and Is there any chance of a coffee?,
respectively. The resolution of such ellipsis is dependent on the context, since an
utterance of the expression in (31 .b) could be interpreted as asserting that any number
of entities are in here, depending on the circumstances.
Just restoring ellipses is not sufficient to enable the interpretation of (31) to take
place, however. The context must also provide referents for all the deictic
expressions. The term deixis is derived from the Classical Greek verb deiknumi
meaning 'to point out' or 'show' (the Latin translation being demonstrativus, hence
the term demonstrative pronoun). The purpose of deictic elements is to link aspects
of the meaning of the sentence containing them to the context in which it is uttered,
either spatially, temporally or in terms of the participants in the discourse. Deixis may
be realised as words, particles or morphological processes. For example, in (31) the
indexical pronoun / in (31.b) (restored from context) refers to Jo, but in (31.e) it
refers to Fiona, i.e. the respective speakers of the expressions. Similarly, the pronoun
22
A note on method
you in (31 .e) refers to the hearer in the discourse who happens to be Jo in this instance,
but could be someone else in another context. Furthermore, context is required to
indicate the location indicated by the term here in (31 .b) (i.e. near the speaker) and the
time when Fiona bought the milk in (31.e) (the day before the day on which the
utterance of the sentence is made). Grammatical tense is also deictic and serves to
refer to times relative to the context of utterance at which the situations described by
any declarative sentences are deemed to take place. Thus, in (31 .b), (31 .c) and (31 .d),
the relevant time is the present (the time of utterance) whilst in both sentences in
(31.e) (ignoring the interjection) the relevant time is prior to the time of utterance.
Other aspects of the dialogue in (31) also require the assistance of the context to
enable interpretation to take place. For example, the pronoun some in (31 .e) is to be
interpreted as some milk, picking up on the mention of milk in the previous utterance.
The anaphoric pronoun it in the final sentence refers to the milk that Fiona bought
yesterday, information that is not explicitly stated in the utterance. In addition to
anaphoric resolution of this sort, the context resolves the ambiguity of the homonym
PUNCH which is interpreted as the drink and not the action because of the fact that
the subject of this latter part of the conversation is milk, something that is drunk.
This short example indicates the extent to which interpretation relies on the
context of utterance. The relation between context and interpretation is, however,
extremely complex. Although tiêre are certain aspects that have been incorporated
into truth-conditional semantics (as, for example, tense, cf. Chapter 8, and certain sorts
of anaphora), exactly how the context resolves ambiguities, anaphoric dependencies
and ellipsis and the way participants in a discourse tie up referential expressions to
particular individuals remains obscure. In what follows, it is recognised that the
context is almost always required to fix the reference of expressions (and
disambiguate ambiguous utterances), but we will proceed on the assumption that a
theory of truth-conditional interpretation accounts for the meaning of sentences
considered as abstract units of the language system. The definition of the meaning of
a sentence is given in the discussion above in terms of truth-conditions and
truth-conditions are associated with propositional content. Since the latter is
independent of context, then truth-conditional meaning may also be treated
independently of context, even though, as we have seen, the actual interpretation of
a specific utterance requires information provided by the context. Whether or not a
theory of context can be constructed that can interact with truth-conditional meaning
to provide a fuller account of utterance interpretation than is yet available is a topic
for future research.
1.4 A note on method

The scene has now been set for the development of the theory that is to be the subject
of the rest of this book. A sketch has been given of those aspects of meaning that form
the central domain of inquiry, of some of those that do not, and of the main tenets of
the theory to be adopted below. Before starting on the long road to Chapter 10,
however, something must be said about the methodology adopted in the following
chapters.
In order to avoid problems with context, like those discussed in the last section,
all interpretation is carried out, not directly on the sentences of a language or even
23
1 Introduction
on their structural descriptions, but on a logical language into which the object
language is translated. The logical translation of a declarative sentence is intended
to represent the proposition expressed by the utterance of that sentence on some
particular occasion. As it fully represents all the information supplied by the context
of utterance, it represents only one reading of an ambiguous sentence at a time, the
one intended on a particular occasion. For this reason, the translation language must
be a disambiguated language, i.e. permit no ambiguity. Furthermore, translation
resolves all anaphoric, deictic and referential uncertainties of the utterance of the
sentence, thus making sure that each expression has a determinate denotation.
Essentially, the translation of a sentence can be thought of as the representation of the
proposition derived from the sentence plus its context of utterance. It acts, therefore,
as an intermediate step between the syntactic analysis of an expression and its
truth-conditional semantic interpretation. Whether this level of representation is
necessary, or not, is a controversial issue. Richard Montague assumed that it was
merely convenient rather than necessary, but later linguistic research has indicated
that, at least as far as anaphoric and deictic reference is concerned, a representation
intermediate between syntax and interpretation is required. It is assumed henceforth
that translation is an intrinsic part of any semantic analysis. Interpretation is carried
out on the logical translation of a sentence, and so provides an indirect interpretation
of the latter.
It is usual within formal semantics to provide not only an explicit semantic
theory for an object language, but also an explicit syntax. Only by doing this can it be
guaranteed that the principle of compositionality is being maintained. In studying the
formal semantics of human languages, it would be a difficult task to specify a
complete (and formally rigorous) definition of the syntax of a particular language
before tackling the semantics. Hence, it is customary to define grammar fragments
that determine only a subset of the set of grammatical sentences of the object language
with the semantic interpretation specified for these sentences (via their translation into
a logical language). Once this interpretation has been successfully defined, the
coverage of the fragment is extended to take in more constructions. In this way, it is
hoped that ultimately a complete formal grammar of a language, containing both an
explicit syntax and semantics, can be obtained (this hope is, however, still a very long
way from realisation). This method will be pursued in Chapters 2 to 6 of this book,
beginning with a formal analysis of a fragment of English including proper names and
simple intransitive, transitive and ditransitive sentences. This is expanded into an
explicit account of co-ordination, passives and noun phrases. In the final chapters, the
syntactic aspect of the analysis is not so explicit, because of the debates about the
proper syntactic treatment of such things as tense, modality and finite and non-finite
complementation. Some discussion will, however, be given of some of the difficulties
presented by these constructions for the syntax.
It is commonplace nowadays to assume that interpretation depends on the
surface constituent structure of a sentence, although different theories adopt a more
or less abstract interpretation of this level of syntactic representation. It is not the
intention of this book to support or criticise any particular theory of syntax, however.
The semantic method outlined below can be adapted to many different syntactic
theories and each one has its own merits and leads to its own problems from the
semantic point of view. To maintain relative neutrality between theories, the syntactic
24
Further reading
framework that is adopted in this book is that of simple context-free phrase

structure grammar. This type of grammar has advantages and disadvantages, but has
the merit of being well known and easily describable. Each grammar fragment in the
following pages, therefore, takes the form of a set of context-free phrase structure
rules together with an appropriate lexicon. The translation of each sentence generated
by the grammar into a disambiguated logical language is carried out in parallel with
its derivation according to the specified phrase structure rules, in accordance with the
rule-to-rule hypothesis in (5) above.
Finally, before we move on to the real work of the book, it should be noted that
until time is introduced into the model in Chapter 8, every example sentence is given
in its simple past tense form. It is usual within most introductions to logic in
linguistics to put sentences into the present tense. Unfortunately, the present tense in
English is rarely used to report events neutrally, but has aspectual and futurate
interpretations that can sometimes obscure the intended sense of an example. No such
problems are associated with the simple past, however, and this is used throughout
Chapters 2 to 6. The proper interpretation of this tense is not, however, discussed until
Chapter 8.
1.5 Further reading
Each of the chapters in this book ends with a section of further reading. These give
references to other introductory texts that cover all or some of the topics discussed in
each chapter and also to one or two more advanced books or articles on specific topics
that have been mentioned in the text. Not all the topics that are discussed in the text
are given further reading and the references that are given are very selective, only
those works that the current author thinks are accessible and/or of reasonable
importance being given. Although these references, and the accompanying
bibliography, are in no way comprehensive, it is hoped that readers will use them to
get more details of specific issues and to treat them as suitable starting points for more
protracted research.
There are a number of basic introductions to general linguistic semantics that
make good background reading for this book: see especially Lyons (1981) and
Hurford and Heasley (1983), but Palmer (1981) and Leech (1974) may also be
consulted, although the latter two books are somewhat idiosyncratic. Other
introductory textbooks to general linguistic semantics and logic for linguists are
Chierchia and McConnell-Ginet (1990), Kempson (1977), All wood, Andersen and
Dahl (1977) and McCawley (1981) which will be referenced below for specific topics.
The criteria of adequacy for semantic theory in (1) above are introduced and
discussed in Kempson (1977: ch. 1), Chierchia and McConnell-Ginet (1990: ch. 1)
discusses the domain of semantics, and the distinction between semantics and
pragmatics is discussed in Levinson (1983: 1-35). Compositionality is discussed in
Partee (1984) and Partee, ter Meulen and Wall (1990: 317-338) (although both
presuppose some familiarity with formal semantic concepts). The logical meaning
relations of entailment, paraphrase and contradiction are discussed mainly in
textbooks on logic: see Guttenplan (1986: 1-41) for a good and simple introduction
to the notion of logical inference (other references are given in Chapter 7). Levinson
25
1 Introduction
(1983: ch. 3) provides an introduction to implicature and Ch. 4 of that book discusses
presupposition, as does McCawley (1981: ch. 9). A more recent discussion of the
phenomenon with extensive references is Burton-Roberts (1989) and, for a somewhat
different view, see also Chierchia and McConnell-Ginet (1990: 286-317). The
distinction between homonymy and polysemy is discussed at length in Lyons (1977:
550-573) and ambiguity (and vagueness, a topic not touched on here) is discussed in
Kempson (1977: ch. 8), McCawley (1981: 5-11) and Lyons (1977: 396-409). On
denotation, reference and sense, see Chierchia and McConnell-Ginet (1990: 46-60;
77-88) and Lyons (1977: 174-215). For a technical discussion of the difference
between extension and intension, see Carnap (1956: 16-32) and McCawley (1981:
401-406). On propositions, see Lyons (1981:106-109, passim), Hurford and Heasley
(1983: 15-24) and Allwood, Andersen and Dahl (1977: ch. 3). For an outline of the
sort of truth-conditional semantics pursued in this book, see Dowty, Wall and Peters
(1981: ch. 1) (which gives an introduction to Montague semantics, but at a much more
difficult level than that given in this book), Partee, ter Meulen and Wall (1990: ch. 5;
200-203) and for a general discussion of mathematical versus psychological
approaches to meaning, see Partee (1979). On truth and truth-conditions, consult
Lyons (1977: 167-173), Kempson (1977: ch. 3), Chierchia and McConnell-Ginet
(1990: 61-65), Lycan (1984: 13-43), and, for a philosophical view, essays 1 to 5 in
Davidson (1984). Speech act theory is discussed in Levinson (1983: ch. 5) and for
context, see Lyons (1981: ch. 9) and Levinson (1983: passim). Kempson (1977: chs.
4 & 5) also discusses language use and pragmatics, and relevance theory is discussed
in Sperber and Wilson (1986), Wilson and Sperber (1988) and Blakemore (1987).
26
2 Predicates and arguments
2.1 Translating English into a logical language

In this chapter and the next, we will lay the foundations on which a good deal of
logical semantics is built. In accordance with the discussion in Chapter 1, we first
define a logical language into which sentences of English are translated in order to
circumvent the problems of ambiguity and underdeterminacy found in the object
language. Having defined the translation language, and specified the procedure for
translating simple English sentences into it, our attention will turn to the interpretation
of these logical expressions in terms of their truth-conditions, thus providing an
indirect interpretation of the corresponding English sentences.
2.7.7 The syntax of LP

Like all languages, natural or artificial, logical languages have a syntax, i.e. a set of
rules for constructing composite expressions from simpler ones. The logical language
described in this chapter, called LP, contains expressions that fall into one of four
logical categories: individuals, predicates, formulae and operators (or connectives).
Expressions in each of the first three categories can be further subdivided into two
sorts: constants, which have a fixed interpretation, and variables, which do not.
These two sorts of expression correspond, roughly, to content words (e.g. table, run,
Ethel) and pronominal expressions (e.g. she, they) in natural languages, respectively.
This chapter deals only with constants, but variables will become increasingly
important in later chapters.
Sentences in natural languages translate into formulae in LP which have the
logical category t (as sentences have the syntactic category S). Such expressions are
constructed by combining individual and predicate expressions in much the same way
that sentences are composed of combinations of noun phrases and verbs. Just as
certain noun phrases in a sentence can be described as the complements of the main
verb, so individual expressions, of logical category e, perform the role of arguments
of the predicates in a formula. Predicates in LP may take zero or more individuals
as arguments and the number of arguments that a predicate takes is referred to as the
valency of the predicate. Predicate constants and variables are subcategorised
according to their valency, just as verbs are subcategorised according to whether
they are intransitive, transitive or ditransitive, etc. All predicates are classed as
members of the logical category Pred and their subcategorisation according to the
number of individual arguments they take is shown by the use of a subscripted
number to the right of the category symbol. Thus, for example, the category Predi
contains all those predicates that take just one individual as an argument, the category
Pred2 contains all those predicates that take two individual arguments, and so on.
Unlike verbs in natural languages, however, there is no upper limit on the number
of arguments a predicate in LP can take; thus the language may contain an infinite
number of subcategories of predicates, Pred«, where n is any number between zero
27
and infinity. For the purposes of translating from English into LP, however, we can
safely ignore all predicates that have a valency higher than 3. Thus, only four of the
possible subcategories of Pred are used below: Pred0, Predi, Pred2 and Pred3.
The basic syntactic rule for constructing formulae in L P is one that combines
a predicate expression with the number of individual arguments it requires. Thus, an
expression in Pred! combines with one individual expression to make a formula, an
expression in Pred2 combines with two individual expressions to make a formula and,
in general, an expression in Pred,, combines with n individual expressions to make a
formula, where n is some positive whole number. The predicate is written to the left
of the arguments which are enclosed within round brackets and separated from each
other by commas. The appropriate rule appears in (1) where the superscripted
number appearing with the individual category symbol, e, does not represent a
subcategory of individual, but simply the number of the individual in the argument
structure of the predicate. This ensures that predicates appear with the right
number of arguments for their valency. The symbol < in (1) is read as is less than, or
equal to, so that in the rule n may be zero or any number greater than zero. Pred0 is,
therefore, a valid subcategory of Lp, but one that takes no arguments to make a
formula. Hence, Predo is equivalent to a formula, i.e. Pred0 = t. We shall meet some
English verbs that may be associated with this category in the next section.
(1) t -> PrednCe1^2,...,^), 0 < n.

Given the syntactic rule in (1) and some (abstract) expressions in LP (a, b, c, d being
individual constants, P a one-place predicate, Q a two-place predicate and R a three-
place predicate), then all the expressions in (2) are well-formed formulae (often
abbreviated as wff) in Lp, but those in (3) are not.
(2) a. P(a). b. Q(c,d). c. R(b,a,d).
d. R(c,b,c). e. P(b). f. Q(b,b).
(3) a. Q(c). b. R(b,a,b,b). c. P(a,a)
d. a(P). e. Rabc. f. P(e).
2.1.2 A grammar fragment for English

Having defined the (very simple) syntax of Lp, we are in a position to translate
sentences of English into it. To show how translation works, we shall provide a
grammar for a small fragment of English and then develop a translation procedure
that associates every sentence generated by the grammar with at least one logical
expression. As discussed at the end of Chapter 1, grammars for parts of a language are
called grammar fragments and are often used in formal semantics to develop specific
and precise theories about the meanings of the expressions they analyse. The grammar
developed in this chapter and in Chapter 3 covers only a very small part of English
and a syntactically very unexciting part at that. Although we begin by analysing a
rather small and uninteresting set of sentences, by the time we get to the end of the
book, we will have covered a quite considerable number of English constructions and
have armed ourselves with the machinery to tackle many more. From the apparent
triviality of the initial examples, a powerful semantic theory of a range of
constructions in English will be developed.
28
Translating English into a logical language
The type of syntactic grammar adopted in this book is a simple context-free

phrase structure grammar. There are many other forms of grammar, of course, that
may well be syntactically more adequate than this, but context-free phrase structure
grammars have the advantage of being easy to understand and simple to use. For
example, it is always possible using a context-free phrase structure grammar to
decide in a finite number of steps whether some string of words is generated by the
grammar and it is therefore quite easy to check whether it generates ill-formed
sentences of the object language. On the other hand, the well-known disadvantages
of context-free phrase structure grammars are not particularly relevant for our
purposes, since we are interested in semantics rather than syntax and the points made
should be relevant to almost any syntactic theory. The adoption of phrase structure
grammar in this book is for expository purposes only and the syntactic analyses
presented are not intended to be definitive. Readers are invited to try to apply the
semantic methods used in later chapters to the analyses of the same constructions
determined by their favourite syntactic theory.
The sentences that are generated by the grammar in this chapter are of the
simplest sort. They contain between one and three noun phrases, consisting of either
a single proper noun or a common noun plus the definite article, the, together with
a single verb, subcategorised according to the number of noun phrases it can appear
with. (4) gives examples of five sentence types which contain intransitive (4.a),
transitive (4.b) and ditransitive (4.c & d) verbs. The final sentence, (4.e), contains an
impersonal verb which appears with the meaningless subject pronoun, it.
(4) a. Chester ran.
b. Chester ate the cake.
c. Ethel gave the cake to Jo.
d. Ethel gave Chester the cake.
e. It rained.
The sentences in (4) (and many, but finitely many, more besides) can all be
generated by the grammar in (5), which is called Gu in conjunction with the simple
lexicon in (6). Common abbreviations for categories are adopted. Thus, S stands for
sentence, NP for noun phrase, Npr for proper noun, N for common noun, Vj for
intransitive verb, V, for transitive verb and Vdt for ditransitive verb. Impersonal verbs
are assigned to the category Vo. A syntactic feature, symbolised as [+FIN], indicates
that a verb (of any subcategory) is finite (and, until tense is introduced in Chapter 8,
in the past tense, as mentioned in Chapter 1). In Chapter 3, non-finite verbs, i.e. verbs
in their bare stem or citation forms, are introduced using the feature [-FIN]. In rules
3 d , 5 d and 7 d , below, the items to, it and the are introduced directly in the
syntactic rule and do not appear in the lexicon. These, and some other expressions, are
treated in this way, either because they do not contribute significantly to the semantics
of a phrase and so have no translation in Lp, as with to and if, or because their
contribution to the semantics of an expression is stated directly in the translation of
the rule itself (as with didn't below). The definite article the is treated in the same way
for expository reasons only. It is introduced here in order to increase the interest of the
grammar fragment which would otherwise only contain proper names as exemplars
of NPs. The semantics of the will be dealt with more fully in Chapter 6, while the
means to rectify the other departures from the principle of compositionality will also
29
be provided in later chapters when other theoretical machinery has been introduced.
Another non-standard feature of the rules in (5) is the fact that the noun phrases
are given subscripted numbers. These numbers are strictly unnecessary from the
syntactic point of view, but essential for translation and interpretation purposes.
They may be thought of as marking the grammatical function of the noun phrase
within the sentence: so that NP, functions as the syntactic subject, NP2 as the
syntactic direct object and NP3 as the syntactic indirect object. For the sake of
variation, three, not very sophisticated, rules giving the passive counterparts of the
transitive and ditransitive rules 2G,, 3G, and 4G, are given in (7).
(5) a. 1G, S -» NP,
b. 2G, S -> NP, Vt[+FIN] NP2.
c. 3G, S -» NP, Vdt[+FIN] NP2 to NP3.
d. 4G, S -» NP, Vdt[+FIN] NP2 NP3.
e. 5G, S-MtV 0 [+FIN].
f. 6G, NP->N p r .
g. 7G NP->theN.
(6) a. N pr -> {Prudence, Ethel, Chester, Jo, Bertie, Fiona}.
b. N —» {book, cake, cat, golfer, dog, lecturer, student, singer).
c. Vj[+FIN] —> {ran, laughed, sang, howled, screamed).
d. Vf+FIN] -> {read, poisoned, ate, liked, loathed, kicked).
e. Vdt[+FIN]-> {gave}.
f. V0[+FIN] -» {rained, snowed).
(7) a. 8G1: S -> NPI was Vt[PAS] by NP2.
b. 9G,: S -> NP, was Vdt[PAS] to NP2 by NP3.
c. 10G,: S -> NP, was Vdt[PAS] NP2 by NP3.
d. VJPAS] —» {read, poisoned, eaten, liked, loathed, kicked).
e. Vdt[PAS] -> {given}.
A grammar like G, is said to generate a sentence, £ , if (and only if) there is a
derivation of £ from the symbol S using the syntactic rules and the lexicon. The
derivation of a sentence is the procedure whereby, starting with S, one category in
a string is expanded at a time using a rule of the grammar until the string contains
only words in the lexicon. The derivation of the sentence Ethel poisoned the cat
using the grammar, G,, is shown in (8) where each line in the derivation has a
number on the right hand side showing the rule used to derive the new string of
symbols from the line before. The symbol Lex indicates that lexical substitution has
taken place and this is represented in (8.e) as having taken place all at once.
Equivalently, the generation of a sentence can be shown by a phrase structure
tree as in (9).
(8) a. S.
b. NP, V,[+FIN] NP2. 2G,
c. NprV,[+FIN]NP2. 6G,
d. Npr Vf+FIN] the N. 7G,
e. Ethel poisoned the cat. Lex
30
(9)
Ethel poisoned cat
The notion of derivation and the use of phrase structure trees are important,
because, as we will see in the next section, translation from English into LPis not done
on the strings of words that make up sentences, but on the syntactic structures of
those sentences. Before going on to translation, however, let us first add another two
rules to the grammar, both of which introduce the third person singular past tense form
of the copular verb be. This verb has a number of semantic functions, one of which
is to equate two entities, to indicate that they are identical. The rule in (10) allows the
generation by Gx of sentences containing this use of the copula (often called equative
sentences) like those in (11).
(10) wasNP2.
(11) a. Fiona was the singer.
b. Chester was the dog.
c. Jo was Jo.
Another copular construction in English involves predicative adjectives. Here,
the copula does not seem to have any real semantic function (except to carry
information about tense which we are ignoring for the moment), but serves instead
to provide a means of associating an adjective with the subject NP, as in (12).
(12) a. Chester was crazy.
b. The singer was messy.
Syntactically we can analyse the sentences in (12) as consisting of a noun phrase
followed by the verb was followed by an adjective. This is provided by the rule in
(13.a) with the additions to the lexicon in (13.b).
(13) a. 12G,: S -> NP was A.
b. A —> {happy, crazy, messy, disgusting, wealthy}.
31
2.1.3 The translation procedure

Having defined this small grammar, we can proceed to translate the sentences it
generates into logical expressions in LP. First of all, it is necessary to associate the
syntactic categories in G! with their corresponding logical categories in LP. As already
mentioned, all proper nouns and definite noun phrases are associated with individual
constants, i.e. they translate into LP as expressions of category e. For the moment,
common nouns are assigned to no logical category as they have no direct
interpretation in this fragment. This situation will be rectified in Chapter 6 when the
interpretation of noun phrases is tackled in more detail.
Verbs, which combine with noun phrases to form sentences, are naturally
associated with the general semantic category of predicate. As verbs may take
different numbers of noun phrases to make grammatical sentences, so they translate
into different subcategories of predicate according to the number of noun phrases,
including subjects, that appear with them in sentences of English. Hence, intransitive
verbs, which take no objects but do have a subject, translate into one-place predicates.
Transitive verbs, on the other hand, which subcategorise for a direct object and appear
with a subject, thus translate into two-place predicates. Ditransitive verbs have three
dependent noun phrases including their subjects and so translate into three-place
predicates in LP. Although impersonal verbs do appear with a syntactic subject in
English, the dummy subject pronoun, it, has no semantic significance and has no
denotation. Hence, it may be ignored in the logical translation and impersonal verbs
translate into zero place predicates.
Predicative adjectives semantically behave in a similar fashion to verbs. For
example, they may be used to ascribe properties to entities, as in the sentence Chester
was disgusting where the adjective ascribes the property of being disgusting to the
entity denoted by the word Chester. Such adjectives translate them into 1-place
predicates. (We ignore two-place predicative adjectives like, afraid of Jo, here.)
Finally, the logical category assigned to S is t, the category of formulae. The
correspondences between categories of d and those of Lp are summarised in (14),
where the symbol => means translates into.
(14) a. NP=>e.
b. Npr=> e.
c. V o ^ Predo.
d. v l= > Pred,.
e.
f.
v,=> Pred2.
Pred3.
g. A=>]
h. S => t.
The category assignments given in (14) determine the way the translation of
expressions containing members of the different syntactic categories are carried
out. According to the LP construction rule in (1), above (repeated below), an n-place
predicate combines with n individual expressions to make a formula.
(1) t^Pred n (e 1 ,e 2 ,...,e n ).
32
Hence, a predicate that translates an intransitive verb combines with the individual
expression that translates its subject, while a predicate translating a transitive verb
combines with the translations of the subject and object noun phrases to make a
formula, and so on. This is not sufficient, however, to get the right translations for any
English expression generated by grammar d , because the order in which arguments
appear when combined with a predicate has yet to be specified. For example, the
transitive verb construction analysed in d by rule 2 d (repeated in (15.a)) translates
into an expression containing a two place predicate translating the verb and the two
individuals which translate the subject and object noun phrases. There are two
possible well-formed formulae that meet this description depending on the relative
orders of the two individual expressions. These alternatives are shown in (15.b) and
(15.c) where the superscript prime, \ following the syntactic category symbol stands
for the LP expression that translates an English expression analysed by that symbol
in some actual derivation. A symbol like V,', therefore, stands in place of the
translation of the actual transitive verb that is introduced under the node Vt in some
particular sentence. In translation rules, V,' thus operates as a variable over the
translations of all transitive verbs in the grammar. Just as the usual category symbols
allow the statement of regular rules of combination without the need to specify their
internal structure, so this notation allows the generalisation of translation rules
regardless of the particular expressions that actually appear in any given sentence.
(15) a. 2G,: S -> NP, Vt[+FIN] NP,.
b. Vt'(NPr,NP2').
c. Vtf(NP2\NP,').
In Lp, as in English, the order in which arguments appear is semantically
significant, so that, as we will see in Section 2.2.3, (15.b) is interpreted differently
from (15.c). Since ordinary transitive sentences like Ethel poisoned Chester are not
ambiguous in English, only one of the translations in (15) must be chosen as the
translation of rule 2 d . Thus, in addition to the logical category assignment in (14),
we also need rules to tell us how to combine predicates and their arguments to obtain
the translations of particular English constructions.
In Chapter 1, one of the criteria for adequacy for a semantic theory is that it
should adhere to the principle of compositionality. This requires the interpretation
of an expression to be a function of the meanings of its component parts and the way
they are put together. In other words, to interpret an expression we need to have
access to the syntactic rules by which it is constructed. Because we are interpreting
natural languages indirectly, via translation into LP, we must show compositionality
in the translation procedure and thus take syntactic information into account as we
build up Lp representations. This can be done by providing a translation rule for each
syntactic rule in the grammar, so that 2 d , for example, is paired with only one of the
possible translations in (15). As discussed in Chapter 1, this approach to
compositionality adheres to the rule-to-rule hypothesis because each rule in the syntax
is paired with a corresponding translation rule and, thus, indirectly with a rule of
semantic interpretation, as we will see below. In accordance with this hypothesis,
therefore, every rule in d is matched with a corresponding translation rule. These
appear in (16) with numbers that parallel those of the syntactic rules of Section 2.1.2.
Thus, T l d translates rule I d , T2d translates rule 2, and so on.
33
(16) a. TIG,: S => W(NP').

b. T2G,: S =» Vt'(NP,\NP2').
c. T3G,: S=*Vdt'(NPi\NP2\NP3f).
d. T4G,: S=>Vdt'(NP,>,NP3',NP2').
e. T5G,: S => V o \
f. T6G,: NP=>N pr \
g. T7G,: NP=»the-N\
h. T8G,: S => Vt'(NP2\NPD.
i. T9G,: S=>Vdt'(NP3',NPr,NP2').
j. T10G,: S => VdtJ(NP3',NP2',NP,').
This leaves the two copular rules, 1 I d and \2GU to be translated. In the latter
rule, which introduces predicate adjectives, the copula has no significant semantic
effect (other than to signal tense) and so is ignored in the translation, leaving the one-
place predicate translating the adjective to apply directly to the translation of the
subject, as shown in (17).
(17) T12G,: A'(NP').
In the equative construction, rule HGi, however, the copula does have semantic
effect, as it identifies the two entities named by the subject and complement noun
phrases and so must have a representation in LP. Since the word was is introduced
directly by the rule, it does not appear in the lexicon and so its semantic effect must
be captured directly in the translation of the rule. One of the reasons for treating the
equative copula in this way is that it has an interpretation defined in the logic itself,
as we will see later in this chapter, which differentiates it from normal lexical items,
which translate as arbitrary constants. The logical nature of the meaning of was in
this rule is directly shown in its translation. Instead of following the usual method for
translating words discussed below, it is represented by a new sort of expression in Lp,
a logical operator which appears between the LP expressions that are its arguments.
The logical operator associated with the equative copula is =, the equals sign,
requiring the new syntactic rule for Lp shown in (18.a). This yields the translation of
1 lGi shown in (18.b), completing the set of translation rules for Gi.
(18) a. t->(e = e).
b. TUG,: (NPr=NP 2 ').
Before these rules can be used to translate sentences of English generated by
d , however, some way must be specified of translating English words into
expressions of LP. Firstly, in order to bring out the fact that different inflectional
forms of the same 'word' do not differ in their fundamental meaning (ignoring tense
and number information), each word is stripped of any inflectional markers. In
translating into Lp, therefore, we are interested only in lexemes and not word forms
as such. Thus, the English word forms give, gives, given and gave are translated into
Lp as a single form because they are all forms of the same lexeme, the same meaning
unit. Because Lp is an artificial language, it is possible to translate lexemes into
arbitrary symbols which then function as basic words or constants in the language.
This is usually done, for example, in exercises in predicate logic where the relation
34
between the symbols of the logic and the lexemes of English (or any other language)
is not significant. Hence, the symbol P might translate a verb like run, runs, ran, Q
could stand for eat, eats, ate, and so on. Associating English words with completely
arbitrary symbols in this way, however, would mean that someone using LP would
have to learn vocabulary lists that give the translations of all English words, just like
learning a foreign natural language. This is possible, of course, but unnecessarily
tedious and serves no real purpose as there are no native speakers of LP to tell us what
the words of their language translate as. Thus, instead of translating lexemes of
English into arbitrary symbols that need to be learned for each lexeme, we can state
a simple rule that provides the LP symbol that translates any lexeme in the object
language. This rule translates any word in English as its citation form (which is taken
to be the base form of the expression without any inflectional affixes; cf. Huddleston
(1984:102)), with capital letters replaced by their lower-case counterparts and
followed by a prime,'. The latter indicates that the symbol is not a word in the object
language, e.g. English, but an expression in the translation language, i.e. LP. For
example, the English word forms gives, give, gave, given are all associated with the
citation form give and the translation of any of these word forms into LP is thus give \
More examples appear in (19). This way of translating content words into LP is, of
course, just one of many possible ways of representing the logical constants associated
with such expressions in a logical language. Other writers use other means of
representing translations: for example, Lyons (1977) uses the base form of a word in
double quotes, e.g. "give", while Hurford and Heasley (1983) puts the form into
capitals, e.g. GIVE, and other representation are possible. The one adopted here is the
one adopted by Montague and is the one very commonly used by formal semanticists.
It is very important in translating into LP to bear in mind that an expression
followed by a prime is not an expression of English but an expression of LP. They
are translation equivalents, but not the same word. It is essential for the reader to
keep in mind the fact that LP expressions are not English, even if they appear to be
similar.
(19) a. Ethel => ether.
b. cake => cake'.
c. gave =»give\
d. sang =>• sing'.
e. ate=> eat'.
f. eaten => eat'.
g. messy => messy'.
Because LPis an unambiguous language, it is necessary to be able to distinguish
the different meanings of homonyms. Homonyms are different lexemes which are
associated with the same word forms but which have different meanings. For example,
the English noun punch may describe a kind of drink or a physical action. In Lp, these
meanings must be kept separate, so that simple translation into a citation form plus
prime is not quite sufficient to provide the procedure for translating words in English
into constants of LP. To resolve this, we can resort to the simple expedient of
associating each constant in Lp with a superscripted number, so that gives translates
into give1' and not just give'. Where a word form is homonymous, it translates into
a number of different constants in Lp which have the same basic form but differ in
35
their superscript. For example, the two senses of the noun punch are represented as
punch1' (the drink) and punch2' (the action). In this way, the translation of a
homonymous word in any sentence results in a unique representation in Lp, ensuring,
as required, that the translation is unambiguous. The same approach can be taken for
proper names that may have more than one referent. If, for example, there were two
people called Jo, one man and one woman, they would be distinguished in LP as jo1'
and jo2' and so on. The selection of which Jo is meant or which meaning of a
homonym is intended is determined by context and no attempt is therefore made here
to provide a way of getting to the intended referent or meaning. All that is necessary
for the purpose of interpretation is that this disambiguation is done by the context and
that Lp representations invoke only one of a range of alternatives. In general, of
course, most word forms are not homonymous and so there is no need to distinguish
their different senses. In these cases, the superscript associated with the translation
is omitted in the exposition that follows for ease of representation, since there is no
possible ambiguity. (As discussed in Chapter 1, polysemy is not incorporated into the
current system and is thus ignored here and elsewhere.)
The full rule for translating words of English into LP is given in (20). This
provides the means for translating any word in the object language into an appropriate
constant in the logical language together with a means of showing all possible
alternative readings of the word.
(20) Lexical translation rule (TLex): If Wis a set of inflectionally related word
forms {w h w2y..., wm} of which wt is the citation form, then the translation
of any word, wh in W, is W/Zi \ where n is greater than or equal to 1.
We are now in a position to translate any sentence generated by the grammar,
d , into Lp. The translation procedure is entirely mechanical and is based on the
syntactic derivation of the object language sentence. The procedure provides each
line in the derivation of an English sentence with a translation of that line as
determined by the translation counterpart of the syntactic rule used to derive that
line. For example, if there is a line in a derivation NP Wh with translation Vj'(NP')
then the NP can be expanded as Npr using Rule 6 d . This rule is associated with the
translation rule in (16.f) which equates the translation of the noun phrase with that of
the proper noun it contains. The next line in the derivation is therefore Npr Vs with
translation Vj'(Npr'). Hence, just as one line in a syntactic derivation is derived from
the previous one by substituting the string on the right of a syntactic rule for an
instance of the symbol on its left, so the translation of the sentence is derived by
substituting the translation of the expanded symbol with the translation of the syntactic
rule that was used.
To illustrate this further, (21) shows the derivation and parallel translation of
the sentence Jo laughed according to G h The first line shows the start symbol S which
tells us that we are deriving a sentence. This symbol is then rewritten as the string in
(21.c) using rule 1G! which has the translation in (21.d) according to TIG i in (16.a).
The NP node in (21 .c) is then expanded using rule 6G,, giving the string in (21 .e) and
translation in (21.f), as discussed above. Now we can substitute lexical items for all
the symbols in (21 .e) to give (21 .g). At the same time we apply the lexical translation
rule in (20) to each word in the English sentence and substitute the result in the
appropriate place in the translation in (21.f), yielding (21.h) as the translation of the
36
English sentence. The numbers on the right of the examples in (21) (and below)
indicate the rules that license each step in the derivation. The translation procedure
is formally defined in (22), and, as mentioned above, it is entirely mechanical. If the
rules are applied correctly, any sentence generated by Gi can be paired with its
translation in LP. A more complex translation than that in (21) is shown in (23) to give
a better idea of the method.
(21) a. S.
b. S\
c. NPVi. 1G,
d. V,f(NPf). TIG,
e. NPTW, 6G,
f. V^Np/). T6G,
g- Jo laughed. Lex
h. laugh'(jo'). TLex
(22) Translation procedure: If i is a line in a syntactic derivation with a
translation i' and A is a category in i with translation A' in i \ then if the
next line in the derivation i+1 is obtained from i by expanding A as a string
X by Rule n or Lex, then the translation of i+1 is obtained by replacing
A' in i' by X' the translation of X obtained by applying Translation Rule
n or TLex.
(23) a. S
=>S\
b. NP, was Vt[PAS] by NP2 8G,
=> Vl>(NP2>,NP1>).
c. The N was VfPAS] by NP2 7G,
=»V t '(NP,\the-N').
d. The N was VJTAS] by Npr 6G,
=>Vt'(Npr',the-N').
e. The cat was poisoned by Ethel Lex
=> poison' (ether ,the-cat').
Exercise 2.1:
Show the derivations and translations of the following English sentences using the
format of (23):
i. Ethel poisoned the cat.
ii. The student gave the cake to the lecturer.
iii. The dog was crazy.
The procedure described in (22) and illustrated in (23) is called a top down
derivation and translation, but a sentence can also be translated bottom up. With this
method a sentence is translated by building a phrase structure tree, starting with
the translations of the lexical items and then combining them in the way specified by
different syntactic rules. This results in a translation tree that parallels the syntactic
one and the translation of the whole sentence appears at the top. This can be shown
37
(24)
poison' (ethel' ,the_cat')
The cat was poisoned by Ethel
by annotating each node in a phrase structure tree with its associated translation as
done in (24) which gives the bottom-up version of the sentence The cat was
poisoned by Ethel that was derived by the top-down method in (23). Trees like this
are referred to as analysis trees in this book. Although the phrase structure tree
representation is often more perspicuous and reveals the parallelism of the translation
to the syntax very clearly, the derivation procedure illustrated in (23) is the safer
method and readers are recommended to stick to that way of translating until they are
confident of their translating skills. Once the technique has been mastered, however,
therigorousderivational method may be dispensed with, provided that it is clear that
the translation given for each phrase in the sentence is licensed by some rule. If there
is any doubt, the appropriate translation rule should be referred to.
Exercise 2.2:
Draw analysis trees like that in (24) for the sentences in Exercise 2.1.
2.2 Interpreting Lp
We have now devised a procedure for translating (a restricted set of) sentences in
English into a formal language, LP. As we have seen, formulae in Lp are unambiguous
and have all contextual uncertainties resolved in order to provide a representation of
the semantic content of the sentences in the object language which they translate.
Translation into a logical language in this way is a useful method of representing the
intended meaning of an English sentence on some occasion of utterance, but does not
in itself constitute a theory of semantics. What needs to be done to complete the
38
Interpreting LP
project is to interpret the formulae of Lp in terms of their truth-conditions, as

discussed in Chapter 1. In this way, the English sentences generated by d are
provided with an indirect interpretation.
There are two parts to the interpretation procedure. In the first place,
interpretation takes place, not in a vacuum, but with respect to a representation of
some state-of-affairs. Such a representation is called a model and models have two
parts: an ontology and a denotation assignment function or naming function. The
word ontology comes from the Greek word meaning to be or exist and refers to that
branch of philosophy (a subpart of metaphysics) that has to do with the study of what
exists in the world or some part of the world. The ontology of the model provides the
basis for interpretation in that it defines what exists and so what can be talked about.
If something is not recognised as existing in the model then it cannot be referred to
or discussed by any language that the model is being used to interpret. This means, of
course, that a model must contain abstract, as well as material, entities, otherwise we
could never talk about things like love or virtue, etc. The assumption that abstract
things exist just as concrete ones do may seem strange, but the differences between
them may be captured by structuring the ontology in different ways. For example, it
is possible to put different entities (the things that exist) into different sorts according
to their basic properties. A basic distinction can be made between all entities
depending on whether they are of an abstract or concrete sort. The semantic properties
of the words that are associated with these two sorts of entity may differ from each
other in predictable ways. Exactly what sorts of entity should be recognised as
making up adequate ontologies for natural language interpretation is a vexed and
controversial issue. Indeed, much current debate in formal semantics centres around
just this question. The problem will, however, not be addressed in this book. The
models that are defined in this and later chapters all contain only concrete, physically
manifested, entities. For present purposes, therefore, it is not necessary to worry about
any fundamental differences between the entities in a model.
The important thing that we are concerned with is the relation between the
entities in the model and expressions of the language being interpreted. Once the
things that exist in the state-of-affairs being modelled have been determined, it is
necessary to associate the basic expressions of the language being interpreted with the
entities that constitute the ontology. The basic expressions of the object language are
here taken to mean the expressions of the language belonging to the major parts of
speech (e.g. nouns, verbs, adjectives) in contrast to the grammatical (or functional
expressions (e.g. determiners, conjunctions, auxiliary verbs)). In LP, the basic
expressions are the individual and predicate constants associated with the nouns,
verbs and adjectives of the language being translated in contrast with the purely
logical expressions like operators and quantifiers (see later chapters). Logical
expressions take their interpretation, not from the model, but from the theory of
interpretation, the model theory itself, thus reflecting the fact that such expressions
have an invariant interpretation independent of the particular situation being modelled.
The denotations of constant expressions, however, may change from model to model.
For example, the name Prudence could be associated with a cat in one model, but a
human being in another, or two entities could be singing in one model, but not in
another, and so on. On the other hand, the equative copula was does not change its
function of identifying two entities from situation to situation. Thus, the interpretation
39
of a constant expression must be defined for each model, but those of grammatical
expressions are fixed by the theory of interpretation.
The association of constants with entities in the model is done by the denotation
assignment function. As discussed in Chapter 1, the denotation of an expression is the
relation between the expression and things that exist in the world. Since the world for
our purposes is given by the model of the world or situation, what is denoted by an
expression in Lp is something in the model. The denotation of an expression has two
aspects: an extension, the sorts of things in the world that an expression can be used
to refer to, and an intension, the concept that determines the extension of the
expression. A discussion of the latter aspect of denotation is left until the final four
chapters of the book and the semantics defined in Chapters 2 to 6 is purely
extensional. Hence, the denotation assignment functions of the models in these
chapters associate each constant in the logic with its extension. In the discussion that
follows, therefore, the denotation of any expression is given by its extension, and what
particular expressions denote are referred to as their extensions. The general term
denotation is used when talking about the relation between classes of expressions in
the object language and the sorts of thing (entities, sets or relations) in the model that
those expressions are associated with. This latter term is neutral between intension and
extension, and, while Chapters 2 to 6 deal only with extensional denotations, what is
said about denotation in these chapters is relevant to intensional denotation as it is
defined in Chapters 8 to 10.
Since the relation between a constant (or content word in English) and the
entities it denotes is conventional and arbitrary, the denotation assignment function
provides the necessary foundation for talking about the entities in the model by
specifying precisely what is happening in the situation being modelled. For example,
consider a simple situation in which there are three entities, a book, a chair and a
child, where the latter is sitting on the chair reading the book. A model of this situation
has an ontology consisting of the three entities involved in it: the book, the chair and
the child. The denotation assignment function of the model specifies that the English
words book, chair and child (extensionally) denote the book, the chair and the child,
respectively. It also specifies that the transitive verbs read and sit on (extensionally)
denote a relation between the child and the book in the first case and the child and the
chair in the second and that the verbs sit and read both denote things that the child is
doing, and so on.
Together, the specification of an ontology and an assignment of denotations to
basic expressions in a language thus gives a complete description of some state of
affairs. From this complete description we can ascertain the truth or falsity of any
formula in the language to be interpreted. How this is done is determined by the
model theory which constitutes the second part of the interpretation procedure. This
provides a definition of how the denotations of non-basic (i.e. composite) expressions
in Lp are to be determined from the denotations of their constituent parts in accordance
with the principle of compositionality. Ultimately, it defines what it is for a formula
to be true with respect to some model by specifying the truth-conditions for formulae
in Lp. It thus constitutes the central part of the theory of semantics pursued in this
book.
The rest of this chapter is given over to defining a model of a fairly simple (but
non-trivial) situation, and providing an informal account of the model theory. The
40
Interpreting LP
interpretation procedure thus defined may be used to interpret any formula in LP.
Where these formulae translate sentences generated by d , the procedure provides an
indirect interpretation of this fragment of the grammar of English.
2.2.1 Individuals and identity

As discussed above, model-theoretic interpretation provides the definition of the
truth-conditions of formulae with respect to a model of some state-of-affairs. In order
to interpret Lp, therefore, we must first specify a model and then detail a set of rules
for ascertaining the truth or falsity of formulae with respect to this model. The model
defined in this chapter is a very simple one, containing eight entities: two men, two
women, a cake, a cat, a dog and a book. The ontology of this model, called Mi, could
thus be represented pictorially as in (25). Using pictures to represent the entities that
exist in particular models helps to bring out the way that the theory of semantics
presented in this book associates linguistic expressions with non-linguistic objects.
However, they are somewhat inconvenient (and expensive) to have in a textbook and
the entities in models will thus be represented using symbols, rather than pictures.
Any symbols may be used to represent these entities, provided only that there is no
ambiguity in the symbols used. A unique symbol should thus represent each separate
entity in the model. For example, one might represent the eight entities in M, by using
the first eight Arabic positive whole numbers, as in (26.a), or the last eight letters of
the Roman alphabet in their uppercase form, as in (26.b).
The entities in M,
41
(26) a. 1 2 3 4 5 6 7
b. S T U V W X Y
The fact that there are eight different symbols in (26.a) and (26.b) indicates that
there are eight different entities in the model whose properties are determined by the
denotation assignment function. However, since the entities in the model M! are two
men, two women, a cat, a dog, a book and a cake, it is more convenient for mnemonic
purposes to represent these entities by using English words in boldface uppercase
letters with subscripts differentiating entities where necessary, as in (27). The English
words in upper case letters in (27) are intended to represent actual entities and are
not expressions of English. This is an important point and it should be borne in mind
throughout the book that symbols like MANi only ever stand for entities and never for
expressions in English, Lp or any other language.
(27) The entities in M,:
MAN2 WOMAN! WOMAN2 CAT DOG BOOK CAKE.
Entities in the world are typically referred to by using proper names or definite
descriptions, i.e. types of noun phrase. In the translation procedure developed earlier,
such noun phrases are translated into individual constants in LP, which are therefore
analysed as denoting entities in the model. The denotation assignment function in
model M! thus associates each individual constant in Lp with a single entity in the
model, the entity that the constant denotes. Only one entity is associated with each
individual constant because LP is an unambiguous language and so each constant
must have an unambiguous denotation in a particular model. In M,, all the entities
except for the book and the cake have proper names. One of the women is called
Fiona, the other woman is called Ethel, the men are called Jo and Bertie, the dog is
called Chester and the cat is called Prudence. So, part of the denotation assignment
function associates the LP constants translating these names with the appropriate
entities, as shown in (28).
(28) a. jo' DENOTES MANL

b. bertie' DENOTES MAN2.
c. ethel' DENOTES WOMAN,
d. fiona' DENOTES WOMAN2
e. Chester' DENOTES DOG.
f. prudence' DENOTES CAT.
The grammar d also contains simple definite NPs so that we can refer to the
other entities in Mi despite the fact that they do not have proper names. Although
definite NPs in English are composite phrases (consisting at least of the definite
article the and a common noun), the translation procedure treats them as basic
expressions very much like proper names and translates them as individual constants
in Lp. Since individual constants denote entities in the model, the denotation
assignment function directly associates particular definite noun phrases with
particular entities, ensuring that each definite description picks out only one entity. For
example, the expression the-book' is associated with BOOK and the-cake* with
CAKE by the denotation assignment function.
42
Interpreting LP
Although an individual constant may be associated with only one entity in the
model, entities may themselves be identified by more than one expression in a
language. In Mi, the entity WOMAN 2 is denoted not only by the expression fwna'
but also by the expressions the-lecturer' and the-singer\ while the Lp expression
the-studenf denotes the entity, MAN, and DOG is denoted by the-dog' as well as
Chester'. (29) gives the extensions in M, for all the referring expressions in LP, thus
indirectly associating all noun phrases generated by G, with some entity in the model.
(29) a. jo' DENOTES MAN,.

b. the-student' DENOTES MAN,.
c. bertie' DENOTES MAN2.
d. ethel' DENOTES WOMAN,
e. the-golfer' DENOTES WOMAN,
f. fiona' DENOTES WOMAN2
g. the-singer' DENOTES WOMAN 2
h. the-lecturer' DENOTES WOMAN,
i. the-cat' DENOTES CAT.
j. prudence' DENOTES CAT.
k. Chester' DENOTES DOG.
1. the-dog' DENOTES DOG.
m. the-cake' DENOTES CAKE.
n. the-book' DENOTES BOOK.
We are already in a position to assess the truth or falsity of some formulae that
translate some of the sentences generated by the grammar fragment G,: those that
contain the equative copula, be. Sentences like Bertie is the golfer are used to assert
that two descriptions refer to the same entity. Hence, we interpret formulae that
translate such sentences, e.g. bertie' = the-golfer', as being true only where the entity
denoted by bertie' and that denoted by the-golfer are one and the same. This statement
describes the truth-conditions for formulae containing the identity operator = which
can be stated more formally as (30) which forms part of the model theory. (The
subscripts on the NPs are given by the translation rule for equative sentences in (18.b),
above.)
(30) Truth-conditions for identity: A formula of the form (NP,' = NP 2 ') is true
if, and only if, the entity denoted by NP,' is the same entity as that denoted
byNP 2 '.
The expression if and only /fused in (30) is not a common one in everyday
English but is used a lot in semantics which borrows the use from logic. The
conjunction of two sentences using this expression is interpreted to mean that the two
sentences are semantically equivalent. The first if in the conjunction expresses the idea
that the truth of the proposition expressed by the second sentence is sufficient to
guarantee the truth of the proposition expressed by the first, while the only //"expresses
the idea that the truth of the proposition expressed by the second sentence is necessary
for the proposition expressed by the first to be true. The second sentence in the
conjunction thus provides the necessary and sufficient conditions that guarantee the
truth of the proposition expressed by the first sentence. In logical terms, as we shall
see in Chapter 3, the semantic equivalence of two formulae that translate two
43
sentences is guaranteed if both formulae have the same truth value. Hence, given a
true equivalence, if one of the formulae is true then so is the other and if one is false,
then so is the other. Consequently, if Chester* = the-dog' is true then the entity
denoted by Chester* is exactly the same as that denoted by the-dog' and conversely,
if the entities denoted by Chester' and the-dog' are identical then the formula Chester'
= the-dog' is true. The phrase if and only if thus imposes a strong relation between
the two sentences it connects. The expression if and only if is often abbreviated to iff
and this will be the practice in giving numbered definitions from now on.
Exercise 2.3:
Translate the following sentences into LP and then ascertain their truth or falsity with
respect to Mx:
i. Ethel was the golfer.
ii. The student was the singer.
2.2.2 A little light set theoiy

In addition to expressions that denote entities, languages contain expressions that
describe properties that entities have and the relations that they enter into. In English,
properties and relations are generally expressed by verbs and adjectives whose
equivalents in LP are, as we have seen, n-place predicates. In order to complete the
interpretation procedure for LP, therefore, it is necessary to determine the sorts of
extensional denotations that predicates have and to give a definition of the
truth-conditions for formulae containing them in such a way as to capture our
intuitions about the properties and relations English verbs and adjectives describe. In
order to do this, we need to look at set theory, on which the semantic theory in this
book is based. This section provides a short introduction to the necessary concepts,
and further information can be obtained from the references at the end of this chapter.
A set is a collection, or group, of items which are called the elements or
members of that set. This collection may consist of concrete objects like the
computer I am typing this book on, the mug of coffee I have just finished, the
Wallace Monument, and so on, or of abstract items like the meridian, the number
99, the fourth of April 1991, or a set of symbols or indeed any combination of
abstract, concrete and symbolic elements. Finite sets (i.e. those containing a finite
number of elements) may be symbolised by putting the names of their members,
separated by commas, within curly brackets, { and }. This is called the list notation
for sets and there can be either closed sets where all its members are listed or open
sets where only part of the set is listed. Examples of sets in the list notation are given
in (31). (31 .a) gives the closed set consisting of all and only thefirstfiveletters of the
Roman alphabet. (31.b) defines an open set which contains these symbols and more
besides and (31 .c) lists the set of entities in the model M,.
(31) a. {a,b,c,d,e}.
b. {a,b,c,d,e,...}.
c. {MAN,, MAN2, WOMAN,, WOMAN2, CAT, DOG, BOOK, CAKE}.
44
Interpreting LP
In order to show that some element, a, is a member of a particular set, the

symbol e , which is read as is a member of is written between the element, a, on its
left and the set on its right. Thus, the expression a e {a,b,c,d,e} asserts that a is a
member of the set consisting of all and only the first five letters of the Roman
alphabet. The complementary relation is not a member of is symbolised by £ and so
the expression MANj £ {CAT, DOG, BOOK} asserts that the entity, MANi is not
a member of the set consisting of the entities, CAT, DOG and BOOK. An element
must either be a member of a specific set or not be a member of that set. It cannot be
both a member and a non-member of a set nor can it be neither a member nor a
non-member of a set.
Sets may have any number of members. Those with just one member are often
called singleton sets (or just singletons). For example, the set {MAN^ is the set
consisting only of the entity called Jo in the model, Mi. It is important to remember
that a singleton set is not the same as the entity that is its sole member. Thus, the set
consisting only of the letter a, i.e. {a}, is not the same as the letter a. A singleton set
is not, however, the smallest possible set, as a set may contain no members at all.
There is, in fact, only one set with no elements and it is called the empty or null set,
written as 0 . It may be thought that having a set with nothing in it is somewhat
pointless and that it can hardly be described as a set at all. But, like the number zero,
it is a very useful notion to have when stating general rules that apply to all sets
irrespective of the number of members they have, including, of course, none. Because
the empty set contains no members, statements of the form x € 0 (where x is some
element) are never true while those of the form x £ 0 are always true.
Sets may share some or all of their members with other sets. In certain cases, all
the members of one set, A, are members of another, B. Sets like A are called subsets
of sets like B and sets like B are called supersets of sets like A. To symbolise the
subset relation we use the symbols c o r D . The different symbols have the same
interpretation but differ in the direction in which they are read. Both A c B and B 3
A mean that A is a subset of B, but we can read the first as A is a subset ofB and the
second as B is a superset of A. A formal definition of the notion of subset and
superset is given in (32):
(32) Given two sets A and B, A is a subset of B (A <z B) if, and only if, every
member of A is a member of B, in which case B is a superset of A (B 3 A).
According to (32), the sets {a,b,c} and {c,d,e} are both subsets of {a,b,c,d,e} but
{a,b,f} is not a subset of this, because it contains one element, f, that is not in
{a,b,c,d,e}. This definition has two important consequences. In the first place, it
entails that every set is a subset of itself, because (32) just requires all the members
of the subset to be in the superset. Where two sets have exactly the same members,
and are thus identical, the definition is trivially satisfied. In order to specify that the
subset is actually smaller than the superset, the symbols =) and c are used. A c B is
read A is a proper subset of B while B z> A is read as B is a proper superset of A. For
example, {a,b,c} is a proper subset of {a,b,c,d} because the latter set contains one
element, d, not in the former and so the statement, {a,b,c} c {a,b,c,d} is true.
However, while {a,b,c} c {a,b,c} is true, {a,b,c} cz {a,b,c} is not, because the two
sets contain exactly the same elements.
45
The other result of the definition of subset and superset in (32) is that the empty
set is a subset of every set. One can see that this must be so by considering what
would have to be the case for it not to be a subset of some set. From (32), we deduce
that A cannot be a subset of B, if (and only if) there is some element in A that is not
in B. Since the empty set contains no members, then it is not possible for it to have
a member not in any other set. Hence, 0 c A is a true statement, whatever set A
happens to be.
Some sets share members with others without being subsets or supersets of each
other. For example, the set {a,b,c,d} shares the elements c and d with the set {c,d,e,f}
and thus we say that the intersection of these two sets, written {a,b,c,d} n {c,d,e,f},
is the set {c,d}. The definition of this relation given in (33) has three important
consequences: firstly, where two sets have no common elements, their intersection is
the null set, 0 ; secondly, where two sets are identical, their intersection is equal to
both sets; and, finally, the intersection of two sets is a subset of them both.
(33) The intersection of two sets A and B, A n B , is the set containing exactly
those members of A that are members of B.
We may also join sets together to form larger sets. This is known as set union and the
union of two sets, A and B, written A u B, is the smallest set containing all the
members of both A and B. Thus, the union of the sets {a,b,c,d} and {c,d,e,f} is that
set that contains the six elements that appear in both these sets, i.e. {a,b,c,d,e,f}. The
definition of set union in (34) has two consequences: the union of a set with the empty
set is the same as the original set; and both sets joined by the union operation are
subsets of the resultant set.
(34) The union of two sets A and B, A u B, is the set which contains all and only
those elements which are in A or which are in B, or both.
There is one further property of sets that it is necessary to know at the moment.
This is a basic assumption of set theory that two sets with the same members are
identical. For example, the listed set {z,y,x,w,v,u} is identical to the set described as
the last six letters of the English alphabet while the set consisting of the first 10 prime
numbers {1,2,3,5,7,11,13,17,19,23} is identical to the set {23,19,17,13,11,7,5,3,2,1}
which is identical to the set {1,11,3,23,7,13,5,17,2,19}. This property is called the
Axiom of Extension and is defined in (35).
(35) Axiom of extension: Two sets A and B are identical iff they have the same
members.
(35) is called an axiom because it is a primitive of set theory and its truth cannot be
derived from the truth of other statements in the theory. The term extension has the
same meaning as the term discussed above and in Chapter 1 with respect to
denotation, i.e. it refers to the elements in a set and not to the property or properties
that are common to those elements. The axiom of extension thus says that whatever
name is given to a set and however a set is represented, it is fully defined by the
elements it contains. Therefore, if some set of elements has two names or two
different representations, this does not mean that it is two different sets. If two sets
have precisely the same members then they are extensionally identical. This is similar
to the notion of identity between two entities discussed in Section 2.2.1 above. No
46
Interpreting LP
matter how an entity is named (or indeed represented), if two names or definite
descriptions denote the same entity then they are extensionally identical. The
interpretation of the logical language LP thus does not depend on the expressions used
to describe entities in a model, but on what those expressions denote. This important
property is the basis of extensional systems like the semantic theory that is developed
in Chapters 2 to 6, but fails in the intensional ones developed in Chapters 8 to 9 (see
the discussion in Sections 9.1 and 10.3, in particular). Until then, however, the Axiom
of Extension is assumed to hold and this has two important consequences. In the first
place, (35) entails that sets are unordered. This means that no matter what order the
elements appear in a listed set, provided that the different orderings involve all the
same elements, then the differently ordered lists define the same set. Thus,
{z,y,x,w,v,u} is the same set as {u,v,w,x,y,z} which is the same as
{z,u,y,v,x,w}. Secondly, the Axiom of Extension guarantees that a set with repeated
specifications of the same element is identical to the same set with only one listing of
the element. For example, the set {u,v,w,w,x,x,x} is identical to {u,v,w,w,x} which
is identical to {u,v,w,x}.
Exercise 2.4:
Given the sets defined below, which of the following statements are true?
A ={1,3,5,7} B = {a,c,e,j*,i} C={l,a,3,5,e,7}
D = {7,3,1,5} E = (e a i c4,g} F= {{1},{1,3},{ 1,3,5},{1,3,5,7}}
l.A = B. 2. E = B. 3. D = the first four positive odd numbers.
4. e G C. 5. a G D. 6. {1} G A.
7 . D G F. 8. A G C 9. {1,3} 2 A.
IO.DCZC. ll.DcF. 12. E c B.
13. B e E. 14.AnD = ^L. 15. A u D = D.
16.CnB = 0
2.2.3 Interpreting predicates

After that brief interlude, we can now go back to the interpretation procedure and
assign extensions to predicates and specify the truth-conditions of formulae
containing them. We begin by looking at one-place predicates.
Let us assume that the situation being modelled by Mi is one of which the
statements in (36) are all true and other statements containing the intransitive verbs
and adjectives in Gi are all false. This situation can be represented by dividing the
set of entities in the model into subsets according to whether they ran, or laughed, or
were crazy, and so on. These sets provide the extensions of the one-place predicates
that translate the intransitive verbs and predicative adjectives of G b Thus, while
individual constants denote entities in the model, one-place predicates denote sets of
entities, just those entities of which it is true to say that they ran, laughed, were crazy,
etc. Hence, the denotation assignment function of a model associates each one-
place predicate in Lp with some set of entities. The assignment for Mj according to
the situation described in (36) is given in (37).
47
(36) a. Jo was happy and laughed.

b. Ethel was happy and laughed.
c. Fiona sang and was happy.
d. Bertie was wealthy.
e. The dog ran and howled.
f. The cat ran.
g. The cake was disgusting.
h. No-one screamed or was crazy or messy.
(37) a. run' DENOTES {DOG,CAT}.
b. laugh' DENOTES {MAN,, WOMANi}.
c. howl' DENOTES {DOG}.
d. sing' DENOTES {WOMAN2}.
e. scream' DENOTES 0.
f. crazy' DENOTES 0.
h. disgusting' DENOTES {CAKE}.
i. wealthy' DENOTES {MAN2}.
k. happy' DENOTES {MAN^WOMAN^WOMAN,}.
1. messy' DENOTES 0.
The generalisation that all one-place predicates extensionally denote sets can
only be maintained if we accept the existence of the empty set. According to (37), in
M] 0 functions as the extensions of crazy \ messy' and scream'. It might be thought
that one could interpret predicates that are true of no entities in the model by leaving
them out of the denotation assignment function. This is not possible, however. To
interpret a sentence, it is necessary to know what each of the content words in the
sentence means. This is what the denotation assignment function does. It makes sure
that each basic expression in the language is matched with something in the model
and thus represents what the particular constants denote. If no extension is assigned
to some basic expression, then that expression and every expression that contains it
is strictly speaking meaningless. It has no denotation and therefore no sentence or
formula containing the item can be assigned a truth value. For example, a
(pseudo-)English sentence like Fiona glipped the egg could be a well-formed
expression in English (since it conforms to the phonological and syntactic rules of the
language). It has no interpretation, however, because the item glipped is an invented
word and thus has no denotation. It is very important, then, that when defining a
model, every constant (predicate or individual) in LP (and thus indirectly every
content word in the grammar under consideration) receives a denotation, even if
nothing in the model has the property represented by the predicate. Hence, the empty
set is used to indicate that nothing in the model is in the extension of particular
expressions, thus enabling the interpretation procedure to function in the most general
manner.
Once the extension of a one-place predicate is determined, it is an easy matter
to determine whether a formula containing that predicate is true or false. All that is
necessary is to see whether the entity denoted by the argument of the one-place
predicate is in the set of entities it denotes. If it is, then the formula is true, if not, then
it is false. Thus, the formula run '(the-dog') is true in M, because the entity denoted
by the-dog\ i.e. DOG, is, according to (37), a member of the set of entities denoted
48
Interpreting LP
by the predicate, run\ i.e. {CAT,DOG}. By the same reasoning, run'(bertie') is

false with respect to Mu because MAN2 is not a member of {CATJDOG}, i.e. Bertie
is not one of the entities that are running. This provides the definition of the truth-
conditions for formulae containing one-place predicates which is restated more
formally in (38).
(38) Truth-conditions for one-place predicates: A formula W (NP') or A' (NP')
is true with respect to a model iff the entity denoted by NP' in the model is
a member of the set of entities denoted by \Y or A' in the model. The
formula is false otherwise.
From (38) it follows that a formula like messy'(the-cat') is true if and only if the
entity CAT, the extension of the argument expression, the-cat \ is a member of the
extension of the predicate messy*. Since the extension of the predicate is, according
to (37), 0 , the empty set, nothing in the model is messy and, because CAT e 0 is
false, then so is the formula messy'(the-cat'). Again this illustrates the usefulness of
treating 0 as a set in the same way as other sets. If we did not do this, we would have
to have separate truth-conditions for formulae containing predicates that are true of
no entities in a model, clearly missing a generalisation.
Exercise 2.5:
Translate the following sentences into LP and ascertain whether they are true or
false with respect to the model M,.
i. The cat laughed.
ii. Jo was happy.
iii. Fiona ran.
Loosely speaking, intransitive verbs and adjectives (i.e. one-place predicates in

Lp) ascribe properties to entities, but not only do entities have properties, they also
enter into relations with each other. So, for example, two entities, say Jo and Ethel,
might like each other, but not like a third, say Chester. In other words, there is a
relation of liking between Ethel and Jo and vice versa but not one between Jo and
Chester or Ethel and Chester. Let us further specify the situation described by the
model Mi by relating the entities in the model to each other using the transitive verbs
in the grammar, d , as in (39).
(39) a. Jo liked Ethel, Fiona, Bertie and himself.
b. Ethel liked Jo and Fiona.
c. Fiona liked the cat and Ethel.
d. The women, the men and the cat all loathed the dog and Fiona loathed Jo.
e. Chester ate the cake.
f. The cake poisoned the dog.
g. Bertie and Ethel both read the book,
h. Nothing kicked anything else.
We can represent these relations, and others like them, as sets of pairs of
entities, according to whether the first entity bears a particular relation to the second.
49
So, for example, we can represent the liking relation in Mi by grouping together into
pairs Jo and Ethel, Jo and Fiona, Jo and Bertie, Jo and himself, Ethel and Jo, Ethel and
Fiona, and so on. In other words, two-place predicates translating transitive verbs
denote sets of pairs of entities. Thus, like one-place predicates, two-place predicates
denote sets, but sets with members that are not single entities but pairs of entities.
Relations denoted by the verbs in Gx specify a direction in which the relation goes.
Hence, in the definition of the relation denoted by like* in Mi the pair Jo and Ethel
must be distinguished from the pair Ethel and Jo, because in the former Jo is the one
doing the liking while in the latter it is Ethel who does it. Thus, in the same way that
Jo liked Ethel means something different from Ethel liked Jo, we must make sure
that the pairs that make up the extensions of two-place predicates are properly
ordered. The extensions of two-place predicates are, therefore, defined as sets of
ordered pairs of entities. Ordered pairs of elements are conventionally written
between angle brackets, <, >, and separated by a comma. Hence, the two ordered
pairs corresponding to Jo and Ethel and Ethel and Jo are written as
<MAN1,WOMAN1> and <WOMANi,MANi>. Because order is significant, these
pairs represent different things: the first pair shows a relation between MANi and
WOMANi and the latter a relation between WOMAN! and MANi. As a rough and
ready principle that will suffice for the moment: the first member of an ordered pair
corresponds to the entity denoted by the first argument of a predicate (the translation
of the syntactic subject of an active sentence), and the second member corresponds
to the entity denoted by the second argument of the predicate (the translation of the
syntactic direct object of an active sentence). Letting two-place predicates denote sets
of ordered pairs of entities, we may further specify the denotation assignment function
for Mi as in (40).
(40) a. like' DENOTES {<MAN1,WOMAN1>,
<MAN1,MANa>,
<MAN,,WOMAN2>,
<MAN1,MAN1>,
<WOMAN!, WOMAN2>,
<WOMAN!,MANj>,
<WOMAN2,CAT>,
<WOMAN2, WOMAN,>}.
b. loathe' DENOTES {<MAN!,DOG>,
<MAN2,DOG>,
<WOMAN,,DOG>,
<WOMAN2,DOG>,
<WOMAN2,MANX>,
<CAT,DOG>}.
c. poison' DENOTES {<CAKE,DOG>}.
d. eat' DENOTES {<DOG,CAKE>}.
e. read' DENOTES {<WOMAN1,BOOK>,
<MAN2,BOOK>}.
f. kick' DENOTES 0.
The truth-conditions of a formula containing a two-place predicate are similar
to those involving one-place predicates, in that set-membership is the important
50
Interpreting LP
criterion. Such a formula is true with respect to some model, if the ordered pair
formed by taking the entity denoted by the first argument of the predicate as the first
element and that denoted by the second argument of the predicate as the second
element is in the set of ordered pairs denoted by the two-place predicate. For example,
to compute the truth value of the formula like '(the-student'fiona') with respect to Mi
the entities denoted by the two arguments, the-student' and fiona', are identified.
According to the denotation assignment in (29), the first expression denotes MAN!
and the second denotes WOMAN2. The relevant ordered pair that is formed from
these two entities is <MAN!,WOMAN2> because this reflects the order of the
arguments in the formula. The extension of the predicate like' is then identified to see
if the ordered pair is a member of this set. (40) shows that this is the case and that
<MANi,WOMAN2> is indeed in the extension of like'. Hence, the formula
like '(the-student 'fiona') is true with respect to M,. On the other hand, the formula that
translates the sentence The student is liked by Fiona is false in Mi, because the
ordered pair <WOMAN 2 ,MAN!> is not in the extension of like1. The translation
rules ensure that it is not the pair <MAN1,WOMAN2> which is checked for being in
the extension of the predicate. This is because the translation of the sentence is
like '(fiona ', the-student') and not like '(the-student * fiona'). The interpretation rule for
two-place predicates given in (41) ensures that the order of arguments in a
predicate-argument representation is significant and so the differences in the
translation rules of active and passive sentences are semantically significant.
(41) Truth-conditions for two-place predicates: A formula \V(NP,\NP 2 ')
is true with respect to a model iff the ordered pair <Ei,E>> is in the set of
ordered pairs denoted by V,' in the model, where Ei is the entity denoted
by NP,' and E2 is the entity denoted by NP 2 ' in the model. The formula is
false otherwise.
Exercise 2.6:
Translate the following sentences into LP and ascertain whether the formulae they
are associated with are true or false with respect to the model, M } .
i. The cat was liked by the lecturer.
ii. Ethel kicked the student.
iii. The cake poisoned the cat.
If transitive verbs describe relations between two entities, ditransitive verbs

describe relations between three entities. For example, in the relation of giving there
has to be an entity who does the giving, an entity that is given, and an entity to whom
the latter is given. Thus, the extensions of three-place predicates that translate such
verbs are represented as sets of ordered triples of entities. An ordered triple is like
an ordered pair except that it contains three members, not two, but again the order is
significant and fixed. The order of the entities in each of the triples that make up the
extension of a three-place predicate follows the order of the arguments of the
predicate. Thus, the entity denoted by the first argument forms the first element in the
triple, that denoted by the second argument forms the second element and that denoted
51
by the third argument forms the third element.

The only ditransitive verb in fragment d is gave, given and it receives the
extension in (42) which represents a situation in which Jo gave the cake to the dog,
Fiona gave the cake to Jo, Jo gave the book to Bertie, Bertie gave it to Ethel and Jo
gave Fiona the cat.
(42) give' DENOTES {<WOMAN2,CAKE,MAN1>,

<MAN1 ,CAKE,DOG>,
<MAN1,BOOK,MAN2>,
<MAN2,BOOK,WOMAN1>,
<MAN1,CAT,WOMAN2>}.
The definition of the truth-conditions for formulae containing three-place predicates
follows the same pattern as for the other predicates. The formula
givey(the-student\prudence\the-lecturery) is true if (and only if) the ordered triple
<MANi,CAT,WOMAN 2 >, representing the entities denoted by the three arguments
of the predicate in order, is in the set of ordered triples denoted by give'.
(43) Truth-conditions for three-place predicates: A formula
V d t'(NPi\NP 2 \NP 3 > ) is true if and only if the ordered triple <E!,E2,E3> is
in the set of ordered triples denoted by V dt \ where Ei is the entity denoted
by NPi\ E 2 is the entity denoted by NP 2 ' and E3 is the entity denoted by NP 3 '.
The formula is false otherwise.
Exercise 2.7:
Translate the following sentences into Lp and say whether they are true or false with
respect to the model Mi.
i. The golfer gave the book to the golfer.
ii. The student gave the lecturer the book.
iii. Ethel was given the book by Bertie.
Exercise 2.8:
Write a general rule giving the truth-conditions for formulae containing any n-place
predicate along the lines of (41) and (43).
2.2.4 Finishing up
We have now given the extensions for almost all the basic expressions in L P and
provided the truth-conditions for almost all the simple formulae derived by
translating the English sentences generated by the grammar d . The only thing
that remains is to interpret the zero-place predicates, rain' and snow'. Unlike the
other predicates, these do not denote sets, because they have no arguments. Instead
they are directly assigned truth values, because they are, in effect, basic formulae.
The denotation assignment function of the model must, therefore, assign them either
52
Interpreting LP
the value true or the value false, thus directly specifying their truth as part of the
model. No separate statement of their truth-conditions in the theory is thus required
to interpret them. For the sake of completeness, then, let us assume that in the
situation modelled by Mi it was raining but not snowing. The denotation assignment
function of Mi is thus completed by the statements in (44).
(44) a. rain' DENOTES true.
b. snow' DENOTES false.
All basic expressions of the grammar fragment, the constants in LP that translate
the lexemes in English, have now been assigned a denotation. Furthermore, the rules
defining the truth-conditions of the simple formulae resulting from the translation of
sentences in G^ together constitute a complete theory of interpretation for those
formulae in LP. Hence, the truth or falsity of the formulae that translate any of the
simple sentences generated by G! can be ascertained with respect to Mj or indeed any
other model that assigns denotations to the constants of LP. The interpretation
procedure outlined above remains the same for all possible models of situations that
can be described by sentences generated by Gx. Hence, it forms a theory of
interpretation independent of particular situations represented by models. In the next
chapter, the interpretation discussed in this section is put on a more formal footing, but
before this is done the grammar fragment G{ will be extended to generate simple
conjoined sentences.
2.3 Further reading

The language Lp is based on part of predicate logic and there are many introductions
to this logical system, although they all begin with an introduction to propositional
logic which is discussed in Chapter 3. The basic syntax of this language is introduced
in Allwood, Andersen and Dahl (1977: 58-61) and the semantics is given in Allwood,
Andersen and Dahl (1977: 72-75). The most accessible logic textbook is Guttenplan
(1986), of which pp. 166-175 are the most relevant here. Most introductions to
classical phrase structure grammar are to be found in older introductory textbooks to
linguistics (see, particularly, Bach (1974)), but Borsley (1990) discusses more modern
approaches to such grammars with respect also to transformational and unification
grammars. Formal grammar theory and a formal account of phrase structure grammar
can be found in Partee, ter Meulen and Wall (1990:433-454; 492-505). Introductions
to set theory are again legion: see, inter alia, Allwood, Andersen and Dahl (1977: chs.
1 & 2), Lyons (1977:154-161), Partee, ter Meulen and Wall (1990:3-30), and Martin
(1987: ch. 2).
53
Negation and Co-ordination
3.1 Compound sentences

The grammar presented in Chapter 2 generates some very basic sentences of English
and the translation procedure enables each one to be associated with at least one
representation in Lp (more than one, if it contains a homonym). While the number
of sentences generated by the grammar, Gu is relatively large (and can be made
larger if more words are added to the lexicon), the language it generates is still finite.
One of the properties that all natural languages are assumed to have is that they
contain an infinite number of sentences. Since one of the goals of a theory of
semantics, as we saw in Chapter 1, is to pair each sentence in a natural language with
an interpretation, the theory must contain the means to provide an infinite number of
interpretations. Furthermore, because natural languages are here being interpreted
indirectly via a logical representation, the logical translation language must itself be
infinite.
The reason that natural languages are infinite is that they are recursive. This
means that expressions in certain categories may contain other expressions of the same
category. For example, sentences may contain other sentences conjoined by the
expressions and or or, or they may be connected by if...then, or a sentence may
contain one or more repetitions of the expression it is not the case that. Some
examples based on the grammar Gi are given in (1).
(1) a. Ethel poisoned the cat and the dog howled.
b. The lecturer gave the cake to the dog or the student gave the cake to the
lecturer and the lecturer laughed or the dog howled.
c. If the dog ate the cake, then the student poisoned the dog or the lecturer
poisoned the dog.
d. If it was not the case that it rained or it snowed, then it was not the case that
it was not the case that it rained and it was not the case that it snowed.
The somewhat formal sentences in (1) (and an infinite number of others) can
be generated by adding three recursive rules to the grammar, d , and a set of
conjunctions to the lexicon. These are given in (2). The first rule, 13Gi, introduces the
conjunctions and and or, the second, 14Gi, analyses if...then sentences and the third,
15G,, provides a type of recursive negative sentence.
(2) a. 13G,:S->SCONJS.
b. 14G,:S->ifS,,thenS 2 .
c. 15G,: S -» it is not the case that S.
d. CONJ^{and, or}.
54
Complex formulae
The negative sentences generated by Rule 15G, are not elegant and in order to
obtain more natural negative sentences, we can incorporate a set of non-recursive
negation rules that provide the negative counterparts of the basic sentence rules of
Chapter 2. Essentially, all the rules in G, that introduce a finite verb (Vi[+FIN],
V,[+FTN], Vdt[+FIN] and V0[+FIN]) have a negative counterpart containing the word
didn't which appears after the subject and before the non-finite, [-FIN], form of the
verb, as in (3). The non-finite form of a verb in English is its base (or citation) form,
without any tense or participial suffixes or other morphological modification.
(3) a. lG,(neg): S ->NP didn't V,[-FIN].
b. 2G,(neg): S ->NP, didn't VtNP2.
c. 3G, (neg): S -> NP, didn't Vdt[-FIN] NP2 to NP3.
d. 4G, (neg): S -> NP, didn't Vdt[-FIN] NP3 NP2.
e. 5G,(neg): S -> it didn't V0[-FIN].
All the rules introducing the copula, on the other hand, have negative counterparts
with was replaced by wasn't, as in (4).
(4) a. 8G,(neg): S -> NP, wasn't VJPAS] by NP2.
b. 9G, (neg): S -> NP, wasn't Vdt[PAS] to NP2 by NP3.
c. 10G, (neg): S -» NP, wasn't Vdt[PAS] NP2 by NP3.
d. 11G, (neg): S -> NP, wasn't NP2.
e. 12G, (neg): S -> NP wasn't A.
It would be better from the syntactic point of view, of course, if these negative
rules could be directly related to their positive counterparts. The machinery for doing
this will be introduced in Chapter 5, but for the moment we adopt the rules in (3) and
(4) without any attempt at providing syntactic adequacy. However, the grammar now
generates more natural sounding negative sentences like those in (5).
(5) a. If the dog didn't eat the cake, then Ethel wasn't happy.
b. It didn't rain and it didn't snow.
c. It was not the case that Bertie didn't sing.
3.2 Complex formulae

Having defined a set of negation and co-ordination rules, it is necessary to specify the
way the sentences that are generated by those rules are translated into LP and
ultimately interpreted. Because the new syntactic rules are recursive, they allow the
generation of an infinite number of sentences, as already mentioned. It is in situations
like these that the reasons for the rule-to-rule translation procedure set up in Chapter
2 become apparent. The translation of a finite language may be achieved by listing the
sentences generated by the grammar with their appropriate translations, but that of an
infinite language cannot be achieved in the same way. The pairing up of syntactic and
translation rules in the procedure discussed in the last chapter has the advantage that,
provided that both the object language and the logical translation language can be
defined by recursive rules, any sentence in the language can be paired with its
translation, no matter how long it is. Indeed, it is theoretically possible to provide the
55
3 Negation and co-ordination
translation of an infinitely long sentence in this way, even though no human being
could produce such a sentence. Hence, the rule-to-rule hypothesis not only helps
maintain compositionality, but also ensures that the translation of any sentence in the
object language generated by a particular grammar has a well-formed and determinate
translation (and hence interpretation).
In order to translate the syntactic rules in Section 3.1, the syntax of L P itself has
to be extended to make it recursive. Two new L P syntactic rules are given in (6.a) and
(6.c) introducing two new logical categories, Op2 for two-place propositional
operators, connecting two formulae and Opi for one-place propositional operators
which modify only one formula. These categories contain only a few, logical,
expressions which are given in (6.b) and (6.d). The operators in (6.b) and (6.d) have
particular logical functions, shown by their common names: & is the conjunction
operator, v is the disjunction operator, —> is the symbol for material implication,
<-> is the equivalence operator and ~ is the negation operator.
(6) a. t -> t Op 2 1.
b. Op 2 -> {->,v ,&,<-> }.
c. t -> Op,(t).
d. Op,->{-}.
The expressions in (6.b) and (6.d) are used to translate the English conjunctions
introduced by the rules in (2.a) and (2.b) and in translating negative sentences. Like
the copula in equative sentences, conjunctions and negative expressions are not
translated using the general rule for translating lexical items given in (21) of Chapter
2. Instead, the relationship between the logical expressions and their English
counterparts is specified directly by the rules in (7).
(7) a. and => &.
b. or => v .
c. if...then => —> .
d. it is not the case that => ~.
Given these correspondences and the LP construction rules in (6), rules 13Gi,
14Gi and 15Gi have the obvious translations given in (8). In these rules, the brackets
that are introduced are important to the translation because they ensure that the logical
expressions are unambiguous, as we will see below.
(8) a. T13G,: S =» (S* CONP S').
b. T14G,: S =» (S,' -> S A
c. T15G,: S => ~(S').
The translation rules for all the negative rules in (4) and (5) are formed by putting the
negation operator, ~, in front of the translation of the positive version of the rule. For
example, lGi (neg) has the translation ~(V7(NP')), 8G! (neg) has the translation
-(Vt>(NP2,NPi)), and 11G, (neg) has the translation ~(NP, = NP2) and so on.
Translation proceeds as described in Chapter 2: for each line in the derivation of
a sentence, there is a corresponding translation of that line into LP. An example
derivation is given in (9) of the compound sentence The cake was eaten by the dog
and if Ethel laughed, then Ethel poisoned the dog.
56
Complex formulae
(9) a. S
=>S\
b. S CONJ S 13G,
=> (S' CONJ' S').
c. S CONJ if S, then S2 14G,
=>(S'CONr(S,'->S2')).
d. NPi was Vt[PAS] by NP2 CONJ if S, then S2 8G,
=> (V.XNP^NP,') CONJ' (Sr -> S2')).
e. NP, was V,[PAS] by NP2 CONJ if NP V, then S2 1G,
=> (V t '(NP 2 ',NPr) CONJ' (Vi'(NP') -> S,)).
f. NP, was V,[PAS] by NP2 CONJ if NP V, then NP3 Vt NP4 2G,
=> (V, f (NP,',NP,') CONJ' (V,'(NP') -> Vt(NP3',NP4'))).
g. the N was VJTAS] by NP2 CONJ if NP V, then NP3 V, NP4 7G,
=> (Vt'(NP2',the-N') CONJ' (W(NP') -> Vt(NP3',NP4'))).
h. the N was VfPAS] by Npr CONJ if NP V» then NP3 Vt NP4 6G,
=> (Vt'(Npr',the-N') CONJ' (W(NP') -> Vt(NP3',NP4'))).
i. the N is Vt[PAS] by Npr CONJ if Npr V, then NP3 Vt NP4 6G,
=> (Vt'(Npr',the-N') CONJ' (W(N pr ') -> V,(NP3\NP4'))).
j. the N is Vt[PAS] by Npr CONJ if Npr V, then the N Vt NP4 7G,
=» (Vt'(Npr',the-N') CONJ' Cv7(Npr') -> Vtfhe-N\NP 4 '))).
k. the N is Vt[PAS] by Npr CONJ if Npr Vj then the N V, Npr 6G,
=> (Vt'(Npr',the-N') CONJ' (V^N,/) -> V,(the-N',Npr'))).
1. The cake was eaten by Chester and if Ethel laughed, then Ethel poisoned the
dog Lex
=> (eat'(Chester',the-cake') & (laugh'(ethel') -> poison'(ethel',the-dog'))).
As before, derivations may also be represented by analysis trees, i.e. phrase
structure trees annotated with their translations. (10), below, shows the tree associated
with the sentence If it didn 7 rain, it snowed.
Exercise 3.1:
Translate the sentences in (1 .a), (1 .c), (5.a) and (5.c), using either the derivational or
the analysis tree method.
In the grammar of the Chapter 2, the only source of ambiguity was homonymy,
but logically more interesting forms of ambiguity arise when sentential co-ordination
and negation are introduced into the grammar. For example, both (1 l.a) and (1 l.b),
which are generated by Gi, are structurally ambiguous. That is, they each have two
translations depending on the two syntactic structures that analyse them.
57
(10)
(~(rain') —> snow')
If it didn't rain then it snowed
(11) a. It is not the case that the dog howled and Bertie screamed,
b. The cat ate the cake or the dog ate the cake and Fiona laughed.
The ambiguity in (11 .a) depends on the relative scopesof the conjunction and and the
negation it is not the case that. On one interpretation, the latter has scope over the
conjunction of the dog howled and Bertie screamed whereas on the second
interpretation the reverse is the case: the conjunction has scope over the negation
which only has scope over the dog howled. In the grammar fragment, d , the two
translations that correspond to these two interpretations are derived by applying the
rules 13Gi and 15Gj in different orders. If the latter is applied first, then the reading
of the sentence is that the dog did not howl or Bertie did not scream as in (12). The
interpretation in which the dog did not howl but Bertie did scream is the one given to
the translation derived by applying rule 13Gi before 15Gi as in (13).
As we will see in Section 3.3, the two LP expressions in (12.h) and (13.h) have
different truth-conditions and so give rise to the two interpretations described above.
The difference between the two expressions is shown by the bracketing of the
component formulae which indicate the relative scopes of the two logical operators,
~ and &. In the first expression, (12.h), ~(howV(the-dog') & scream''(ether)) the
connective, &, comes inside the brackets that are introduced by rule 15d in (12.b).
These brackets define the domain over which the negation has effect. This is called
the scope of the negation and the conjunction operator comes inside this. In the
second expression, (13.h), (-(howV(the-dog')) & scream'(ethel')) the reverse is the
case. The negation operator now comes within the scope of the conjunction, because
it appears within the brackets introduced by 13G, in (13.b). The scope of a connective
is crucial to the interpretation of any formula containing it. Where formulae contain
different operators and connectives, different readings result from the relative scopes
of the latter. A second example appears in (14) where the two translations of (11 .b)
are given. The first formula has the disjunction operator within the scope of the
58
Complex formulae
conjunction, giving the reading where Fiona is laughing and either the dog or the cat
ate the cake. The second reading reverses the scopes of the connectives to give the
reading where the cat ate the cake and either the dog ate the cake or Fiona laughed.
It is, therefore, very important to include the brackets introduced by rules 13G,, 14Gi
and 15Gi, whenever there is more than one conjunction or negation in the sentence.
Without them, the LP expressions translating the sentences remain ambiguous,
contrary to the requirement that LP be an unambiguous language. Certain pairs of
brackets, e.g. the very outermost ones in a formula, do not really serve a useful
purpose and will often be omitted and, later on in the book, when the reader has more
familiarity with the logical system, other sets of brackets will occasionally be omitted
where there is no risk of ambiguity.
(12) a. S
b. it is not the case that S \5GX

=» ~(S').
c. it is not the case that S CONJ S 13d
=> ~((S' CONJ' S')).
d. it is not the case that NP V, CONJ S 1G,
=>~(Cv7(NP')CONJ'S')).
e. it is not the case that Npr V{ CONJ S 6d
=» ~((W(the-N') CONJ' S')).
f. it is not the case that Npr V, CONJ NP V, 1G,
=> ~((Vi'(the-N') CONJ' W(NP'))).
g. it is not the case that Npr V, CONJ Npr V, 6G,
=> -((WCthe-N') CONJ' W(Npr'))).
h. it is not the case that the dog howled and Bertie screamed Lex
=> ~((howl'(the-dog') & scream'(bertie'))).
(13) a. S
=>S'.
b. S CONJ S 13G,
=» (S' CONJ' S').
c. it is not the case that S CONJ S 15G,
=>(~(S') CONJ' S').
d. it is not the case that NPVj CONJ S 1G,
=>(~CvY(NP'))CONJ'S').
e. it is not the case that Npr Vi CONJ S 6G,
=> (~CV7(the-N')) CONJ' S').
f. it is not the case that Npr V; CONJ NP Vi 1G,
=> (-CvY(the-N')) CONJ' W(NP')).
g. it is not the case that Npr V{ CONJ Npr Vj 6Gi
=» (-(ViXthe-N')) CONJ' W(N pr ')).
h. it is not the case that the dog howled and Bertie screamed Lex
=> (~(howl'(the-dog')) & scream'(bertie')).
59
(14) a. ((eat'(the-cat',the-cake') v eat'(the-dog\the-cake')) & laugh'(fiona')).

b. (eat'(the-cat',the-cake') v (eat'(the-dog',the-cake') & laugh'(fiona'))).
Exercise 3.2:
Letting the translation of Fiona laughed be represented by the symbol p and that of
Bertie screamedbe represented by the symbol q, how many different translations into
Lp do each of the following have?
i. It is not the case that it is not the case that Fiona laughed or Bertie
screamed.
ii. Fiona laughed or it was not the case that Fiona laughed and Bertie
screamed.
3.3 Interpretation
We must now turn our attention to the interpretation of formulae that contain the
logical connectives or the negation operator. In defining the truth-conditions of
complex formulae that contain these expressions, the internal structure of the
component formulae is irrelevant as it does not affect the way the truth or falsity of
the formula is computed. All that is significant is whether the component formulae of
a complex formula are true or false, not what the predicates and arguments that make
up these formulae denote in the model. For this reason, ~, &, v , and -> are known as
truth-conditional connectives. This section shows how the connectives in LP
contribute to the truth-conditional meaning of the formulae in which they appear,
taking each in turn and beginning with the negation operator ~.
3.3.1 Negation
The simplest of the five connectives is the negation operator, ~, used in the
translations of the negated sentence rules in d . This operator takes only a single
formula in its scope and reverses the truth value of that formula. In other words, if a
formula (J) is false ~((|)) is true and if $ is true then ~((|)) is false. This provides the
truth-conditions for negative formulae, which can be stated as in (15).
(15) Truth-conditions for negation: A formula ~(S') is true if, and only if, S'
is false. ~(S') is false otherwise.
Using (15) it is possible to work out the truth value of negative formulae with respect
to particular models. For example, the truth value of the formula ~(run'(chester'))
with respect to Mi can be worked out as follows:
(16) a. run'(Chester') is true in Mi iff the extension of Chester', i.e. DOG, is a
member of the set of entities denoted by run', i.e. {DOG,CAT}.
b. Since DOG e {DOG,CAT} is true, run'(Chester') is true in M, and so
-(run'(chester')) is false in Mi.
Logicians have devised another way of representing the truth-conditions for
operators like ~, and connectives like &, etc., which is, perhaps, clearer than a verbal
60
Interpretation
statement. This method sets out a truth table giving all the possible combinations
of truth and falsity for all the component formulae and showing the truth value of the
resultant complex formula containing the appropriate connective. To show the truth
conditions of the connectives using the truth table method, the component
formulae are represented by propositional variables. Variables do not have a
fixed interpretation in some model, like constants, but range over extensions of the
appropriate sort. Given that formulae denote truth values, propositional variables
range over the values true and false. Such variables are usually represented by lower
case letters from the middle of the alphabet: p (mnemonic for proposition) is used
where only one variable is required, p and q are used when two are needed and r is
added if a third is required.
To draw a truth table for a connective, you take an appropriate number of
propositional variables, one in the case of negation and two for the other
connectives. Underneath the variables you write the letter t or the letter f standing
for truth and falsity, respectively, and put all possible combinations of the letters on
different lines. If there is one component formula in the complex formula, then
there are only two lines, one for t and one for f, while if there are two formulae
there are four lines, for cases where both formulae are true, where both are false and
where they differ in truth value. Each combination of truth and falsity is associated
with the appropriate truth value for the complex proposition involving the operator or
connective. Hence, for negation, there are two lines in the truth table and the
resulting complex formula has the opposite truth value to the component formula.
This is shown in (17) where the variable p stands for the component formula.
(17) Truth table for negation
p ~p
t f
f t
The use of variables in truth tables means that we can replace them by actual
formulae to get the particular truth value of a compound formula very easily. All
that needs to be done is to take a component formula, e.g. (etheV - the-golfer') that
translates Ethel was the golfer and replace all occurrences of a variable in a table
with this formula. Once the truth value of the component formula (or formulae)
with respect to some model has been determined, it is possible just to go to the line in
the truth table with that truth value and read off the truth value for the complex
formula. Thus, to find out the truth value of ~(ethel' - the-golfer'), we need
first to determine the truth or falsity of (etheV = the-golfer') with respect to M, by
checking whether the expression etheV denotes an entity identical to that denoted by
the expression the-golfer' in that model. Since this is the case, (etheV = the-golfer')
is true with respect to Mi. Looking at the line in the truth table for negation where the
propositional variable has the value t, we find the truth value of the negated formula
shown to the right, i.e. f. Hence, -(etheV = the-golfer') is false with respect to the
model Mi.
61
Negation in L P is recursive and the truth-conditions given in (15) allow the truth
value of formulae that contain more than one or two embedded formulae to be
computed in a mechanical fashion. For example, the translation of the sentence // is
not the case that the lecturer didn't scream contains two instances of the negation
operator, i.e. ~(~(scream'(the-lecturer'))). To find out the truth value of this formula
with respect to some model, we need to find out the truth of the most deeply
embedded formula, scream'(the-lecturer'). The truth-conditions for - then provide
the truth value of -(scream'(the-lecturer')) and a re-application of these truth-
conditions yields the value of the formula -(-(scream'(the-lecturer'))). Thus, the
truth value of this particular formula with respect to M, can be worked out as in (18).
(18) a. scream'(the-lecturer*) is false in M, because WOMAN 2 , the extension of
the-lecturer', is not a member of 0 , the extension of scream' in Mi.
b. Hence, -(scream'(the-lecturer')) is true in M, by the truth-conditions for
negation.
c. Hence, -(-(scream'(the-lecturer'))) is false in M, again by the
truth-conditions for negation.
A more succinct way of showing these steps is again to use a truth table. There
are a number of ways of showing this, but we shall adopt the clearest. Here you
write a propositional variable p on the left with its two possible truth values, then
you write the next embedded formula to its right with the corresponding truth
values, and so on, until you reach the largest formula. At every stage you use the
truth table for negation to give you the truth values for the current column by treating
the previous column as the base formula. All you have to do then is find out the
actual truth value of the formula that you are substituting for /?, and look along
that line to get the truth value of the whole compound formula. The truth-conditions
for a doubly negated formula can thus be represented by the truth table in (19) which
combines two instances of the truth table for negation.
(19)
p -p ~(-p)
t f t
f t f
The truth table for -(-/?) is identical to that for p itself. This means that in
logical form double negatives give rise to the same interpretation as the
corresponding positive. In other words, -(-/?) has exactly the same truth value as
/?, whatever p happens to be. Thus, something that may not have been completely
obvious from just looking at the truth-conditions for negation becomes obvious when
we use truth tables. In translating sentences with double negatives in them, however,
it is often the case that only single negation is intended. For example, translating
Chester never did nothing with two negatives, the formula expressed is semantically
equivalent to Chester always did something. It is often the case, however, that
sentences containing multiple negatives in many languages serve only to express a
single negation, the extra negatives merely serving as reinforcements of this and not
62
Interpretation
as independent negations. For example, the expression Chester never did nothing in
many dialects of English is interpreted as Chester never did anything; the second
negative expression nothing does not independently negate but emphasises the
negation of the expression never. In the logical language Lp, on the other hand,
multiple negatives are always interpreted as independent negations. Thus, ~(~(p)) is
always equivalent to /?, never to ~(p). Hence, if a sentence in some natural language
is analysed as containing multiple negative expressions with only a single negation
intended, that sentence must be translated into LP with only one negation operator.
3.3.2 Conjunction
Specifying the truth-conditions for formulae containing &, the logical translation of
and, involves reference to the truth values of the two formulae that the expression
conjoins: such a formula is true only when both component formulae are true and is
false in every other situation. This is guaranteed by the truth-conditions stated in (20)
which can also be represented by the truth table in (21).
(20) Truth-conditions for conjunction: A formula (S,' & S 2 ') is true iff S,' is
true and S2' is also true. (Si' & S2') is false otherwise.
(21) Truth Table for &
p q p&q
t t t
t f f
f t f
f f f
To interpret a formula containing & thus involves finding out the truth values of the
formulae it conjoins and then checking the truth table to ascertain the truth or falsity
of the complex expression. For example, the translation of The dog howled and the cat
ate the dog with respect to Mi is (howV(the- dogf) & eatf(the-cat\the-dog')). The
steps in interpretation of this formula with respect to M, are set out in (22), a situation
that corresponds to the third line of the truth table in (21).
(22) a. howl'(the-dog') is true in M, because the entity denoted by the-dog' in M,
is a member of the extension of howl' in Mi, i.e. DOG e {DOG}.
b. eat' (the-cat' ,the-dog') is false in M i because the ordered pair formed from
the entities denoted by the-cat' and the-dog', respectively, are not in the set
of ordered pairs denoted by eat' in Mi, i.e. <CAT,DOG> £
{<DOG,CAKE>}.
c. Hence, (howl'(the-dog') & eat'(the-cat', the-dog')) is false in Mi by the
truth-conditions for &.
More complex examples can be constructed using negation and conjunction.
Let us work through an example in detail, interpreting the sentence It is not the
case that Ethel loathed Chester and Ethel wasn't happy. The structure of one of the
63
formulae this ambiguous sentence can be translated into is set out in tree form in (23)
which is parallel, of course, to the structure of the translated sentence.
(23)
( (loathe'(ethel\ Chester') (happy'(Chester'))) )
The interpretation of this formula begins with the specification of the

truth-conditions of the whole complex formula and works downwards to those of the
simplex component formulae. Once we have reached this level, the truth of the whole
expression can be ascertained by finding the extensions of the constants in the
expression in some model. Thus, the interpretation procedure of (22) with respect to
the model Mj follows the steps laid out in (24) which begins by specifying the
truth-conditions of the outermost operator, i.e. negation, and then moves on to those
for the conjunction connective, &. Since one of the conjuncts is negative, the truth-
conditions for negation are repeated. Finally, the truth of the simplex propositions is
checked with respect to the model by applying the truth-conditions for two-place and
one-place predicates, as set out in the last chapter. Having ascertained these truth
values, those of the more complex formulae are worked out until we finally get to the
operator with the widest scope, where the truth of the whole formula can be computed.
(24) a. -((loathe'(ethel',Chester') & -(happy'(ethel')))). is true in M, iff
(loathe'(ethel',Chester') & -(happy'(ethel'))) is false (by (15)).
b. (loathe'(ethel',Chester') & ~(happy'(ethel'))) is true iff
loathe'(ethel',Chester') is true and -(happy'(ethel')) is also true (by (20)).
c. loathe' (ethel' ,chester') is true iff the ordered pair consisting of the entities
denoted by ethel' and Chester', respectively, is in the set denoted by loathe'.
d. Since, ethel' denotes WOMAN, and Chester' denotes DOG and loathe'
denotes {<MAN1,DOG>,<MAN2,DOG>,<WOMAN,,DOG>,
<WOMAN 2 ,DOG>,<WOMAN 2 ,MAN 1 >}, loathe'(ethel',Chester') is true
in M,.
e. -(happy'(ethel')) is true iff happy'(ethel') is false (by (15)).
f. happy'(ethel') is true iff the entity denoted by ethel' is in the set denoted
by happy'.
64
Interpretation
g. Since the entity denoted by ethel' is WOMANi and the extension of happy'
is {WOMAN,,WOMAN*MAN,}, happy'(ethel') is true in M,.
h. Hence, -(happy'(ethel')) is false in Mi.
i. hence, (loathe'(ethel',Chester') & -(happy'(ethel'))) is false in M,.
j. hence, -((loathe'(ethel',Chester') & -(happy'(ethel )))) is true in M,.
As before, the truth table method can be used to get the same result. Firstly,
the structure of the formula is made more transparent by replacing the simplex
formulae by the variables p and q to get the formula -(/? &~(q)). Then, a truth table
is set up showing all the combinations of the truth values of the basic formulae and
columns are constructed showing the truth values of each of the more complex
formulae within the expression. The final column gives the truth value of the whole
expression determined from the combination of truth values of the basic formulae.
Once the truth values of the simplex formulae have been ascertained with respect to
a particular model, the final value of the expression can be read off the appropriate
line. Thus, line 1 in (25) gives the truth value of the expression in M, where p is
associated with the formula loathe'(ether,Chester') and q is associated with the
formula happy '(ethely).
(25)
p q ~q p&~q ~(p&~q)
t t f f t
t f t t f
f t f f t
f f t f t
Exercise 3.3:
Work through the interpretation of the other formula of LP that translates It was not
the case that Ethel loathed Chester and Ethel wasn 't happy, using both the method
illustrated in (24) and the truth table method illustrated in (25).
According to Gi, a sentence containing two or more instances of the conjunction

and has two or more derivations. Hence, they have more than two translations,
because of the parallelism between syntactic derivation and translation. Thus, Bertie
laughed and Fiona laughed and the dog howled has the two translations in (26) where
the difference resides in the comparative scopes of the two instances of &, shown by
the bracketing.
(26) a. (laugh'(bertie') & (laugh'(fiona') & howl'(the-dog'))).
b. ((laugh'(bertie') & laugh'(fiona')) & howl'(the-dog')).
This ambiguity does not, however, give rise to any difference in interpretation.
The truth-conditions for (26.a) and (26.b) are identical: three formulae conjoined
by & give rise to a true formula if, and only if, all three component formulae are
true. This can be proved by constructing truth tables for the two readings,
65
representing the basic formulae as p, q and r. Because the pattern of truth and falsity
is identical in the first and final columns, the different formulae must have the same
truth-conditions: they are true or false in the same situations. This is shown in the
truth tables in (27) and (28).
(27)
p q r (q&r) (p&(q&r))
t t t t t
t t f f f
t f t f f
f t t t f
t f f f f
f t f f f
f f t f f
f f f f f
(28)
p q r (P&q) «p&q)&r)
t t t t t
t t f t f
t f t f f
f t t f f
t f f f f
f t f f f
f f t f f
f f f f f
In fact, no matter how many formulae are strung together using the conjunction
connective, the actual bracketing of the conjoined formulae does not matter to the
final truth value (provided that no other connective appears in the expression).
Thus, for any formulae/?, q and r, (p&(q& r)) has the same truth value as ((p & q)
& r). This property is called associativity and & is known as an associative operator.
Associativity is a property also found with certain arithmetical functions like addition
and multiplication (but not subtraction and division). Thus, for example, (3 + 2) + 4
= 3 + (2 + 4) = 9 and (2 x 3) x 4 = 2 x (3 x 4) = 24 or, in general, (x + y) + z = x +
(y + z) and (x x y) x z = x x (y x z). This is an important property which, as we shall
see in Chapter 7, enables us to make some useful inferences from formulae with this
structure. Nothing more will be said about this here, but remember that in working
out the truth value of a formula containing a string of formulae all connected by & (or
sentences conjoined by and) the different derivations do not affect the truth-
conditions; the ambiguity is only apparent, not real. In other words, our translation
procedure overdisambiguates, as it were. But this is not a problem, because the
interpretation procedure ensures that no semantic ambiguity is produced.
66
Interpretation
3.3.3 Disjunction
The next connective is v which we have used to translate English or. This is
known as the logical disjunction operator and a formula containing this connective
is true when at least one of the component formulae is true:
(29) Truth-conditions for disjunction: A formula of the form (ST v S>') is
true, iff Si* is true or S 2 ' is true. (S/ v S2') is false otherwise.
(30) Truth table for disjunction
p q pvq
t t t
t f t
f t t
f f f
It follows from (29) that the sentence The lecturer poisoned the cat or Chester
howled is translated into a formula that is true in M H as can be seen from the
detailed interpretation in (31).
(31) a. (poison' (the-lecturer' ,the-cat') v howl' (Chester')) is true with respect to Mi
iff poison'(the-lecturer',the-cat') is true or howl'(Chester') is true.
b. poison'(the-lecturer',the-cat') is true iff the ordered pair consisting of the
entity denoted by the-lecturer' and that denoted by the-cat' is a member of
the set denoted by poison' in M,.
c. Since the entity denoted by the-lecturer' is WOMAN2, that denoted by
the-cat' is CAT and the extension of poison' is {<CAKE,DOG>},
poison'(the-lecturer',the-cat') is false in M L
d. howl'(Chester') is true iff the entity denoted by Chester' is a member of the
set denoted by howl'.
e. Since the entity denoted by Chester' is DOG and that of howl' is {DOG},
howl'(Chester') is true in Mj.
f. Since howl'(Chester') is true in Mi, (poison'(the-lecturer',the-cat') v
howl'(Chester')) is true with respect to Mi.
You may have already noticed that there are some odd things about this
treatment of disjunction, particularly in the fact that a disjunctive formula is true if
both of its disjuncts are true. For example, under this interpretation the formula
translating the sentence The student was crazy or the student was happy is true in a
model in which both disjuncts are true. Yet we often tend, when using disjunctions,
to assume that one of the disjuncts must be false if the other is true. Such
disjunctions are said to have an exclusive interpretation, one which disallows
situations where both disjuncts are true. For example, on hearing the sentence in (32),
a speaker of English would normally assume that the jacket being referred to was not
simultaneously red and blue, a situation that is not excluded by the truth of the formula
in (32.b), according to the truth-conditions for the inclusive disjunction operator v.
67
(32) a. The jacket was blue or red.

b. (blue* (the-jacket') v red' (the-jacket*)).
The fact that sentences like (32) receive an interpretation that excludes the
situation in which both disjuncts are true can be captured directly in the semantics by
interpreting or, not as v in (30) above, but as a different operator ve which has the
truth-conditions specified in (33) which gives rise to the truth table in (34).
(33) Truth-conditions for Exclusive Disjunction: A formula of the form
(ST v e S2') is true, iff S/ is true and S2' is false or S2' is true and S/ is
false. (Si' ve S2') is false otherwise.
(34) Truth table for exclusive disjunction
p q pveq
t t f
t f t
f t t
f f f
There is, however, a compelling reason for not translating the English conjunction or
as ve and thus interpreting all disjunctive sentences exclusively. While many uses of
disjunction do imply that only one of the disjuncts holds, there are many occasions
where we do not want to preclude a situation in which both disjoined formulae are
true. For example, the propositions expressed by the sentences in (35) are not
obviously false of situations where both of the disjuncts are true, i.e. where there are
students who have a private income and rich parents, or where a job applicant has
two Higher and five 'O' grade passes, or where the lecturer does not want to rule out
the possibility of a zealous student reading both articles.
(35) a. Students who have a private income or rich parents are better off at
University than those who do not.
b. Applicants for the job must have two Higher, or five Ordinary, grade passes.
c. You must read the article on presupposition or the article on implicature by
next Wednesday.
The sentences in (35) (and many others like them) show that exclusive disjunction
on its own makes too strong a claim about the truth-conditions of English or, since
inclusive, as well as exclusive, readings of disjunctive sentences must be allowed for.
For this reason, or is translated as v and interpreted as inclusive disjunction. The
problem of accounting for the exclusive readings of or as illustrated in (32) will be
addressed in Chapter 7 where different sorts of inference are discussed.
As can be seen from checking truth tables, disjunction, both inclusive and
exclusive, like conjunction, is associative. Thus, (p v (q v r)) has the same
truth-conditions as ((p v q) v r): they are both true if any one of the component
formulae is true. (The reader is invited to show that this is indeed so by drawing the
appropriate truth tables.) However, when or is used to combine clauses which are
68
Interpretation
negative or contain and, genuine ambiguities may result. For example, the expression
It is not the case that it rained or it snowed and the dog howled has no fewer than five
different interpretations, depending on the scopes of the negation, the disjunction and
the conjunction. Lettingp stand for the formula rain\ q stand for snow* and r stand
for howl'(the-dog'), we get the schematic formulae in (36). The two truth tables in
(37) and (38) show that at least the readings in (36.a) and (36.b) have different
truth-conditions. A formula of the form of (36.a) is true only when p and q are both
false and r is true, while one with the form in (36.b) is true in all situations except
where r is true and either p or q or both are also true.
(36) a. (~((pvq))&r).
b. ~(((pvq)&r)).
c. ~((pv(q&r))).
d. ((~(p)vq)&r).
e. (~(p)v(q&r)).
(37)
p q r (pvq) ((pvq) &r) ~(((pvq))&r))
t t t t t f
t t f t f t
t f t t t f
f t t t t f
t f f t f t
f t f t f t
f f t f f t
f f f f f t
(38)
P q r (pvq) ~((pvq)) (~((pvq))&r))
t t t t f f
t t f t f f
t f t t f f
f t t t f f
t f f t f f
f t f t f f
f f t f t t
f f f f t f
69
Exercise 3.4:
Write out the truth tables for the other formulae in (36). Which, if any, of these
interpretations do you think would be the most likely ones to be given to an utterance
of the expression It is not the case that it rained or it snowed and the dog howled by
a speaker of English and why do you think this might be the case? Does the
expression It didn 't rain or snow and the dog howled have the possible translations
into LP shown in (36)? If not, why not?
There is another property that should be mentioned with respect to disjunction

and that is that the relative order of the disjuncts is not truth-conditionally significant.
Thus, the sentence It rained or it snowed has the same truth-conditions as It snowed
or it rained. This is reflected in the truth-conditions for disjunction given above,
which do not mention the order of the formulae, thus ensuring that (p v q) is
truth-conditionally equivalent to (q v pj.This property is technically known as
commutativity and disjunction is a commutative operator. Commutativity does
indeed seem to be a property of English disjunctive sentences, as shown by the fact
that Bertie laughed or Fiona sang may be used to describe the same situations as
Fiona sang or Bertie laughed. On the other hand, most speakers of English consider
there to be a semantic difference between Bertie laughed and Fiona sang and Fiona
sang and Bertie laughed. However, the conjunction operator also turns out to be
commutative in Lp, since (p & q) is truth-conditionally equivalent to (q & p), as
readers can check for themselves. This potential problem will be put to one side for
the moment and taken up again in Chapter 7.
3.3.4 Implication
The final connective used in the grammar fragment of Section 3.1 is the one that we
have used to translate English if...then sentences. This is —>, the logicians' material
implication. A formula consisting of two formulae, p and q, conjoined by -> is
interpreted as true, provided that the second formula, q, called the consequent, is true
whenever the first formula, p, called the antecedent, is also true. A formula of the
form p —> q is thus false only when p is true and q is false. The truth-conditions for
formulae containing -> are given in (39) and its associated truth table is given in (40).
(39) Truth-conditions for material implication: A formula ( S / -> S 2 ') istrue
iff Si* is false or S 2 ' is true. (Si' -» S2') is false, otherwise.
(40) Truth table for material implication:
p q p->q
t t t
t f f
f t t
f f t
70
Interpretation
A good way to help you remember the truth table is to remember the
sentence in (41).
(41) If Jo is over six feet tall, then Jo is over four feet tall.
There are only three possible ways that someone's height may relate to being over (i.e.
greater than) four feet or over six feet: either they are over six foot and hence over
four foot, or they are under six foot but over four or they are under four foot. No-one,
however, can be under four foot but over six. These possibilities are illustrated in the
diagram in (42) where the variable/? stands for the formula translating Jo is over six
foot tall and q stands for Jo is over four foot tall Notice that the fourth situation
cannot be represented diagrammatically and this gives you the situation in which the
conditional would have to be false.
(42) If Jo is over six feet tall, then Jo is over four feet tall
6' - -
4'
Material implication is neither associative nor commutative. If it rained, then

Bertie was happy does not mean the same as If Bertie was happy, then it rained. This
is guaranteed by the truth-conditions in (39), as in the former case the formula rain'
—» happy'(ber'tie') is true provided that it is not raining, but Bertie is happy, while
happy'(bertie') —> rain' is true with respect to the latter situation. The
non-associativity of -> can be shown by looking at the truth tables associated with
the two schematic formulae ((p -> q) -» r) and (p -> (q -> r)). As can be seen by
comparing (43) and (44), the formula ((p -> q) -» r) is false in three situations, i.e.
where r is false and either p and q are both true or where p is false. The latter formula
71
r
(P -> (o -> > M is, however, only false where p and q are both true, but r is false. The
definition of the truth-conditional meaning of if... then thus ensures that the indirect
interpretation of If if it rained then the dog howled then Bertie screamed (a sentence
that is better expressed in English by If when it rained, then the dog howled, then
Bertie screamed) is truth-conditionally distinct from If it rained then if the dog howled
then Bertie screamed.
(43)
p q r (p-»q) ((p->q) -»r)
t t t t t
t t f t f
t f t f t
f t t t t
t f f f t
f t f t f
f f t t t
f f f t f
(44)
p q r (q->r) (p -> (q -> r))
t t t t t
t t f f f
t f t t t
f t t t t
t f f t t
f t f f t
f f t t t
f f f t t
There are, however, strange things about this interpretation of if...then. One of
the peculiar aspects has to do with the situations in which it is possible to assert
a true conditional formula. The only situation in which it is possible for a
conditional to be false according to the truth table in (40) seems intuitively plausible.
Consider the sentence If it rained, then it didn't snow. A situation in which it was
raining and it was also snowing would be a counterexample to the conditional
statement and sufficient to falsify it. This is, of course, exactly what the truth-
conditions for -»tell us, because the antecedent would be true but the consequent
false. Hence, the conditional formula must be false. The assertion of this sentence in
other situations involving rain and snow would give rise, however, to the expression
of a true statement which appears much less intuitively satisfying. In a situation in
which it was raining and it wasn't snowing, the sentence would express a true
proposition. The situation could be said to be a confirming instance of the
conditional, but it would be most peculiar to utter such a sentence unless the speaker
were committed to there being a causal connection between rain and a lack of snow.
72
Interpretation
Knowing simply that two formulae happen to be true does not justify a belief in
any sort of a connection between them, something that would be conveyed by the
assertion of a conditional containing them. Even more peculiar would be the
assertion of Ifit rained, then it didn 7 snow with respect to a situation in which it was
not raining but was also not snowing. The truth table for material implication,
however, ensures that the conditional formula is true because the formula expressed
by the antecedent is false.
These peculiarities may be explained, not by the truth-conditions of if .then, but
from what we consider to be an appropriate utterance of a conditional sentence. In
other words, associating material implication with natural language conditionals
merely gives the central core of the meaning of if ..then, its truth-conditions. It has
nothing to say about the appropriateness or otherwise of the assertion of a conditional
statement. Hence, irrespective of the connectivity of antecedent and consequent,
someone who utters if it rained then it didn 't snow with respect to a situation in which
it did indeed rain and did not snow, makes a true assertion, the situation being a
confirming instance of the conditional. With respect to situations in which it didn't
snow and also did not rain, the sentence may also be truthfully asserted, because there
are many situations in which it didn't snow (e.g. if the sun was shining). The
sentence could also be truthfully asserted with respect to the last possible situation,
one in which it is not raining but it is snowing. If the sentence is asserted as a general
rule about the weather, a day on which rain didn't fall, but snow did does not
disconfirm the rule. Thus, it would be possible for someone to say See, it's true. If it
rained, then it didn't snow. Isn't this a true statement? It is certainly not a false one,
and given that the theory of truth put forward in this book is bivalent, and so only
allows formulae to have one of two possible truth values, the sentence must express
a true proposition with respect to this situation. The truth table in (40) does, therefore,
provide some of the meaning of simple conditionals in English, and for the moment
let us accept it as a reasonable interpretation of the if.then sentences generated by d .
We will, however, return to a discussion of the adequacy of material implication as
the interpretation of natural language conditionals in Chapter 7.
3.3.5 Equivalence
Logicians usually operate with another logical connective which is not in GM but
which we have been using in the definitions of truth-conditions. This is the
equivalence operator which is represented in LP as <-> and in English by the phrase
if, and only if. We could introduce this into the grammar by putting the English
phrase into the lexicon as a CONJ so that it could be introduced via the same
syntactic rule, Rule 13, that introduces and and or. Although the phrase is little used
in everyday English, adopting it into the grammar fragment allows us at least to see
what contribution it makes to the truth-conditions of the definitions that are given. The
truth-conditions of the equivalence connective, <-> , are captured by requiring a
formula p <-> q to be true just in case p and q have the same truth value. The
truth-conditions for this connective are spelled out in (45) and its corresponding truth
table is given in (46).
73
(45) Truth-conditions for equivalence: A formula (S/ <-» S2') is true iff
either S!' is true and S2' is true or S,' is false and S2' is false. (Si' <-> S2')
is false, otherwise.
(46) Truth Table for equivalence
p q p<->q
t t t
t f f
f t f
f f t
Notice that equivalence has the same truth-conditions as a double material

implication conjoined by &, as you can see by comparing the truth table in (47) with
that in (46). The final columns in these two tables show the same patterns of truth and
falsity for the same combinations of truth and falsity of the component formulae, thus
proving their semantic equivalence.
(47)
p q (p->q) (q->p) ((p -> q) & (q -> p))
t t t t t
t f f t f
f t t f f
f f t t t
Exercise 3.5:
(46) and (47) show that a formula of the form (p <-» q) is truth-conditionally
equivalent to one of the form ((p -> q) & (q-> p)). Write out the truth table for the
formula ((p <^ q) <-» ((p -» q) &(q-^> p))). What, if anything, is odd about this truth
table?
3.4 Formal interpretation

The specification of the truth-conditions for formulae containing the connectives
completes the interpretation of the logical language LP used to translate the set of
English sentences generated by Gh This informal definition of the interpretation of LP
provides (indirectly) the truth-conditional semantics of a small fragment of English.
We end this chapter by providing a more rigorous definition of the theory set up so
far. This is primarily concerned with the replacement of informal English as the
metalanguage and the definition of models and the theory of interpretation by a more
formal language. The metalanguage that is defined in this section is still partly based
74
Formal interpretation
on English, but is much less open to misinterpretation and provides a properly

rigorous and unambiguous specification of the whole of the interpretation procedure.
Formalisation of this sort is, as in any discipline, an aid to understanding and
shows precisely what the theory being formalised predicts. It is to this task that we
now address ourselves. The notation and the points made in this section are of
necessity somewhat technical, but no concepts are introduced that have not already
been informally presented. Those unused to formal theories should try not to overreact
to the symbolic nature of the formalism. A symbol is not something mystical or
arcane, but a tool to aid analysis and understanding. When presented with a string of
symbols, readers should take care to recall what each symbol is intended to represent
and to assign the symbol the intended interpretation. Unlike words or phrases in
natural languages, each symbol in a formalism is unambiguous and the pedantry of the
metalanguage is designed so that there can be no possibility of ambiguity. The
formalism thus provides an unambiguous andrigorousspecification of what has been
said about truth-conditional meaning in this chapter and in Chapter 2 which allows the
predictions of the theory to be checked.
3.4.1 Formal models

As we have already seen, there are two parts to the theory of interpretation that
we are developing: the construction of models which are the representations of
situations or states of affairs, and the theory of interpretation, called the model theory,
which provides the means of relating all the expressions in some object language to
denotations in these models. As discussed in Chapter 2, a model consists of an
ontology and a function that assigns extensions to basic expressions in the object
language. The ontology is defined as a set of entities, usually symbolised as the set A,
which contains the entities that exist in the state of affairs being modelled. Extensions
are assigned to basic, or constant, expressions of the object language by a function F
which associates each constant expression with some entity in A or some subset of
A or some set of ordered n-tuples in A, depending on the logical category of the
constant expression. Crucially, F is a function, something that assigns a unique
denotation to each constant in the object language. More is said about functions in
Chapter 4, but their importance lies in the fact that they always provide a unique
value (here an extension in the model) for a particular argument (here a basic
expression in the object language, LP). This ensures that the interpretation of some
Lp expression is properly unambiguous. Thus, F always applies to some constant a
in the object language to yield a unique extension, A {delta). If F were not a function,
it would be possible for a constant to be assigned two different extensions. In such a
case, we could not necessarily decide whether a formula involving the
expression was true or not and, therefore, could get no information from it.
For example, if a predicate like scream' denoted two different sets, say 0 and
{MANi,WOMAN2},then a formula like scream '(jo') could be either true or false
depending on which extension is chosen. On a particular occasion of utterance,
however, the sentence Jo screamed expresses a proposition that is either true or false,
but not both, assuming that Jo has a unique extension in the context. Furthermore,
although we may come across homonymous or polysemous expressions in speech,
these are interpreted with the aid of the context as if they were properly unambiguous.
75
In other words, they are treated as denoting a unique extension in the given context.
The fact that F is a function does not, of course, prevent two lexemes from having the
same extension, like two synonyms e.g. mercury and quicksilver. Items that are not
synonymous may also have the same extensions. In the real world, this is true of the
words unicorn and gryphon on the assumption that they denote nothing (i.e. nothing
that exists in our world is either a unicorn or a gryphon). The significance of this will
be examined further in Chapters 9 and 10.
As discussed in Chapter 2, the denotation assignment function must also specify
a denotation of every constant in the object language, even if there is nothing in its
extension. This is because, even if nothing in a model has a certain property or
stands in a certain relation to anything else, we can use expressions with null
extensions to describe some state of affairs and to make true statements using the
negation operator. An expression like Jo threw the cat to Fiona, however, is strictly
both ungrammatical and meaningless in the fragment generated by d , because
threw is not a word in the lexicon of the grammar. Such an expression would be
grammatical, of course, if the ditransitive verb threw were introduced into the lexicon.
If this is done, however, its translation into Lp, throw \ must be included in the
specification of the denotation function assignment in any model used to interpret LP.
Otherwise formulae containing this constant would remain meaningless.
Any model for interpreting expressions in LP can be defined as an ordered pair
consisting of the ontology and the denotation assignment function, i.e. M = «A,F».
The model Mi may thus be formally defined as in (44). This specifies everything that
exists in the model, i.e. the ontology A, and the extensions of all the constants in LP
that are associated with lexemes in Gi. The situation thus represented involves two
men, two women, a dog, a cat, a book and a cake, with the properties and relations
assigned informally in Chapter 2. F t is specified as a set of statements of the form
Fi(a) = a where a is the argument of the function, a constant in LP and a is the
value of the function applied to this argument, i.e. its extension in the model M,. It is
important to remember that the upper case words in (48) and elsewhere represent real
world entities and are not words of English or Lp.
(48) M, = «A,F,», where
a. A = {MANi,WOMAN2,MAN2,WOMANi,DOG,CAT,BOOK,CAKE}
b. F,=
F^the-student') = MAN!,
F,(bertie') = MAN2,
F,(ethel') = WOMAN!,
F,(the-golfer') = WOMANi,
F,(fiona') = WOMAN2,
F^the-singer') = WOMAN2,
F^the-lecturer') = WOMAN2,
F,(the-cat') = CAT,
F,(prudence') = CAT,
F,(chester') = DOG,
F1(the-dog>) = DOG,
F,(trie-book') = BOOK,
76
F,(the-cake') = CAKE,
F,(ran>) = {DOG,CAT}>
F,(laugh') = {MAN!, WOMAN,},
F,(howlf) = {DOG},
F^crazy') = 0 ,
F,(happy') = {WOMAN^WOMAN^MANi} ,
F,(disgusting') = {CAKE},
F^scream') = 0 ,
Fi(messy') = 0 ,
F,(sing')= {WOMAN,},
F,(like') = {<MAN lfWOMANi>, <MAN lf WOMANi>,
<WOMAN!,WOMAN2>, <WOMAN2,CAT>, <MANi,MANi>,
< WOMAN2, WOMAN^, <MAN!,MAN2>}
F,(loathe') = {<MAN,,DOG>, <MAN!,DOG>, <WOMAN!,DOG>,
<WOMAN2,DOG>, <WOMAN 2 ,MAN t >, <CAT,DOG>},
F,(poison') = {<CAKE,DOG>},
F,(eat') = {<DOG,CAKE>},
F,(kick') = 0 ,
F,(read') = {<WOMAN1,BOOK>, <MANi,BOOK>},
F,(give') = {<WOMAN2,CAKE,MAN2>, <MAN!,CAKE,DOG>,
<MAN, ,BOOK,MAN2>, <MAN2,BOOK,WOMAN2>,
<MAN! ,C AT, WOMAN 2 >},
Fi(rain') = 1,
Fi(snow') = 0}.
The last two entries in the specification of F! may seem a bit strange. The
symbols 1 and 0 are not in A and in Chapter 2 the expressions rain' and snow' were
directly associated directly with a truth value, their extensions in the model M, being
true and false, respectively. It is not just these two constants that are defined as having
truth values as their extensions, but all formulae are ultimately assigned a value, true
or false, by the interpretation procedure. What is meant by the terms true and false in
semantics is not, however, quite the same as what is meant by the same words in
English. In order to emphasise this the English words, true and false are replaced in
model-theoretic semantics by two neutral symbols which are traditionally shown as
0 and 1, for falsity and truth, respectively. It does not really matter what these
symbols actually are - they could be a cup of coffee and a biscuit - as long as they
have a specific interpretation in the theory. This interpretation is defined by the
theory of interpretation itself (ie. by the set of truth-conditions): a formula denotes 1
just in case it corresponds to the structure of some model according to the rules of
interpretation, in which case we say it is true, or it denotes 0 if doesn't correspond, in
which case we say it is false. This statement may seem somewhat peculiar and
perhaps even circular, but it is not. True and false are informal (English) terms which
help us remember what we are getting at in our semantic analysis, whereas 0 and
1 are precise (logical) terms, whose interpretation is defined by the theory. For this
latter reason, the symbols 1 and 0 are not included in the ontology of the model, but
are part of the theory of interpretation. If they were members of A, some models
might not contain them and so have no means of representing truth and falsity.
77
However, it is part of the theory that all formulae have determinate truth values and
so it is in the theoretical part of the interpretation procedure that they are defined.
3.4.2 Model theory

We have now given a formal definition of the structure that models have and given
a specification of one such model, M,. The next thing to be done is to provide a
formalisation of the theory of interpretation which defines how formulae are
interpreted with respect to some model, the model theory. The model theory provides
a means of determining the extensions of all expressions in LP, whether simple or
complex. Formally, the denotation of an expression is represented by putting it within
square brackets, [.]. Thus, for example, [the-caf] represents the denotation of
the-caf and [scream'(jo}) v laugh'(fwna')] represents the denotation of the formula
that translates Jo screamed or Fiona laughed. As we have seen, however,
expressions do not have denotations tout court, but only with respect to a
particular model. For this reason, the name of the model being used to interpret the
expression is written as a superscript to the right of the square brackets. For example,
the denotation of laugh'(fiona') with respect to the model, Mi, is written as
[laugh'(fiona')]m and that of the-caf is [the-cat']m.
Formally, the model theory is stated as a recursive definition of the notion of
denotation in a model, i.e. of [oc]M where a is any expression in the object language.
The base of the recursive definition is given by the equation of the model-theoretic
denotation of constants with the extension assigned to that constant by F in the
relevant model (see (49.1)). This means that while the denotation function, F, can
vary from model to model, it always provides the basis of the interpretation of
complex expressions. In other words, the truth value of a sentence is determined by
the denotations of the words it contains. For example, because the expression the-caf
is a constant in LP, its extension with respect to M,, [the-caf ]M\ is the result of
applying F l 5 the denotation assignment function in Mi, to the expression the-caf. By
(48), Vx(the-caf) is CAT, so [the-caf]m is also CAT. From the denotations of the
constants supplied by F in a particular model, the denotations of all other expressions
in the language are built up by a series of recursive clauses which define how to build
up the denotations of complex expressions from the denotations of their parts, in
accordance with the principle of compositionality. Anything not specified by the base
and recursive clauses does not count as a possible denotation in the model.
The recursive clauses state formally the truth-conditions of formulae in L P by
defining what must be the case for a formula with a particular structure to be true in
any model. Thus, the truth-conditions for formulae constructed from an n-place
predicate and n arguments is formally set out in general form in (49.2) and requires
that the n-tuple formed by taking the extensions of each of the arguments in order
to be in the extension of the predicate for the formula to be true. (49.3) states that
an equative formula is true only when the entities denoted by the two individual
constants being equated are one and the same entity. The next four clauses in the
model theory in (49) deal with the interpretations of complex formulae given the
extensions of their component parts: a negative formula is true if the component
formula is false; two formulae must be true for their conjunction by & to be true; at
least one of two formulae must be true for their disjunction by v to be true; for
78
conditionals to be true, either the antecedent must be false or the consequent true;
and two formulae are equivalent only if they have the same truth value. This
completes the formal definition of the model theory used to interpret LP. (49) does
nothing more than formally state the model theory that has been informally presented
in Chapter 2 and the earlier part of this chapter. No new information about
interpretation is introduced.
(49) Given a model, M = «A,F», then
1. For any constant a , [a ] M is F ( a ) .
2. I f / i s an n-place predicate, and ah...,an are individual constants, then
\f(alf...,an)]M is l,iff<[«y]M,...,[«n]M>G [/]M. Otherwise, \f(a,,...,a,)]M is 0.
3. If a and b are individual constants, then [a = b]M is 1 iff [a]M is extensionally
identical to [b]M. Otherwise, [a = b]M is 0.
4. If 0 is a formula, then [~((())]M is 1, iff [(|>]M is 0. Otherwise, [~(<|>)]M is 0.
5. If (|> and \j/ are formulae, then [<> | & \j/]M is 1, iff [c|)]M is 1 and [\|/]M is 1.
Otherwise, [()) & \|/ ] M is 0.
6. If <|) and \\f arc formulae, then [($ v \|/ )] M is 1, iff [<()]M is 1 or [\j/]M is 1.
Otherwise, [(()) v \j/ )]M is 0.
7. If (|> and \\f are formulae, then [((() -» \|/)]M is 1, iff [())]M is 0 or [\|/]M is 1.
Otherwise, [((() -^ \|/)] M is 0.
8. If <|) and \jr are formulae, then [$ <-> \|/] M is 1, iff [c|)]M is [\|/]M. Otherwise,
[ty <-> \\f ] M is 0.
To show how the model theory works, we shall look at the interpretation of the
formula that translates the sentence If Jo was the student, then the lecturer loathed Jo,
(jo'= the-student'—> loathe'(the-lecturer', jo')). To specify the interpretation of this
Lp formula, we begin by defining the truth-conditions of the complex formula
according to the model theory and then find out whether these are met in the model
under consideration. If they are, the formula is true. If not, then it is false. Since the
formula is a conditional, the first step in interpreting it is given by the truth-
conditions for material implication set out in (49.7). In other words, we replace <|) and
\|/ in this clause with the component formulae to get (50).
(50) [(jo' = the-student' -> loathe'(the-lecturer',jo'))] M1 is 1, iff [jo' =
the-student'] M1 is 0 or [loathe'(the-lecturer'jo')] M1 is 1.
It is then necessary to interpret the component formulae to see whether these
truth-conditions are met. The antecedent formula is equative, the truth-conditions of
which are provided by the clause in (49.3). We now come to constants of the language
Lp whose extensions are provided by the model. So, the base clause (49.1) is applied
twice to get the extension of jo' and the-studenf in M,, allowing us to determine the
truth value of the antecedent formula with respect to the model, as shown in (51).
(51) a. [jo' = the-student'] M! is 1 iff [jo']M1 is extensionally identical to
[the-studenf ] M1 (by (49.3)).
b. [jo']M1 = F,(jo') = MAN! (by (49.1) and definition of F, in (48.b)).
c. [the-studenf ] M1 = F,(the-studenf) = MAN! (by (49.1) and definition of F,).
d. Therefore, as MANX is identical to MAN,, [jo = the-studenf ] M1 is 1.
79
Because the antecedent is true with respect to the model, we cannot yet tell
whether the conditional is true and must find out the truth value of the consequent.
This contains a two-place predicate, loathe' and two arguments, the-lecturer' and
jo '.By the clause in (49.2) the truth-conditions of this formula are met if entities
denoted by the-lecturer' and jo' form an ordered pair which is in the extension of
loathe' in M,, as set out in (52). Since this is the case, it follows from (50) that the
whole conditional is true with respect to Mi. Hence, we have indirectly interpreted the
English sentence IfJo was the student, then the lecturer loathed Jo as being true of the
situation modelled by Mi, as required.
(52) a. [loathe'(the-lecturer'jo')]M1 is 1 iff<[the-lecturer']M1,[jo']M1>e [loathe']M1
(by (49.2)).
b. [loathe']"1 = F,(loathe') = {<MAN1,DOG>, <MAN2,DOG>,
<WOMAN1,DOG>, <WOMAN2,DOG>, <WOMAN2,MAN,>,
<CAT,DOG>} (by (49.1) and the definition of F, in (48.b)).
c. [the-lecturer']M1 = F,(the-lecturer') = WOMAN2(by (49.1) and F, in (48.b)).
d. [jo']M1 = F,(jo') = MAN,(by (49.1) and F, in (48.b)).
e. Because <WOMAN2,MAN!> e {<MAN,,DOG>, <MAN2,DOG>,
<WOMAN 1 ,DOG>, <WOMAN2,DOG>, <WOMAN2,MAN1>,
<CAT,DOG>}, [loathe'(the-lecturer',jo')]M1 is 1.
This process, which may seem rather tedious (as indeed it is!), is a rigorous
specification of the truth-conditional meaning of the formula that translates the
sentence IfJo was the student, then the lecturer loathed Jo. Each step is automatic
and the final conclusion is reached just by following the rules set out in (49). Thus,
the model theory makes absolutely explicit how the.interpretation of formulae that
translate sentences generated by G{ is to be carried out. Note that the theory is not
intended to represent the steps that we, as speakers of English, carry out in our heads
when interpreting expressions generated by this small grammar. It does, however,
provide an automatic procedure for determining the truth or falsity of any formula
associated with a sentence in d with respect to the model Mi and thus constitutes
a theory of the truth-conditional meaning of the grammar fragment. (49) thus provides
an explicit theory of semantic interpretation whose precision enables us to check its
predictions and see if they do indeed match up with our intuitions about what English
sentences mean. In the following chapters, it will be shown that the theory in (49) is
not sufficient in itself to provide a full theory of interpretation for English (or any
other natural language), but it does provide a firm foundation for the more adequate
theories to be discussed later.
Exercise 3.6:
Translate the following sentences into LP and determine their truth or falsity with
respect to Mi using the formal method shown in (50) to (52):
i. Jo didn't give the cat to the lecturer.
ii. The dog ate the cat or the dog ate the cake.
iii. If Chester was messy, then the cat didn't loathe the dog.
80
Further reading
3.5 Further reading

The logical operators ~, &, v , -» and <-^ and their truth-conditional interpretations
form the basis of propositional logic and there are many introductions to this system:
see, in particular, Guttenplan (1986: 42-88) (which also discusses translation on pp.
105-124), Allwood, Andersen and Dahl (1977: ch. 4), Lyons (1977:138-150), Partee,
ter Meulen and Wall (1990:99-107), Martin (1987: 8-30) and McCawley (1981: chs.
2 & 3) which gives a discussion of the syntax and semantics of propositional logic and
inference, but uses a different notation that is potentially confusing to the novice.
(Barwise and Etchemendy (1987) offers an introductory text and a computer program
for learning predicate logic on a Macintosh computer.) The formal approach to
interpretation discussed in Section 3.4 is discussed in Dowty, Wall and Peters (1981:
ch. 2) which also touches on matters that will become important in later chapters.
81
4 Type Theory
4.1 Verb phrases and other constituents

One of the conditions of adequacy for a semantic theory set up in Chapter 1 is that it
conform to the Principle of Compositionality. This principle requires the meaning of
a sentence to be derived from the meaning of its parts and the way they are put
together. The interpretation procedure for the grammar fragment set up in the last two
chapters adheres to this principle insofar as the translations of sentences, and thereby
their interpretations, are derived from the translations of their parts and the syntactic
rules used to combine them. Thus, for example, the translation of the sentence Ethel
kicked the student is derived from the translations of the two noun phrases Ethel and
the student and the verb kicked. These are combined using the translation rule for
transitive sentences to give kick'(etheV,the-student'). The truth or falsity of the
resulting formula can then be directly ascertained by checking whether the ordered
pair of entities denoted by the subject and object in that order is in the set of ordered
pairs denoted by the predicate, kick'.
Unfortunately, in the theory of Chapters 2 and 3, compositionality is maintained
only at the expense of the syntax. The 'flat' structure of the predicate-argument syntax
of Lp and its interpretation requires a flat sentence structure in the English syntax in
order to maintain a direct correspondence between syntax and translation, and thus a
transparent relation between elements in the interpretation and constituents of the
English sentence. There is, however, strong syntactic evidence that many languages,
including English, do not have a flat sentence structure, but one that contains a number
of hierarchically defined constituents. In particular, English sentences are usually
analysed as containing verb phrases consisting of the verb and its complements and
modifiers. According to the principle of compositionality, all constituents within a
sentence must have a well-formed translation and thus be assigned a denotation by the
model. Hence, if the sentence Ethel kicked the student were analysed as containing a
verb phrase, a properly compositional semantics would need to assign a logical
representation and denotation to the phrase kicked the student. It is not too difficult to
see what sort of thing such a phrase denotes: verb phrases denote the same things as
intransitive verbs, i.e. sets of entities, because the simplest verb phrases consist only
of a single intransitive verb. Hence, in the same way that the verb ran may be
associated with a predicate run' denoting the set of things that have the property of
running, so the verb phrase kicked the student ought to be associated with some
predicate denoting the set of all entities that kicked the student. In formal terms, the
extension of this verb phrase should be derived from the extension of predicate
representing the verb kick' by taking the first element of each of the ordered pairs in
this set where the second element of the pair is the entity denoted by the direct
object, [the-student']M in this case. There is, however, no expression in LP that could
be assigned such an interpretation. It is not possible, for example, to represent verb
phrases by leaving out the subject expression from a formula. Expressions like
kick'(the-student') or kick'(_,the-student') are not well-formed expressions of LP
because they do not conform to the construction rule for formulae given in Chapter
82
A typed logical language
2, repeated in (1), below, which requires n-place predicates to have n arguments to

make a formula. The expression kick' is a two-place predicate, and so must combine
with two individual expressions to make a well-formed formula.
(1) t = Pred n (e\...,e n ).
There are also, of course, many other constituents that the logical language L P
cannot represent. For example, adverbs like slowly or prepositional phrases like in the
garden have no obvious representation or interpretation within the theory of the
previous chapters. The word slowly in the sentence Chester ran slowly cannot be
treated as a simple predicate because such expressions do not define properties of
individuals. It makes no sense to translate Chester ran slowly as run'(chester') &
slowly'(chester') because the idea of Chester 'being slowly' is not coherent. Even if
the adverb were translated as if it were equivalent to its adjectival base form to give
run '(chester') & slow '(chester'), this would not provide an adequate interpretation for
Chester ran slowly. This is because it is Chester's running that is described as slow,
not Chester himself. It is possible that Chester's slow running is still fast and thus to
predicate slow of Chester would not capture the meaning intended to be conveyed by
the use of the adverb. Adverbs and prepositional phrases (in some of their uses)
predicate properties, not of individuals, but of predicates themselves. Since LP only
contains expressions that are predicates, individuals or connectives, it is not capable
of representing higher order concepts like predicates of predicates.
4.2 A typed logical language

There is no way to represent verb phrases or verb phrase modifiers in LP in such a
way as to give their intuitively correct denotations. A new logical language is
therefore needed to represent more adequately the semantic structure of natural
languages like English. This logical language must have more flexibility than L P and,
in particular, it should have a more complex system of semantic categories, to allow
formulae to be given more structure in order to better represent the rich constituent
structure of natural languages. From now on, we will distinguish logical semantic
categories, like e and t, from syntactic ones, like NP, S, etc., by referring to the former
as semantic types (or just types). The theory behind them is referred to as Type
Theory and the new logical language is called Ltype. Although L ^ is a very different
language from L? in many ways, it contains many of its features. In particular, it
contains the propositional operators and the notion of predicate/argument structure,
albeit in a slightly modified form. The interpretation of the language is also similar in
that it is based on set theory, although again this is somewhat modified. We now turn
to the definition of the syntax of the language Ltype and see how translation into this
language maintains the constituent structure of a sentence.
4.2.1 Semantic types

In Lp, there are only five logical types: individual constant (e), formula (t), one-place
connective (i.e. ~), two-place connective (i.e. &, v , —> , <->) and n-place predicate
(Predn). These types are all primitive in LP, i.e. they have no internal structure and
cannot be further decomposed. It is possible, however, to define them in such a way
83
4 Type theory
that only two of the types are primitive and the other three are defined in terms of
these. In order to do this, we must consider a different way of viewing the construction
of logical formulae.
The main formula construction rule in (1) is a rule schema that collapses all the
rules in (2) and (infinitely) many more.
(2) a. t -> Predo.
b. t -> Pred,(e).
c. t -> Pred2(e1,e2).
d. t ->
In most phrase structure grammars, the combinatorial properties of expressions of
particular categories, like e and Predn in (2) (or NP and S in the grammar fragment
d ) , have to be inferred from the set of rules that govern their distribution. For
example, from the grammar fragment G h we can extract the information that transitive
verbs combine with two noun phrases to make a sentence or that a conjunction
combines with two sentences to make another sentence. In the same way, we can
think of a two-place predicate as an expression that combines with two individual
constants to give a formula, or a connective like & as an expression that combines
with two formulae to make a complex formula. Such information, however, instead
of being encoded within a set of rules, could as well be directly encoded in syntactic
categories or semantic types. For example, instead of treating a two-place predicate
like kick'as having a primitive type, Pred2, we may give it a complex type which itself
contains the information that the expression combines with two individuals to make
a formula. Such complex types thus provide information about the number and types
of the arguments with which particular expressions combine and the type of the
expression that results after the combination. Hence, the type of an expression directly
encodes what other types of expression it can combine with and we only need to know
exactly how the combination is effected to perform the required operation. It is as if
we have an equation like that in (3) where we need to replace the e's with particular
individual expressions (like etheV and the-student') and to know what the combination
operation x means to get to the resulting expression which is a formula of type t.
(3) kick' x e x e = t.
Generalising this idea and making it more formal, we can define non-primitive
types as ordered pairs of types. The first element of the pair indicates the required
type of the expression which expressions of the complex type combine with, the input
type and the second element gives the type of the resulting expression, the output
type. Thus, a one-place predicate, like scream\ which combines with an individual
expression (an expression of type e) to give a formula (an expression of type t), has
the complex type <e,t>. Similarly, the negation operator, ~, which combines with a
formula to make a formula that has the complex type <t,t>.
(4) Complex types: <input type,output typo.
What, however, of expressions that combine with more than one argument, like
n-place predicates and two-place connectives? It would be possible to allow types to
have an ordered list of types as their first element, instead of just a single type. This
list would give the types of all the expressions with which another expression
84
combines and the order in which they are combined. On this definition, a two-place
predicate would have two instances of type e as its input list and t as the type of the
output, i.e. have a type <[e,e],t>. A two-place connective, like &, would have two
instances oft as its input and t again as its output, i.e. <[t,t],t> and so on. It would
then be a simple matter to state a rule that combines an expression with a complex
type with expressions of all of its input types at the same time: a two-place predicate
would combine with two expressions of type e to give a formula (as in the equation
in (3)), and a two-place connective would combine with two formulae to give a
formula and so on. This approach would, however, maintain the basic syntax of Lp,
since it combines a predicate expression with its argument expressions all in one go.
As such it is subject to the same problems in defining representations for constituents
like VP as Lp itself.
Instead of defining complex types as having an ordered list as the first element,
however, it is possible to restrict the number of input types to one, allowing only
binary types and so only binary combination of logical expressions into more
complex expressions. This may seem far too strict. If complex types only have one
possible input type, how can n-place predicates be defined? The answer is quite
simple. The types for one-place predicates, <e,t>, and the negation operator, ~, <t,t>
have a primitive type, t, as its output type. The output type could, however, be
complex, not primitive. For example, an expression could combine with an individual
constant to give an expression that itself requires an individual constant in order to
make a formula. The type of such an expression would therefore be <e,<e,t» and,
since an expression of this type ultimately requires two individual expressions to make
a formula, we may associate this type with two-place predicates. The combination of
such expressions with their arguments would not be a one stage process as shown in
(3) but a two stage process as shown in (5). First, the predicate is combined with one
expression of type e to make an expression of type <e,t> as in (5.a) and then this
expression is combined with another expression of type e to yield an expression of
type t, i.e. a formula.
(5) a. <e,<e,t» x e = <e,t>.
b. <e,t> x e = t.
It is not only the result types that can be complex but argument ones as
well. For example, an expression may combine with a one-place predicate to give an
expression of the same type. As we have seen, one-place predicates have the type
<e,t> because they combine with an individual to make a formula. A predicate
modifier would, therefore, have the type «e,t>,<e,t», giving the type of a VP
adverb, like slowly. In fact, the theory imposes no constraints on the complexity of
semantic types, beyond the fact that types must be binary: there must be only one
input type and one output type. Given this lack of constraint, there is an infinite
number of types which must, therefore, be defined recursively. The base of the
recursion is given by the statement of what constitutes a primitive type, i.e. individuals
and formulae, and the recursive clause defines complex types based on these, as fully
specified in (6).
85
4 Type theory
(6) Types for Ltype:

a. e is a type,
b. t is a type,
c. If a is a type and b is a type, then <a,b> is a type.
From (6), we can construct types like <e,t>, and <t,t>. From the latter we can
construct types like «e,t>,<e,t», «e,t>,t>, <t,<t,t». From these we can construct
more complex types like «e,t>,«e,t>,t», <«e,t>,t>,«e,t>,<e,t>» and so on. All
the types in (7.a) are thus well-formed according to the definition in (6) but those in
(7.b) are not.
(7) a.
b. <e,e,t>, <e,<oc,t», «e,e,e>,t>, <Predn>.
Now that the system of types has been given, the syntactic rules that combine
expressions with different types must be specified in order to complete the definition
of the language L ^ . One of the advantages of encoding combinatorial properties
directly in the types themselves is that it is not necessary to state individual rules for
combining expressions of different types. Instead, only one syntactic rule needs to be
stated. Because types are binary, the rule combines at most (and at least) two
expressions. The rule of functional application (or RFA for short), defined in (8),
combines an expression with type, <a,b>, with an expression of type a to yield an
expression of type b. The expression with the type <a,b> is called the functor and the
expression with which it combines is called the argument. The result of applying the
functor to its argument is obtained by writing the functor to the left of the argument,
which is enclosed in parentheses. So, in general, if/is the functor expression and a is
the argument, then the functional application of/to a is f(a) (which is often read as
/applied to a or justfofa).
(8) Rule of functional application (RFA): If/ is an expression of type <a,b>,
and a an expression of type a, then f(a) is an expression of type b.
The rule of functional application can be thought of as performing some sort
of logical cancellation analogous to the sort of cancelling that goes on in arithmetical
multiplication. If the cancelling yields a single result at the end, then the combination
of expressions is well-formed. If no cancelling takes place, however, or if it is
anomalous, then the resulting expression is ungrammatical in L ^ . As an arithmetical
illustration, consider the simple multiplication in (9.a). The result of this
multiplication is determined by cancelling out the identical multiplier and divisor to
leave the unique uncancelled dividend as the result. In (9.b), on the other hand,
because multiplier and divisor do not cancel each other out, the result is not equal to
the dividend but must be computed.
(9) a. 2/4x4 = 2.
b. 2/4 x 3 * 2.
Similarly, to get the type of two expressions when they are combined by RFA, the
type of the argument is cancelled with the identical first element of the type of the
functor, as illustrated schematically in (lO.a). The type of the resulting expression is
given by the (uncancelled) output type of the functor category. If no cancellation can
86
take place, because the argument type and the first element of the functor's type are
not identical, then the result of the combination has no type in L ^ and so is not
a well-formed expression in the language, e.g. (lO.b). To see whether a combination
of expressions is well-formed, the types of the functor and the argument need to be
checked: if the input type of the functor can be cancelled with the type of the argument
expression then the functor can be applied to the argument. If not, then there is no
well-formed expression in L ^ that combines just those two expressions.
(10) a. f + a = f(a).
<e,t> x e =t
b. g + a *g(a).
<t,t> x e
As a more concrete example, consider how the three expressions ~, scream' and
bertie* may be combined. In (11 .a), the functor expression, scream' of type <e,t>, is
applied to the argument, bertie\ of type e, to get an expression of type t (a formula).
The negation operator (type <t,t>) can now be applied to this expression to get the
resulting formula in (ll.b). No other combination of these expressions yields a
well-formed expression of L ^ , as illustrated in (1 l.c) and (1 l.d).
(11) a. scream' + bertie' = scream'(bertie').
<e,t> x e = t
b. - + scream'(bertie' ) = -(scream'(bertie')).
<t,t> x t = t
c. ~ + scream'.
<t,t> x <e,t>
d. ~ + bertie'.
<t,t> + e
The formula in (ll.b) looks the same as an equivalent combination of these three
expressions in LP. However, this is only true of a very restricted number of formulae
in Ltype. As we will see in Section 4.2.2, formulae containing two-place predicates look
rather different from their LP counterparts and contain in particular a subexpression
that corresponds to a verb phrase in the syntax of the English sentence being
translated. It is to the matter of translating English into L ^ that we now turn.
Exercise 4.1:
Given the following abstract expressions with their types, state whether the complex
expressions that follow are well-formed according to RFA. If they are, what are the
types of the resulting expressions?
f is an expression of type <e,t>. a is an expression of type <e>.
h is an expression of type <t,t>. j is an expression of type «e,t>,t>.
g is an expression of type « e , t > , < e , t » .
i. f(a) ii. g(f) iii. g(a)
iv. h(f) v. j(f) vi.
87
4 Type theory
4.2.2 Translating verb phrases

To see how the language defined in Section 4.2.1 is able to represent the translations
of verb phrases, consider how the sentence Ethel kicked the student might be
translated into Ly^. In a grammar of English descriptively more adequate than Gu
there will be a rule that combines a transitive verb, e.g. kicked, with a noun phrase,
e.g. the student, to give a verb phrase, kicked the student. L ^ contains a parallel
means of combining two-place predicates, like kick' with its arguments. In Ltype a
two-place predicate is analysed as having a type that requires an individual expression
to yield an expression that needs an individual expression to make a formula, i.e.
<e,<e,t», as has already been seen above. An expression of such a type, like kick\
will combine, according to the RFA, with an individual expression like the-studenf
to give an expression of type <e,t>, i.e. kick1 (the-studenf). The parallelism between
the syntactic construction of a verb phrase in English and the combination of a
two-place functor with an argument in L ^ is shown in (12). (12.a) gives the English
expressions that are combined according to the (backwards) phrase structure rule in
(12.b), while in (12.c), the L ^ counterparts are combined according to the RFA
shown by the valid 'cancellation' in (12.d).
(12) a. kicked + the student = kicked the student.
b. Vt NP <r- VR
c. kick' + the-student' => kick'(the-student').
d. <e,<e,t» x e = <e,t>.
Similarly, the construction of the sentence Ethel kicked the student from the verb
phrase in (12.a) and the noun phrase Ethel is paralleled by the functional application
of the one-place predicate in (12.c) with an individual constant, ethel\ as shown in
(13).
(13) a. Ethel + kicked the student = Ethel kicked the student.
b. NP VP <- S.
c. ethel' + kick'(the-student') =» (kick'(the-studenf ))(ethel').
d. e x <e,t> = t.
Notice that the translation of the sentence into the formula in L ^ given in (13.c)
is not equivalent to the translation of the same sentence into LP. Translation into the
latter would result in the expression kick'(ethel\the-studenf). Here the translation
of the syntactic subject appears as the first argument of the predicate and that of the
(direct) object as the second. In the 1 ^ expression, on the other hand, this order is
reversed. The translation of the object is combined with the predicate before the
subject and so appears to the left of the translation of the latter. This better accords
with traditional grammar and with traditional semantic notions of subject and
predicate which both include the direct object as part of the predicate. It is possible,
following Dowty (1982), to identify the positions of the translations of noun phrases
in a formula in Ltype that translates some English sentence with the grammatical
functions those noun phrases perform in that sentence. In general, the more oblique
the grammatical function of the noun phrase the closer its translation appears to the
translation of the verb. Thus, the translation of an indirect object (if present) appears
next to the predicate, being combined first. The translation of the direct object comes
88
next, while the translations of syntactic subjects are always combined last of all. (14)
gives a schematic representation of the relative order of the translations of noun
phrases performing the three functions within some sentence. Obviously, if not all
three functions are represented in a sentence, the relative ordering remains the same:
the translation of the object always precedes that of the subject in the
functor-argument structure of the predicate. If this is borne in mind while
translating sentences of English into Ltype, the trap of mimicking the order of
arguments in L P translations can be avoided.
(14) (((verb' (indirect-object* ))(object' ))(subject')).
Let us now put the procedure for translating from English into L ^ on a firmer
footing by specifying the new grammar fragment and stating its translation rules. As
in the previous chapters, the new grammar, G2, is defined in terms of phrase structure
rules with the translation rules operating in parallel to the syntactic ones, in
accordance with the rule-to-rule hypothesis. The grammar G2 has much the same
coverage as G] of Chapters 2 and 3, but with a more adequate constituent analysis.
Discussion and specification of rules for accounting for passive, copular and
co-ordination constructions will, however, be left until Chapter 5.
Since it is now possible to have verb phrase constituents, the basic finite sentence
of English can be constructed by the familiar rule in (15.a), the first rule of the new
grammar G2. Verb phrase rules can also be stated in their more usual form and
intransitive, transitive and ditransitive VPs are analysed by the rules in (15.b) to
(15.e). The variable a in these rules ranges over the values + and - so that these rules
generate finite ([+FIN]) and non-finite ([-FIN]) verb phrases. The NP, PP and
impersonal sentence rules are the same in G2 as in G, and are repeated in (16).
(15) a. Rule 1G2: S ->NP VP[+FIN].
b. Rule 2G,: VP[ocFIN] -> VJocFIN].
c. Rule3G 2 : VP[aFIN] -> V t [aFIN] NP.
d. Rule 4G2: VP[aFIN] -> Vdt[aFIN] NP, NP2.
e. Rule 5G2: VP[aFIN] -> Vdt[aFIN] NP PPto.
(16) a. Rule6G 2 : NP->N p r .
b. Rule7G 2 : N P - K h e N .
c. Rule 8G2: PPto -» to NP.
g. Rule 9G2: S —> it V0[+FIN].
As mentioned above, conjunction rules will be omitted for the time being, but,
at the cost of compositionality, we can continue to include one of the negation rules
of Gi in G2 as (17.a). The category Neg which is rewritten as it is not the case that,
cf. (17.b), is treated as a single unanalysed constituent. The other negation rules of Gi
can be collapsed into the single rule in (17.c). This introduces the negated auxiliary
didn't followed by a non-finite VP. Because the VP rules given above can be
expanded as finite or non-finite, nothing else is required to generate negative
sentences.
(17) a. Rule 10G 2 :S-> Neg S.
b. Neg —> {it is not the case that}.
c. Rule 11G2: S -> NP didn't VP[-FIN].
89
4 Type theory
In order to begin translating English expressions generated by G2 into L ^ , the

syntactic categories of G2 must first be linked up with corresponding logical types.
This is done by a type assignment function (often also called a type mapping) that
associates each syntactic category with a unique type. The reason that each syntactic
category must be assigned only a single type is to ensure that no ambiguity is present
in the logical translation and that each syntactic rule is paired with a unique translation
rule. Although each syntactic category must be associated with a unique type, it is not
the case that that type is necessarily unique to that category, as can be seen from (18),
below. The type assignment function is called TYPE; its domain is the set of
syntactic categories and its range is the set of types. (See Section 4.3 for a definition
of these terms.) The part of this function that assigns types to phrasal categories is
given in (18). Notice that although there are six categories they only map into four
types, since both NP and PP,O translate into expressions with the type of an individual,
i.e. e, as in d .
(18) a. TYPE(S) = t.
b. TYPE(VP) = <e,t>.
c. TYPE(NP) = e.
d. TYPE(PPto) = e.
Types for lexical categories are slightly more interesting. Vf has the same type as
VP, i.e. that of a one-place predicate, <e,t>. The type for a proper name remains that
of an individual constant and that for an impersonal verb is t, but the types of the other
three categories, Neg, Vdt and Vt, are more complex. Transitive verbs, as we have
seen, translate into expressions with a type whose argument type is an individual
constant and whose resultant type is that of a one-place predicate (19.b). Ditransitive
verbs have a type that takes an individual constant to yield something of the type of
a transitive verb (19.c). The negation operator, ~, has the type <t,t> because it
combines with a formula to give another formula. These lexical types are summarised
in (19). (The type of a common noun, N, will be ignored until Chapter 6.)
(19) a. TYPE(Vi) = <e,t>.
b. TYPE(Vt) =
c. TYPE(Vdt) =
d. TYPE(Vo) = t.
e. TYPE(Npr) = e.
f. TYPE(Neg) = <t,t>.
The translation rules for G2 are of two sorts. In the first place, there are rules
where there is only a single category (as in Rules 2G2 and 6G2) or where there is a
meaningless element plus one other category (as with the preposition to in Rule 8G2
and dummy it in rule 9G2) on the righthand side of the rule. In all of these cases, the
translation of the phrase is identical to that of the meaningful category, cf. (2O.a),
(2O.b) and (2O.c). The rule introducing definite NPs, 7G2, is given the
non-compositional translation in (2O.d) which is identical to its counterpart in G,. This
is for expedience only and will be rectified in Chapter 6 when we look in more detail
at the semantics of noun phrases.
90
(20) a. T2G2: W.
b. T6G2: Npr\
c. T9G2: Vo\
d. T7G2: the-N'.
In the other cases, the translation is given by the functional application
of the translation of a category with a functor type over the translation of one (or
more) expressions of the argument types. Thus, the translation of the transitive VP
rule in (15.c) is provided by applying the translation of the verb, of type < e , < e , t » to
that of the NP of type e, to give an expression of the correct type for a VP, i.e. <e,t>.
One thing that becomes apparent with the translation of the ditransitive VP rules
is that Ltype, instead of being too flat like LP, appears, from a syntactic point of view,
to be too structured. This is because the logical language constructs complex
expressions only through binary combination while VPs, in the analysis of English
at any rate, are usually given a flatter structure. This is not, however, a major
problem and too much structure is preferable to too little (provided that it is the correct
structure, of course). For our own purposes, it is a simple matter to state for both of
the ditransitive VPs the order of combination of the objects. This, of course, follows
the pattern set out in (14) above, but in the double object construction in (21.c) the
NP immediately following the verb is interpreted as its indirect object. This allows the
paraphrase relationship between sentences like Bertie gave Fiona the book and Bertie
gave the book to Fiona to be captured directly within the translation rules. The lexicon
of d may be carried over to G2 and lexical items are translated in the same way as
outlined in Chapter 2.
(21) a. TIG,: VP'(NP').

b. T3G2: V.'(NP').
c. T4G2: V dt '(NP,')(NP 2 f ).
d. T5G2: V dt '(PP to ')(NP').
e. T10G2: NEG'(S').
f. T11G 2 :~(VP'(NP')).
As before, the translation of a sentence proceeds via a syntactic derivation
using the rule-to-rule translation algorithm of Chapter 2. Thus, every time a syntactic
rule is applied expanding a syntactic category, the parallel translation rule also applies
to give the corresponding translation. An example derivation of the sentence Fiona
didn 'tgive Chester the cake appears in (22) where the numbers on the righthand side
of the derivation indicate the syntactic and translation rule used in each step. The
derivation in (22) has a corresponding phrase structure analysis which appears in
(23). In this analysis tree, each node is labelled with a syntactic category, its type
and a translation constructed from the translations of the nodes it dominates.
91
4 Type theory
(22) a. S
b. NP, didn't VP[-FIN] 11G2
c. Npr didn't VP[-FIN] 6G2

' /
d. Npr didn't Vdt[-FIN] NP 2 NP 3 4G2
=> ~((V dt '(NP 2 ')(NP 3 '))(N pr ')).
e. Npr didn't Vdt[-FIN] Npr NP3 6G2
=> ~((V dt '(N pr ')(NP 3 '))(N pr ')).
f. Npr didn't Vdt[-FIN] Npr the N 7G2
=> ~((V dt '(N pr ')(the-N'))(N pr ')).
g. Fiona didn't give Chester the cake Lex
=> ~((give(chester')(the-cake'))(fiona')).
(23)
S
t
-((give' (Chester' )(the-cake' ))(fiona'))
VP
<e,t>
(give' (Chester' )(the-cake'))
Fiona didn't give Chester the cake
Exercise 4.2:
Draw analysis trees, i.e. phrase structures trees annotated with the translations and
types of each node as illustrated in (23), for the following sentences:
i. It is not the case that Chester liked Jo.
ii. The cat ate the cake.
in. Ethel didn't scream.
92
More set theory
4.3 More set theory

As (23) shows, the adoption of L^. as the translation language allows us to provide
a much better syntactic analysis of (a subset of) English sentences. In particular, of
course, our syntax now contains a verb phrase constituent that has a well-formed
translation, one that is of the same type as an intransitive verb, i.e. <e,t>. As these are
desirable properties according to the discussion in Section 4.1, it seems that our
original problem is solved, since we now have a means of representing VPs in the
translation language. This is not sufficient, however, as we still need to provide an
interpretation of these constituents that conforms to our intuitions about the meaning
of the sentences in which they appear. Before this can be done, however, we need to
take another look at sets and relations and to discuss further the notion of function
introduced in Chapter 3.
4.3.1 Relations and functions

As we saw in Chapter 2, sets consist of collections of elements called its members
which may be single items (entities), ordered n-tuples or even sets themselves. Sets
that contain ordered pairs of elements are said to specify relations which have so far
been represented by sets of ordered pairs consisting of <(extension of)
subject,(ex tension of) object>. Thus, the relation denoted by loathe' in the model
Mi is represented by the set in (24).
(24) {<MANbDOG>, <WOMAN!,DOG>, <WOMAN2,DOG>,
<WOMAN2,MANi>, <CAT,DOG>}.
This relation could, however, also be represented diagrammatically by drawing arrows
between individual elements showing that the first element bears a particular relation
(e.g. of loathing) to the second. This is shown in the diagram in (25) where X
represents the set of loathers and Y the set of the loathed. The arrows show who
loathed whom.
(25)
X Y
MANj. ^rMAN1
MAN2 ^ v . ^ ^ ^ MAN 2
WOMANX>^^><^ WOMANX
WOM AN2 < C ^ \N . WOMAN2
:
CAT _ 2 l ^ ^ £ ^ ^ CAT
DOG
BOOK BOOK
CAKE CAKE
93
4 Type theory
In (25), both thefirstand second elements of the relation are taken from the same set,
i.e. the set, A, of entities in the model, M b but this is, of course, not necessary. A
relation can be stated between the elements of any two sets, even sets that have
complex members like ordered n-tuples or other sets. For example, we could represent
the relation being loathed by in Mj as a relation between entities and the sets of
entities that loathe them, as in (26). In talking about relations, we refer to the set of
first elements of the ordered pairs in the relation (X in (25) and (26)) as the domain
of the relation and the set of the second elements (Y in (25) and (26)) as its range. In
(25), the set A forms both the domain and the range while in (26) it forms the domain
only and the range is a set of subsets of A.
(26)
X (domain) Y (range)
{CATÔMAN! ,MAN1,MAN2}
{WOMAN2}
Certain relations have a special property: they assign every element in the
domain to a single element in the range. Relations of this sort are called functions
and are very important in logic and mathematics. Their importance has to do with the
fact that they are complete (every element in the domain is assigned some value by the
relation) and unambiguous (each element in the domain is assigned one and only one
element in the range). Thus, functions, unlike relations in general, admit of no gaps
in the domain and provide one and only one value for each such element. In (27), four
relations between two sets are represented. The domain contains four letters of the
alphabet and the range contains the first five positive whole numbers. Of these
relations, those in (27.a) and (27.b) both represent functions, but those in (27.c) and
(27.d) do not. (27.c) does not assign a value in the range to every element in the
domain, while (27.d) assigns some members of the domain two values in the range.
< 27 > a. b.
94
More set theory
c. d.
There is some terminology that is associated with mathematical functions that

should be learnt at this point. In the first place, assuming that F is a function with
domain D and range R, F is said to map D to R Secondly, if a is an element in the
domain D, then F applied to a is written as F(a) and read as *F of a1, a is then said to
be the argument of F. Thirdly, if a is an element in the domain D and b is an element
in the range R, and F assigns b to a then b is called the value of F applied to a
(written F(a) = b). Fourthly, if a is an element in the domain D and b is an element
in the range R, and F assigns b to a then we say that F maps a onto b. Finally, if the
domain and range are the same, i.e. D, then F is said to be an operation on D.
The two characteristic properties of functions, that of providing a value for every
element in the domain and ensuring that that value is unique, make them very useful
in linguistics, and especially in semantics. We have already come across things that
can be modelled by a function. For example, in the truth-conditional semantics we
have been developing, every formula is either true or false and cannot be both.
Hence, we could represent our model theory as a function from formulae (the
domain) to truth values (the range) with respect to some model. A very small part
of this function is given in (28) for the formulae that translate sentences in G, with
respect to Mi.
(28)
laugh'(ethel')
run'(the-dog')
scream'(Chester')
~(eat' (the-dog' ,the-cake
poison'(jo',the-cat')
Another example of a function that has already been introduced is the denotation
assignment function, F, used in the specification of models. As we have seen, this is
a function from basic expressions to some element or set of n-tuples of elements in A,
the set of entities in the model. As discussed in Chapter 3, F assigns every constant
in the object language a unique denotation in the model, since otherwise sentences
containing words whose translations have no denotation would be meaningless.
Furthermore, F is required not to assign more than one denotation to some lexeme, as
this would make the translation language semantically ambiguous. Since both of these
properties are properties of functions, denotation assignment is best treated as a
95
4 Type theory
function of the appropriate sort. In fact, most aspects of the interpretation procedure
of Chapter 3 can be reformulated in terms of functions over some appropriate domain
and from now on functions start to play an increasingly important part in the theory.
In particular, the interpretation of L ^ is defined in terms of functions and it is to a
discussion of the model theory required to interpret this language that we now turn.
4.3.2 Sets again

The last piece of groundwork that needs to be done before we can re-examine the
model theory involves a re-characterisation of sets. Until now we have been
describing sets by listing their members. It is possible, however, to describe them
using a function that picks out just those entities that are members of the set and
rejects anything else. In other words, given each element of some domain, a function
that defines a set gives yes or no as a value according to whether the element is in the
set or not. As an example, consider the set assigned as the extension of the expression
run' in M,, i.e. {DOG,CAT}. This can be represented as the function given in (29),
which maps DOG and CAT onto yes and everything else in the domain (i.e. A -
{DOG,CAT})onto/io.
(29) [run']*1
no
Functions that map all the elements in a domain onto one of two values, e.g. yes and
no, are called characteristic functions of sets of elements of the domain, because
they characterise the sets they are associated with. Formally, the characteristic
function of a set, Z, can be defined as a function from a domain, A, into the set {0,1}
(the truth values) so that every element in Z is mapped onto 1 (i.e. yes) and everything
else is mapped onto 0 (i.e. no). Thus, every subset, Z, of a set of individuals, A,
can be associated with a characteristic function, fi, which has the following formal
definition:
(30) Characteristic function of a set: For all a e A, fz(a) = 1, if a e Z ,
andfi(a) = 0, i f a g Z .
The characteristic function, fi, of a set, therefore, divides the domain A into two
subsets of A: the set Z and every element that is in A but not in Z_. The latter set is
called the complement of Z relative to A. This is often written as Zand is equivalent
to A - Z , as shown in the diagram in (31).
96
More set theory
(31) A set and its complement
The diagram in (29) above represents the characteristic function of the set
{DOG,CAT} with respect to M, and (32) provides two more illustrations of the
concept. (32.a) represents the characteristic function, fN, of the set N, where N is the
set of even numbers with respect to the set consisting of the first nine positive whole
numbers, i.e. N = {2,4,6,8}. (32.b) represents the characteristic function of the set, B,
of bilabial consonants in the set {b, d, g, t, k, p, v, (3, 0 }.
(32)
a. fN b. fiB
97
4 Type theory
Exercise 4.3:
Draw diagrams representing the characteristic functions of the following sets in the
manner of (30) and (32) above:
i. the set of all square numbers in the positive integers between 1 and 20.
ii. the set {a,e,i,o,u} with respect to the Roman alphabet.
iii. the extension of happy' and scream' in the model Mj with respect to A.
Instead of talking about a single function with a particular domain and range, we
often need to refer to all the functions that have the same domain and the same range.
For example, it is often necessary to refer to all possible subsets of the set of entities,
A, in some model, but we do not want to have to laboriously write out all the actual
subsets of A, (e.g. in M,, 0 , {DOG}, {CAT,DOG}, {DOGJVIAN!}, etc.), or to
draw diagrams representing all the characteristic functions of these sets. There is a
way of referring to a set of functions which all have the same domain and range: the
name of the set defining the domain is written as a righthand superscript to that of the
set defining the range. Hence, the set of all functions with domain A and range B is
written as B A . For example, the characteristic functions of all subsets of A have A as
their domain and {0,1} as their range and so we can refer to all such functions as
{0,1}*. This is an important piece of notation that will be used to a considerable
extent in the following chapters, so it is necessary to learn its significance.
(33) RANGE°°MAIN =def the set of all functions from DOMAIN to RANGE.
4.4 Interpreting LtyPe

We are now in a position to provide an interpretation for L ^ and begin by looking
at the sorts of denotations that expressions of particular types have in a model. The
model theory of Chapter 3 will then be revised to take the revisions in denotation and
the syntax of expressions into account.
4.4.1 Denotation
As discussed in Chapter 2, the extension of an expression is the entity, property or
relation that the expression is associated with in some model. Although it is not useful
to specify the actual denotation of some expression in every possible model, it is
useful to define the sort of denotation it should have in any possible model. The
specification of the sorts of denotations associated with expressions of different types
was done in Lp by directly assigning possible denotations to the types. For example,
one-place predicates were defined as denoting subsets of A, two-place predicates as
denoting sets of ordered pairs of A, and so on. Because L ^ contains an infinite
number of different types, however, we cannot list the types and the sorts of
denotation expressions of that type should have in some model, but must provide a
recursive definition of such denotations. To refer to the set of possible denotations that
expressions of some type, T , can have, we use the symbol Dv which is defined in a
98
Interpreting L,ype
way that is parallel to the definitions given to the types themselves. Firstly, the sort
of denotations assigned to expressions of the primitive types, e and t, is given and then
a general rule is specified for determining the sorts of denotations expressions of
complex types have.
The basis for the interpretation of Ltype is the same as for the LP: i.e. a set A of
entities given by a model and the set of truth values, {0,1}- As before, individual
constants, i.e. expressions of type e, denote members of A and formulae, i.e.
expressions of type t, denote truth values. In other words, the set of possible
denotations for expressions of type e, De, is the set A and the set of possible
denotations for expressions of type t, D t , is {0,1}. This provides the base of the
recursive definition of D ^ , the sets of possible denotations of expressions in LryPe. The
denotations of expressions of complex types are defined in terms of the possible
denotations of their input types and those of their output types. Because L ^ is an
unambiguous language, the denotation of an expression must be properly and uniquely
determined within a model. Furthermore, the Principle of Compositionality requires
the denotation of a composite expression to be determined by the denotations of its
parts. Hence, because L ^ allows only binary combination of expressions, consisting
of a functor and a single argument, the functor must denote a function that when
applied to its argument yields a unique value, the denotation of the complex
expression. Since functors must, by definition, combine with an expression of a
particular type to yield an expression of another type, they must denote functions from
the denotations of the type of their arguments to denotations of the types of expression
that result after combination. For example, one-place predicates in L ^ have the type
<e,t>, as we have already seen. Thus, they combine with expressions of type e to yield
expressions of type t. Hence, by the above reasoning, expressions of type <e,t> must
denote functions from domain De to range Dt. As we saw at the end of Section 4.3, the
set of all functions from a domain A to a range B is written B A , so that we can define
D<et> as DtDe, the set of all functions from De to Dt, as shown schematically in (34).
(34) D<et>
The set of possible denotations of expressions of type e, De, is the set A and the set of
possible denotations of expressions of type t, D,, is the set of truth values, {0,1}.
Hence, D<e.t> is the set of all functions from the set of entities A to the set of truth
values {0,1}, i.e. { 0 , l } \ As we saw in Section 4.3.2, functions from A to {0,1} are
characteristic functions of subsets of A and so we maintain the intuitively appealing
idea from L P that one-place predicates denote such sets, although they are now being
defined in terms of functions.
Generalising this approach, we may interpret the denotations of expressions with
a complex type, <a,b>, as functions from the set of possible denotations of the input
(first) type, a, to the set of possible denotations of the output (second) type, b. In other
words, D^fe is the set of all functions from Da to Db, written DaDb and shown
99
4 Type theory
schematically in (35). The full recursive definition of the possible denotations of

expressions of some type is given in (36) which directly parallels the recursive
definition of types given in (6), above.
(35) D ^
(36) The denotations of types:

a. De is A.
b. D t is{0,l}.
c. If a and b are types then D<ab> is DbDa.
This definition thus provides a transparent relationship between the types of L ^ and
the sort of denotations they are associated with in the model. However complicated
the type is, the sort of denotation it can have can be worked out by checking the
possible denotations of its argument and result types. For example, the denotation of
an expression of type «e,t>,t> is a function from D<et> to D t , i.e. DtD<et>. But D<et> is
also a function, from De to D, so D«et>,t> is {0,1 J ^ w , the set of all functions from
functions from entities to truth values to truth values. This is shown more
perspicuously by the schematic diagram in (37).
(37) D<<et>t>
D
t
Exercise 4.4:
Given the definition in (37), what are the sets of possible denotations associated
with the following types?
ii. iii. iv.
Let us now take a more concrete example and look at the sort of denotation that
is assigned to a two-place predicate, i.e. an expression of type <e,<e,t». According
to (37), an expression of type < e , < e , t » denotes a function from domain, D e , of
entities to a range, D<e.,>, itself a set of functions from De to D,. Thus, D<e,<e,,» is worked
out according to the steps in (38), with (38.e) giving the set of possible denotations of
100
Interpreting L^
expressions of this type as functions from entities to functions from entities to truth
values. The reference on each line is to the line of the recursive definition in (37) that
provides the justification for each step.
(38) a. D<e,<e,t» = D ^ . (37.c)
b. D < e ^=D t D e . (37.c)
c. Dt={0,l}. (37.b)
d. De=A. (37.a)
e. D <e<et>> =({0,l} A ) A .
A two-place predicate therefore denotes a function that, when applied to an

entity, yields a function that picks out the set of entities that bear a particular relation
to the first. As an example, consider a simple situation where Jo and Ethel (from M J
are both kicking the dog, Chester, and the dog is kicking Jo. No-one else in A (of Mi)
is doing any kicking at all. This relation, which provides the extension of the
two-place predicate kick' with respect to a model (let us call it M2), can be represented
by a function that associates each element in A with the set of elements in A that kick
it. Thus, assuming the association of entities and names of Mi, DOG is associated
with the set {MAN^WOMAN!}, MANi is associated with the set {DOG} and all the
other entities in A are associated with the empty set 0 , as shown in (39).
(39)
{DOG}
MAN 2
WOMAN X
WOMAN 2
CAT
DOG
CAKE
{MAN^WOMANj}
BOOK
As we have seen, sets of entities in A can be defined as characteristic functions,

functions from A to {0,1}- Hence, the extension of kick* with respect to M2 should
more properly be represented as a function from elements in A to the characteristic
functions that pick out the sets of entities that kick them. This is shown in (40), where
f, represents the characteristic function of the set {DOG}, f2 represents that of 0 and
f3 that of {MAN t , WOMAN,}.
The complex function in (40) defines a relation between entities in the set A,
something that can also be represented as a set of ordered pairs, as we have seen. To
recover this set, each entity in the characteristic functions, fh f2 and f3, in (40) is made
the first element of an ordered pair if it is mapped onto 1. The second element of each
pair is provided by the entity that is associated with the relevant characteristic
function. Thus, in f, DOG is mapped onto 1 and so forms the first element of an
101
4 Type theory
(40) [kick'] M2
MAN
WOMANX
WOMAN 2
CAT
DOG
CAKE
BOOK
MAN X
MAN 2
MAN 2
WOMAN X
WOMAN 2
WOMAN 2
CAT CAT
DOG DOG
CAKE CAKE
BOOK BOOK
MAN 2
WOMAN X
WOMAN 2
CAT
DOG
CAKE
BOOK
ordered pair with MAN1? the entity mapped onto f,, to give the pair <DOGJMANi>.
Nothing in f2 is mapped onto 1 and so we turn straight to f3 where both MANi and
WOMANi are mapped onto 1. These thus form the first elements of two ordered pairs
whose second element is DOG, since (40) maps the latter onto f3. This yields the set
102
Interpreting L^
{<MAN1,DOG>,<WOMAN!J)OG>,<DOG,MAN1>} which is the extension

assigned to kick' in M, by the denotation assignment function of Chapter 3. In other
words, functions from entities to characteristic functions are semantically
equivalent to sets of ordered pairs. Hence, the sort of denotation assigned to
two-place predicates has not been significantly changed but just reformulated to allow
for the interpretation of VPs.
The extensions of three-place predicates have the same pattern as the
extensions of two-place ones, except that they are functions from entities to functions
from entities to characteristic functions. In other words, they denote functions from
entities to the sorts of thing denoted by two-place predicates. To illustrate, let's take
a quite complex example using the ditransitive verb, gave, that translates into an
expression in Ltype of type <e,<e,<e,t>». In Mi, Fiona gave the cake to Bertie who
gave it to the dog. Jo also gave the book to Bertie who gave it to Fiona who was also
given the cat by Jo. This situation can be represented by the function in (41) which
looks somewhat complicated but it just defines in terms of functions the set of ordered
triples, { < W O M A N 2 , C A K E , M A N 2 > , <MAN2,CAKE,DOG>,
<MAN1,BOOK,MAN2>, <MAN2JBOOK,WOMAN, >, <MAN1,CAT,WOMAN2>},
the extension assigned to give' in Mj. This can be checked by taking those entities that
are mapped onto 1 by the characteristic functions, f4 to f7 and letting them form the
first elements of ordered triples. The second elements are provided by the entities
mapped onto these functions and the third elements are given by the leftmost entities
that are mapped onto the latter. The order of the mapping from entities to functions
in these diagrammatic representations of functions follows the order of the arguments
in Ltype expressions containing the appropriate functors. In other words, in (41) the
leftmost entities are those denoted by the indirect object, the second set of entities
gives those denoted by the direct object and the third set those denoted by the subject.
Such diagrams thus go from most oblique object on the left to least oblique on the
right.
4.4.2 Revising the theory

It is not difficult to redefine the model structure so that it can be used to interpret
expressions in the language, L ^ . First of all, the denotations in a model of all
predicates must be redefined as functions of some sort rather than as sets. For
expressions of types e (e.g. etheV) and t (e.g. rain'), there is no change from previous
models. These are assigned entities and truth values, respectively, as before. The
denotation assignment function, F, however, assigns appropriate functions to all
expressions with complex types, instead of sets of n-tuples of entities. Thus, one-place
predicates, of type <e,t>, are assigned characteristic functions over A (as in (43),
below); two-place predicates, of type <e,<e,t», are assigned functions from entities
to characteristic functions (as in (40)) and three-place predicates, of type
<e,<e,<e,t>», are assigned functions from entities to functions from entities to
characteristic functions (as in (41)). As we have seen, this is equivalent to assigning
sets of entities, sets of ordered pairs and sets of ordered triples, respectively, but
allows the interpretation of expressions other than basic expressions and formulae,
as we will see directly.
103
4 Type theory
(41) [give']M2
u
MANi MANj - ^ ^
MAN 2 \^\ MAN2 — - ^ ^ 1
WOMANjT\ \ WOMAN! v \
WOMAN2\\\ \ ^*l WOMAN2 \ Y\
f CAT >. \\\\^^ / CAT <s. \
DOG DOG ^ ^
/ \N^\v\ / CAKE — — -W 0
BOOK - ^ ^ ^
///1 , Yi %
MAN! '
MAN2 •—-~--AlIII^X^__ MAN2 A 1
WOMAN! TJ^5§^^ WOMAN! » \ \
W O M A N 2 ^ ^ ^ ^ ^ ^ WOMAN2 s V\ \
CAT CAT - ^ ^ \
^f ^ ^ ^ / A >^^
MANj ^Jj / // DOG—~//' \\^vi DOG - C ^ ^
CAKE — —^ 0
MAN2 ^ / / I CAKE 'y' I/7J1 4 BOOK ^ ^ ^
WOMANj / / i 1 BOOK^ yy/\// 1
WOMAN2 \
CAT — ~ - y
/
\/ /
/
M A N //
y/f/Iki
I //I niln MAN! ^ — - ^
k,
DOG -> Y / l/ y A / nM
CAKE
BOOK
\ f \
W N
\
\
MAN2
\ ZZl^///// \
DOG
/I/
/ ///HI
///Ink MAN2
WOMANj V
I\
CAT v ^ ^ >
DOG - ^ \
r
^ 1
\ CAKE //HI CAKE — —

\ BOOK v //HI BOOK ^^^
\ MANX
^rll// TrU
II MANi ^ ^
f
7
\ MAN2 //1 l\ MAN2 A .• * • 1
\ WOMAN! — / / / / / ^ ^ WOMAN! \
\ WOMAN2 ' / / / ^ ^ WOMAN2 v \
CAT — Z ^ ^ ^ \ ^ CAT v \
DOG-^^^^^^/ DOG C ^ v .
CAKE - _ ^
CAKE - ^ ^ ^
BOOK
/
BOOK ^^~ 0 °
104
Interpreting L^
Given a model containing a set of entities A, as before, and a denotation

assignment function, F, revised to assign functions rather than sets, the interpretation
of all expressions in Ltype (excluding formulae containing the connectives) can be
specified by two clauses. The first clause equates the interpretation of any basic
expression with the denotation assigned to it in the model (i.e. by F). This gives the
basis of the interpretation which is built on by the recursive application of the second
clause which provides the means of interpreting all complex expressions formed from
these. This clause derives the denotation of a well-formed expression in L ^ consisting
of a functor expression,/, and an argument expression, a, from the application of the
denotation of/ in some model to that of a in the same model. In other words, the
(41) [give']M2
denotation in a model of an expression/fa) is as the value of a function denoted b y /

applied to the denotation of an argument a. The two clauses of the model theory
needed to interpret L ^ are formally specified in (42).
(42) Given a model, M = «A,F», then
1. For any constant a, [a]M = F(a).
2. If/is of type <a,b> and the type of a is a, then \f(a)]M = [/]M(MM).
The simplicity of the model theory results directly from the adoption of mathematical
functions as the denotations of complex types. If the output type of the functor (the
second member of the ordered pair of types defining the complex type) is basic then
the value obtained when the function is applied to some argument is either a truth
value or an entity (usually the former). If, on the other hand, it is a complex type,
then the denotation resulting from the application of the function to some argument
is itself a function.
To make this more concrete, let us take over the extension of the predicate laugh *
in the model, Mi, into M2 (assuming that M2 has the same ontology, i.e. A, and the
same assignment of entities to individual constants as MO. The situation modelled is
thus one where only Bertie and Fiona are laughing, which is represented by the
characteristic function in (43).
(43) [laugh']
MAN 2
WOMANX
WOMAN2
CAT
DOG
CAKE
BOOK
105
4 Type theory
Given this function, it is simple to work out the extension (i.e. truth value) of a
formula containing laugh'. Once the extension of the subject expression has been
ascertained, the function is applied to it and the automatic result is a truth value, the
value of the function denoted by the predicate applied to the argument. So, for
example, to find out the truth value of a formula laugh'(jo') (translating the
sentence Jo laughed) we go through the steps in (45), where the number on the
righthand side refers to the clause in the model theory or the representation of the
extension that justifies the step taken. We see from the last line that the formula is
true in Mi, since the function denoted by laugh' in M2 returns the value 1 when
applied to MANi which is the extension of jo' in the same model.
(44) a. [laugh'Oo')] M2 is[laugh'] M2 ([jo'] M2 )- (42.2)
b. [laugh']M2 is F2(laugh') is the function in (43) (call it f43). (42.1)
c. [jo']M2 is F,(jo') = MANi. (42.1)
d. f43(MAN0 is 1. (43)
e. Hence, [laugh'(jo')]M2 is 1.
More complex formulae are interpreted in the same way, except that more
applications of the rule in (42.2) are involved. The interpretation of a formula
containing a two-place predicate like kick' or a three-place predicate like give'
proceeds by reducing the complexity of the function until a truth value is arrived at.
In the former case, the function denoted by the predicate is applied to the entity
denoted by the direct object, the first argument combined with the predicate, to give
a characteristic function. Applying this function to the entity denoted by the subject
yields a truth value, the truth value of the whole formula. With the translations of
ditransitive verbs, the function denoted by the predicate is first applied to the
extension of the indirect object and then to that of the direct object to get the
characteristic function that yields the truth value of the formula when applied to the
extension of the subject.
As an example, consider the interpretation of the formula,
(kick'(the-dog '))(the-singer') translating the sentence The singer kicked the dog. First
of all, the rule in (42.2) is applied to the formula to unpack the function/argument
structure, starting with the rightmost argument and working inwards until the constant
expressions are reached, as shown in (45).
(45) a. [(kick'(the-dog'))(the-singer')] M2 .
b. [kick'(the-dog')]M2([the-singer']M2). (42.2)
c. ([kick>]M2([the-dog']M2))([the-singer']M2). (42.2)
The constants in (45.c) can then be then 'cashed out1 in terms of their extensions in
the model M2 as specified by F. The value of each function is computed at every step,
starting from the functor and working outwards. Thus, we first retrieve the extension
of kick' in M2, i.e. the function in (40) (which we may call f40, for convenience). Then,
this is applied to the extension of the-dog\ i.e. DOG, to get the extension of the
expression kick'(the-dog') which is the characteristic function f3 in (40). Applying the
latter to the extension of the-singer\ i.e. WOMAN 2 , gives a truth value, in this case
0. Hence, the formula (kick '(the-dog'))(the-singer') is false with respect to M2. These
automatic steps are spelled out in (46) which should be compared to the function
denoted by kick' given in (40) to see how the final result is obtained.
106
Adverbs
(46) a. [kick'] M2 = F2(kick') = f40.

b. [the-dog']M2 = F2(the-dog') = DOG.
c. [kick']M2([the-dog']M2) = ^ 0 (DOG) = f3.
d. [the-singer']M2 = F2(the-singer') = WOMAN 2 .
e. ([kick']M2([the-dog']M2))([the-singer']M2) = f3(WOMAN2) = 0.
We have now properly solved the problem posed in Section 4.1: how to
incorporate verb phrases into the grammar. In our example sentence, the VP, kicked
the dog, has the translation kick'(the-dog') whose extension is given in (46.c) as f3 in
(41). This results from the application of the function denoted by the two-place
predicate to the extension of the direct object. Since this is a characteristic function
over A, and such characteristic functions define sets of entities, we have neatly
captured the intuition that verb phrases should pick out sets, the sets of entities that
bear the appropriate relation to their subjects. Indeed, the adoption of Lvpe gives us a
powerful logical language into which to translate natural language sentences, one
that can reflect the syntactic structure of the object language. By taking the
denotations of L ^ expressions to be either truth values, entities or more or less
complex functions based on these two sets, we can ensure that the syntactic structure
of an expression is maintained in the interpretation, as shown in the earlier
examples. Thus, in addition to having a Ltype representation, a VP also has a
model-theoretic interpretation, i.e. as some function from entities to truth values.
Hence, each independently meaningful expression generated by the grammar is
associated with an extensional meaning and so compositionality is maintained.
Exercise 4.5:
1. Assuming that the negation operator, ~, is of type <t,t> what is the actual
function that it denotes?
2. Using the step-by-step interpretation method outlined above, work out the
truth value of the translation of the sentence The student didn 'tgive the book
to Ethel with respect to M2 (i.e. using the function drawn in (41)), assuming
the interpretation of negation given in Chapter 3.
4.5 Adverbs
The grammar fragment G2 as yet covers only the same ground as the earlier G!
(although without the recursion allowed by conjunction) but our new theory of
translation and interpretation allows us to extend the grammar in a number of new
and interesting directions. In particular, we are now in a position to tackle certain
sorts of adverbial modification, including simple English adverbs of manner like
slowly, happily, pathetically, etc. Such adverbs modify verbs, or more correctly,
verb phrases and so may be syntactically analysed as combining with a VP to give
another VP, as in the rule in (47).
107
4 Type theory
(47) a. Rule 12G2: VP[aFIN] -> VP[aFIN] Adv.

b. Adv -> {slowly, happily, stupidly, crazily}.
Since VPs have the type of one-place predicates, i.e. <e,t>, adverbs introduced by this
rule must be associated with Ltype expressions that combine with one-place predicates
as arguments to yield one-place predicates. In other words, the type of an adverb
is «e,t>,<e,t». This type assignment entails that the translation rule associated with
Rule 12G2 is given by the functional application of the translation of the adverb to that
of the VP, as in (48). A verb phrase like walked slowly is thus translated into L,ype as
slowly'(walk*).
(48) TRule 12G2: Adv'(VP').
The question that must be asked whenever we augment the grammar with a new
category and an associated type is Does the type assignment make sense
semantically?. In the case of manner adverbs, the suggestion made above does accord
with intuition. VPs with simple -ly adverbials in English pick out a subset of the set
of entities denoted by the VP without the adverbial. So, for example, the set of things
that walk slowly in any situation is a subset of the set of things that walk in that
situation. The semantic effect of slowly can thus be analysed as a function that maps
a set, the extension of the translation of a VP, onto a subset of that set. This
parallels the semantic relation between adjectives like slow or happy to their common
nouns in phrases like a happy student. Such adjectives pick out a subset of the set of
entities denoted by the common nouns, so that, for example, the phrase happy student
denotes a subset of the set of all students, i.e. the happy ones. The relation between
such adjectives and common nouns will be discussed in Chapter 6, but the functions
that are denoted by both manner adverbials and their adjectives have characteristic
functions as both their domain and range. In other words, such expressions have
denotations in D^,^0^^ which is the set of denotations associated with the type
«e,t>,<e,t», as shown in (49).
(49) D«e,t>,<e,t»
D D.<e,t>
<e,t>
{1,0} A {1,0}
According to the model theory for interpreting expressions in L ^ , the

interpretation of an expression like slowly'(walk') is given by applying the function
denoted by slowly' to that denoted by walk\ i.e. [s low ly'(walk f)]M is
[slowly'^[walk']1*) or more generally as in (50).
(50) [Adv'(VP')]M = [Adv']M([VP']M).
108
Adverbs
As an example, let us assume that in M2 Bertie, Ethel and the cat are the only
entities that are walking, a situation that can be represented as the function in (51),
supplying the extension of walk' in M2. The things that are slowly walking in M2 are
Bertie and the cat, giving us the extension of slowly '(walk') in (52). The effect of
applying [slowly *]M2 to [walk']M\ then, is to pick but a subset of the set of things that
walk, i.e. the set of things that walk slowly. This means that part of the function
denoted by slowly' is that given in (53). Note that this is only a part because the
function applies to other characteristic functions of sets (i.e. VP extensions) to yield
other results.
(51) [walk']M2
MANj
MAN2
WOMAN!
WOMAN2
CAT
DOG
CAKE
BOOK
(52) [slowly'(walk')] M2
109
4 Type theory
(53) [slowly '] M 2
MAN 2 MAN 2
WOMANX WOMANX
WOMAN 2 WOMAN 2
CAT CAT
DOG DOG
CAKE
BOOK BOOK
Exercise 4.6:
Assume that in M2, the man, Ethel, the cat, the dog and the bike are all moving, and
that the man, the cat and the bike are all moving slowly.
i. Draw the characteristic function denoted by [move '] M2 .
ii. Draw the characteristic function denoted by [slowly'(move')]M2.
iii. Draw the function now associated with slowly' in M 2 , assuming that
nothing else is happening slowly and representing all characteristic
functions apart from those associated with move' and walk' as <|).
There is much more that could be said about the extensions of adverbs. In
particular, we need to constrain the functions denoted by adverbs like slowly so that
they do pick out subsets of the sets denoted by their associated VPs in every model
(cf. subsequent chapters, particularly Chapter 7, for details of how to do this).
Furthermore, in a more extensive fragment, we would need to provide the means for
interpreting other VP modifiers like prepositional phrases, other classes of adverbials
and sentential modifiers (like necessarily, unfortunately, etc.). Such constructions
present their own problems in semantics (which may or may not necessitate the
extension or revision of certain aspects of the theory presented so far), but the
introduction of a typed logical language and its transparent relation to possible
denotations at least allows for the expressions to be representable in the logical
translation language. Hence, for the rest of this book, L ^ will form the basis for all
revisions to the translation language required by the development of the grammar
fragment and the basis of the model theory will remain the binary application of
functions to their arguments in the way that we have seen.
110
Further reading
4.6 Further reading

The idea that the combinatory properties of expressions can be encoded directly in
their categorial definition is central to categorial grammar which evolved out of the
work of certain logicians, most particularly Ajdukiewicz (1937) and Bar-Hillel (1953).
A simple introduction can be found in Allwood, Andersen and Dahl (1977: 132-135)
and more detailed discussions can be found in the articles in Oehrle, Bach and
Wheeler (1988). Montague himself used categorial grammar for his syntactic analysis
in Montague (1973) which is therefore also used in Dowty, Wall and Peters (1981).
The definition of logical types can be found in Dowty, Wall and Peters (1981: 83-97)
and Partee, ter Meulen and Wall (1990: 338-341). Discussions of type theory,
translation and the syntax of natural languages can be found in Dowty (1982) and
Gazdar, Klein, Pullum and Sag (1985: ch. 9). The notion of a function is introduced
in Partee, ter Meulen and Wall (1990: 30-33), Allwood, Andersen and Dahl (1977:
9-13) and Chierchia and McConnell-Ginet (1990: 438-440). An early discussion of
the semantic representation of adverbs can be found in Thomason and Stalnaker
(1973). Cresswell (1985a) contains a collection of papers on adverbial and
prepositional phrases, primarily with respect to temporal and spatial modification; and
see also Parsons (1980). The discussion in the latter two references is somewhat
technical and uses concepts that have not yet been presented or discussed in the
present work and readers are advised to consult this book only after they have tackled
subsequent chapters (particularly Chapters 8 and 9). The application of type theory to
natural languages is discussed in Parsons (1979). There is a current debate about
whether higher order types are necessary for the analysis of the semantics of natural
languages. Representative articles on this subject are Thomason (1989), Turner
(1989), and Chierchia and Turner (1989), but these are very technical and should not
be tackled until the reader has read most of this book.
Ill
The Lambda Operator
5.1 The passive

In discussing the translation from English into L ^ in Chapter 4, rules for
generating and interpreting simple passives were omitted. Although it is possible to
define the extension of a passive verb phrase like kicked by Jo as the characteristic
function of the set of things that Jo kicks, it is not possible with the apparatus we
currently have to link this function directly with the extension of the active verb kick,
kicks, kicked. The appropriate relationship between the two voices is that, in the
relation denoted by the passive, the entity denoted by the object of the preposition
by corresponds to the entity denoted by the subject of the active and the entity
denoted by the passive subject corresponds to that denoted by the object in the active.
This correspondence was handled in Chapter 2 directly in the translation for the
passive rule by switching around the individual constants translating the two noun
phrases in the passive rule to yield an identical translation to that of the active. So,
for example, Jo kicked Chester and Chester was kicked by Jo are both translated
into Lp as kick'fjo',Chester'). Unfortunately, this simple expedient is no longer open
to us because of the existence in G2 of a verb phrase constituent. This prevents subject
and complement NPs from being ordered with respect to each other in a translation
rule because they are no longer introduced by the same syntactic rule.
To see the problem that passives pose, together with other constructions that
involve a change in grammatical functions, let us devise an extension of G2 that not
only generates passive sentences but also ensures that active and passive pairs of
sentences are semantic paraphrases. The appropriate syntactic rules to generate
passives are easy to define and appear in (1). In accordance with the general policy of
this book, no attempt is made to draw out the syntactic regularities shown by
different constructions. The rules in (1) clearly lose the obvious syntactic relations
with their active counterparts and this problem may be resolved in a number of ways
according to the system of syntax with which the theory of semantics is associated.
Readers who are familiar with one of the many other systems of generative grammar
currently used by linguists are invited to adapt the following grammar and its
interpretation to suit the requirements of the theory they know. The important thing
from the point of view of the semantics, however, is that, no matter what the
appropriate syntactic relationship between active and passive happens to be and
however it is formalised, active and passive pairs are truth-conditionally
synonymous and this, at least, must be shown by an adequate semantic theory.
(1) a. 13G2:VP->wasVP[PAS].
b. 14G2: VP[PAS] -> Vt[PAS] PPby.
c. 15G2: VP[PAS] -> Vdt[PAS] PPto PPby.
d. 16G2: VP[PAS] -> Vdt[PAS] NP PPby.
e. 17G 2 :PP by -»byNP.
f. VJPAS] = {kicked, loathed, poisoned, read, eaten, liked}.
g. Vdt[PAS] = {given}.
112
The passive
For ease of exposition, it is assumed that the prepositions in passive by phrases

function simply as syntactic case markers, like indirect objects marked by the
preposition to, and the syntactic problems posed by this assumption will be ignored.
The syntactic category, PPby, thus has the same type and translation as the noun phrase
it contains. Given this, it is possible to assign to the translations of passive verbs the
same type as that assigned to their active counterparts, as shown in (2). These give
rise to the set of translation rules in (3) for the syntactic rules in (1).
(2) a. TYPE(VJTAS]) = <e,<e,t». (= TYPE(Vt))
b. TYPE(Vdt[PAS]) = <e,<e,<e,t>». (= TYPE(Vdt))
(3) a. T13G2: VP[PAS]\
b. T14G2: Vt[PAS]'(PPby').
c. T15G2: Vdt[PAS]'(PPto')(PPby').
d. T16G2: Vdt[PAS]'(NP')(PPb/).
e. T17G2:NP\
The question that must now be addressed is: how to translate/interpret passive
verbs to ensure that the rules in (3) yield the appropriate truth-conditions? If
we stick to the rules for translating lexical items that we have been using until now,
we should translate passive verbs exactly like their active counterparts, because
translation is based on the lexeme underlying the word form. Given the usual
assumption that passive morphology in English is inflectional, the lexeme associated
with the formkicked, passive and active past tense, is KICK, so both ought to translate
as kick' (as in Chapter 2). However, if we do this we will end up with the incorrect
interpretations for passive sentences: they will come out with different truth-
conditions from the corresponding active sentences. To see why this is so, consider
the diagram in (4) which represents part of the extension for kick' in the model, M2.
The translation of a sentence like Ethel kicked the dog comes out as true here
with the translation of the VP, kick'(the-dog'), having the extension shown by the
function f3 which picks out the set of entities kicking the dog, i.e. {MAN!,
WOMANi}. If the passive participle is translated by the expression kick', however,
the translation of the corresponding passive The dog was kicked by Ethel comes out
as false, because the VP translation, by T17G2, is kick'(ethel') which denotes the
function represented by f2, i.e. the characteristic function of the null set. This, of
course, picks out the set of things that kick Ethel, not that kick the dog, but,
intuitively, the phrase be kicked by Ethel does not denote this set but the set of things
that Ethel kicked. There is unfortunately no way of representing this set in L ^ directly
without assigning an extension to the passive participle kicked that is unconnected
with that of the active verb kicked. The required function is that shown in (5).
Although the two diagrams in (4) and (5) represent different functions, there
is an obvious relation between them. The function in (5) switches around the entities
denoted by the first and second arguments of the function in (4). Unfortunately, this
generalisation is not actually expressed in the grammar. If we give kick' and kicked'
separate extensions as specified by the denotation assignment function, F, we lose
the semantic generalisation about the passive-active relation and cannot guarantee
that the extensions of passive participles do not depart radically from the extensions
of their active counterparts. This would mean that the grammar was descriptively
113
5 The lambda operator
inadequate, because it failed to capture an intuition about the relation between active
and passive. To maintain the adequacy of our theory, therefore, some way must be
found of relating the extensions of active and passive verbs.
(4)
MAN JL
WOMANj
DOG
MAN X
WOMAN!
DOG DOG
MANt
WOMANt
DOG"
(5)
MAN
WOMANj
DOG
WOMAN 2
DOG
WOMANX
DOG
114
Introducing the lambda operator
5.2 Introducing the lambda operator

To solve this problem, we need to take another look at sets. As we have seen, a set
can be specified by listing of all its members or by defining a characteristic function
for that set. These two methods of defining a set are easy to do if the set being
defined is finite, but if the set is not finite but infinite a simple listing is not
sufficient because, by definition, the list or the function will never be complete. It is
possible, however, to define infinite sets by describing in general terms the property
that picks out all the members of the set. Thus, instead of giving the definition of a set
by listing its members, we can give a definition that describes the general property that
determines which entities are members of that set. For example, the set of all natural
numbers is an infinite set for which we can give a partial listing, e.g.
{1,2,3,4,5,6,7,8,9,10,...} (the three dots before the right hand curly bracket
conventionally indicates that the set is open and not fully listed), but we could also
describe this set as the set consisting of all the natural numbers. Slightly more
formally, we can define it as the set of all things that are natural numbers, or more
formally still, as the set of all x such thatx is a natural number. This latter definition
selects a variable and specifies a property that any entity substituting for the variable
must have in order to be a member of the set in question. Traditionally, individual
variables (i.e. variables ranging over individual entities) are written as lower case
members of the end of the alphabet, i.e. x, y, z, which sometimes appear with
subscripts, e.g. x b y243, z15, etc., if a large number of such variables are required. The
intensional description of a set may be symbolised by writing some individual variable
before a vertical stroke, I, which both precede a quasi-English expression where the
variable takes the place of some noun phrase in the expression, e.g. x is a natural
number. The resulting expression is enclosed in curly brackets, {•} to indicate that
what is being defined is a set. So, the set of natural numbers may be defined as fx I
x is a natural number}. Although this is the most convenient way of specifying an
infinite set, we can, of course, use it to specify a finite one. For example, we might
specify the sets in (6), which are all presumably finite, and which could be used to
describe the extensions of the English phrases kicked Chester, was kicked by Jo and
was a drunken lecturer, respectively.
(6) a. {x I x kicked Chester}.

b. {y I Jo kicked y}.
c. {z I z was drunk and z was a lecturer}.
The semi-English set descriptions in (6) may easily be translated into formulae in
Ltype, but with the formulae containing a variable in place of one of the argument
individuals. Hence, x kicked Chester can be represented in L ^ as
(kick'(chester'))(x), (6.b) can be represented as (kick'(y)Xjo') and (6.c) as drunk'(z)
& lecturer'(z).
Formulae containing variables are semantically incomplete, since variables do
not have a fixed interpretation like constants but take their meaning from being
associated with specific entities in specific situations, rather like pronouns in natural
languages. Thus, a formula like (kick'(chester'))(x) does not denote a truth value until
the variable, x, is associated with some entity in the model. Thus, to interpret such a
formula, some entity, say MAN2, must be assigned to the variable x and the formula
115
(kick'(chester'))(x) thus gets assigned the same truth value in the model as
(kick'(Chesterf))(bertie')y since bertie' denotes MAN2. Interpreting formulae
containing variables of this sort is thus like trying to ascertain the meaning of a
sentence containing a pronoun, like Chester was kicked by him. The pronoun must
be associated with a referent through the context before interpretation can proceed.
Formulae in L ^ (and other logical languages) which contain a free variable, i.e. a
variable not associated with some operator (see below), are referred to as
propositional functions or open formulae.
5.2.1 Extending L^
Each of the descriptions in (6) provides the intensional property that determines the
membership of some set of entities. The semantic theory developed so far has little to
say about intensional things such as properties, but we may instead interpret them
extensionally as equivalent to the sets they pick out. In other words, we may equate
the set descriptions in (6) with one-place predicates in L ^ which denote characteristic
functions of sets of entities, as we saw in Chapter 4. Thus, given any entity, a, in some
domain, if a substitutes for the variable in a set description to give a true statement,
then a is in the set denoted by that description. If the resulting statement is false,
however, then a is not in the set.
The propositional functions used at the end of the last section to translate the
descriptions in (6) do not, however, denote sets. They are expressions of type t and
not of type <e,t>. Such expressions can be turned into one-place predicates by the
use of a special logical symbol, written X (lambda), called the lambda operator. This
operator picks up, or, more technically, abstracts on, a variable contained in a
propositional function and turns the expression into something that denotes a
characteristic function. The lambda operator is paired with the variable it abstracts
upon which it is said to bind. For example, Xx abstracts on the x variable and binds
all instances of x in some propositional function while Xy abstracts on and binds the
y variable, and so on. The operator together with the variable it selects is written to the
left of a propositional function which is itself enclosed in square brackets. For
example, given the propositional function (kick'(Chester'))(x) we can form the
lambda expression Xx [(kick'(chester'))(x)] which is a well-formed L ^ expression
of type <e,t>. The fact that all instances of the variable abstracted upon by the
lambda operator which are contained in the propositional function are bound by
the operator is an important point that will be returned to below. Generally, given
a propositional function <> | , containing an instance of the variable x, of type e,
Xx [<))] is a well-formed expression of type <e,t>. This requires the addition to the
syntax of L ^ of the rule in (7) which allows expressions like those in (8), but not
those in (9):
(7) If $ is an expression of type t containing an unbound instance of a
variable x of type e, then Xx[((>] is a well-formed expression of type
116
(8) a. Xy [(kick'(y))(jo')].
b. Xz [drunk'(z) & lecturer* (z)].
c. Xx [-(crazy'(x))].
d. Xx [(kick'(y))(x)].
e. Xz [(kick'(z))(z)].
(9) a. Xx [(kick'(y))Go)].
b. Xy [(kick' (Chester' ))(jo)].
c. kk
In effect, the lambda operator turns a propositional function containing a free
individual variable into a one-place predicate which denotes the characteristic function
of the set of entities that satisfy the propositional function. An entity is said to satisfy
a propositional function, $, containing a variable, JC, if the association of x with that
entity gives rise to a true formula. A lambda expression A,x [(()] is thus interpreted as
if it were a set-theoretic expression {x I <J>}, extensionally denoting the set of all x
such that <|). For example, the one-place predicate XK [(kick'(chester'))(x)] denotes
the set of all x such that x kicked Chester, while the predicate Xy [(kick'(y))(jof)] is
interpreted as the set of all y such that Jo kicked y and Xz [drunk (z) & lecturer (z)]
denotes the set of all z such that z was drunk and z was a lecturer. Indeed, because
propositional functions are formulae with at least one free variable, ^-expressions can
be set-denoting expressions of indefinitely complex internal structure. An expression
like Xx [(((loathe'(x))(jof) v (like(x))(joy)) & ((x = the-cakey) -> (eaty(x))(joy)))] is
perfectly well-formed in Ltype and denotes the set of all x such that either Jo loathed x
or he liked x and if x was the cake then Jo ate x. Thus, ^-expressions provide a
representation in LtyPe of the sorts of set descriptions introduced above.
Exercise 5.1:
Give an informal description of the sets denoted by the following lambda
expressions:
i. Xy [(like'(y))(jo') & ~((like'(y))(ethel'))].
ii. Xz [(give'(z)(the-cake'))(z)].
iii. A,x[x=jo'].
iv. Xz [student' (z) & (like' (jo' ))(z)].
Because lambda expressions of the sort given above are of type <e,t>, they can
combine with expressions of type e by the rule of functional application to yield
an expression of type t. Applying the predicate Xx [(kicky(chester'))(x)] to the
individual constant 70' we get the expression Xx [(kick'(chester'))(x)](jo'). This
formula, which may be read as Jo is an x such that x kicks Chester, has identical
truth-conditions to the one obtained by replacing the bound variable x in the
propositional function following the lambda operator by the argument expression,
jo* and removing the lambda operator. Thus, Xx [(kick'(chester'))(x)](jo') is
equivalent to (kicky(chestery))(jo'). The process of replacing variables by constants
and removing the lambda operator is known as lambda conversion because it
converts a complex lambda expression into a simple one.
117
(10) Lambda Conversion: If Xx [0] is a lambda expression of type <e,t> and

a is an expression of type e, then Xx [<|)](a) is an expression of type t and
is truth-conditionally equivalent to the expression <|>* where <|>* is derived
from 0 by replacing every occurrence of x in § bound by X by the
expression a.
The rule for lambda conversion in (10) requires all instances of the variable
bound by lambda to be replaced by an instance of the argument expression. If this
were not done, the expression after conversion would still contain a variable and so
would not be a formula but a propositional function that does not directly denote a
truth value. The equivalence would, therefore, not hold. For example,
Xz[drunk'(z) & lecturer'(z)J(ethel') is equivalent to drunk'(ether) &
lecturer'(ethel'), but not to drunk'(ethel') & lecturer'(z) nor drunk'(z) &
lecturer (etheV), which are both propositional functions and not full formulae.
A second very important point about the use of the lambda operator is that, in
the definition of lambda abstraction given in (7) above, mention was made of the fact
that the variable abstracted on by the lambda operator must itself not be bound, i.e.
not abstracted on by any other lambda operator. As an illustration, consider the
lambda expression Xx [(like'(x))(fwna')] where the x variable is bound by the
lambda operator. This can be applied to an individual constant to yield a formula,
hc[(like'(x))(fiona')](ethel') which in turn could be coordinated with a propositional
function like laugh(x) to give the expression Xx [(like'(x))(fiona')](ethel') &
laugh'(x). The second instance of the variable x in this expression is not bound by the
lambda operator, as can be seen from the application of lambda conversion to the
expression. The rule of lambda conversion in (10) allows the variable bound by X to
be replaced by the argument expression of the X-expression to yield a
truth-conditionally equivalent formula. In the case of the expression
Xx[(like'(x))(fiona')](ethel') & laugh'(x), ethel' is the argument of the functor
Xx[(likef(x))(fwna')] so that the result of lambda conversion is (like'(ethel'))(fiona')
6 laugh '(x), a propositional function. Thus, if we now take the original expression
before lambda conversion XJC [(like '(x))(fiona ')](ethel') & laugh '(x), we can abstract
on the variable x again to get the one-place predicate XJC [XX [(like '(x))(fiona')](ethel')
& laugh'(x)]. The variable x in (like'(x))(fiona') in the latter expression is not
bound by the outermost lambda operator, but only by the inner one. This means that
the extension of the expression Xx [Xx [(like'(x))(fwna')](ethel') & laugh'(x)] is
the set of all x such that Ethel is in the set of things such that Jo liked x and x
laughed, but not the set of all x such that Jo liked x (who is Ethel) and x laughed.
If this expression is applied to an individual constant like jo' to yield
Xx[Xx[(like '(x))(fwna ')](ethel') & laugh '(x)](jo') and is then lambda converted, the
resulting expression is [Xx [(like'(x))(fiona')](ethel') & laugh'(jo')] and not
[Xx[(like '(jo '))(fwna ')](ethel') & laugh '(jo')]. The latter would have a very strange
extension something like Ethel is an x such that Fiona liked Jo and Jo laughed, i.e.
no properties are predicated of Ethel at all. It is necessary, therefore, to be careful
when using variables. Even if two variables look the same in an expression, they may
not behave in the same way in the expression concerned, as they may be bound by
different operators and so behave differently semantically. For your own lambda
expressions, try to differentiate variables as much as possible, writing the same
118
symbol for different instances of variables only when they are bound by the same
operator. This helps in keeping track of what conversions to make and ensuring that
constants are not accidentally put where they shouldn't go. Using the symbol » to
stand for X-converts into the statements in (11) are all true and valid instances of
lambda conversion, according to (10), while those in (12) are not.
(11) a. Xy [(kick'(y))Go')](chester')
» (kick'(chester'))(jo').
b. Xx [(kick'(x))(x)](jo')
» (kick'(jo'))Go').
c. Xx [Xy [(kick'(y))(x)](jo')](bertie')
» Xy [(kick'(y))(bertie')]Go')
»(kick'(jo')Xbertie').
(12) a. Xx [(kick'(x))(x)](jo')
( ( ) ) ( j )
b. Xx [Ax [howl'(x)](the-cat')](the-dog')
» Xx [howr(the-dog')](the-cat').
c. Xy [(like'(jo'))(x)](bertie')
» (like'(jo'))(bertie').
d. Xz [crazy'(z)](etheP) v drunk'(z)
» crazy'(ethel') v drunk'(ethel').
Exercise 5.2:
1. Explain why are the A-conversions in (12) wrong.
2. Convert the following lambda expressions plus arguments into their equivalent
formulae:
i. Xy [(like'(y))Go') & ~((Uke'(y))(ethel'))](the-dog').
ii. Xz [(give'(z)(the-cat'))(z)](bertie').
iii. ^x [~(x = jo')](bertie').
iv. Xy [(like'(the-cat')Xy) v ~(like'(the-dog'))(y)](bertie').
v. Xx [Xy [crazy'(x) & (like'(x))(y)]Go') v -(crazy'(x))](bertie').
5.2.2 Interpreting lambda expressions

What lambda expressions denote has already been briefly discussed above.
Essentially the idea is that by abstracting on a variable using the lambda operator,
an expression is formed that denotes a set, the set of things that have the property
specified by the propositional function in the scope of the operator. So, an
expression like Xx [(Ukey(x))(etheV)] denotes the set of things that Ethel liked. It is
now time to put this on a more formal footing and specify precisely how the
denotation of a lambda expression can be determined from the denotation of the
propositional function it is made from. Although the fundamental idea behind lambda
abstraction is a simple one, i.e. that it is a set-forming operator, the formal definition
is unfortunately quite complicated, but, as with all formal definitions given in this
book, it does no more than restate explicitly the informal definition given earlier.
119
To see how the formal semantics of lambda abstraction works, let us look at
what happens in lambda conversion. The equivalence between <|>* and Xx [ty ] in (10)
provides the interpretation of the lambda expression when applied to an appropriate
argument. As we have seen, an expression like Xx[(kick'(x))(jo')](chester') is, by the
rule of lambda conversion, equivalent to (kick'(Chester '))(jo')9 i.e. the formula that
is obtained when (all occurrences of) the abstracted variable are replaced by the
argument expression. In yet other words, the truth-conditions of a lambda expression
Xx [<))] (where <>
| is a propositional function containing x) applied to an individual
argument a are identical to those of the expression (()* derived by substituting a for all
free occurrences of x in ty . From the semantic point of view what this equivalence
expresses is that the value of the function denoted by Xx[ty] applied to the denotation
of a in a model is identical to that denoted by <|>* in the same model. This can be
more formally represented as (13), but at the moment we have no way of formally
expressing what (|>* is, or, more exactly, what its semantics is. To do this, the model
theory needs to be extended to cope properly with the existence of variables in Ltype
expressions.
(13) [Xx [((>](a)]M if, and only if, [c|)*]M.
In a number of places in this book, it has been said that variables are like
pronouns: if the referent of a pronoun is unknown, there is no way that an expression
containing it can be interpreted (assigned a truth value). For example, to know
whether She didn 7 like him expresses a true proposition, one has to know who she and
him refer to. In a similar way, the truth value of a formula containing a variable
cannot be ascertained until the variable is associated with some entity in the model.
In an expression Xx [(|>](a), the lambda operator binding a variable x gives the
information that x is to be associated with the entity denoted by a, the argument that
the lambda expression applies to. For example, Xx [(kicky(x))(jo')](chesterf) is a
formula that associates x in the model with [Chester']M2, i.e. DOG. On the other hand,
the equation of x with DOG is not appropriate for every occurrence of x in any
formula, since variables are not uniquely associated with particular entities in any
particular model. The point of a variable is that it can be associated with different
things in different circumstances, unlike constants, whose denotations are fixed by
the denotation assignment function of a model. The association of variables with
entities in a model is something in addition to, and separate from, the model itself.
Because variables do not have a fixed denotation, they are associated with
entities in a model not by the denotation assignment function F, but by a variable
assignment function, usually represented by the symbol g. This function assigns to
each variable an entity in the model as its value. It is usually assumed that the
number of variables available is infinite, but of course we are only ever interested
in a small number of variables and in particular only the ones appearing in the L^pe
expressions we are interpreting. Because of this, and because infinite functions pose
no real problem in interpretation, we shall ignore all but a very few of the infinite
number of variables mentioned in any variable assignment function. Formally,
then, g is an infinite function but in practice we will deal with only a very small part
of it. Because the number of variables is infinite, the number of variable assignment
functions for any set of entities in a model is infinite and varies quite considerably.
For instance, there is a function that maps every variable onto only a single entity
120
in the model, one that maps every seventeenth variable onto the same entity or one
that maps variables onto entities in strict rotation. These are all perfectly valid
functions but not really very useful for our purposes. The sorts of function we will be
dealing with are very restricted, and the only important thing to bear in mind is that
because g is a function, every variable is assigned one and only one entity as its value.
From now on, therefore, expressions in L ^ are interpreted with respect not just to
a model, M, but also to a variable assignment g. Denotations of expressions are
written between square brackets with the names of the model and the variable
assignment function as superscripts outside the right bracket:
(14) If a is an expression in L^,*, then [a]M» is its denotation with respect to a
model, M, and a variable assignment g.
We need to look at the interpretations of expressions with respect to the same
model, but different assignments of values to variables and, in particular, with
respect to similar variable assignments. For example, we might have a (partial)
variable assignment like that in (15) and we could also have one that is almost
identical except that x is assigned DOG as value as in (16)
(15) g
MANj
y MAN2
z \VOMANj
x
WOMAN2
i CAT
DOG
z
i CAKE
X
2
BOOK
Z
2
X
3
In the case of the variable assignments in (15) and (16), we say that g' is exactly like
g except for the fact that x is assigned DOG as value. The relation between g and
g' can be symbolised as: g' = g000'*, where the notation a/x indicates that a is
assigned as the value of x in the appropriate function. Hence, g000^ is that function
exactly like g but with the value DOG assigned to x, i.e. g\ This notation is useful in
discussing the interpretations of expressions with respect to the same model but
slightly different variable assignments; in particular, variable assignments that may
differ only in the value of one variable.
121
(16) g'
X
\
y MAN2
z \
\
WOMAN2
x
l
CAT
DOG
z
l CAKE
X
2
BOOK
Z
2
X
3
For expressions that contain no variables, the choice of variable assignment

function is irrelevant. If there are no variables, it obviously makes no difference to
the interpretation what variables are assigned to what entities. Hence, the
interpretation of [laugh'(ethel')]M* is identical to [laugh'(etheV)]u*\ where g' is not
identical to g. Such expressions are still assumed to be interpreted with respect to an
assignment of values to variables in order to make the theory more general, i.e.
without having to refer to whether or not an expression contains a variable or not. Just
bear in mind that the assignment is significant only when the interpreted expression
does contain a variable.
The introduction of the variable assignment function provides a way of formally
characterising ())* in (13) in terms of the semantics rather than by the syntactic method
of substituting the argument expression of a X-expression for every occurrence of
the abstracted variable. Given the use of variable assignment functions, the
interpretation of an expression like Xx [(kick'(x))(jo')](chesterf) with respect to a
model, M 2 , and a variable assignment, g, can be defined as being identical to the
interpretation of the expression (kick'(x))(jo') with respect to the same model M 2 but
a different variable assignment function g'. The assignment function required is that
one exactly like g except that the variable bound by X, x in this instance, is assigned
as value the extension of the argument Chester' in M2. In the example given,
[chester']m is the entity DOG, so that g' maps the variable x onto this entity (i.e. g'
is gi*>G/x) Because x is associated with an entity, DOG, the truth value of
[(kick'(x))(jo 'j] M2 g' is determined by first applying the function denoted by kick' in M2
(with respect to g') to the entity DOG and then applying the resulting function to
[jo']M2*\ This is, of course, equivalent to the truth value that is assigned to the
expression (kick'(chester')Xjo') in M2, as required. More generally, then, we can
restate the equivalence in (13) more accurately as in (17), where (() is a propositional
function, and a is an individual constant.
(17) Lambda conversion: [Xx [<|>](a)]M-8 <-> [(|)]M.g[a]M/x
122
The equivalence in (17) needs, however, to be shown to be true. In order to do

this, it is necessary to define how the denotation of a lambda expression is determined
and how the application of this denotation to that of its argument guarantees the truth
of (17). It has already been established that the type of a lambda expression, Ax[(|) ],
is<e,t>and so it must denote a characteristic function (i.e. a function from entities
to truth values), but what function? To find this out, let us consider a concrete
example. The function denoted by the (active) verb liked in M2 (more strictly the
extension of its translation, like') is shown in (19), overleaf. This is equivalent to the
extension of the expression in Mu i.e. the set of ordered pairs {<MAN 1 ,WOMAN 1 >,
<MAN1,WOMAN2>, <MAN1,MAN1>, <MAN1,MAN2>, <WOMAN1,MAN1>,
<WOMAN1,WOMAN2>, <WOMAN2,CAT>, <WOMAN 2 ,WOMAN 1 >}.
It is now our task to work out the denotation of the lambda expression
foc[(like'(x))(jo')] based on this function. Clearly, the characteristic function denoted
by this expression must pick out the set of entities that Jo liked in the model, i.e.
WOMANi, WOMAN 2 ,MAN 2 and himself, M A N L This function can be defined
formally by using different variable assignment functions and checking the truth
value of the expression (like '(x))(bertie') with respect to the model and the
appropriate variable assignment function. To get the function denoted by the lambda
expression, therefore, we ascertain the truth value of [(like '(x))(jo ')]M2<* for every
assignment function g' exactly like g except that x is assigned to a different member
of A, the set of entities in the model. If the truth value of the expression for some
assignment of a value to x is 1 (i.e. true), then in the function denoted by the lambda
expression, the entity assigned to x is mapped onto 1. Otherwise, it is mapped onto
0. To see how this works, let us go through the construction of the characteristic
function denoted by Xx [(like'(x)Xjo')] with respect to M2, step by step.
In the first place, let g be the assignment function in (15). Here x is mapped onto
M A ^ and the value of [(like '(x))(jo ')]m•«is 1 because Jo did like himself. This result
can be checked by comparing the function in (19), where MAN t is mapped onto a
function that maps MANi onto 1. Hence, the information in (18) must be part of
the function denoted by [he [(like '(x))(jo ')]]M2*.
(18)
Next, we look at the truth value of (like '(x))(bertie') with respect to a variable
assignment function exactly like g except that x is mapped on to some other
member of A, say MAN2. This function is gMAN2/* and, as shown by (19),
[(like YJtJX>')]M2'gMAN2/x is 1 and so the function [Xx[(like'(x))(joy)]]M2* maps MAN 2
onto 1 as well, giving (20).
123
(19)
MANj
MAN 2
WOMANj
WOMAN 2
CAT
DOG
CAKE
BOOK
MANj MAN
MAN 2
WOMANj
WOMAN 2
CAT
DOG
CAKE
BOOK
MANj
MAN 2
WOMANj
WOMAN 2
CAT
DOG
BOOK
MANj
MAN 2
WOMANX
WOMAN 2
CAT
DOG
BOOK
124
(20)
This process is repeated with a new variable assignment function like g but where
x is mapped onto WOMAN! to get the value of [(like '(x))(jo') ]M2,8WOMANI/X w hich again
is 1. This is repeated again and again until the value of [(like '(x))(jo')]M2*u* has been
ascertained for every member of A. This yields the complete characteristic function
in (21) which provides the extension of the expression he [(like'(x)Xjo')] with
respect to the model M2 and the variable assignment function, g.
(21)
MAN X
MAN 2
WOMAN!
WOMAN 2
CAT
DOG
CAKE
BOOK
This description of the way in which the denotations of lambda expressions

containing abstracted individual variables are constructed provides the basis for the
formal definition for ascertaining the denotations of such ^-expressions. The
definition is added to the model theory used to interpret Ltype as the new clause which
appears in (22). This restates in a more general and precise fashion how the function
in (21) was determined and introduces no new concepts.
(22) For any model M = «A,F», if ty is an expression of type t containing an
unbound instance of the individual variable x (of type e), then [tac [§]]M<*
is that characteristic function K such that for all individuals, a , in A, K(a)
= 1 if, and only if, [<)>]**.*/* = l and K(a) = 0, otherwise.
(22) guarantees that the equivalence in (13) holds for any model. The reason is
that, according to (13), the truth value of an expression Xx[<|) ](a) with respect to M
and g is given by assigning the value of [a]M«, say a, to x and ascertaining the truth
value of [c|)]M ga/x. Since the denotation of ^x [(j)], according to (22), includes a mapping
from [a]M'8 to [<|>]M-8«*, the equivalence always holds. Hence, the informal syntactic
process of ^-conversion can be carried out validly, because the semantics of X
-expressions in (22) entails that Xx [<|>](a) is always truth-conditionally equivalent to
(J)3*, no matter what model is being used for the purposes of interpretation.
125
Exercise 5.3:
Using the definition in (22) and assuming that the model M2 is the same as M|, except
for parts already specified, work out the extensions of the following L ^ lambda
expressions:
i. Xy [drunk* (y) & lecturer' (y)].
ii. Xz [(give'(bertie')(z))(jo')].
iii. Xx [laugh' (x)].
iv. Xy [~((y = the-lecturer') v (y = the-student'))].
5.2.3 The passive again

The lambda operator, as should be clear from the above discussion, provides an
elegant way of capturing the semantic relation between active and passive verbs, or,
more precisely, verb phrases. As we have already noted, the syntactic subject of a
passive sentence corresponds semantically to the direct object of the active. In order
to capture this relation, all that needs to be done is to translate passive VPs by a
lambda expression that abstracts on a variable in the appropriate position. So, for
example, we can translate the passive VP kicked by Jo as XK [(kick'(x))(jo')] which
denotes the set of things that Jo kicked. Assuming that copular be adds nothing to the
truth-conditions of the passive sentence (beyond tense), this is the expression that is
functionally applied to the translation of the subject. Hence, Chester was kicked by
Jo translates into L ^ . as Xx [(kick'(x))(jo')](Chester'). By lambda conversion, this is
equivalent to (kicky(chestery))(jo') which is, of course, the translation of the active
sentence Jo kicked Chester, exactly as required.
To complete the analysis of the simple passive in G2, here are the translation
rules for the syntactic rules 13G2 to 17G2. The rule in (23.b) is for the passive of
transitive verbs, and so the abstracted variable x appears in the object position nearest
the verb. The two rules for ditransitive passives in (23.c) and (23.d) differ in the place
where the variable appears. In the former rule, the X-operator binds the second, direct
object, position while in the latter it binds the indirect object position, nearest the verb,
capturing the difference between ordinary ditransitive passives like the book was given
to Ethel by Bertie and the ones associated with double object ditransitive sentences
like Ethel was given the book by Berth.
(23) a. T13&: VP[PAS]\
b. T14G2: Xx [V t [PAS]'(x)(PP by ')].
c. T15G,: Xx [Vdt[PAS]'(PPto')(x)(PPby')].
d. T16G2: Xx [V dt [PAS]'(x)(NP')(PP by ')].
e. T17G 2 :NP\
(24) presents a sample derivation and translation of the ditransitive passive
sentence, Ethel was given the book by Bertie. That this translation is equivalent to
that of the active sentence Bertie gave the book to Ethel can be seen by deriving the
translation of the latter. This is left as an exercise to the reader.
126
Generalising lambda expressions
(24) a. S
=>S\
b. NP VP[+FIN] 1G2
=> VP'(NP').
c. N pr VP[+FIN] 6G2
=> V F ( N P / ) .
d. N pr wasVP[PAS] 13G2
=> VP[PAS]'(N P /).
e. Npr was Vdtpas NP PPby 16G2
=> Ax [v d , pas '(x)(NP')(PP b ;)](N p ;).
f. Npr was Vdt.pas the N PPby 7G 2
=> Ax [V d , pa ;(x)(the-N')(PP b ;)](N p ;).
g. Npr was V ^ ^ the N by NP 17G2
=> Ax [V d , pas Xx)(the-N')(NP')](N pr ').
h. Npr was Vdt.pas the N by Npt2 6G2
=> Ax [V d , pas '(x)(the-N')(N pr2 ')](N p /).
i. Ethel was given the book by Bertie Lex
=> Ax [give\x)(the-book')(bertie')](ether)
» give'(ethel')(the-book')(bertie').
Exercise 5.4:
Translate the following passive sentences into L ^ and compare their translations
with their active counterparts:
i. The cake was eaten by Chester.
ii. The book was given to Fiona by Ethel.
5.3 Generalising lambda expressions

Because propositional functions may contain more than one free variable, it is
possible to construct expressions that have more than one instance of the lambda
operator in them. According to the formulation we gave above, however, this is not
possible because we have restricted lambda abstraction to binding individual
variables in propositional functions. There is, however, no sound reason to adhere to
this restriction and the theory benefits from generalising the definition of lambda
abstraction to allow the lambda operator to bind a variable of any type contained
within an expression of any type.
For example, instead of introducing the lambda operator in the translation rules
for the passive VPs, it is possible to state the semantics of the passive participle
directly by manipulating the argument structure of the verb using multiple lambda
abstraction. For the passive of a transitive verb, this allows the first and second
arguments to be permuted in the translation. For example, it is possible to derive the
semantics of the passive participle kicked from that of its active counterpart kicked,
kicks, kick by creating an open formula from the translation of the active verb using
two variables. This would give the propositional function (kick'(y))(x). Because the
direct object argument (the second to last argument to be combined with by a
127
predicate) appears last in the passive, surfacing as the syntactic subject, we abstract
on the y variable first to give the expression Xy [(kick'(y))(x)] oftype<e,t>. This
expression contains the free variable x associated with the subject, or final, argument
of the verb which appears in the passive as the object of the preposition by.
Abstracting on this variable gives an expression of type <e,<e,t»,
Xx[Xy[(kick '(y))(x)]]. Using this approach, the translation of a passive VP like kicked
by Ethel comes out as Xx [Xy [(kick'(y))(x)]](ether), which by lambda conversion is
equivalent to Xy [(kick'(y))(ethel')]y which has the form of the output of the
translation rule T14G2 in (23 .b). The translation of the sentence Prudence was kicked
by Ethel is thus equivalent to the application of this latter expression to the translation
of the subject, i.e. Xy l(kick'(y))(ethel')](prudence') which, of course, is equivalent
to (kick'(prudence'))(ethel') by lambda conversion, as shown in (25).
(25) a. kicked
=*Ax[Ay[(kick'(y))(x)]].
b. kicked by Ethel
=* Xx [Xy [(kick'(y))(x)]](ethel').
c. Prudence was kicked by Ethel
=» (Xx [Xy [(kick'(y))(x)]](ether))(prudence').
d. » Xy [(kick'(y)Xether)Kprudence').
e. » (kick'(prudence' ))(ethel').
If lambda abstraction is allowed to operate over expressions of any type in this
way, then the rule of lambda conversion must be relaxed to allow this. All that needs
to be done is to say that a lambda expression, Ax [<()], applied to an individual
argument is semantically equivalent to $ with the denotation of the argument as
value of the free variable x, as in (26).
(26) Lambda conversion: If <)| is an expression of type a containing a free
instance of x, a variable of typee, and a is an expression of type e, then
[Ax[(|)](a)]M8<-> [(j)]1^* where a is [a]M-«.
When converting lambda expressions with multiple abstractions, it is very
important to make sure that the correct variable is replaced by the correct argument.
(26) requires that lambda conversion be successive. Conversion starts from the
leftmost lambda operator and the first argument after the square brackets
demarcating the scope of the operator, then moves on to the next lambda and the
argument it combines with and so on until there are no more lambdas left. Thus in
(25.c), the leftmost lambda abstracts on the variable x and the first argument after the
square brackets is etheV'. Hence, it is this argument that must replace instances of x.
The second argument prudence' becomes the first argument after this lambda
conversion and because the leftmost X is now associated with the variable v,
prudence' replaces the instance of y to give the correct expression after X-conversion,
i.e. (kick*(prudence'))(etheV). A further example is given in (27) which shows the
necessary steps in the conversion of the triple abstraction in the lambda expression
((Xx[Xy[Xz[(give'(x)(z))(y)]]](fiona'))(bertie'))(the-book')y the translation of the
sentence The book was given to Fiona by Bertie. In this example, each variable is
given a different subscript which also appears with the arguments that are to replace
them. Although strictly speaking unnecessary, this co-indexing of variable and
128
argument shows the dependencies between them very clearly and allows the order of
X-conversion to be more easily traced.
(27) a. » ((Xx, [Xy2 [Az3 [give'(x1)(z3)(y2)]]](fiona)1))(bertie'2))(the-book)3)
b. » (Xy2 [Az3 [give'(fiona' i)(^(y2)]](beitie'2))(the-book'3)
c. » Xz3[give'(fiona'i)(z3)(bertie%)](the-book%)
d. »[give'(fiona',)(the-book'3)(bertie'2)].
Exercise 5.5:
Which of the following lambda conversions are well-formed?
i. Xx [ Xz [(like'(x))(z)](jo')](bertie')
» (like'(jo')Xbertie').
ii. (Xx [XL [(Hke'(x))(z)]]Oo'))(bertie')
» (like'(jo')Xbertie').
iii. (Xx [Xy [(like'(x)Xy) & laugh'(y)]](the-cat'))(ethel')
» (like'(the-caf))(ether) & laugh'(ethel').
iv. Xx [scream'(x)] v Xx [-(laugh'(x))](jo')
» scream'(jo') v -(laugh'(jo')),
v. Xx [scream'(x) v -(laugh'(x))](jo')
» scream'(jo') v-(laugh'(jo')).
•Exercise 5.6:
Construct a lexical rule for passivisation in English along the lines suggested in the
text above.
•Exercise 5.7
Another process by which grammatical functions are altered in English is the one
often referred to as dative shift whereby a prepositional indirect object becomes the
direct object and the direct object appears in second place, as shown in the relation
between a and b below. Using the X-operator, construct a rule that relates the semantic
structure of the non-dative-shifted verb in i to that in ii, making sure that the
paraphrase relation that holds between these sentences is maintained.
i. Ethel gave the cake to the dog.
ii. Ethel gave the dog the cake.
Not only is it possible to abstract over expressions of any type, it is also

possible to abstract on variables of any type. Thus, we can generalise the definition
of lambda abstraction given in (7) to derive lambda expressions of all types. For
example, variables of type t, usually written using the letters p, q and r, may be
abstracted upon in an expression like snow' <->/? to create an expression Xp [snow1
<-» p] of type <t,t>. This denotes the set of all truth values that are identical to that
assigned to snow' in a model. Alternatively, a one-place predicate variable, P, may
apply to an individual constant like the-caf to give the expression P(the-cat'). The
free predicate variable can then be abstracted upon to yield the expression,
)dP[P(the-cat')], of type «e,t>,t> which denotes the set of all sets of which the cat is
129
a member. Indeed, Ltype contains variables of every type and allows any of these to
be bound by the lambda operator in any expression of whatever type. Hence, the
rules for constructing well-formed expressions in Ltype include the general syntactic
rule for lambda expressions in (28).
(28) Lambda Abstraction: If <|) is an expression of type b, and u is an unbound
variable of type a contained in <|>, then hi [(()] is a well-formed
expression of type <a,b>.
The type of a lambda expression is derived from taking the type of the abstracted
variable as the argument type and the type of the expression abstracted over as the
result type. Thus, taking p and q to be variables of type t, an expression like Xp[p
-» q] is of type <t,t>, because p and p -> q are both of type t. The expression
Xq[Xp[p -» q]]y on the other hand, is of type <t,<t,t», because q is of type t and
^ ~* ql is of type <t,<t,t», and so on.
Exercise 5.8:
Given the variables below, what are the types of the lambda expressions that follow?
(Where a is a type, the set of variables of type a can be symbolised as Var a .)
P,Q e Var<e.t> p,q e Vart R e Var<e,<e,t»
x,y,z G Vare A e Var«e.t>.<et>>
i. XR [(RGo')Xbertie')]. ii. Xx [Xy [XL [(give'(x)(y))(z)]]].
iii. X? [XA [A(P)]]. iv. Xp [p <-> rain'],
v. XQ [Q(ether)]. vi. Xp [Xq [p <-> q]].
The interpretation of these generalised lambda expressions is similar to that for

expressions where the abstracted variable is an entity contained in a propositional
function. The denotations, however, have different domains and ranges. For example,
the expression XP [P(chester')] is of type «e,t>,t> and so denotes a function from
sets (denotations of type <e,t>) to truth values. This function maps every set of
entities onto 1 if the extension of Chester* is in the set, i.e. if [P(chester')]M&/p is 1
where s is some subset of A. Otherwise [XP [P(chester')J]M* maps s onto 0. In other
words, the expression XP [P(chester')] denotes the set of all subsets of A of which
Chester is a member. A more complex example is given by the denotation of an
expression that has more than one instance of lambda operator, e.g.
Xx[Xy[(like'(y))(x)]]. This expression denotes a set of ordered pairs: the set of all
pairs of entities <oc,p> such that (3 is liked by ot, or, more formally, {<(3,oc> I <a,(3>
e [like ']M'8}. More examples of informally described denotations of complex lambda
expressions appear in (29) where the variables have the types given in exercise 5.8.
130
(29) a. Xx [(like'(x))(jo') & (hate'(x))(ethel')] denotes the set of all x such that Jo
and Ethel hate x.
b. XP [P(bertie') -» P(chester')] denotes the set of all sets P such that if Jo
is in P then Chester is also in P.
c. XR [(R(chester')Xbertie')] denotes the set of all relations R that hold
between Chester and Bertie.
d. Xy [Xx [give'(y)(tne-book')(x)]] denotes the set of all ordered pairs <x,y>
such that x gives the book to y.
e. XP [Xx [P(x)]] denotes the set of ordered pairs of entities and sets <x,P>
such that x is in the set P.
f. A,R[A,x [(R(x))(chester')]] denotes the set of ordered pairs of entities and
sets of ordered pairs <x,R> such that Chester stands in the relation R to x.
g. XP [XQ [Q(ethel') & P(ethel')]] denotes the set of ordered pairs of sets of
entities <Q,P> such that Ethel is a member of both sets.
Exercise 5.9:
Give informal descriptions of the denotations of the lambda expressions given in
exercise 5.8 along the lines of those in (29).
The formal definition for the denotations of generalised lambda expressions is

very similar to that given in (22) for simple lambda expressions involving only
individual variables and propositional functions. Any lambda expression Xu [(()]
denotes a function from the denotations of the type of the abstracted variable, u, to
denotations of the type of the expression that is abstracted over, i.e. <|). The actual
function denoted by Xu [(()] is one that when applied to any object in the domain of
the function (i.e. of the type of the abstracted variable) yields a semantic object that
is identical to the interpretation of the expression abstracted upon, (|), with respect
to a variable assignment where the abstracted variable, u, is associated with the
object denoted by the argument. The formal definition is given in (30), which forms
part of the model theory interpreting L ^ .
(30) If u is a variable of type a and (() an expression of type b containing a
free instance of u, then [Xu [(|>]]M«8 is that function h from Da into Db such
that for all elements a in Da, h(a) is [(j>]M-»l/u.
To make this definition clearer, consider the interpretation of the expression

Xx[Xy[(like'(y))(x)]]. Informally, as we have seen, this denotes the set of ordered
pairs of entities such that the first is liked by the second. Formally, it denotes a
function from entities to characteristic functions where each member of the set of
entities, A, is mapped onto the characteristic function obtained from the extension
of Xy [(Hkef(y))(x)] wherex is associated with the relevant element in A. For
example, if the function [Xx [Xy [(like'(y))(x)]]]M2s is applied to the entity DOG,
then the result is [Xy [(likeXy))00]]M2'8DOG*. which denotes the set of entities that the
dog likes. Since the latter is truth-conditionally equivalent to the expression
Xy[(like'(y))(chester')], we get the effects of lambda conversion. Hence, it again
follows from the definition of the interpretation of lambda expressions in (30) that
131
lambda conversion holds and Xx [Xy [(like'(y))(x)]](chester') is equivalent to

Xy[(like'(y))(chester')]. Thus, the generalised version of the definition of lambda
conversion in (17) given below is valid for all lambda expressions whatever the type
of the variable abstracted upon and whatever the type of the expression abstracted
over. (But see Chapter 10 for some environments in which it is not valid.)
(31) [XU [<|>](a)]M 8 = [(|>]M.g[a]M/x
To complete the example of generalised lambda abstraction, let us work through

the construction of the function denoted by Xx [Xy [(like '(y))(x)]] in M2. To do this,
it is necessary to construct the denotation of the lambda expression Xy [(like y(y))(x)]
for all values of x. In M2, the extension of like' is the function in (19) above. We
begin constructing the function [Xx [Xy[(like'(y))(x)]J]M2* by taking an arbitrary
member of the set of entities, A, say MAN^ and working out the value of
[Xy[(like'(y))(x)]]M2*MAt*]/x. The latter is, of course, the characteristic function that
picks out the set of things Jo liked, i.e. himself, Bertie, Fiona and Ethel. This function
can be laboriously constructed by working out the value of [(like'(y))(x)]M2*'a * for
all members a of A where g' isg MAN1/ \ Working out the values of (like'(y))(x) with
respect to M2 and the variable assignment function noted above (which may also be
written as gM^m**) yields the function in (32).
(32) [Xy [(like'(y))(x)]]M2-*MAN1/x
MAN 2
WOMAN!
MANt WOMAN^
Next we take another member of A, say DOG, and compute the value of
[Xy[(liket(y))(x)]YA^yoGtx, the set of things that the dog liked. This gives another part
of the extension of X x [X y [(like '(y))(x)]]. Stepping through this procedure for the
remaining members of A we get the function in (33) which reverses the relations
between the entities in the diagram in (19), as can be checked by interpreting the
function in (33) in terms of sets of ordered pairs.
132
(33)
MAN 2
WOMANX
WOMAN 2
CAT
DOG
CAKE
BOOK
MAN
WOMAN
WOMAN 2
MAN CAT
DOG
CAKE
BOOK
WOMAN 2
CAT MAN
DOG MAN
CAKE WOMAN
BOOK
WOMAN 2
CAT
DOG
CAKE
BOOK
MAN 2
WOMANX
WOMAN2
CAT
DOG
CAKE
BOOK
133
(34) [Xy [(like'(y))(x)]]M2'81**5'*
DOG
To further illustrate the formal interpretation of generalised lambda expressions,

let us take a look at a second example expression, one that abstracts on a variable
over one-place predicates, e.g. XP [P(ethel')]. As mentioned above, this expression
denotes the set of all sets of which Ethel is a member. In functional terms, this
is construed as a function from characteristic functions to truth values, where each
characteristic function, k, in the domain is mapped onto 1 if (and only if) WOMAN!
(the extension of [ethel']m*) is mapped onto 1 in k itself, i.e. if [P(ethel')]M2^k/p is
true. For example, in M2 (assuming the denotations in M! are carried over into M2)
the sets that contain Ethel include the sets denoted by the expressions happy' and
laugh' but not the sets denoted by run' or XK [(like'(x))(bertie')]. Hence, the
function in (35) forms part of [XP [P(etheV)]]U2*. The domain of the functions that
expressions like XP [P(ethel')] denote include, not only the extensions of simple
one-place predicates specified by the denotation function of a particular model,
but all the characteristic functions denoted by some expression in the logical language.
This includes the sets of entities liked by Jo, those given the book by the man, those
that were loathed by the cake and more abstruse sets like the set of all entities that
were eaten by the dog and did not poison it. (35) thus only represents a part of the
function denoted by AP [P(ethel')] in M2, but gives enough information to show what
this function looks like.
Exercise 5.10:
Give the formal definitions of the denotations of the following lambda expressions
in M2 and represent these in the way shown in the examples in (33) and (35):
i. Xp [p <-»rain'].
ii. Xz [Xy [Ax [give'(x)(y)(z)]]].
iii. XP [slowly'(P)].
134
(35) [AP [P(ethel'))]]M2*
[ happy']M2'8
MAN2 _ T ^^ 1
WOMANj ~ " ^ 5 ^
WOMAN2 * ^ \
CAT - ^ ^ ^\
DOG — ^ ^ ^ ^\
CAKE T^^r
BOOK - ^ ~
[ laugh' ] M2 '§
MAN
1 ^-^^_
MAN2 — ^ ^^ i
WOMANX ^ ^ T
WOMAN2 ^ \
CAT ^ ^ ^ ^ V ^\
DOG " ~ ^ ^ ^\
CAKE ^ ^ Z a* o
BOOK —"
MANj v
MAN2 . \
WOMANj v \ \ >
WOMAN 2 >X<Â
CAT ONV\
/
DOG \\
LAKE —W 0
BOOK
[\x[like'(x)(bertie')]]]*
MANj v
MAN2 A 1
WOMANj \ \ \
WOMAN2 \ V \ V
CAT - ^ ^ ^ s > ^
DOG ^ ^ ^ ^ ^
CAKE ~ ^ ^ ^ ^ ^= ^ 0
BOOK ~-~~^
135
5.4 Reviewing co-ordination

So far, G2 contains no means of generating compound sentences. This section will
remedy the fault and show how lambda abstraction and an interpretation of
propositional connectives in terms of functions can provide a more adequate account
of English co-ordinate structures. The syntactic analysis of co-ordinate sentences in
Chapter 3 is not the best that can be made. First of all, there is some evidence that
a conjunction like and in English and the clause that follows it form a single binary
constituent, rather than that they both combine with the preceding clause to give a
flat ternary branching structure. Secondly, and more importantly, there are a number
of constructions not handled in G^ in particular conditional sentences where the
antecedent follows the consequent, e.g. Ethel sang, if Chester howled and
non-sentential co-ordination, e.g. Bertie and Fiona loathed Chester. It is not my
intention to discuss the syntax of co-ordination in great detail, but a set of rules is
presented below that remedy the above inadequacies. This analysis is a simplified
version of the GPSG account presented in Weisler et al. (1986) and Gazdar et al.
(1985), one of the better analyses of constituent co-ordination in the generative
literature. In keeping with the spirit of the rest of the book, I will ignore the
alternatives to this account that may be proposed.
5.4.1 Sentential co-ordination

We begin with a revision of sentential co-ordination. Syntactically the new analysis
involves a rule that expands S as two sentential nodes one of which is marked with
a syntactic feature of co-ordination. Another rule then realises this category, S[CONJ],
as a conjunction plus an ordinary sentence. The new rules are given in (36) and a
sample phrase structure tree generated by them appears in (37):
(36) a. 18G 2 :S-»S S[CONJ].
b. 19G2: S[CONJ] -> CONJ S.
c. CONJ -» {and, or, if and only if, if}.
(37)
The dog ate the cake or Ethel didn't laugh
136
Reviewing co-ordination
The interpretation of the logical connectives presented in Chapter 3 is, as we will

see in Chapter 7, broadly adequate as an account of the semantics of the basic
English conjunctions. In view of this, the truth-conditional interpretation of the
connectives is maintained in the new grammar, as, for the sake of convenience, is the
logical syntax of compound formulae familiar from LP. Strictly speaking, in L ^ the
logical connectives which take two formulae to yield a formula have the type
<t,<t,t», denoting functions from truth values to functions from truth values to truth
values. If the connectives are treated as having this type, then by the usual rule of
functional application we should get expressions like (&(p)) of type <t,t> and
(&(p))(q) of type t, where/? and q are variables of type t. However, this notation is not
as perspicuous as the infix notation used in L P and, for the sake of clarity, we import
into Ltype the syntactic rule for generating complex formulae from L P repeated in (38).
(38) a. t -> t Op 2 1.
b. O P 2 e {&,v ,->,<-> }.
This decision, however, poses a problem for the syntactic analysis proposed above,
because the new co-ordination rules introduce the conjunctions as combining with
only one of the two co-ordinated clauses. According to the construction rule in (38),
however, these connectives require two formulae at a time to make a formula, not
one. There is thus no simple translation of rule 19G2 in (36.b), since expressions like
(& S') are not well-formed in LP. It is possible, however, to use lambda abstraction
to help out here, so that instead of translating and simply as &, or as v , if as —> and
if, and only if as <-» , we can translate them as complex lambda expressions that
combine with one formula at a time.
In order to obtain appropriate translations of the conjunctions, we first replace
each of the formulae that they combine with by a different propositional variable, to
get p & qy p <-> qy p v qy and p -> q. These propositional functions are turned into
expressions of the correct type (i.e. <t,<t,t») by using the lambda operator to bind the
variables. This provides the revised translations of the connectives shown in (39).
(39) a. and => Xp [Aq [p & q]].
b. or => Xp [Xq [p v q]].
c. if =»A.p[Aq[p-»q]].
d. if, and only if => Xp [Xq [p <-> q]].
Adopting these translations allows the translation of rule 19G2 to combine the
translations of the conjunction and the clause it combines with simply by functional
application. The translation rule in (4O.a) thus induces translations like Xp [Xq [p
v q]](laugh '(ethel')) for the phrase or Ethel laughed, and Xp [Xq [p —> q]](rain')
for If it rained, expressions which can be reduced by ^-conversion to
Xq[laugh '(ethel') v q] and X q[rain' —» q], respectively. It follows from this that the
type of S[CONJ] must be <t,t> and so the translation of rule 18G2 results directly
from the application of the translation of the sentence containing the co-ordinating
morpheme to that of the higher sentence, yielding a full formula. The translation rules
and appropriate type assignment appear in (40) and give rise to the translation tree in
(41) which parallels the syntactic analysis in (37). The translation of the topmost
sentence node in (41) can be simplified by lambda conversion to the expression in
137
(42.c), the latter being identical to the LP translation of the same sentence. The fact
that the order of the formulae is permuted in these examples from the order found in
previous chapters is semantically insignificant because of the commutativity of v and
&, as discussed in Chapter 3.
(40) a. T19G2: CONJ'(S').
b. TYPE(S[CONJ]) =
c. T18G2: S[CONJ]'(S').
(41)
(Xp [Xq [p V q]]~(laugh'(ether)))((eat'(the-cake'))(the-dog>))
(eat' (the-cake' ))(the-dog')) Xp [Xq [p V q]]~(laugh'(ethel'))
Xp[Xq[pVq]] -(laugh '(ethel'))
(42) a. (Xp [Xq [p v q]]~(laugh'(ether)))((eat'(the-cake'))(the-dog')).

b. » Xq [-(laugh'(ethel')) v q]((eat'(the-cake'))(the-dog')).
c. » [-(laugh'(ethel')) v (eat'(the-cake')Xthe-dog')].
A further example of this approach is shown in the analysis tree of the sentence
It rained if it didn 't snow in (45). This tree assumes a negative counterpart of the rule
introducing impersonal verbs, i.e. S —> it didn't V0[-FIN] which translates as ~(V0)
(call this rule 9'G2). Again although the translation of the topmost sentence node
looks complicated, it reduces to the more familiar logical expression in (43.c).
(43) a. (Xp [Xq [p -> q]](-snow'))(rain').
b. » Xq [-snow' —> q](rain').
c. » -snow' -»rain'.
The grammar does not yet contain a rule for generating sentences containing if...then.
For the sake of completeness, this is remedied by the rules in (44), which specifically
mention the appropriate morphemes. A more elegant solution could be proposed but
this is sufficient for our purposes.
(44) a. 20G,: S -> S[if] S[then].
b. 21G 2 :S[if]-»ifS.
c. 22G2: S[then] -> then S.
138
(45)
(Xp [Xq [p -> q]](~snow'))(rain')
S[CONJ]
Xp [Iq [p -> q]](~snow')
VQ[+FIN] CONJ
rain' ^P \
rained if It didn't VQ[-FIN]
snow
snow
Assuming that the sentences generated by these rules are truth-conditionally

equivalent to those generated by rules 17G2 and 18G2, where CONJ rewrites as //, we
get the translation rules in (46).
(46) a. T20G2: S[if]'(S[then]').
b. T21G2: if (S').
c. T22G2: then'(S').
The translation rule for 22G2 assumes that the expression then is also translated and
so interpreted, just as if it were an ordinary conjunction morpheme. Unlike the others,
however, the word then does not add any semantic information to the sentence. It
should, therefore, be given a translation (and interpretation) that reflects this.
According to the translation rules in (46), then' combines with a formula to yield an
expression that forms an argument to if(S'). The latter, as we have already seen, is an
expression of type <t,t> and so S[then] must be a formula and then' must therefore be
an expression of type <t,t>, as well. Expressions of this type denote functions from
truth values to truth values. One of the four possible functions of this sort is the one
that maps falsity onto falsity and truth onto truth as shown in (47). This function,
which is called the identity function over truth values because it maps a truth value
onto itself, is denoted by a lambda expression of type <t,t> that applies to a formula
to give an expression that is truth-conditionally equivalent to the argument formula.
This expression is Xp [p] which applies to a formula like crazy'(etheV) to give
139
Xp[p](crazy'(ethel')), an expression that ^-converts into crazy'(etheU). This

truth-conditionally empty expression, which denotes the function represented in (47),
thus provides an appropriate translation of the meaningless element then.
(47) [Xp [p]]M*
As an example of the way the rules and translations in (44) and (46) work, the
derivation of the sentence If it didn't snow, then it rained is given in (48). The
translation, as can be seen from the lambda conversion equivalents, turns out to be
truth-conditionally synonymous with It rained, if it didn't snow, as desired.
(48) a. S
=>S\
b. S[if] S[then] 20G2
=> S[if]'(S[then]').
c. ifSS[then] 21G,
=> (Ap [Xq [p -> q]](S'))(S[then]').
d. if S then S 22G,
e. if it didn't V0[-FIN] then S 9'G2

=> (Ap [Xq [p -> q]](~V0'))(Xr [r](S')).
f. if it didn't V0[-FIN] then it V0[+FIN] 9'G,
=» (Ap [Xq [p -> q]](~V0'))(Xr [r](V0')).
g. if it didn't snow then it rained Lex
=> (Ap [Xq [p -> q]](~snow'))(Xr [r](rain'))
h. » Xq [(-snow') —> q](A,r [r](rain'))
i. » [(-snow') -> (Xr [r](rain'))]
j. » [(-snow') -» (rain')].
Exercise 5.11:
Translate the following formulae into L ^ using grammar G2, giving the full
translations and their converted equivalents:
i. The dog ate the cake and the cake poisoned the dog.
ii. Ethel was crazy or the student didn't like Ethel and Chester didn't like
Ethel,
iii. Jo laughed, if Chester howled or the cat yowled.
The truth-conditional connectives are interpreted in the same way as in Chapter

3 and, as in the model theory of that chapter, the truth-conditions of compound
formulae are specified directly in the model theory used to interpret L^pe. The full
specification of this theory in (49) includes clauses for interpreting lambda abstraction
and negation, as well as for functional application and the connectives. This will
140
remain the basis of the interpretation procedure throughout the rest of the book, just
as Ltype will remain the basis of the translation language. There will, however, be
additions to the theory in later chapters as more data are covered.
(49) Given a model, M = «A,F» and assignment of values to variables g,
then:
1. For any item a in the lexicon, [a] M «= F(a ).
2. If a is of type a a n d / is of <a,b>, then \f(a)]M* = (/]M<[tf]M<8).
3. If u is a variable of type a and 0 an expression of type b containing a free
occurrence of u, then [Xu/<t> ]] M 8 is that function h from Da into Db such
that for all objects, a in Da, h(a) is [c|)]M8a/u.
4. If <>| is a formula, then [~(<t>)]M* is 1, iff [c|)]M-8 is 0. Otherwise, [~((|>)]M-8 is 0.
5. If (|> and \\f are formulae, then [(|) & \j/]M« is 1, iff [(|>]M'8 is 1 and [\|/]M8 is 1.
Otherwise, [$ & \\f]M* is 0.
6. If (() and \\f are formulae, then [<() v \j/]M-8 is 1, iff [<|)]M'8 is 1 or [\j/]M-8 is 1.
Otherwise, [$ v \|/]M<8 is 0.
7. If ()) and \|/ are formulae, then [(() —> \|/] M s is 1, iff [<|)]M'8 is 0 or [\\f]M<& is 1.
Otherwise, [(|) -> \j/]M•« is 0.
8. If (() and \j/ are formulae, then [()) <-^ \j/]M 8 is 1, iff [c))]M-s is [\j/]Ms. Otherwise,
[()) <-> \j/]M'8 is 0.
As discussed in Chapter 3, the meanings of the connectives are specified directly
in the model theory above because their interpretation does not vary from model to
model. Their interpretations can, however, be modelled as functions in the same way
as other denotations in the grammar. For example, \p [Xq [p & q]] is that function
that when given a true formula yields a function which yields truth when given
another true formula and yields falsity on all other occasions. The functions denoted
by all the connectives, &, v ,—>,<-> and ~, are set out diagrammatically in (50) to
(54) and provide precisely the same information as the truth tables of Chapter 3.
(50) Xp[X
141
(51) Xp[Xq[pvq]]
1 ^ 1
0 ^ 0
W, -a
1 ^ 1
0 • ^
(52)
1 >fc 1
A W A
1 ^ 1
0 ^
0
(53)
1 >fc, 1
n W, A
0 _^
^ ^ 0
142
(54) ~p
5.4.2 Co-ordinating other categories

In the last section, sentential co-ordination was (re-)introduced into the grammar, but
the lambda operator allows an extension of the grammar to include some types of NP
and VP co-ordination, while still maintaining a propositional interpretation of the
conjunctions themselves. The sorts of VP and NP co-ordination that we can now
handle are those like (55.a & b) which are truth-conditionally identical to the
sentences in (55.c & d), respectively.
(55) a. Jo loathed the dog and liked the cat.
b. The student and the lecturer loathed Chester.
c. Jo loathed the dog and Jo liked the cat.
d. The student loathed Chester and the lecturer loathed Chester.
The phrase structure rules to generate VPs and NPs conjoined by and and or are
relatively straightforward and the appropriate rules are given in (56). In these rules,
(p stands for any features associated with the VP (e.g. [+/-FIN]) which are passed
down onto both conjuncts and the PL marking on the mother NP node is for
agreement purposes. In the latter case, conjunction is restricted to subject NPs (again
to avoid unnecessary confusion) and so some slight revision needs to be made to rule
1G2 to allow for singular and plural verbs. The revised Rule 1G2 appears in (56.e),
but the other revisions to make sure that the right forms of verbs appear in the right
places are again left as an exercise for the reader. Because the conjunctions if and if
and only if do not appear in VP and NP co-ordination the rules must be restricted to
the conjunctions and and or which can be done by specifying the value of the
conjunction.
(56) a. 23G2: VP[q>] -» VP[<p] VP[cp,CONJ:a] a e {and, or}.
b. 24G2: NP[+PL] -> NP NP[CONJ:a] a e {and, or}.
c. 25G2: VP[cp,CONJ:a] -> a VP[q>].
d. 26G2: NP[CONJ:a] -> a NP.
e. 1' G2: S -> NP[aPL] VP[+FIN,aPL].
Since the truth-conditions of sentences like (55.a & b) are identical to those of
(55.c & d), respectively, the conjunctions and and or should have the same
interpretation in sentences containing NP and VP co-ordination as in sentential
co-ordination. But this gives rise to a problem in the translation of rule 23G2 and
24G 2 :the conjunctions are of type <t,<t,t» and so cannot directly combine with
expressions of type <e,t> (the type of a VP) or of type e (the type of a NP). To see
143
how the lambda operator can solve this apparent problem, let us work through the
analysis of (55.a) backwards, as it were, undoing lambda conversions as we go
along. First of all here is the syntactic analysis of the sentence according to G2. (The
numerical subscripts on the VP nodes are for ease of reference.)
(57)
NP[-PL]
Jo Vt[+FIN,-PL] NP[-PL] and VP4[+FIN,-PL]
loathed the N[-PL] Vt[+FIN,-PL] NP[-PL]
dog liked the N[-PL]
cat
As mentioned above Jo loathed the dog and liked the cat is truth-conditionally
equivalent to Jo loathed the dog and Jo liked the cat. The translation of the latter
sentence is shown in (58.a) and, because of the equivalence noted above, this
translation can be analysed as the ^-converted equivalent of that of the S node in (57).
According to the translation rule T1G2, (58.a) must result from X-converting an
expression derived from the application of the translation of VP! to that of the subject
NP as in (58.b). Tofindthefirststage of the ^-conversion the result of which is shown
in (58.b), the rule is applied backwards to (58.a) in order to arrive at an expression
that is truth-conditionally equivalent to VPj. Since the argument of VP/ is jo\ each
instance of the individual expression jo' in (58.a) is replaced by an individual variable
x which is then abstracted on to give the expression in (59.c). This expression when
applied to the argument jo* gives an expression equivalent to (58.a).
(58) a. (loathe'(the-dog'))(jo') & (like'(the-cat'))(jo').
b. VP,'(NP') = VPr(jo')
» (loathe*(the-dog'))(jo') & (like'(the-cat'))(jo').
144
(59) VP,'
» A,x [(loathe'(the-dog')Xx) & (like'(the-cat'))(x)].
The syntax of the co-ordinate VP consists of two VPs, one of which contains a
conjunction morpheme. In order to obtain the correct type for VPj ', the latter should
translate into a functor over VP expressions, i.e. of type «e,t>,<e,t» so that the
translation rule T23G2 must be derived from the application of VP[CONJ]' to VP' as
shown in (6O.b). Thus, the expression in (59), translating VP,, must be derived from
the application of VP3' to VP2 (cf. (6O.c)). Since VP2 translates directly into the
expression loathe'(the-dog'), we can apply backwards X-conversion to (59) by
replacing the latter with a variable P, of type <e,t>, and abstracting on this to give the
expression in (6O.f), of type «e,t>,<e,t» as required.
(60) a. TYPE(VP[CONJ:oc]') = «e,t>,<e,t».
b. T23G2: VP[CONJ:oc]'(VF).
c. VP3[CONJ:and]'(VP2')
» Xx [(loathe'(the-dog'))(x) & (like'(the-cat'))(x)].
d. VP2'=loathe'(the-dog').
e. VP/
» XP [Xx [P(x) & (like'(the-cat'))(x)]](loatheJ(the-dog0).
f. VP3[CONJ:and]'
» XP \Xx [P(x) & (like'(tne-cat'))(x)]].
The translation in (6O.f) is derived in turn from the combination of the translations
of the conjunction and and the VP liked the cat. The translation of the former is A/?
[Xq [p & q]], as we have already seen, and that of VP4 is like'(the-cat'). However,
because the type of the translation of the conjunction is <t,<t,t» and that of the VP
is <e,t>, the combination of these expressions cannot result directly from the
functional application of one expression to the other. Hence, the translation of the rule
expanding VP[CONJ:oc], rule 25G2, must directly specify the way these translations
are combined. As can be seen from (6O.f), the rule must introduce and abstract upon
the predicate variable P and the individual variable x, since these do not appear in the
translations of and or liked the cat Furthermore, the expressions P(x) and
like'(the-cat') must be associated with the p and the q variables in the translation of
the conjunction (the order is immaterial because of the commutativity of & as
mentioned in Chapter 3). Hence, (6O.f) must be derived by ^-conversion from the
expression in (61 .a) and the appropriate translation rule for VP[CONJ:a] must be that
in(61.b).
(61) a. VP3' = X? [Ax [ftp [Xq [p & q]](P(x)))((like'(the-cat)))(x))]]
» X? [Xx [P(x) & (like'(the-cat'))(x)]].
b. T25G2: AP [Ax [(oc'(P(y)))(VP'(y))]].
(62) shows that the rules that we have come up with for VP co-ordination by applying
backwards A,-conversion work in the way intended by giving the derivation and
translation of the example sentence Jo loathed the dog and liked the cat. The X-
conversion of the resulting complex expression is carried out in (63) to show its
equivalence to (58.a) as required.
145
(62) a. S
=>S\
b. NP VP,[+FIN,-PL] VG->
=> VP,'(NF).
c. NprVP,[+FIN,-PL] 6G2
=> VP/CNp/).
d. Npr VP2 [+FIN,-PL] VP3[CONJ:and,+FIN,-PL] 23G2
=> (VP3[CONJ:and]'(VP2'))(Npr').
e. Npr VJ>FIN,-PL] NP VP3[CONJ:and,+FIN,-PL] 3G,
=> (VP3[CONJ:and]XV;(NF)))(Np/).
f. Npr Vf+FINrPL] NP and VP4[+FIN,-PL] 25G,
=> (AP [Xx [(Xp [Xq [p & q]](P(x)))(VP4'(x))]](Vt'(NP')))(Np/).
g. Npr V{+FIN,-PL] NP and Vt NP 3G,
=> (XP [A.x [(Ap [Xq [p & q]](P(x)))((V;(NP'))(x))]](Vt'(NP*)))(Np;).
h. Npr Vf+FINrPL] the N and Vt NP 7G2
=> (XP [Xx [(Xp [Xq [p & q]](P(x)))((V;(NP'))(x))]](Vt'(the-N')))(Np/).
i. Npr VJ>FIN,-PL] the N and Vt the N 7G2
=> (XP [Xx [(Xp [Xq [p &q]](P(x)))((V;(the-N'))(x))]](Vt'(the-N')))(Np/).
j. Jo loathed the dog and liked the cat Lex
=> (XP [Xx [(Xp [Xq [p & q]](P(x)))((like'(the-cat'))(x))]]
(loathe'(the-dog')))(jo').
(63) a. (XP [Xx [(Xp [Xq [p & q]](P(x)))((like'(the-cat'))(x))]]
(loathe'(the-dog')))(jo').
b. » (XP [Xx [Xq [P(x) & q]((like)(the-cat)))(x))]](loatheXthe-dogO))Go').
c. » (XP [Xx [P(x) & (like'(the-cat)))(x)]](loathe>(the-dog')))Go>).
d. » Xx [(loathe'(tne-dog'))(x) & (like'(the-cat'))(x)](jo').
e. » [(loathe'(the-dog'))Go') & (like'(the-cat'))Go')].
This method of arriving at translation rules, while perhaps not strictly
interesting from a purely semantic viewpoint, does illustrate one of the advantages
of the semantic theory being advanced here. It is possible, using this formal theory,
to work out the meanings of expressions in a language from the interpretation
of expressions with which they are synonymous and from the interpretations of the
entailments of the phrase. So from the equivalence of (58.a) and (58.c) we were able
to work out that the translation of the VP and liked the cat is XP [Xx [P(x) &
(like*{the-caf))(%)]], something that was not previously obvious. By working
backwards like this, we can be sure that entailment and paraphrase relations are
maintained in the semantics and that constituents of complex sentences are
assigned appropriate denotations.
146
Exercise 5.12:
It is also possible to co-ordinate transitive verbs as in the following sentences:
i. Ethel loathed and detested Jo.
ii. The farmer shot and killed the dog.
iii. The student loved or loathed the lecturer.
State a syntactic rule and an appropriate translation rule like that for VP co-ordination
above to generate co-ordinate transitive verbs, making sure that the translations given
to the above sentences are truth-conditionally equivalent to the respective co-ordinate
sentences that follow:
iv. Ethel loathed Jo and Ethel detested Jo.
v. The farmer shot the dog and the farmer killed the dog.
vi. The student loved the lecturer or the student loathed the lecturer.
A similar process can be gone through to obtain the translations oftheNP

co-ordination rules which are given in (56.b) and (56.d). We begin by noticing the
truth-conditional equivalence between the sentences in (64.a) and (64.b), the latter of
which has the simple translation in (64.c).
(64) a. Ethel or Bertie was the golfer.

b. Ethel was the golfer or Bertie is the golfer.
c. (ether = the-golfer') v (bertie* = the-golfer').
The formula in (64.c) is truth-conditionally equivalent to some expression constructed
from the translations of the VP was the golfer and the NP Ethel or Bertie. The former
is given by the L ^ expression in (65.a) (denoting the set of entities that are identical
to the golfer). An LtyPe expression that combines with this to give an expression that
X -converts to (64.c) is given in (65.b) which may be taken to be the translation of the
subject NP.
(65) a. was the golfer => Xx [x = the-golfer'].
b. Ethel or Bertie => XP [P(ethel') v P(bertie')].
(65.b) must be derived from the translations of or Bertie and Ethel, i.e. ethel\ while
the former is derived from the translations of or, i.e. A,p [^q [p v q]], and Bertie, i.e.
bertie \ Without going into further details, the appropriate translation rules and type
assignments are given in (66) and the derivation of the NP Ethel or Bertie according
to these is given in (67). (68) provides the appropriate ^-conversion showing the
equivalence of the translation to (65.b).
(66) a. TYPE(NP[CONJ: a]') = <e,«e,t>,t».
b. TR 24G2: NP[CONJ:a ]'(NP').
c. TR 26G2: Ax [XV [(a'(P(x)))(P'(NP'))]].
147
(67) a. NP
=>NP\
b. NP NP[CONJ:or] 24G2
=> NP[CONJ]'(NF).
c. NprNP[CONJ:or] 7G2
=> NP[CONJ]'(Npr').
d. N pr orNP 26G2
=> Ax [XP [(Xp [Xq [p v q]](P(x)))(P(NF))]](Npr').
e. N pr orN pr 7G2
=> Xx [XP [(Xp [Xq [pvq]](P(x)))(P(Npr'))]](Npr').
f. Ethel or Bertie Lex
=> Xx [XP [Xp [Xq [p v q]](P(x)))(P(bertie'))]](etheD.
(68) a. Xx [XP [(Xp [Xq [p v q]](P(x)))(P(bertie'))]](ether).
b. » Xx [Xp [Xq [P(x) v q](P(bertie'))]](ethel').
c. » X,x [XP [P(x) v P(bertie')]](ethel').
d. » XP [P(ethel') v P(bertie')].
There is an important thing to notice about the expressions that result from the
application of the rules in (66) (and the equivalent expressions in (68)) and that is their
type. In (68.d), we have a formula containing two instances of a predicate variable, P,
bound by the X operator. Since the type of the variable is <e,t> and that of the
expression in its scope is t, the type of the whole expression is «e,t>,t>, denoting a
function from sets to truth values. This is not, however, the type previously assigned
to noun phrases, i.e. e, which we have been assuming is appropriate up until now. This
leads us into our next topic, the interpretation of noun phrases generally, which is
taken up in Chapter 6.
*Exercise5.13:
It is often argued nowadays that auxiliary verbs in English like does/didn 7 form a
constituent with its dependent content VP (e.g. like the cat) (cf. Pullum and Wilson
(1977), Gazdar, Klein and Sag (1981)). However, G2 assigns a flat structure to
sentences negated by didn 't. Revise the grammar to capture the fact that didn 7 VP
is a constituent, giving didn't a translation that ensures that a sentence like Ethel
didn't like the cat is a paraphrase of It is not the case that Ethel likes the cat.
*Exercise5.14:
G2 has three rules for VP that introduce the positive copula was and an additional two
sentential rules introducing the negative copula wasn 't. Furthermore, the grammar
gives no translation to the copula, encoding the appropriate translations directly
into the rules themselves. Revise the grammar so that negative and positive copulas
are introduced by the same rules (hint: use a feature NEG in the syntax) giving the
positive and negative copulas appropriate translations. In your answer, you should
have noticed that you have to give the copula appearing with an NP complement
a different translation to that appearing with A or VP[PAS]. Propose a way to
modify the grammar so that the interpretation of the copula remains constant but
148
Further reading
the translation of a NP in predicate position has a different translation to its

non-predicative counterpart, one that is of the same type as a predicative adjective
(hint: you may find the identity operator useful here).
5.5 Further reading

A logical account of the lambda operator can be found in Church (1956), and
linguistic introductions to the operator are found in Partee, ter Meulen and Wall (1990:
338-351), Dowty, Wall and Peters (1981: 98-104), Allwood, Andersen and Dahl
(1977: 155-157), Chierchia and McConnell-Ginet (1990: 318-330) and McCawley
(1981: 395-401). Applications of the ^-operator are not hard to find. Dowty (1982)
uses it to discuss cross-linguistic analyses of passive formation and other
grammatical-function-changing operations. The approach to co-ordination discussed
in Section 5.4 owes a lot to the discussion in Montague (1973) and the specific
analyses of Sag, Wasow, Gazdar and Weisler (1985) and Gazdar, Klein, Pullum and
Sag (1985: ch. 8). (A more sophisticated approach to the co-ordination of all syntactic
categories within a slightly different framework can be found in Keenan and Faltz
(1985) and Partee and Rooth (1983).) Partee, ter Meulen and Wall (1990: 351-367)
has a discussion of linguistic applications of the ^-operator including passive, phrasal
co-ordination, VP-deletion, and relative clauses. These constructions are also
discussed in Chierchia and McConnell-Ginet (1990: 330-348) which also discusses
VP anaphora. The interpretation of unbounded dependencies in Gazdar, Klein, Pullum
and Sag (1985: ch. 7; 229-236) relies heavily on the use of this device. (See also
Cresswell (1973).)
149
6 Quantification
6.1 The variety of noun phrases

We have so far in this book been looking at the meaning of sentences primarily in
terms of the properties that entities have and the relations that hold between them.
This has meant a concentration on the interpretations of verbs and verb phrases, with
the meanings of the phrases that serve as their arguments, noun phrases, taking second
place. Indeed, only two sorts of English noun phrase have been analysed in the
grammar fragments so far: proper names and simple definite noun phrases. But there
are, of course, many other types of noun phrase in English, including indefinite noun
phrases like those in (La & b), quantified noun phrases as in (l.c & d), noun phrases
containing adjectives or relative clauses (l.e & f), noun phrases with possessive
modifiers (l.g), and many more.
(1) a. a book.
b. some cat.
c. every dog.
d. each person.
e. the happy student.
f. any student who gives a good report.
g. Ethers friend's dog.
The noun phrases in (1) all require a more sophisticated interpretation than the
one supplied for proper names and definite descriptions in earlier chapters. Both of
these expressions have been translated as expressions of type e, denoting entities in
the model. Such an analysis is, however, not ultimately tenable for proper nouns, is
suspect for definite noun phrases and cannot be sustained at all for the other types of
noun phrase in (1). For example, that noun phrases like (La) and (l.c) cannot denote
a specific entity in the model can be seen from a consideration of the meaning of a
sentence like Every student read a book. If the phrase a book denoted a single entity
unique in any model, say Jane Eyre, then the proposition expressed by this sentence
is true only if all students read the same book. But this is not necessarily what the
sentence means. It may well be true that all the students did read Jane Eyre, but the
proposition expressed is also true if one student read Agnes Grey, one read Villette,
one read Wuthering Heights, and so on. Hence, the phrase a book can pick out
different entities on different occasions, provided only that they are all books. Thus,
an indefinite noun phrase cannot be directly associated with a unique entity in the
model and so cannot be translated into an expression of type e. For similar reasons the
subject NP every student also cannot denote a single individual in the model. The
example sentence is not verified by the fact that one entity, say Jo, was a student and
read Jane Eyre, but is true only if all those entities in the model that are students read
the book.
The appearance of nominal modifiers in noun phrases (as in (l.e) and (l.f)) also
provides evidence against letting NPs denote individuals. In earlier chapters, each
150
Introducing the logical quantifiers
definite description is associated with a unique entity in the model by the denotation
assignment function. Thus, the dog is specified as extensionally denoting DOG and
the student MAN^ etc. If this is maintained for a larger fragment of English including
modifiers, the denotation assignment function of each model must be extended to
provide a unique extension for every modified and non-modified definite NP in the
object language. Hence, not only will an extension be assigned to the student, but also
to the happy student, the tall, happy student, the happy student with a cold, and so on.
One argument against this approach is that, if attributive modification is recursive, as
usually assumed in the generative literature, then the denotation assignment function
will be required to assign an extension to an infinite number of phrases, and will thus
itself be infinite. A more cogent argument against treating modified definite NPs as
entity-denoting expressions concerns the fact that there are clear semantic relations
between all these noun phrases. The extension of the phrase happy student is related
to those of happy and student and the extension of tall happy student is related to those
of tall and happy student and so on. The information provided by nominal modifiers
in definite noun phrases narrows down the domain in which the referent of the definite
description is sought. In a roomful of students, the entity being referred to by the
phrase the student with green hair can be identified by looking at the entities in the
room that are students and seeing which one also has green hair. If the only entity in
the room with these two properties happens to be Fiona, we can identify her as the
referent of the NP without having previously known that the student with green hair
is actually Fiona. In other words, the semantics of definite noun phrases, like those of
other NPs, is compositional, the meaning of the whole being derived from the
meaning of the parts. Hence, even definite NPs must have more semantic structure
than that provided by an individual constant. The approach to NP denotation of
Chapters 2 to 5 is thus inadequate and must be revised to account for the points just
raised. It is the purpose of this chapter to provide the basis for such a theory which
is compositional and assigns meanings to full noun phrases according to the meanings
of the adjectives, nouns and determiners that they contain.
6.2 Introducing the logical quantifiers

We begin our analysis of noun phrases by looking at a now traditional semantic
treatment of NPs like those in (l.a) to (l.d). This analysis comes directly from the
logical tradition and has been adopted by many linguists as a basis for the
interpretation of universal NPs like everyone, all gardeners, every dog, etc. and
existential ones like someone, some gardener, a dog, etc. This analysis goes back to
the work of Gottlob Frege, a German philosopher of the nineteenth century, who first
provided the now standard technical means of analysing sentences containing
quantified noun phrases (QNPs), which have been the object of philosophical study
in the Western tradition since Aristotle. Since this book is about the semantics of
natural languages, many of the logical aspects of this analysis are ignored and this
chapter is concerned only with the contribution it makes to the understanding of
quantified noun phrases in English and other human languages that have been
seriously investigated. As with the analyses of other constructions in this book,
interpretation is not carried out on English sentences directly but on logical
representations of these. Hence to provide a semantic analysis of quantified sentences,
151
6 Quantification
we must first provide a logical language which can represent them and then define the
interpretation of that language. The language that is used to translate English with
noun phrases like those in (1) is based on L ^ , but contains two new logical operators
and is called LQ. In this section, we define this language and discuss its interpretation.
6.2.1 The quantifiers

To begin the analysis of quantified sentences, consider the sentence Someone liked
Jo. This provides the information that some entity liked (at some time in the past)
some, possibly different, entity called Jo and it may thus be paraphrased by the formal
English sentence in (2).
(2) For some entity, that entity liked Jo.
Replacing the word entity in (2) by a variable gives the quasi-logical sentence in
(3) which brings out the fact that the something being talked about is the thing that
liked Jo.
(3) For some z, z liked Jo.
The second part of this expression is easily represented in L ^ as (like'(jo'))(z), a
propositional function containing a free variable, z. Such an expression is not suitable
as the translation of (3), however, as, although it is of the right type to translate a
sentence, i.e. t, it does not denote a truth value because it contains an unbound
variable. To make the propositional function into a formula the free variable must be
bound by some operator which, unlike the lambda operator, must not alter the type of
the logical expression; which must be a formula, since the expression translates a
sentence and not a predicate. The relevant operator is 3 (or V, in some notations) and
is known as the existential operator. Like the lambda operator, 3 is written before
the variable it binds and comes to the left of square brackets which enclose the
propositional function over which the operator ranges. Ignoring for the moment the
fact that the English expression someone is used only of people (or, less commonly,
of other animate entities), the sentence in (2) can be translated into the formula in (4).
(4) 3z [(like'Go')Xz)].
Care is needed in specifying the right scope for the existential operator, as it is
for lambda. Thus, neither (5.b) nor (5.c) are correct representations of the sentence in
(5.a). In the former, (like (jo ))(x) is not in the scope of the operator while in (5.c) it
is in its scope, but the variable x is not bound by the operator. The correct
representation is given in (5.d).
(5) a. Someone liked Ethel and Jo.
b. 3x [(like'(ethel'))(x)] &
c. 3y [(like'(ethel'))(y) &
d. 3z [(like'(ethel'))(z) &
Associated with the existential operator is the universal operator. This is used
in the translations of sentences that contain noun phrases with all and every and is
written as V (or A). This is again prefixed to a propositional function containing the
variable it is associated with and has all the usual properties in terms of variable
152
binding and scope. The representation of (6.a) can be given the quasi-logical
translation in (6.b) which translates into the fully logical expression in (6.c).
(6) a. Everyone liked Fiona.
b. For all x, x liked Fiona.
c. Vx [(like'(fiona'))(x)].
The operators, 3 and V, are more generally known as quantifiers and their use
has now become standard in logic, forming the backbone of the predicate calculus and
more complex logical languages. If we add the quantifiers to the typed language with
the X-operator of Chapter 5, we get the language LQ, which is defined by the same
rules as L ^ with the additional rule introducing quantified formulae given in (7).
(7) Quantifier rule: If (() is an expression of type t containing a free instance
of a variable u of type e, then 3u [(()] and Vu [())] are expressions of type t.
In writing a syntactic rule for generating quantified formulae, logicians do not specify
that (> should contain an unbound instance of the variable to be bound by the
quantifier, as this does not affect the semantics. As we will see below, an expression
like Vx [(like'(fiona'))(jo')] is truth-conditionally equivalent to (like'(fiona'))(jo').
However, the string most nearly equivalent to the former expression in English
Everyone Jo liked Fiona is ungrammatical (as are the equivalent expressions in many
other languages). As the semantic theory being developed in this book is intended for
the interpretation of natural languages, the rule in (7) has been written specifically to
disallow quantifiers from combining with formulae that are not propositional
functions. In LQ, therefore, expressions like Vx [sing*(jo')] are not well-formed, even
though they are perfectly acceptable in ordinary predicate logic.
The definition in (7) allows quantifiers to bind more than one instance of a
variable in its scope (as in (5.d)), and more than one quantifier to bind distinct free
variables in propositional functions. For example, (8) shows four well-formed
formulae that can be constructed from the propositional function (like '(x))(y) using the
two quantifiers, V and B, with an English sentence that it can be used to translate
following each logical expression.
(8) a. 3x [3y
(Someone liked someone.)
b. Vx [Vy [(like'(y))(x)]].
(Everyone liked everyone.)
c. Vx [3y [(like'(y))(x)]].
(Everyone liked someone.)
d. 3x[Vy[(like'(y))(x)]].
(Someone liked everyone.)
Just as with the lambda operator, different formulae can be constructed by binding
variables in different orders. Thus, there are four other well-formed formulae based
on the propositional function (like '(y))(x) which parallel the expressions given in (8).
These are given in (9) which again has examples of English sentences that the logical
expressions can be used to translate.
153
6 Quantification
(9) a. 3y [3x
(Someone liked someone.)
b. Vy [Vx [(like'(y))(x)]].
(Everyone liked everyone.)
c. 3y [Vx [(like'(y)Xx)]].
(There was someone that everyone liked.)
d. Vy [3x [(like'(y)Xx)]].
(Everyone was liked by someone.)
Where two formulae differ in the relative orders of quantifiers binding the same
variables, they show a difference in the scope of the quantifiers. Thus, the formulae
in (9.a to d) reverse the quantifier scopes of the corresponding formulae in (8). As
there are situations in which different orders of quantifiers change the truth-conditions
of a formula, it is necessary to be careful to put them in the right order. This topic
will be taken up in Section 6.4.2, but for the moment look at the logical expressions
in (8) and (9) carefully and compare them with their English counterparts. Where the
latter are the same for more than one of the expressions, the relative order of the
quantifiers is not truth-conditionally significant. With the others, the differences in the
order in which multiple quantifiers appear in expressions is significant.
Exercise 6.1:
Pair the English sentences in i to iv with one of the LQ expressions in a to h which
represents a possible translation into LQ.
i. Someone liked herself.
ii. Someone liked Jo and hated Fiona.
iii. Someone was liked by everyone.
iv. Everyone gave something to someone.
a. 3x [(like'(jo'))(x)] & (hate'(fiona'))(x).
b. 3x [By [(like'(y))(x)]].
c. 3x[Vy[(like'(y))(x)]].
d. 3x [(like'(jo'))(x) & (hate'(fiona'))(x)].
e. 3x [Vy
f. Vx [3y
g. 3x [(
h. Vx [3y [3z
6.2.2 Interpreting LQ
To understand the interpretation of expressions containing quantifiers, we must
consider the conditions under which the formulae translating quantified sentences are
true. We have already seen that a sentence like Fiona liked Jo (or more properly the
formula it translates into) is true if (and only if) the entity conventionally referred to
by the word Fiona is a member of the set of entities denoted by the verb phrase liked
Jo. Once the entity denoted by Fiona has been identified, therefore, and providing the
154
extension of liked Jo is also determined, the truth or falsity of the proposition

expressed by the sentence can be assessed. On the other hand, the utterance of a
sentence like Someone liked Jo (in the interpretation with which we are concerned)
does not assert that a specific entity is in the extension of the verb phrase, but just
states that some entity in the universe of discourse is in that extension. In other words,
if the set denoted by an expression like like'(jo') contains at least one element, then
the formula is true, otherwise it is not. Similarly, the proposition expressed by the
sentence Jo liked someone is true if, and only if, there is some entity in the universe
of discourse such that Jo liked it (or, equivalently for our purposes, if the extension of
was liked by Jo contains at least one element).
On a simple level, one can think about the truth of the propositions expressed by
sentences containing someone as being guaranteed if there is a sentence in the object
language that contains a name or a definite description in place of the quantified
pronoun that can be uttered to make a true statement. Thus, the truth of the proposition
expressed by the sentence Someone liked Jo is guaranteed by that of Fiona liked Jo
or Ethel liked Jo or any of the other sentences formed in this way. This is rather how
the formal definition of the semantics of these sentences is given, except that it does
not involve the substitution of names or definite descriptions in place of the quantified
pronoun (which can give rise to problems in intensional contexts; see Chapters 9 and
10). The formal interpretation of a quantified formula involves the assignment of
different values to the variable bound by the quantifier in the formula and checking
the truth or falsity of the propositional function in its scope with respect to the value
assignment.
In Chapter 5, the operation of lambda-conversion was discussed in terms of
looking at different ways of associating variables with constants in a model. The truth
value of any formula formed by applying a lambda expression of type <e,t> to a
constant of type e in L ^ is the same as that of the propositional function in the
lambda expression when the abstracted variable is assigned the same value as the
constant. In other words, [Xu [(|>](a)]M8 is equivalent to [(flM-gMM* ( s e e Chapter 5). The
interpretation of a formula containing the quantifier 3 also requires us to look at
different assignments of values to variables, except that we may need to look at more
than one. Essentially an existentially quantified formula, 3x [<|>], is true if there is some
assignment of values to variables which makes the propositional function, <|>, true. (10)
expresses this formally.
(10) Truth-conditions for existential formulae: If $ is an expression of type
t containing a free occurrence of a variable, u, of type e, then [3u [<|>]]M8 is
1, iff there is some value assignment g', exactly like g except perhaps for
the individual assigned to u, such that [§]M<* is 1.
As an example of how this works, consider the interpretation of the L Q formula
representing the sentence Someone liked Jo, i.e. 3z[(like '(jo r))(z)]. According to (10),
this is true if there is some value that can be assigned to z that makes (like '(jo '))(z)
true. Assume that the model against which we want to assess the truth of the
proposition is M2 of Chapter 5 and in particular that the extension of like' is that
shown in the diagram in (18) of that chapter. The interpretation of 3z [(like'(jo'))(z)]
with respect to M2 and the value assignment g in (11) requires us to look at value
assignments like g! to g4 in (12), which are all exactly like g except for the value
155
6 Quantification
assigned to z. From the extension of like9 in M 2 shown by the diagram (18) in

Chapter 5, we can see that the value assignment ga is the one that shows that 3z
[(like '(jo '))(z)l is true. This is because g3 assigns the value MAN! to z and the function
denoted by like'(jo') in M2 maps this entity onto 1, showing that Jo liked himself.
Hence, since the variable assignment, g3, is exactly like g except for the value assigned
to z, [3z [(like '(jo '^(z)7]M2<8 is 1 and we say that g3 satisfies the formula (like '(jo f))(z),
because this value assignment makes the formula true.
(11)
x
y -•MAN2
z -•WOMANj
x
l - • WOMAN2
yi -•CAT
z -•DOG
i
X
2 -•CAKE
-•-BOOK
(12)
WOMANX WOMANX
x • WOMAN2 • WOMAN 2
l
yi
z z -•DOG
i i
X
2 -•CAKE X
2 -•CAKE
V
2 -•BOOK -•BOOK
156
y
WOMAN! z WOMAN X
- • WOMAN2 x \ WOMAN 2
l
-•CAT \ CAT
z -•DOG 7-1 \
i z
l
X
2 -•CAKE X CAKE
2
-•BOOK
The interpretation of the universal quantifier follows the same lines, except that
here we require that every value assigned to z make the propositional function true.
In other words, the truth of the proposition expressed by the sentence Jo liked
everyone on some occasion of utterance is guaranteed, if the propositions expressed
by the sentences Jo liked himself and Jo liked Fiona and Jo liked Ethel, and so on, are
all true on the same occasion. The formal interpretation of universally quantified
formulae is given in (13).
(13) Truth-conditions for universal formulae: If $ is an expression of type t
containing a free occurrence of a variable, u, of type e, then [ Vu[(()]] M-« is 1,
iff there for every value assignment g% exactly like g except perhaps for the
individual assigned to u, [c|>]M-8' is 1.
Clearly, the formula Vz [(like'(z)Xjo')] is false with respect to M2 because there at
least one value assignment that makes (like'(z))(jof) false; e.g. gi which assigns the
value DOG to z and <DOG,MAN!> is not in the denotation of [like '] M2 . Thus, while
an existentially quantified formula is verified by a single value assignment that
satisfies the propositional function containing the bound variable, a universally
quantified one is falsified by a single value assignment that does not satisfy its
propositional function.
157
6 Quantification
Exercise 6.2:
Work out the truth values of the following formulae with respect to model M3, where
this model is exactly like M2 with the addition of the denotation assignments shown
below. For each formula, also give an English sentence that might be translated into
it.
F3(exist') = {MAN1,MAN2,WOMAN1,WOMAN2,CAT,DOG,CAKE,BOOK}.
F3(touch') = {<WOMANi,MANi>, <WOMAN1,WOMAN1>,
<WOMAN1,CAT>, <WOMAN1,WOMAN2>, <WOMAN1,DOG>,
<WOMAN!,CAKE>, <WOMAN!,CAKE>, <WOMAN1,BOOK>}.
F3(like') = F2(like') (= (18) of Chapter 5).
i. Vx [(like'(x)Xx)].
ii. 3x [like'(fiona'Xx) & (like'(x))(fiona')].
iii. Vx [exist'(x)].
iv. Vx [(touch' (x))(ethel')].
V. Vx [3y [(like'(x))(y)]].
6.2.3 Quantification and negation

Before tackling the question of how the logical quantifiers can be incorporated into
a theory of natural language semantics (the topic of Section 6.3), let us look briefly
at the way the negation operator, ~, interacts with the quantifiers, 3 and V. Given a
propositional function like (like '(jo '))(z)> there are four possible formulae that can be
constructed using ~, 3 and V, depending on their relative scopes. (14) gives the
relevant LQ expressions and (15) gives sentences in English that can be translated by
the respective formulae in (14).
(14)a. 3z [~((like'(jo'))(z))].
b. Vz [~((like'(jo'))(z))].
c. ~(3z [(like'(jo'))(z)]).
d. ~(Vz [(like'(jo')Xz)]).
(15) a. Someone doesn't like Jo.
b. No-one liked Jo.
c. No-one liked Jo.
d. Not everyone liked Jo.
Notice that two of the formulae in (14) are taken as translating the same English
sentence. Since the sentence No-one liked Jo is not ambiguous, if the two translations
accurately represent its truth-conditional meaning, they must be equivalent, i.e. have
the same truth-conditions. For (14.b) to be true, the interpretation rule in (13) requires
that the formula -((like '(jo f))(z)) be true for all values assigned to z. But the semantics
of negation, of course, means that it can be true only if there is no assignment of a
value toz that satisfies the propositional function (like '(jo '))(z), i.e. if no-one liked Jo.
The formula ~(Bz [(like '(jo '))(z)])> is true, however, if, and only if, 3k [(like '(jo *))(z)]
is false. Since the latter is only false where there is no value assignment that satisfies
158
A compositional approach
(like'(jo'))(z), the formula in (14.c) is true if, and only if no-one actually liked Jo.
(14.b) and (14.c) are, therefore, truth-conditionally equivalent, giving us two ways of
representing sentences containing the pronoun no-one: either with the universal
quantifier having scope over a negated propositional function (i.e. Vx [-<()]) or with
the negation operator having scope over an existentially quantified formula (i.e. ~(3x
It is important to ensure that the relative scopes of the quantifiers and negation
in the representation for no-one that is being used are correct: universal before
negation or negation before existential. This is because formulae that have the scopes
of the operators reversed have very different interpretations. Where the existential
quantifier has scope over the negation operator, as in (14.a), the formula is true if
there is some assignment of a value to x which satisfies ~((like'(jo'))(z)), i.e. there is
some element in the universe of discourse that does not like Jo. Where the negation
scopes over a universally quantified formula, as in (14.d), on the other hand, the
whole formula is true if, and only if, the formula Vz [(like '(jo '))(z)J is false, i.e. if
there is some value assignment to x that does not satisfy (like '(jo f))(z). These two sets
of truth-conditions are the same and so, like (14.b) and (14.c), (14.a) and (14.d) are
truth-conditionally equivalent. However, these truth-conditions are not the same as
those for the former. A single variable assignment that fails to satisfy (like '(jo '))(z)
which is sufficient to guarantee the truth of 3z [~((Hke'(jo'))(z))] and ~(Vz
[(like'(jo'))(z)]) is not sufficient to guarantee the truth of \/z [~((like'(jo'))(z))] and
~(3z [(like'(jo'))(z)]) which require every value assignment to fail to satisfy the
propositional function. Hence, (14.a) (and (14.d)) may be true in models in which
(14.b) and (14.c) are not and so the formulae in (14.a) and (14.d) are not
truth-conditionally equivalent to those in (14.b) and (14.c). Intuitively this is the
correct result, since the formula that translates No-one liked Jo may be false even
though the formula that translates Someone doesn 't like Jo is true. Relative scope is
thus semantically significant and care must be exercised when translating sentences
that contain negatively quantified noun phrases.
6.3 A compositional approach

The previous section gives an introduction to the logical quantifiers, but does not do
so as part of a compositional theory of natural language semantics. No translation
rules from English to LQ have been given and there has been no attempt to analyse
more complex noun phrases involving the quantifiers. This is remedied in this section
where a type-theoretic analysis of NPs is given that follows Richard Montague's
approach. As we shall see, this analysis is general enough to allow the translation and
interpretation of most universally and existentially quantified noun phrases in
English.
6.3.1 Translating quantifier pronouns

In order to simplify matters, let us first work out a way of translating sentences
containing quantifier pronouns in subject position like those in (16) in line with their
syntactic structure. Obviously, the first thing we need to do is add a new syntactic rule
to the grammar fragment G2 that allows a noun phrase to be rewritten as a pronoun.
159
6 Quantification
This is given in (17.a) and with the addition of the pronouns in (17.b) to the lexicon,
we can now analyse the sentence in (17.a) as having the structure in (18).
(16) a. Everyone liked Jo.
b. Someone sang.
c. No-one laughed.
(17) a. 27G 2 :NP->Pro.
b. Pro = {everyone, someone, no-one, everything, something, nothing,
nobody}.
(18)
Everyone
According to the previous section, the translation of sentence (16a) into LQ is

equivalent to the formula in (19):
(19) Vx [(like'(jo'))(x)].
Assuming that proper names are still translated as individual constants (an assumption
that will be revised in Section 6.3), the translation of the VP in (18) is (20).
(20) VP=»like'(jo').
The rule of functional application requires the translation of the pronoun everyone to
combine with the translation in (20) to give an expression that is equivalent to (19).
There are two ways in which VP' and Pro' may be combined: either Pro' is an
argument of VP' or VP' is an argument of Pro'. As we saw in Section 6.1 above, the
pronoun cannot be translated into an expression with the type of an entity and so
cannot serve as the argument of VP. Hence, it must denote a function from VP
denotations to sentence denotations, and so have the type assignment in (21). (Notice
that this type was proposed for conjoined subject NPs at the end of Chapter 5.)
160
(21) T YPE(Pro) = «e,t>,t>.

The type assignment in (21) requires the translation rule for basic sentences to be
revised to allow the subject to act as a functor over the VP translation. The appropriate
revision appears in (22).
(22) T1G2: S =» NP'(VP').
As we have seen, the translation that is assigned to the sentence Everyone likes
Jo is truth-conditionally equivalent to the formula in (19). Assuming that this
equivalence follows from A,-conversion, we may apply this rule backwards to arrive
at the translation of the unconverted formula and hence recover the translation of the
pronoun everyone. Thus, from (22) we can infer the schematic conversion in (23.a),
since the expression like '(jo') is the translation of the VP. To get the translation of the
pronoun we can replace this expression in the formula in (19) by a predicate variable
to get the propositional function in (23.b). As this expression is of type t and contains
a variable, P, of type <e,t>, we can abstract on this to obtain the expression in (23.c)
which is of type «e,t>,t>, the type required for subject NPs. The resulting expression
provides an appropriate translation of the pronoun everyone and the application of this
expression to the VP translation like '(jo') gives (23.d) which can be X-converted into
(19), as required.
(23) a. Pro'(like'Go'))
» Vx [(like'Go')Xx)].
b. Vx[P(x)]. (typet)
c. XP [Vx [P(x)]]. (type «e,t>,t>)
d. XP [Vx [P(x)]](like'Go'))
» Vx [(like'Go')Xx)].
Exercise 6.3:
Using the sentences in (16.b) and (16.c), provide translations for the pronouns
someone and no-one along the lines of that given for everyone above.
The domain of denotations for expressions with the type assigned to quantifier
pronouns in (21) (and other quantified noun phrases) is the set of functions from sets
of entities (VP extensions) to truth values (S extensions), as shown in (24). We may
thus think of a QNP as denoting a function that maps the extension of a VP (a
characteristic function) onto the value 1 if it satisfies the property expressed by the
quantifier, and 0 otherwise. Since a function from some denotation type onto the truth
values is a characteristic function, another way of defining a set, QNPs are analysed
as denoting sets of sets of entities in the model.
(24) a. D<<et>tt> = Ey*w> = {0,1}«01>A>.
161
6 Quantification
b.
A {1,0}
As an illustration, consider the denotation of the pronoun someone. As shown in

the answer to Exercise 6.3, this translates into an expression AP [Bx [P(x)]], again of
type «e,t>,t>. According to the interpretation given to the existential quantifier in
(10), the translation of the pronoun combines with a VP translation to give a true
formula, if the extension of VP' contains at least one element. Hence, [A,P [3x
[P(x)]]]M« is that function that maps [VP']Ms onto 1 if [VP']Ms does not denote the
empty set and onto 0 if it does. Hence, we may say that [XP [3x [P(x)]]]M'« is the set
of all subsets of A (the universe of discourse) that are non-null. In set-theoretic terms,
this can be represented as (25.a).
(25) [X? [3x [P(x)]]]M'«= {X c A I X
Exercise 6.4:
What are the extensions of the translations of everyone and no-one in set-theoretic
terms?
6.3.2 Complex NPs

The translation and interpretation of quantifier pronouns presented above provides the
basis of an approach to the treatment of quantified noun phrases of all sorts. It needs,
however, to be extended to cover more complex QNPs like those in (26.a) to (26.c)
and to bring out the fact that the quantifier pronouns ending in -one are applicable
only to human entities and those ending in -thing (primarily) to non-human ones, as
in (26.d) and (26.e).
(26) a. Every student liked Jo.
b. A lecturer screamed.
c. No woman laughed.
d. No-one moved.
e. Nothing moved.
Information is conveyed in these sentences by the common noun in the subject
NP and so this must be represented in their translations. To incorporate the noun
translations, we first need to know their type. Common nouns, like verb phrases,
extensionally denote sets of entities: sets that consist of just those entities of which it
is true to say That is a N, for any common noun N. Thus, the extension of a noun like
chair is the set of all chairs, that of student is the set of all students, and so on. The
162
type of a common noun must, therefore, be that of a set-denoting expression, i.e.

<e,t>. Given this type assignment, the translations of sentences like those in (26) is
based on two open formulae which contain instances of the same variable: one
formula derived by applying the translation of the VP to a variable and the other
derived by applying the translation of the common noun to an instance of the same
variable. For example, the translation of the sentence Every student liked Jo in (26.a)
is derived from some combination of the open formulae student'(x) and (like '(jo '))(x).
The free variable in the open formulae must be bound by a quantifier appropriate to
the determiner and the two open formulae must be combined by some propositional
connective, &, v, -» or <->, to form a well-formed formula with no free variables.
The question of how the two open formulae are connected can be answered by
considering the truth-conditions associated with particular sentences. For example, the
formula associated with the sentence in (26.a) is true if everything in the model that
is a student also liked Jo. This interpretation can be made more explicit by
paraphrasing (26.a) as the semi-formal English sentence in (27.a) which can be
translated directly into LQ as the formula in (27.b).
(27) a. For every x, if x was a student, then x liked Jo.
b. Vx [student'(x) -> (like'(jo'))(x)].
Given the interpretation rule for the universal quantifier, the formula in (27.b) is true
as long as every value assignment to* makes the formula student'(x) -» (like '(jo '))(x)
true. By the semantics of material implication, this is the case if every constant
instantiating the variable in a value assignment is in the extension of like '(jo'), if it is
also in the extension of student'. In other words, the formula in (27.b) is true provided
[student']M>& is a subset of [like'(jo')]M*. This situation can be represented by the
diagram in (28) where the main box represents the set of entities in the model.
(28) Every student liked Jo
[student']M'g
163
6 Quantification
To get the translation of a noun phrase like every student we can apply
backwards lambda-conversion to (27.b). The revised translation rule for basic
sentences in (22.a) requires this formula to be derived by the application of the noun
phrase translation to that of the verb phrase. Since the VP translation is like'(jo') (of
type <e,t>), we can replace this expression in (27.b) with a predicate variable to get
an open formula and then abstract on this to get an expression of type «e,t>,t>, as is
done in (29). This yields (29.d) as the translation of the NP every student.
(29) a. NP'(VP')
» Vx [student'(x) -^ (like'(jo'))(x)].
b. NP'(like'(jo'))
» Vx [student'(x) -> (like'(jo'))(x)].
c. XP[Vx [student' (x) ->P(x)]](like'(jo'))
» Vx [student'(x) -> (like'0o'))(x)].
d. NP => XP [Vx [student'(x) -> P(x)]].
We have not, of course, completed the analysis of the example sentence, as we
still need to show how the NP translation in (29.d) is obtained. Syntactically the
phrase every student is made up of a determiner (every) followed by a noun (student)
so the rule in (3O.a) needs to be added to the grammar G2, as well as a list of
determiners to the lexicon (3O.b).
(30) a. 28G 2 :NP->DetN.
b. Det —> {every, a, some, no,...}.
As we have seen, quantified NPs have the type «e,t>,t> and nouns have the type
<e,t>. This means that determiners must have a type that takes a noun expression into
a noun phrase expression, i.e. « e , t > , « e , t > , t » , yielding the translation rule in (31)
for the rule in (30.a).
(31) T28G 2 :NP=>Det'(N').
Since we are treating Det' as a functor over N', we can apply backwards X-conversion
to (29.d). To get the translation of the determiner every, the translation of the common
noun in (29.d) is replaced by another predicate variable, Q, which again is bound by
the lambda operator, as in (32.c). This gives (32.d) as the appropriate translation for
the determiner every.
(32) a. Det'(N')
»>JP[Vx [student'(x) -> P(x)]].
b. Det'(student')
»XP[Vx [student'(x) ->P(x)]].
c. XQ [XP [Vx [Q(x) -> P(x)]]](student')
»^P[Vx [student'(x) ->P(x)]].
d. every =» XQ [XP [Vx [Q(x) ->P(x)]]].
The analysis tree for (26.a) is shown in (33) with the translations of the
constituent phrases shown as usual on the nodes. Applying lambda-conversion to the
translation of the S node gives the equivalent formula in (27.b), as readers can check
for themselves.
164
(33)
(XQ [XP [Vx [Q(x) -> P(x)]]](student'))(like'(jo'))
XQ [XP [Vx [Q(x) -> P(x)]]](student')
Det N Vt[+FIN] NP
XQ [XP [Vx [Q(x) -> P(x)]]] student' Hke > jo'
every student liked Npr

jo'
Jo
The translations of some (or a) and no are determined in the same fashion, but
there is a difference in the way the open formulae formed from the noun and VP
translations are connected. The existential sentence in (26.b) translates into a formula
that is true if, and only if, there is something that was a lecturer and that something
sang. Hence, a conjunction is used in place of the material implication of the
universally quantified formula. The force of this may be brought out by the
semi-formal English sentence in (34.a) and its translation into LQ in (34.b).
(34) a. For some x, x was a lecturer and x screamed,
b. 3x [lecturer' (x) & scream'(x)].
According to the truth-conditions for formulae containing 3, given in (10), above,
(34.b) is true if (and only if) there is at least one entity in A that satisfies both
lecturer* (x) and scream '(x). Set-theoretically, this requires that the intersection of the
extension of the common noun in the subject NP and that of the verb phrase is not
empty, a situation that can be represented by the diagram in (35).
165
6 Quantification
(35)
A lecturer screamed
[lecturer']M'S [scream']M>8
As before, it is possible to get to the translation of the NP a lecturer and that of

the determiner by applying lambda-conversion backwards. The steps are shown in
(36). (36.c) shows the translation of the NP a lecturer applied to the VP translation
and (36.f) that of the determiner a applied to its head noun and then to the VP
translation. This leaves (36.g) as the translation of the determiner a.
(36) a. NF(VF)
» 3x [lecturer*(x) & scream'(x)].
b. NP'(scream')
» 3x [lecturer'(x) & scream'(x)].
c. XV [3x [lecturer'(x) & P(x)]](scream')
» 3x [lecturer'(x) & scream'(x)].
d. (Det'(N')Xscream')
» X? 0x [lecturer'(x) & P(x)]](scream').
e. (Def (lecturer' ))(scream')
» X? [3x [lecturer'(x) & P(x)]](scream').
f. (XQ ftP [3x [Q(x) & P(x)]]](lecturer'))(scream')
» XV [3x [lecturer'(x) & P(x)]](scream').
g. a ^ Q f t P [ 3 x [ Q ( x ) & P ( x ) ] ] ] .
In translating NPs, one must be careful to associate the correct connective with
the correct quantifier, as a mistake gives rise to a formula with very different
truth-conditions. Replacing & by -> in (34.b), for example, yields (37.a) which, by
the truth-conditions of material implication, is true even if there is nothing in the
extension of lecturer', i.e. a lecturer screamed expresses a true proposition even if
there are no lecturers. Replacing -» by & in (27.b), on the other hand, yields the
166
formula in (37.b) which states that everything in the model was a student that liked Jo,
which is not what the sentence Every student liked Jo means.
(37) a. 3x [lecturer'(x) -> scream'(x)].
b. Vx [student' (x) & (like' Go' ))(x)].
Exercise 6.5:
Give a translation of the NP no dog and of the determiner no, following the steps used
above. NB: There are two possible translations for no. Make sure that the connective
used is the right one for the translation.
Exercise 6.6:
Translate the following sentences into LQ, giving for each one both the logical
expression as it is generated by the translation rules and its equivalent lambda
converted form.
i. Every lecturer sang and laughed.
ii. Some student didn't sing.
iii. If no student sang, then no lecturer screamed.
In this way, sentences containing quantified noun phrases in subject position can
be translated compositionally, but what sort of thing does a quantified noun phrase
denote in a model? As with the quantifier pronouns, quantified NPs have the type
«e,t>,t> and so denote functions from sets to truth values or, equivalently, sets of sets
of entities. Because of the interaction between the extension of the common noun in
the NP and that of the VP, these functions are more complex than those associated
with the quantifier pronouns.
The diagrams in (28) and (35) give a clue as to the denotations of universally and
existentially quantified noun phrases. In (28), the diagram representing the
truth-conditions for the sentence Every student liked Jo shows that the extension of the
common noun, student', must be contained in that of the VP, like'(jo*). We can,
therefore, think of the noun phrase every student as (indirectly) denoting a function
that maps every set of entities that contains the set of students onto 1 and every other
set onto 0. In other words, the phrase denotes the set of all supersets of [student']™*,
as shown in (38.a), which generalises to (38.b) for all noun phrases containing the
determiner every.
(38) a. { X c A l [student']M-*e X}.
b. every N denotes { X c A l [N' }** c X ) .
The extension of every student is called the universal sublimation of the property
of being a student. This identifies the set of all sets of which every student is a
member, as represented by the diagram in (39) where the central circle represents the
extension of student' (or the relevant common noun), each Pn represents the extension
of some other property and the whole thing represents the extension of XP /VJC
[student'(x)-*P(x)]].
167
6 Quantification
Universal sublimation of [student']1"
As can be seen from (35), a formula translating a sentence with an existential

subject NP is true if the extension of the common noun in the subject NP has a
non-null intersection with the extension of the VP. Thus, the extension of XP [3x
[lecturer'(x) & P(x)]], is a function that maps every set in the model onto 1 if its
intersection with [lecturer']™* contains at least one element. The extension of the NP
a lecturer is therefore given the set-theoretic definition in (40) which is read as 'the
set of all subsets of A that have a non-null intersection with the set denoted by
lecturer".
(40) { X c A I X n [lecturer'}™*
(40) is called the existential sublimation of the property of being a lecturer, since it
picks out the set of all properties that some lecturer has. This can be represented by
the diagram in (41) where again the central circle represents the extension of lecturer'
in the model, each Pn represents the extension of some other property and the whole
thing represents the extension of a lecturer. We will return to this sort of
representation for NP denotations in Section 6.6.
168
Existential sublimation of [lecturer']1"
Exercise 6.7:
What is the set-theoretic definition of the extension of no dog? How could the
negative sublimation of dog be represented diagrammatically?
6.3.3 Nominal modifiers

The approach to NP translation and interpretation discussed above is general enough
to allow for the introduction of attributive modifiers into noun phrases. In English,
nouns may be modified by adjectives, prepositional phrases or relative clauses. The
proper analysis of PP modification requires an analysis of the meaning of prepositions,
something that is not discussed in this book and so, in this section, we look only at
how certain simple attributive adjectives can be introduced into our grammar
fragment, leaving the analysis of relative clauses as an exercise for the reader.
Syntactically, attributive adjectives like happy and fat combine with a common
noun to give a common noun phrase (CNP) (equivalent to N1 in X-bar theory). Since
more than one adjective (or other nominal modifier) can appear with a single noun,
it is customary to adopt the recursive rule in (42.a) which distinguishes attributive
adjectives from predicative ones by the use of a feature [PRD], the former being
negatively marked for this property. This requires a slight alteration of the grammar
169
6 Quantification
to realise CNPs as simple nouns and allow determiners to combine with CNPs rather
than with nouns directly. (42.d) gives a list of some attributive adjectives.
(42) a. 29G2: CNP -> A[-PRD] CNP.
b. 30G 2 :CNP->N.
c. 28'G2:NP ->DetCNP.
d. A[-PRD] = {happy, sad, silly, fat}.
Since determiners, of type « e , t > , « e , t > , t » , combine with CNPs to give NPs,
CNPs must have the same type as a common noun, i.e. <e,t>, so that the translation
of the rule in (42.b) must be identical to the translation of the noun. This means that
the type of an attributive adjective must be «e,t>,<e,t» as it takes a one-place
predicate into a one-place predicate. But how should adjectives like happy be
translated? We could, of course, simply write it as happy' and let the denotation
assignment function in each model specify for each property the set of happy entities
with that property. However, this would miss two things. In the first place, it would
lose the relationship between the predicative use of the adjective as in Every student
is happy and its attributive use as in Every happy student. Secondly, we want to
capture the fact that every happy student is a student (and indeed that every happy
student is happy). This does not necessarily follow if each model specifies for each
property what set results from applying the function denoted by an attributive
adjective to that property. Thus, if we take the set of students in a model to be {MANi,
WOMAN2}, for example, it is possible to define [happy']M& as a function that maps
this set into {WOMAN^DOG}. In other words, it might be possible for some model
to pick out a set of happy students that are not actually students! Hence, we need to
ensure in the interpretation that the set of happy students is a subset of the set of
students.
To do this, we take as a starting point the interpretation of the predicative use of
such adjectives from Chapter 3. There, sentences like Fiona was happy were assigned
the same logical structure as sentences containing intransitive verbs like Fiona sang,
i.e. they are translated as one-place predicates combined with an individual constant.
In Ltype (and LQ, of course), this gives predicative adjectives the type of a one-place
predicate, <e,t>, with sets of entities as their extension. Using this type assignment
(and assuming that the copula is pleonastic), the sentence Every student was happy is
translated into LQ as (43), a formula that is true provided that the set of students is a
subset of the set of happy entities.
(43) Vx [student'(x) -> happy'(x)].
A sentence like Every happy student sang, on the other hand, must translate into
a formula that is true if, and only if, the set of happy students is a subset of the set of
singers, or, more explicitly, if the set of entities that were both happy and students is
a subset of the set of entities that sang. A formula that has precisely these
truth-conditions appears in (44.a). If we assume that the equivalence between the
formula in (44.a) and the one translating every happy student sang derives from
X-conversion, the rule can be applied backwards to work out the translations of the
constituents of the sentence. (44.b) shows a ^-expression applied to the VP translation
singy which is equivalent to the formula in (44.a).
170
(44) a. Vx [(happy'(x) & student'(x)) -* sing'(x)].

b. XP [Vx [(happy'(x) & student'(x)) -> P(x)]](sing').
As we have seen, the translation of every is the expression in (32.d) so that (44.b)
must be equivalent to the expression derived by applying XQ [XP /VJC [Q(X) -» P(x)]]]
to the translation of the CNP happy student as shown in (45.a) and (45.b). Because
CNPs translate into one-place predicates, the expression translating happy student
must be of type <e,t>. It must also contain the propositional function happy*{y) &
student'(y) that is factored out of (44.b) when the translation of the determiner is
removed. This propositional function can, of course, be turned into an expression of
the appropriate type by abstracting on the free variable y, as in (45.c). This is,
therefore, the translation of the common noun phrase happy student.
(45) a. every CNP
=> XQ [XP [Vx [Q(x) ->P(x)]]](CNP').
b. » XP [Vx [(happy'(x) & student'(x)) -> P(x)]].
c. happy student
=> Xy [happy'(y) & student' (y)].
We can assume that the expression in (45 .b) results from the functional application
of the translation of the adjective to that of the noun. The former can thus be restored
by replacing the translation of the noun, student*, by another predicate variable, Ph
and abstracting on this to give the expression in (46). This expression is of type
«e,t>,<e,t» and denotes a function that maps the set of entities denoted by the
common noun to the set of entities which appear in the intersection of this set with the
set of entities denoted by the predicative adjective happy \ This interpretation
guarantees that every happy student is happy and every happy student is a student.
(46) happy => XP,[Xy [happy'(y) & P,(y)]].
Attributive adjectives like happy' with the same general translation and
interpretation as shown in (46) are often referred to as intersective adjectives and a
general rule may be written that takes the predicative adjective translation and outputs
its attributive counterpart. This is given in (47) and, as with happy, the translation
yields an expression that combines with that of a common noun to yield an expression
that picks out the set of entities that appear in the extensions of the common noun and
the related predicative adjective.
(47) If a is a predicative adjective with translation a\ then its attributive
counterpart has a translation XP [Xx [a\x) & P(x)]].
Not all adjectives are intersective, however, and even adjectives like happy do
not always have the interpretation captured by the rule in (47). The interpretation of
happy student discussed above entails that any entity in its extension is happy in some
absolute sense. However, there are situations where this is not the appropriate
interpretation. A sentence like The happy student was not a happy cook is not
(necessarily) contradictory, as is predicted from the translation of happy defined by
(47). Furthermore, the translation of a sentence like Jo was a happy student, but an
unhappy lecturer given by the above rule makes incorrect predictions about the
entailments of this sentence. The appropriate translation is something like
171
6 Quantification
(happy'(jo*) & student'(jo')) & -(happy'(jo') & lecturer'(jo')). Such a formula can
only be true if Jo was a student and was happy (in some absolute sense), but was not
a lecturer. The English sentence, however, conveys the information that Jo is a
lecturer, but unhappy in that role, whereas he is happy in the role of a student. This
interpretation requires the extension of happy to depend on that of the common noun
with which it combines and is thus not simply intersective. Thus, while all happy
students are students, there is an interpretation in which they are not happy in any
absolute sense, but only happy as students. Many adjectives like small, good, skilful
and so on, have the same properties as this use of happy, but others like alleged and
fake are even more restricted in their entailments. For example, a fake gun is not a gun
and an alleged murderer is not necessarily a murderer. The interpretation of all these
adjectives goes beyond the extensional semantics that has so far been developed and
requires a definition of intensionality. Although the apparatus for dealing with
intensional expressions is introduced in Chapter 10, the proper treatment of intensional
adjectives remains controversial and so will not be further discussed. The interested
reader is encouraged to pursue the references cited at the end of this chapter for an
idea of how complex adjective interpretation may be. The analysis of intersective
adjectives given here is just a beginning, but it gives a basis from which to start.
* Exercise 6.8:
Devise a way of translating simple relative clauses like that in the following sentence
that guarantees that a student that liked Jo, liked Jo and a student that liked Jo, was a
student.
i. A student that liked Jo kicked the cat.
6.4 Proper names and definite descriptions

We have now seen how quantified NPs can be treated as having the type «e,t>,t>,
but what about proper names and definite descriptions? Proper names clearly pick out
individual entities in the universe of discourse and so their assignment to type e seems
the most reasonable assumption to make, since they thereby directly denote unique
entities in the model. However, if this analysis is maintained, an imbalance in the
treatment of noun phrases is created that is not represented in syntax. In English,
quantified noun phrases and proper names have the same syntactic distribution, which
is, of course, why they are assigned to the same syntactic category, NP. But the
difference in the apparent denotations of proper names and QNPs implies a difference
in syntactic category, since no syntactic category may be assigned more than one type
without giving rise to unacceptable ambiguity in the translation language. If this is
done, however, then clearly different syntactic rules have to be written for
expressions of the same category according to whether they contain QNPs or proper
names, thus missing the obvious syntactic generalisation that they have the same
distribution. Richard Montague's solution to this dilemma is to give both proper
names and quantified NPs the same type, thus allowing them to be treated the same
syntactically. Since, as we have seen, QNPs cannot have the type of an entity, the
172
Proper names and definite descriptions
type of Npr must be raised to the type of a QNP. At the same time, however, we must
make sure that the distinctive property of proper nouns, that, in a given context, they
name a specific entity, is not lost.
As we have seen before, the translation into L ^ of the sentence Fiona sang is
sing'(fiona ')> which is true only in those situations in which the set of singers includes
some specific entity called Fiona. Simply changing the type of proper names to
«e,t>,t> and translating the sentence as fiona'(sing') obscures this simple
interpretation and requires a complex interpretation of the constant fiona' to get the
truth-conditions right. However, we can get a more transparent translation for Fiona
by applying backwards lambda-conversion to the formula that best provides the
semantics for the sentence. Thus, assuming that proper names have the same type as
QNPs, i.e. «e,t>,t>, the formula sing '(fiona') must be equivalent to an expression
where the NP translation is functionally applied to sing', the translation of the VP.
Replacing the latter by a predicate variable we get the propositional function P(fiona')
which may be turned into an expression of type «e,t>,t> by abstraction to give the
translation of Fiona in (48.a) which denotes the set in (48.b), i.e. the set of all sets of
entities of which Fiona is a member.
(48) a. [XP [P(fiona')]]M8
b. { X c A l [fiona']M'Se X}.
As (48 .b) shows, the proper name Fiona denotes a set of sets like the other noun
phrases dealt with so far in this chapter. In this case, the NP denotes the set of all sets
that contain the entity denoted by the constant fiona'. That is, instead of picking out
an entity directly, a proper name picks out the properties that the entity has. This is
shown in the diagram in (49) which represents the individual sublimation of Fiona
where there is just one thing in the intersection of all the sets Pn, i.e. Fiona herself.
At this point, it may be objected that just looking at the properties individuals
have may not be sufficient to identify a single individual uniquely. For example, the
model may contain two distinct entities which both have the same name, are both
female, both with red hair and both students studying artificial intelligence at the same
university. If they both have the same likes and dislikes as well, then how can just
looking at the properties they have tell them apart? The answer is that every distinct
entity has one property not shared by any other individual in the model and this is the
property of being identical to itself. This property, which may be represented in LQ by
the expression Xx[x = c] where c is some individual constant, always denotes the unit
set consisting only of the individual denoted by c (for example, he [x = fiona']
denotes the set {WOMAN2} in M3). Hence, any individual sublimation may be
reduced to a unique individual in the model and so may be taken as equivalent
semantically to an individual constant. NPs, whether quantified or proper, may
therefore be treated in exactly the same way both syntactically and semantically, while
still maintaining the differences between them.
173
6 Quantification
Individual sublimation of [fiona']M
While an expression like XP [P(c)]f where c is an individual constant, translates

an expression in English, the individual constant, c, is itself the translation of no
English expression. No expression in English (or, by hypothesis, any other human
language) directly denotes an entity. All reference is done via reference to the set of
properties that a particular entity has. Amongst other things, this means that definite
descriptions also cannot translate into expressions of type e. This is perhaps not such
a difficult concept to grasp as that of proper names no longer directly denoting
individuals, partly because definite noun phrases in English contain a determiner the
and a common noun of type <e,t>. The definite article in English and other languages
is used in a number of different ways, none of which have an uncontroversial semantic
interpretation, as can be seen by a glance at the articles and books referred to at the
end of the chapter. Here, we look only at one treatment of one aspect of definiteness
which for a long time held a dominant place in the logical analysis of definiteness.
Although there are other treatments and obvious faults with the one to be discussed,
it is an interesting proposal and deserves mention here.
The philosopher Bertrand Russell proposed in (1905) that a noun phrase like the
king of France is logically complex and combines with a predicate to make a true
formula only if there is at least and at most one entity that is the king of France and
satisfies the predicate. Montague followed Russell's lead and gave the definite article
an analysis that translates the sentence in (50.a) by the formula in (50.b). According
to the truth-conditions of V, 3, and the connectives, this formula is true if, and only
174
Proper names and definite descriptions
if, there is at least one entity that kicked Prudence (the cat) and every entity that is fat
and a philosopher is identical to it. This guarantees that the formula is true if, and only
if, there is one, and only one, entity in the universe of discourse that fits the
description of being a fat philosopher.
(50) a. The fat philosopher kicked Prudence.
b. 3x [Vy [(fat'(y) & philosopher'(y)) <-> (x = y)] & (kick'(prudence'))(x)].
Adopting this analysis, the definite NP in the sentence in (50.a) has the translation
in (5 La) and the definite article itself that in (51.b), which can again be derived
through backwards X-con version. (The reader is invited to work through the derivation
of the sentence in (50.a) to show how the translations of the individual words combine
to give a translation that is equivalent to the formula in (50.b).)
(51) a. the fat philosopher
=> AP [3x [Vy [(fat'(y) & philosopher'(y)) <-» (x = y)] & P(x)]].
b. the => XQ [AP [3x [Vy [Q(y) « ( x = y)] & P(x)]]].
As mentioned above, there are numerous problems with Russell's analysis of
definite descriptions. The most obvious fault is that the sentence in (50.a) does not
usually imply that there is only one fat philosopher in the whole of the world. There
may well be many such, but the use of the definite article indicates that only one is
relevant to the current discourse situation. Thus, some means of relating the
uniqueness aspect to the immediate context is necessary to give a more accurate
representation of the way definite descriptions are used. One way in which this may
be done is to restrict the domain of the universal quantifier in the representation of a
definite description to some pragmatically specified set containing only those entities
of the model that are currently in the context of discourse. The translation of the fat
philosopher might then be interpreted with respect not just to a model and a variable
assignment, but also to such a pragmatically determined set. The semantics of the
quantifiers must then be restricted to quantifying over this set (call it D) by, for
example, checking the truth values of quantified formulae only with respect to the
entities in D. Such a revision of the semantics for the universal and existential
quantifiers is given in (52) where only those value assignments that associate variables
with members of D are relevant for ascertaining the truth of quantified formulae.
(52) a. If $ is an expression of type t containing a free occurrence of a variable, u,

of type e, then [3u [$]]M8D is 1, iff there is some value assignment g',
exactly like g except perhaps for the individual assigned to u, such that
[(|)]M.g\D j s i and g'(u) is a member of D.
b. If <>| is an expression of type t containing a free occurrence of a variable, u,
of type e, then [Vu [(()]]M'8D is 1, iff for every value assignment g', exactly
like g except perhaps for the individual assigned to u, [C|>]M-8'.D is 1 and g'(u)
is a member of D.
In terms of the Russellian analysis of the fat philosopher, the revisions in (52)
have the effect of requiring there to be a unique fat philosopher in D, the current
context, but says nothing about the existence of other fat philosophers in the model.
Furthermore, this revision has desired effects in the interpretation of universal
sentences like Everyone was drunk which is not usually interpreted as describing a
175
6 Quantification
situation where everyone in the whole world was drunk, but just some contextually
salient set of people was drunk. As an example of how this interpretation works,
consider a situation in which Jo, the cat and Fiona are fat, Jo and Fiona are
philosophers and Jo kicked the cat. When the sentence in (5O.a) is uttered there are
three entities in the discourse context, Jo, Ethel and Prudence the cat. At this point,
therefore, we interpret the formula in (5O.b) with respect to the model of this situation
(M3), an assignment of values to variables, g, and the set {MAN!, WOMANi,CAT}.
The first two steps in ascertaining the truth of this expression with respect to the
relevant context are given in (53), which follow the definitions in (52).
(53) a. Bx [Vy [(fat'(y) & philosopher'(y)) <-> (x = y)] & (kick'(prudence'))(x)]
is 1 with respect to M3, g and {MANi,CAT,WOMANi} iff there is a value
assignment g'likeg where g'(x)e {MANj,C AT, WOMAN,} such that
Vy [(fat'(y) & philosopher'(y)) <-> (x = y)] & (kick'(prudence'))(x)] is also
1.
b. Vy [(fat'(y) & philosopher'(y)) *-> (x = y)] & (kick'(prudence'))(x)] is 1
with respect to M3, g and {MAN^CAT,WOMAN!} iff for every value
assignment g" exactly like g' where gM(y) e {MAN^CAT,WOMAN,}
[[(fat'(y) & philosopher'(y)) <-> (x = y)] & (kick'(prudence'))(x)] is 1.
Because interpretation is restricted to the set of entities that are salient in the discourse
(i.e. to Jo, Ethel and Prudence), it is easy to show that (5O.b) is true. First of all, we let
g' be a function that assigns MANi to x and WOMANX to y. This satisfies the
expression [(fat'(y) & philosopher'(y)) <-»( x = y)] & (kick'(prudence'))(x) with
respect to M3, because Jo (MANi) kicked Prudence (CAT), and [(fat'(y) &
philosopher'(y))]M^^MMilCATWOMA^ has the same truth value as [x =
y]M3,g\{MANi,cAT,woMANi}( s i n c e M ANi, the value of g' applied to x, is not identical to
WOMANi, the value of g' applied to y, and the latter is neither a philosopher nor fat).
There are then only two other relevant value assignments that we need to look at to
verify (53.b), g'MAN1'*and g'cAT/y Both of these do satisfy [(fat'(y) & philosopher'(y))
<-> (x = y)] & (kick'(prudence'))(x) with respect to M3. The first satisfies the
expression because x and y are assigned to the same entity, MANi, which makes [x
= y]M3*\<MANi.cAT.woMANi> true and satisfies philosopher'(x) &fat'(x) because MANi is
both a philosopher and is fat. Since both sides of the equivalence (fat'(y) &
philosopher '(y)) <-> x = y are true, the whole formula is true, and so the conjunction
with (kick'(prudence '))(x) is also true. The final value assignment with y assigned to
CAT satisfies the expression for the same reason that g' itself does, i.e. because the
cat is not Jo nor is it a fat philosopher. Hence, where the domain of quantification is
restricted to Jo, Ethel and Prudence (5O.b) is true. Notice, however, that if the domain
is unrestricted the formula is false of the situation being described. This is because a
value assignment that assigns WOMAN2 to y (and MANi to x, as before) fails to
satisfy the formula [(fat'(y) & philosopher'(y)) <-> (x - y)] & (kick'(prudence'))(x),
since WOMAN 2 is both a philosopher and fat, but, crucially, is not identical to MANi,
the value of g'(x), and so the two formulae in the equivalence do not have the same
truth value, making the equivalence (and the whole conjunction) false. There is thus
more than one fat philosopher in the domain and so, according to Russell, the
proposition expressed by the sentence The fat philosopher kicked the cat is false,
since the phrase the fat philosopher fails to pick out a unique entity. Hence, the
176
Two problems
introduction of a context set, D, as part of the interpretation enables one to capture

more accurately some of the meaning of the definite article the (and indeed the uses
of the other quantifiers and proper names). The truth values of quantified formulae
differ depending on the set of salient entities in the discourse.
There is much more that can be said about Russell's analysis and Strawson's
criticisms. There is also much more to be said about the meaning of the word the in
its anaphoric use (e.g. Bill is my boss and the bastard really keeps me at it) or in its
generic use (e.g. The tiger is a fierce animal) where it appears to have the force of the
universal quantifier. Indeed, much controversy surrounds the proper treatment of the
definite article and definiteness in general. For the purposes of this book, however,
despite its problems, the Russellian interpretation of the is the one that will be
assumed whenever definite descriptions are used.
Exercise 6.9:
Translate the following sentences into LQ, giving both the full translation as generated
by the translation rules and the simplified equivalent form that results from lambda-
conversion.
i. Jo laughed and cried.
ii. The linguist was not happy.
iii. If Ethel sang, then the dog howled.
6.5 Two problems

There are two problems facing the approach to noun phrase interpretation discussed
above. One is more of a technical problem to do with translation and concerns the way
non-subject noun phrases combine with their verbs to form verb phrases. The second
concerns the ambiguity of sentences that contain more than one quantified noun
phrase. We will look at these in turn.
6.5.1 Type raising

If we give NPs the type «e,t>,t>, it is no longer possible to combine the translations
of object NPs directly with those of their verbs, because the latter translate into
expressions of type <e,<e,t» that cannot functionally apply with expressions of type
«e,t>,t>. To combine NP' and V,', therefore, the type of the latter must be changed
so that its arguments have the correct type; so that transitive verbs must be assigned
the complex type <«e,t>,t>,<e,t» and ditransitive verbs the type
<«e,t>,t>,<«e,t>,t>,<e,t>». This would, of course, be a simple matter if all we
were interested in was getting verb translations to combine with NP ones. But the
point of translation into a logical language is the interpretation it receives. Just
asserting that verb types are now all changed is not sufficient. We need to make sure
that the verbs retain their fundamental interpretation as denoting relations between
entities. To do this, the translations of verbs like like and give must be made more
complex so that their type is the appropriate one to combine with non-subject NPs, but
177
6 Quantification
they still end up identifying relations between entities.

The solution to the problem, as before, can be found in the use of the lambda
operator and, as in previous sections, the easiest way to get to the translation of a
constituent is to start with a formula that expresses the truth-conditions of a particular
sentence that contains that constituent. Consider the sentence in (54.a) whose truth-
conditions require that for every entity that is a student in some model, there is a
lecturer that that student liked. The force of this can be brought out by the semi-formal
English sentence in (54.b) which has the transparent representation in LQ shown in
(54.c).
(54) a. Every student liked a lecturer.
b. For every x, if x is a student then there is a y such that y is a lecturer and x
liked y.
c. Vx [student'(x) -> 3y [lecturer'(y) & (like'(y))(x)]].
As we have seen, the translation into LQ of the subject NP every student is the
A,-expression in (55.a) which combines with the translation of the VP to give a formula
truth-conditionally equivalent to (54.c). The VP itself translates as an expression of
type <e,t> that contains the remnants of the formula in (54.c) once the subject NP
translation in (55.a) has been removed. This remnant is an open formula 3y
[lecturerJ (y) & (like'(y))(x)] and abstracting on the free variable, x, yields the
one-place predicate in (55.b) which is equivalent to the translation of the VP. (The
variable in (55.b) has been changed to z to bring out the fact that it is a different
variable from the x in (54.c) which is supplied by the NP translation in (55.a).)
(55) a. X? [Vx [student'(x) ->P(x)]](VP')
» Vx [student'(x) -> 3y [lecturer'(y) & (like'(y))(x)]].
b. VP => Xz [3y [lecturer'(y) & like'(y)(z)]].
The predicate in (55.b) is equivalent to the expression that results from combination
of the translation of the transitive verb like with that of the NP a lecturer. Assuming
that the verb is the functor (in order to keep the types of subject and object NPs
uniform), Vt' must apply to the translation of the object NP in (56.a) to give an
expression equivalent to (55.b), as shown in (56.b). In performing this ^-conversion,
there must be a step where the NP translation in (56.a) applies to a one-place predicate
which is bound to the variable Q and becomes like'(y)(z) after lambda-conversion.
Since z is bound by the outermost lambda operator and y is supplied by the NP
translation, the latter can be replaced by a new variable x which is abstracted upon to
give the one-place predicate in (56.c) which yields (56.d) as the intermediate
expression equivalent to the VP translation in (55.b).
(56) a. XQ [3y [lecturer'(y) & Q(y)]].
b. V?(XQ [3y [lecturer'(y) & Q(y)]])
» Xz [3y [lecturer'(y) & like'(y)(z)]].
c. Xx [(like'(x))(z)].
d. Xz [XQ {By [lecturer'(y) & Q(y)]](Xx [(like'(x))(z)])].
One final step remains to be done to get to the translation of the verb. (56.c) is, as we
have seen, equivalent to the translation of a verb phrase formed by applying the
translation of the verb like to that of the NP a lecturer. To transform (56.c) into an LQ
178
Two problems
expression with such a structure, the NP translation can be replaced with a variable of
the appropriate sort and placed as an argument after the ^-expression formed by
abstraction on this variable. Taking P to be a variable of type «e,t>,t>, backwards
^-abstraction yields (57. a) as the full (unconverted) form of the VP translation of liked
a lecturer which is truth-conditionally equivalent to (56.c) and (55.b). The expression
in (57.b) is thus the translation of the verb like, an expression that has the correct type,
< « e , t > , t > , < e , t » , denoting a function from NP extensions to VP extensions. The
derivation of the example sentence in (54.a) is given in (58) showing the way the
translation is built up and then reduced to the equivalent expression in (54.c).
(57) a. XP [Xz [P(Xx [(like'(x))(z)])]](XQ [3y [lecturer'(y)& Q(y)]]).
b. like => XP [Xz [P(Xx [(like'(x))(z)])]].
(58) a. S
b. NPVP[+FIN] 1G2
=f> NP'(VP').
c. Det CNP VP[+FIN] 28'G 2
=> (Det'(CNP')XVP').
d. Det N VP[+FIN] 30G2
=> (Det'(N')XVP').
e. Det N V[+FIN] NP 2G2
=> (Det'(N')XV'(NP')).
f. Det N V[+FIN] Det CNP 28'G 2
=> (Det'(N')XV'(Det'(CNP'))).
g. Det N V[+FIN] Det N 30G2
=*(Det'(N'))(V'(Det'(N'))).
h. every student liked a lecturer. Lex
=* (XQ [X? [Vx [Q(x) -*P(x)]]](student'))
(XP [Xz [P(Xx [(like'(x))(z)])]](XQ [XP [3y [Q(y) & P(y)]]](lecturer'))).
i. » (XP [Vx [student'(x) -> P(x)]])
(XP [Xz [P(Xx [(like'(x))(z)])]](XQ [XP [By [Q(y) & P(y)]]](lecturer'))),
j. » (XP [Vx [student'(x) -> P(x)]])
(XP [Xz [P(Xx [(like'(x))(z)])]](XP [By [lecturer'(y) & P(y)]])).
k. » (XP [Vx [student'(x) -> P(x)]])
(Xz [XP 0 y [lecturer'(y) & P(y)]](Xx [(like'(x))(z)])]).
1. » (XP [Vx [student'(x) -^ P(x)]])(Xz [By [lecturer'(y) & like'(y)(z)]]).
m. » Vx [student'(x) -> Xz [3y [lecturer'(y) & (like'(y))(z)]](x)].
n. » Vx [student'(x) -> 3y [lecturer'(y) & (like'(y))(x)]].
This translation for like, which is generalised to cover all other transitive verbs,
may seem unnecessarily complex, but it does allow noun phrases to be given a
uniform treatment whatever position they happen to occupy in a sentence, whilst
maintaining the insight that such verbs denote two-place relations between entities.
Although the gain may seem slight as compared to the increased complexity in
translating verbs, in Chapter 10 we will see that there are verbs which must be
analysed as denoting functions from NP extensions to VP extensions directly, without
there being any direct relation between two entities. Thus, to give a uniform syntactic
179
6 Quantification
treatment to such verbs and verbs that denote simple relations between entities, the
increased complexity in translating the latter is a small price to pay.
Exercise 6.10:
Using translations of the transitive verbs of the sort given to like, translate the
following sentences into LQ, giving the formulae as they are generated by the
translation rules and the equivalent formulae that result from lambda-conversion.
i. Ethel shot a philosopher,
ii. No student liked the lecturer,
iii. A linguist slapped Jo.
* Exercise 6.11:
Owing to the type changes introduced above, the translation rules for passive verb
phrases given in Chapter 5 no longer give therightresults. Show why this should be
so and devise new rules that ensure that the correct interpretation is given to each
passive sentence.
6.5.2 Scope ambiguities

Sentences containing more than one quantified NP are often ambiguous, depending
on how wide the relative scopes of the quantifiers contained in the NPs are taken.
Consider the sentence in (59.a) which could be followed by either of the sentences in
(59.b) or (59.c), without contradiction.
(59) a. Every critic enjoyed a play.
b. Charles enjoyed Macbeth, Philippa enjoyed Bedroom Farce, Wendy
enjoyed Happy Days and Kevin enjoyed all the plays.
c. The play that everyone enjoyed was Salome.
The reading of (59.a) that (59.b) picks up on is the one where each critic indeed
enjoyed some play, but the one enjoyed is not necessarily the same for them all.
(59.c), on the other hand, forces the stronger reading, where it is the same play that is
enjoyed by everyone. The difference between these two interpretations is brought out
by the semi-formal English sentences in (60) which may be represented in LQ by the
formulae in (61).
(60) a. For every x, if x was a critic then there was a y such that y was a play and
x enjoyed y.
b. There was a y such that y was a play and for every x, if x was a critic then
x enjoyed y.
(61) a. Vx [critic'(x) -> 3y [play'(y) & enjoy'(y)(x)]]. (= (59.b)/(60.a))
b. 3y[play'(y)& Vx [critic'(x) -> enjoy'(y)(x)]]. (= (59.c)/(60.b))
The difference between the two formulae in (61) resides in the relative scopes of the
NP translations and the quantifiers and propositional connectives they contain. In
180
Two problems
(61.a), the universal quantifier has scope over the existential one, whilst in (61 .b) the
reverse is the case and, because of the way we have translated universal and existential
NPs, the material implication in (6La) has wider scope than the conjunction, a
situation reversed in (6Lb). This difference is semantically significant, since it is
possible for (6 La) to be true in a situation in which (61.b) is false (although the
reverse is not the case; see below). This is the case in the situation shown in (62)
which represents the relation between the set of critics and the plays that they enjoy
in some model (call it M6), as illustrated by (59.b). (NB: the diagram shows the
relation, not the function, denoted by enjoy'.)
(62)
To prove that the interpretation we have given to the quantifiers does give the
formulae in (61) different truth values with respect to M (and thus assign them
different truth-conditions), let us work through the formal interpretation of the
formulae in turn. The procedure for doing so, however, is somewhat involved, because
of the fact that the truth-conditions of quantified sentences are defined in terms of
value assignments to variables. All that is required to understand the method is a bit
of patience to keep track of the different value assignments, but the effort helps in the
understanding of the way the interpretation rules in Section 6.2.2 work.
According to the truth-conditions for 3 and V in (10) and (13), above, the
formula in (6 La) is true with respect to the model M if, and only if, for every entity
that can be assigned to x there is an entity that can be assigned to y that makes
critic'(x) —» (play'(y) & (enjoy'(y))(x)) true. To show that this is so, we need to look
at all possible assignments for x where x is a member of [critic']M* and see if there is
a value assignment that keeps the entity assigned to x constant, but assigns an entity
to y which is in the set of plays and the set of things that x enjoyed. We do not need
to look any further than those value assignments that associate x with a critic, because,
by the semantics of material implication, critic'(x) —> (play'(y) & (enjoy'(y))(x)) is
always true if x is not a critic, because critic y(x) is false on such a value assignment.
Nor do we need to look at assignments of values to y that are not plays, because these
make play'(y) & enjoy(y)(x) false for any value of x and thus are irrelevant for
ascertaining the truth of the existential formula. We begin by looking at the value
assignment g with respect to which (6 La) is to be interpreted. The relevant part of this
181
6 Quantification
is shown in (63), which specifies only the values assigned to x,y (the only variables
in (61)) and to z, for good measure. There is no need to specify the rest of the function,
as this remains constant throughout the interpretation procedure.
(63) g
With respect to g (and Me), the formula critic'(x) —» (play'(y) & (enjoy'(y))(x)) is
false because c is a critic but the pair <c,b> is not in the extension of [enjoy '] M 6 , where
c is the value assigned to x and b is that assigned to y. However, the existential
formula 3y [play'(y) & (enjoy '(y))(x)] is not thus falsified, because it is still true if
there is some value assignment exactly like g except for the value assigned to y that
satisfies playf(y) & (enjoy '(y))(x). Hence, we must look at the value assignment just
like g except that y is assigned another value, say m. This function is symbolised g"1^,
as we saw in Chapter 5, and the part of this function that we are interested in is shown
in (64). This assignment of values to variables satisfies play'(y) & (enjoy' (y))(x),
because <c,m> is a member of [enjoy']U6*. Hence, the existential formula [3y
[play'(y) & (enjoy'(y))(x)}]M6# is true, thus making [critic'(x) —> 3y [play'(y) &
(enjoy'(y))(x)]]M6'& also true by the truth-conditions of material implication.
(64) g^y
While we have shown that the formula critic '(x) —» 3y [play '(y) & (enjoy f(y))(x)]
is true with respect to M 6 and g, we have not, of course, yet completed the
interpretation of (61 .a), as we need to see what happens with all other assignments of
values to x. Thus, we next look at the value assignment gp* shown in (65). Since the
pair <p,b> is in the extension of enjoy' in Nfc (and p is in [critic']™6* and b is in
[play']M6*)y gp"< again satisfies critic'(x) -> (play'(y) & (enjoy'(y))(x)).
(65) g*
182
Two problems
The next assignment of a value to JC, gWx in (66.a), however, fails to satisfy the
target formula, since Wendy did not enjoy Bedroom Farce, so we need to look at the
value assignments just like this except for the values assigned to y. Assigning the
value of m to y yields the assignment gi**)""* shown in (66.b). This does not help, as
<w,m> is not in the extension of enjoy'. The next assignment gww', however, which
is shown in (66.c) does verify the formula critic '(x) —> 3y [play '(y) & (enjoy '(y))(x)],
because <w,h> is in [enjoy']™6*.
(66)
gw/x # g[w/x]m/y
C. r [w/x]h/y
w w w
m
b y y h
P P P
As there is one more critic in the model, there is one more assignment to JC that
must be looked at, i.e. g*7*. Since Kevin enjoyed all the plays in M 6 , this value
assignment (shown in (67)) satisfies critic'(x) —> (play'(y) & (enjoy'(y))(x)). Thus,
we have now shown that for every value assignment like g but for the value ofJC which
makes critic'(x) true, there is a value assignment exactly alike except possibly for the
value of y that satisfies critic'(x) -^play'(y) & (enjoy'(y))(x). Hence, the formula in
(61 .a) is true with respect to M6 and g. Furthermore, because the value assignment, g,
was chosen arbitrarily, we can say that the formula is also true just with respect to the
model M 6 . In other words, the formula is true irrespective of which value assignment
is chosen.
(67) g*
In ascertaining the truth of the formula in (61 .a), the value of y may be varied for
each value of x in order to satisfy the relevant propositional function. To show that the
formula in (61 .b) is true, on the other hand, it is necessary to find some value for y that
makes the formula play'(y) & (critic(x) —» (enjoy'(y))(x)) true for all values of JC. In
other words, every value for JC must satisfy the formula for some constant value of y.
Since any value for y not in the extension of play' necessarily makes the formula false,
we may restrict our attention only to those values for y which are plays. Furthermore,
we need only look at values of JC which are critics since all other assignments make the
183
6 Quantification
implication true. First, therefore, we may take the value assignment, g, and then look
at all the assignments that vary from this in their value for x. If there is one variant that
makes the target formula false then we move on to look at the next assignment like g
except for the value assigned to y and go through all the values that may be assigned
tox
With respect to g, play'(y) & (critic(x) -» (enjoy'(y))(x)) is false in the model
Me, because <c,b> is not in the extension of enjoy* while c is in that of critic' in the
model. As there is thus at least one critic in the model that fails to enjoy Bedroom
Farce, we may go directly to an examination of another variable assignment, e.g. g1"7*
in (64). This satisfies the formula and so another for x is tested, g^yw* in (68.a),
however, does not satisfy the target formula, so that g^y itself fails to satisfy the
formula. A new assignment to y must therefore be investigated. This is g1^ in (68.b)
which fails to satisfy the target formula, as indeed does the last possible assignment
to y that makes play'(y) true, i.e. g8'* in (68.c), since neither <c,h> nor <c,s> are in
[enjoy']"16*. As no value assigned to y makes critic*(x) & (enjoyt(y))(x) true for all
values of JC, the formula in (61.b) is false with respect to M6. Hence, we have shown
that there are situations like that modelled by M6 in which the formula in (61 .a) is true,
but in which that in (61 .b) is not. Thus, the truth-conditions of the two formulae must
differ and so the sentence Every critic enjoyed a play in (59.a) is shown to be truly
ambiguous.
(68)
a. g[m/y]h/x b. g^ C. g^
b A c c
v • w
m J h V • w
s
P P p
• ^
There are situations, of course, in which (61 .b) is true. One such is shown in (69)
which shows the criticizing relation between critics and plays in a model, M*. The
formula in (61.b) can be shown to be true with respect to M6' and g by the value
assignments in (70) which show all possible values for x when the value ofy is held
constant as s (i.e. the value assignments are all variants of g8^). Since all these
assignments satisfy (play'(y) & critic'(x)) -» (enjoy '(y))(x), (61.b) is also true.
The formula in (61.a) is also true with respect to M6' because, for each critic,
there is a play that they enjoy. It just happens to be the same play in each case. Any
true formula where an existential quantifier has wider scope than a universal entails
the truth of the corresponding formula with the scopes of the quantifiers reversed.
Thus, wheretyis any formula, the formula in (71) is a tautology, allowing Vx [3y<]>]
to be inferred from 3y \\/x§] in all cases. The reading of (59.a) in (61 .b) thus entails
that in (61.a). (See Chapter 7 for further details.)
184
Two problems
(69)
X enjoy Y
C v b
[critic'] M6 ' [play'] M6 '
(70)
a. g^y \y g[s/y]p/x c a[s/y]w/x gls/ylk/x
x • c XV C x • w
y • s y y • s v s
z p P
z > p
(71) 3x [Vy [«]-> Vy 0 x [«].
Exercise 6.12:
Ascertain the truth values of the following two formulae with respect to the model,
Me, and the value assignment g, by examining different value assignments as was done
above. Do the differences in scope of the quantifiers between i and iii and ii and iv
affect the truth-conditions of the formulae?
i. 3x [critic'(x) & 3y [play'(y) & (enjoy'(y))(x)]].
ii. Vx[play'(x) -^ Vy [critic'(y) -Kenjoy'(y))(x)]].
iii. 3y [play'(y) & 3x [critic'(x) & (enjoy'(y))(x)]].
iv. Vy [critic'(y) -> Vx [play'(x) -> (enjoy'(y))(x)]].
The existence of ambiguous sentences like that in (59.a) poses a problem for the
grammar fragment set up so far in that the grammar rules given in Section 6.2 only
generate one translation for the sentence in (59.a), the one where the quantifier in the
subject NP has widest scope. The derivation of the LQ formula that shows this reading
parallels that for the sentence Every student liked a lecturer in (58). The second
reading of (59.a) cannot, however, be generated by the grammar as it stands, because
it requires the quantifier in the object NP to appear to the left of that of the subject. In
185
6 Quantification
other words, the object is semantically combined with the verb after the subject in
some way, although, of course, in the syntax the object combines with the verb before
the subject. This is a problem for the rule-to-rule hypothesis, because this is a semantic
process that has no syntactic reflex. In the other cases of ambiguity that we have come
across in this book, any semantic ambiguity can be seen in the syntactic derivation.
A number of ways have been suggested to obtain different scope readings. In
transformational grammar, noun phrases are moved to sentence initial position after
the surface structure has been derived, but before semantic interpretation is carried out
(i.e. at LF). In monostratal grammars, it has been suggested that the interpretation of
a noun phrase may be stored during the derivation of a sentence and then utilised later
to give the different readings.
No attempt will be made here to formally extend the grammar and derivation
procedures developed so far to cope with scope ambiguities. However, a means of
accounting for wide scope quantification in the semantics can be given using a process
that Montague refers to as quantifying in. This process has the effect of replacing an
argument of a verb by a variable of the appropriate sort and then combining a noun
phrase translation with the X-expression of type t formed by abstracting on this
variable once all other arguments have been combined. Because object noun phrases
have the type «e,t>,t>, as we have seen, a predicate must combine with the
individual sublimation of an individual variable, rather than an individual variable
itself, i.e. XP [P(x)] rather than x. The rule is stated in (72) as an operation on well-
formed formulae and has the effect of associating sentences with more than one
translation (e.g. with wide scope readings for object NPs) and may, therefore, be
thought of an an ambiguating relation.
(72) Quantifying in: If a sentence S is translated by a formula $ containing an
expression a of type «e,t>,t>, then S may also be translated as ot(Xx [<()'])
where <|>' is exactly like $ except that a is replaced by the expression
] where x is a variable of type e not bound in <))' .
As an example, the derivation of the second reading of the sentence in (59.a) can
be shown by the semantic analysis tree in (73). Here, the translation of the direct
object is replaced by the individual sublimation of the variable Xj which combines
with the translation of the transitive verb to give the translation of the VP. The latter
is then combined with the translation of the subject NP to give a formula containing
a free instance of the variable Xj. Finally, the translation of the direct object noun
phrase a play is combined with the predicate formed by lambda abstraction on this
variable. (74) shows the successive lambda-conversions that guarantee that this
derivation is equivalent to the formula in (61.b). Other scope effects will be seen in
Chapters 8 and 10 involving tense and intensionality, but they can all be derived in the
semantics using the rule of quantifying in (72).
186
Generalised quantifiers
(73)
t
XQ [3y [play'(y) & Q C y ) ] ] ^ [XP [Vx [critic'(x) -> P(x)]](AP [A.x2 [P(Xz [enjoy'(z)(x2)])]](AP [PCx
«e,t>,t> t
XQ [3y [play'(y) & Q(y)]] Xx{ [X? [Vx [critic'(x) -> P(x)]](Xi» [Xx2 [Pftz [enjoy'(z)(x2)])]](>P [P( Xl )]))
«e,t>,t> <e,t>
XP [Vx [critic'(x) -> P(x)]] XP [ U 2 [P(Xz [enjoy'(z)(x 2 )])]]aP [PCx
XP [Xx2 [P(Xz [enjoy'(z)(x2)])]]
(74) a. >iQ [3y [play'(y) & Q(y)]](Ax1 [XP [Vx [critic'(x) ^ P( x )]]
(XP [Xx2[P(Xz [(enjoyXz))(x2)])]](XP [PCx,)]))]).
b. » XQ [3y [play'(y) & Q(y)]](Xx1 [XP [Vx [critic'(x) -^ P(x)]]
(Xx2[XP [P(Xl)](Xz [(enjoy'(z))(x2)])])]).
c » XQ p y [play'(y) & Q(y)]](Xx1 [XP [Vx [critic'(x) ^ P(x)]]
(Xx2[Xz[(enjoyXz))(x2)](Xl)])]).
d. » XQ p y [play'(y) & Q(y)]](Xx1 [XP [Vx [critic'(x) -> P(x)]]
(Xx2 [enjoyXxjKx,)])]).
e. » XQ [3y [play'(y) & Q(y)]] (Xx.fVx [critic' (x) -> Xx2[enjoy'(x1)(x2)](x)]]).
f. » XQ p y [play'(y) & Q(y)]](Xx1 [Vx [critic'(x) ^ enjoy'(x1)(x)]]).
g. » 3y [play'(y) & Xx![Vx [critic'(x) -^ enjoy'(x,)(x)]](y)].
h. » 3y [play'(y) & Vx [critic'(x) -> (enjoy'(y))(x)]].
6.6 Generalised quantifiers

Certain English sentences containing quantifiers other than every (all), no and a
(some) can be represented using the universal and existential quantifiers, as shown in
(75).
(75) a. At least two students laughed
=> 3x [3y [student'(x) & student'(y) & ~(x = y) & laugh'(x) &
laugh' (y)]].
b. Both students laughed
=> 3x [3y [Vz [(student'(x) & student'(y) &~(x = y ) ) o ( z = y v z = x)]
& laugh'(x) & laugh'(y)]].
These representations, although truth-conditionally adequate, are very complex and
counter-intuitive and illustrate a weak point in the translation procedure developed
187
6 Quantification
above for every and some: the translation of quantified English sentences introduces
sentential connectives that are just not represented in the English syntax. Thus,
although a compositional translation procedure for such sentences can be provided
using the lambda operator, the interpretation requires a more complex structure for LQ
than is indicated in the object language. Furthermore, there are many quantified
sentences that have no representation at all using the universal and existential
quantifiers, examples of which appear in (76).
(76) a. More students than lecturers laughed.
b. Many linguists sneezed.
c. Few philosophers sang.
d. Most babies cried.
e. More than half the babies cried.
To end this chapter, we take a brief look at a recent development in the theory of
quantification that has shown great promise in helping to solve these problems and
which has led to the development of a series of constraints that can be imposed on the
semantics of natural language quantifiers.
As mentioned in Sections 6.2.1 and 6.2.2, the semantics of universal and
existential noun phrases can be defined in terms of set theory. A NP like a student
denotes the set of sets of entities which have a non-null intersection with the set of
students (cf. (40)) and every student denotes the set of sets of entities of which the set
of students is a subset (cf. (38.b)). It is this analysis of NPs as denoting generalised
quantifiers that has been so fruitful in linguistic semantics in recent years. Instead of
NPs being translated into complex expressions of LQ, their semantics is defined
directly on a simple (lexical) translation. Thus instead of translating every into the
complex expression XQ [XP /VJC [Q(X) -> P(x)]]] it is translated into every' like any
other lexical item. This still has the type « e , t > , « e , t > , t » and so gives rise to NP
translations like every'(student') and sentential translations like
(every'(student'))(laugh') which contains no extra sentential connectives. The
semantics of every' (and a, no, etc.) is then directly defined by the model theory
without the use of variables and value assignments. Each quantifier is associated with
a function which assigns to each subset [N]M« of A (i.e. the extension of each common
noun), a set of subsets of A having a particular property. The definitions for every and
some in these terms is given in (77). The truth value of a formula whose subject NP
contains the quantifiers every or some can be ascertained by seeing whether the VP
denotes a set that is in the set of sets denoted by the subject. Thus, the proposition
expressed by a sentence like Every student laughed is true just in case [laugh']M* is
in {X c A I [student'}™* c X}, i.e. if [student"]™* c [laugh']M*.
(77) a. [every ']M* is that function which assigns to each [N]M-s c A , the set of sets
{XcAl [Nf^cX}.
b. [some '] M * is that function which assigns to each [N]M« c A, the set of sets
(XcAIXn[N]M^0).
Using this set-theoretic method rather than the traditional quantifiers, it is
possible to define the semantics of other quantifiers in terms of the set of sets that they
assign to the extension of each common noun. The interpretation of many
non-classical quantifiers (i.e. quantifiers other than 3 and V) is based on the
188
cardinality of a set, i.e. the number of elements it contains. The cardinality of a set
is symbolised by writing the set or its name within two single vertical lines, i.e. | • |.
Thus, the cardinality of the set {a,b,c}, written | {a,b,c} |, is 3, because the set has three
members, while | A | is 8, because there are eight entities in the universe of discourse,
and so on. Using this notation, we can also refer to the number of entities that have a
certain property in a model. For example, the cardinality of the extension of student'
with respect to some model, M, is written as | [student']M* |. Quantifiers often involve
the comparison of the cardinalities of two sets and the symbol is used to represent the
relation greater than, while > represents the greater than or equal to relation. Thus,
the set-theoretic statement | X | > | Y | means that the set X has more members than the
set Y, while |X | > |Y| means that X has more members or the same number of
members as Y.
Exercise 6.13:
Given the sets A and B specified below, which of the statements that follow are true?
A = {a,b,c,d,e} B = {9,16,32,75,2,62} C = {q,3,s,t,l}
i. |A|>|B|. ii. | B | > | C | . Hi. |C|>|A|. iv. |A|>|C|.
The treatment of the quantifiers (at least) two and both is much simpler using this
method than the predicate logic translations in (75). Firstly, they are both translated
directly according to the lexical translation rule in Chapter 2, as two' and both*. Both
of these quantifiers put constraints on the cardinality of the intersection of two sets,
that denoted by the common noun and that denoted by the VP, but the difference
between them is that two' requires the intersection of these two sets to have two or
more elements while both' requires the cardinality of the set to be exactly two. Under
the analysis of the interpretation of the NP two students in (78.a), the formula
translating Two students laughed is true just in case the intersection of the extension
of student' with that of the VP, laugh', has two or more members (i.e. if the set of
students who laugh has a cardinality of greater than or equal to two). The translation
of Both students laughed, on the other hand, is only true, according to the set-
theoretic interpretation of the subject NP in (78.b), if (and only if) the intersection of
the set of students and the set of laughing entities has exactly two members. These
seem to prescribe the intuitively correct truth-conditions for these sentences in a
considerably more simple fashion than is possible using the classical quantifiers, as
can be seen from a comparison of the interpretations (78.c) and (78.d) with those of
the formulae in (75.a) and (75.b), respectively.
(78) a. [two'(student')]M-8 = { X c A I |X n [student' ]M*\ > 2}.
b. [both'(student' )]M-« = {X c A I | X n [student']M-« | = 2}.
c. [(two'(student'))(laugh')]M-8is 1 iff |[laugh']M» n [student']*1* | > 2.
d. [(both'(student'))(laugh')]M«is 1 iff |[laugh']M8 n [student']M81 = 2.
Using this method, it is also possible to give interpretations to the sentences in
(76) that contain non-classical quantifiers. The first sentence, (76.a), More students
than lecturers laughed, expresses a true proposition just in case the number of students
189
6 Quantification
who laughed is greater than the number of lecturers who laughed. More...than is a
discontinuous determiner taking two common nouns and so has the type
«e,t>,«e,t>,«e,t>,t»>. Noun phrases with this determiner and two common
nouns, Ni and N2, can be interpreted as denoting the set of sets whose intersection with
the extension of N2 has more members than their intersection with the extension of N^
Thus, the NPmore students than lecturers receives the interpretation in (79.a), giving
the formula translating the example sentence in (76.a) the truth-conditions in (79.b),
which ensures that the formula is true if and only if there are more laughing students
than laughing lecturers.
(79) a. [more_than'(lecturer')(student')]M* =
{X c A I |X n [student']1"* | > |X n [lecturer']M*|}.
b. [(more_than' (lecturer' )(student'))(laugh')]M-8 is 1 iff
| [laugh']1"* n [student']M'g | > | [laugh' ]M* n [lecturer' ] M *|.
On one interpretation, the sentence in (76.d), Most babies cried, can be
interpreted as meaning that the number of babies who cried is greater than the number
of babies who didn't cry. The denotation of the subject NP most babies can thus be
defined as in (80).
(80) [most'(baby')]M-g = {X c A I |X n [baby']M* | > |[baby']M* n (A - X)|}.
This interpretation, however, turns out to be truth-conditionally equivalent to the
interpretation that might be given to the NP more than half the babies in the sentence
in (76.e), More than half the babies cried. An appropriate interpretation for this noun
phrase is given in (81) which guarantees that moreJhanjialf'(baby>')(cry'), the
(simplified) translation of (76.e), is true if (and only if) the number of crying babies
is greater than the number of non-crying ones.
(81) [more_than_half (baby')]M* = {XQ AI |X n [baby']M*| > | [baby']M*- X |}.
(81) is, however, equivalent to the interpretation given to most babies in (80), as
the reader can check by showing that, where B,X c A , B n (A - X) = B - X. (76.e)
and (76.d) thus turn out to be paraphrases. Many people, however, might feel that the
sentence most babies cried implies that a significantly greater proportion of babies
cried than didn't, and that (76.e) and (76.d) are not truly paraphrases. For example, in
a situation where there are fifty babies and twenty-six were crying while twenty-
four were not, the interpretation given to most babies in (80) and to more than half the
babies in (81) verifies both sentences, but the sentence with the latter NP is more
appropriate than the one with the former. It is possible that the implication here is a
pragmatic one, determined by appropriate usage rather than truth-conditions so that,
if speakers know that twenty-six out of fifty babies cried they are likely to utter the
sentence More than half the babies criedy rather than Most babies cried. However, it
is possible to incorporate the notion of context dependence more directly into the
semantics by interpreting most with respect to some pragmatically determined
numerical proportion of the number of entities in the extension of the common noun
that is greater than 50 per cent. If we let c represent this proportion, we can give
most'(baby') the interpretation in (82.a). Assuming that most in this context requires
at least 85 per cent of the elements in the extension of the common noun to have the
property denoted by the verb phrase and that there are fifty babies, the sentence in
190
(76.d) has the truth-conditions in (82.b). The truth-conditions of More than half the
babies cried are also met in any situation in which (82.b) holds, so that Most babies
cried entails that More than half the babies cried (cf. (82.c)). The converse, however,
is no longer true, because there are situations (e.g. where only twenty-six out of fifty
babies cried) in which the latter is true but the former is not. Thus, the two sentences
turn out not to be paraphrases under this interpretation, since they do not have exactly
the same truth-conditions.
(82) a. [most'(baby')]M'« =
{ X c A l |X > [baby']M'«| > c • |[baby']M«s|} where c > -5.
b. Let c = -85 and |[baby']M8| = 50, then [most'(baby')(cry')]M'8 is 1 iff
|[cry']*1* n [baby']M'«| > 43.
c. most'(N')(VF) -> more_than_half (N')
Other quantifiers like many and/<?w also need some reference to contextually
determined proportions. For example, the sentence in (76.b), Many linguists sneezed,
means that some number of linguists sneezed, the proportion being determined by the
context. Although it seems to be the case that the relevant proportion for determining
the truth of this should be greater than fifty per cent, this is not always the case. For
example, a sentence like Many civil servants have knighthoods might describe a
situation in which the number of civil servants in Britain who have knighthoods is
proportionately greater than the number of people in other fields in Britain who also
have knighthoods. It could thus be the case that on average a profession could expect
three per cent of its most senior members to be knighted, but that the number of
knighted civil servants is actually five per cent of the number of all civil servants. In
such a situation, the proposition expressed by the sentence Many civil servants have
knighthoods is true, despite the fact that nowhere near half of the number of such
people are knighted. Thus, we may assign to a noun phrase like Many civil servants
the interpretation in (83) where, as before, c is some pragmatically determined
proportion.
(83) [many'(civil-servant')]M-8 =
{ X c A l |X n [civil-servant']M<81 > c • | [civil-servant' ]Ms |}.
The quantifier few, as in (76.c), Few philosophers sang, requires the number of
singing philosophers to be less than fifty per cent, but again the exact proportion of
singing to non-singing philosophers is determined by the context of utterance. It is not
my intention to explore this issue further or to work out exactly how the proportion
c relates to the context for the different quantifiers (references to works that do can be
found at the end of the chapter). The generalised quantifier approach to NP
interpretation does, however, offer greater hope for the interpretation of quantifiers
other than the universal and existential quantifiers than the traditional treatment of
quantification given in the earlier part of this chapter. Indeed, it can be argued, this
approach has the advantage of providing an intuitively and empirically more
satisfactory analysis of many of the examples which were analysed in terms of the
classical universal and existential quantifiers in the earlier sections of this chapter.
Without doubt the theory does provide a fairly simple and coherent account of
quantifiers like two, both and those in the following exercise, which have an
interpretation fixed independently of the context.
191
6 Quantification
•Exercise 6.14:
Provide generalised quantifier interpretations for the determiners the, at most five,
all but three and exactly twelve, using the truth-conditions of the following sentences
as a guide.
i. The linguist screamed.
ii. At most five babies cried.
iii. All but three philosophers are insane.
iv. Exactly twelve students sang.
In addition to providing a framework within which more quantifiers than the

traditional logical quantifiers can be defined, the theory of generalised quantifiers also
allows the statement of constraints on the possible denotations of natural language
quantifiers (a term which may cover just items like every, all, most, few, etc., or all
determiners or the noun phrase itself, depending on the point of view being discussed).
The study of semantic universals is a topic that has not been discussed so far, but
those working within the framework of generalised quantifiers have begun to tackle
this aspect of semantics with some success. The earliest proposals along these lines
have, since the early part of the nineteen-eighties, been discussed and revised quite
extensively, but one universal seems to be particularly robust and to be generally
considered to be valid: the conservativity universal.
A quantifier phrase is called conservative if it has the conservativity property
defined in (84). Here, Q represents the extension of a quantifier, N that of a common
noun and X (which may be thought of as the extension of a VP) is any subset of A, the
universe of discourse. The definition requires the intersection (i.e. the conjunction) of
the extensions of the VP and the N to be a member of the set of sets denoted by a
quantified NP containing the quantifier and the common noun, provided that the
extension of the VP is also contained in this set of sets. The tautologous flavour of the
examples in (85) shows that this property holds for some common quantified NPs in
English. The conservativity universal requires every quantifier in any natural language
to have the property in (84).
(84) A quantifier with extension Q is conservative iff, for N , X c A ,
XeQ(N)H(NnX)eQ(N).
(85) a. Many students laughed iff Many students are students who laughed.
b. Every baby cried iff Every baby is a baby who cried.
c. Few lecturers screamed iff Few lecturers are lecturers who screamed.
(86) Conservativity: Every natural language quantifier is conservative.
This constraint may seem somewhat uninteresting, but it captures a fact about the
denotations of natural language quantifiers that is not logically necessary. It is not, for
example, logically necessary that if every man walks, then every man is a man who
walks, but a fact about the meaning of every (and other natural language quantifiers).
(86) rules out many logically possible quantifiers. For example, it rules out a
quantifier that denotes a function that picks out every subset of A provided that it is
192
not a subset of the extension of the common noun with which it is combined, i.e. one
that denotes the set of sets {X I ~(X c: [N]M'«)}. Since the intersection of any two sets
is a subset of both, such a quantifier does not include X n [N]M-» for any X and so it
is not conservative, and thus by (86) not a natural language quantifier.
Conservativity ensures that the interpretation of a noun phrase containing a
common noun N is not affected by those sets of entities that are not in the extension
of N. If we are interpreting the sentence Every dog barked, we do not need to take
account of cats, sheep, books or anything else in the domain of discourse that are not
dogs. Furthermore, the conservativity property cuts down the number of possible
extensions that quantifiers may have in any model. For example, in a model with just
two elements in its universe of discourse there are 256 logically possible quantifier
extensions (= (24)n), where n = 2, the number of entities in the model). However, if
only conservative extensions are permissible, natural language quantifiers may have
only one of 64 possible extensions (= (23)n). Thus, the space in which quantifiers may
have their meanings located is considerably reduced in number.
As well as identifying other general properties for quantifiers (like their semantic
independence from the identity, as opposed to the number, of entities in the extension
of the common noun), there have also been attempts to characterise the properties of
different subsets of quantifiers. In mathematics, a function, /+, is said to be monotone
increasing if for any two elements, x and y, in the domain where x is greater than y
thenf+(x) is also greater than/+(y). The converse of this is a monotone decreasing
function,/ where for any two elements, x and y, in the domain if x is greater than y
thenf(x) is less thanf(y). In formal semantics, these definitions have been adapted to
deal with the way an increase or a decrease in the number of entities in the extension
of the common noun in a noun phrase or in the extension of the verb phrase affects
truth values. For example, if we know that every philosopher sang in some model,
then adding more singers to the model does not affect the truth value of the formula
every '(philosopher')(sing'), but adding more philosophers to the model may affect its
truth value because the newly introduced philosopher may not be a singer. On the
other hand, taking away one of the philosophers (or picking out a subset of the
philosophers) does not affect the truth value of the formula. If the addition of more
entities to the extension of the common noun or the verb phrase does not affect the
truth value of a formula with the structure Qnt'(N')(VP'), then we say that the
quantifier is monotone increasing. If the subtraction of entities from the relevant
extensions fails to affect the truth value, then we say that Qnt is monotone decreasing.
We may make a further distinction between those quantifiers which are unaffected by
increase or decrease in their common noun extensions (subject monotone) and those
unaffected by increase or decrease of their associated VP extensions (predicate
monotone) to give the four classes in (87). Examples of the different sorts of
monotone quantifiers are given in (88) to (91). Each example contains four sentences
illustrating the four properties in (87) in order. Those sentences that are not
tautologous according to the interpretation of the quantifier are indicated by a
percentage sign. The patterns show that a is both left and right monotone increasing;
every is right monotone increasing, but left monotone decreasing; no is left and right
monotone decreasing and most is just right monotone increasing.
193
6 Quantification
(87) a. A quantifier Qnt is subject monotone increasing iff Qnt Nj VP entails Qnt
N2 VP, where [N,]M* c [N2]M-«
b. A quantifier Qnt is predicate monotone increasing iff Qnt N VPi entails Qnt
N VP2, where [VP,]M'« c [VP2]M'«.
c. A quantifier Qnt is subject monotone decreasing iff Qnt N! VP entails Qnt
N2 VP, where [N2]M* c [N,]M*
d. A quantifier Qnt is predicate monotone decreasing iff Qnt N VPi entails Qnt
N VP2, where [VP2]M.S C [ V P , ] ^
(88) a. If a philosopher with red hair sang, then a philosopher sang.
b. If a philosopher sang well, then a philosopher sang.
c. %If a philosopher sang, then a philosopher with red hair sang.
d. %If a philosopher sang, then a philosopher sang well.
(89) a. %If every philosopher with red hair sang, then every philosopher sang.
b. If every philosopher sang well, then every philosopher sang.
c. If every philosopher sang, then every philosopher with red hair sang.
d. %If every philosopher sang, then every philosopher sang well.
(90) a. %If no philosopher with red hair sang, then no philosopher sang.
b. %If no philosopher sang well, then no philosopher sang.
c. If no philosopher sang, then no philosopher with red hair sang.
d. If no philosopher sang, then no philosopher sang well.
(91) a. %If most philosophers with red hair sang, then most philosophers sang.
b. If most philosophers sang well, then most philosophers sang.
c. %If most philosophers sang, then most philosophers with red hair sang.
d. %If most philosophers sang, then most philosophers sang well.
Not all quantifiers or other determiners are either monotone increasing or
decreasing, however. For example, the pre-modifying quantifier phrase exactly half
in English is not monotone in either subject or predicate position, as illustrated by the
sentences in (92). (This result holds for all English phrases exactly n where n is some
number.)
(92) a. %If exactly half the philosophers with red hair sang, then exactly half the
philosophers sang.
b. %If exactly half the philosophers sang well, then exactly half the
philosophers sang.
c. %If exactly half the philosophers sang, then exactly half the philosophers
with red hair sang.
d. %If exactly half the philosophers sang, then exactly half the philosophers
sang well.
A number of interesting linguistic universals can be stated about noun phrases
according to the different sorts of monotone quantifiers that they contain. One of these
is that noun phrases can be conjoined by and or or only if they have quantifiers with
the same monotone direction. Thus, according to (93) we may conjoin noun phrases
containing no and every (as in (94.a)) or most and every (cf. (94.b)), but not most and
no (94.c) or no and a (94.d).
194
Further reading
(93) Co-ordination constraint: Two NPs can be co-ordinated by conjunction

(and) and disjunction (or) iff they are both monotone increasing or both
monotone decreasing.
(94) a. No student and every lecturer liked the book.
b. Most students and every lecturer liked the book.
c. *No student and most lecturers liked the book.
d. *No student and a lecturer liked the book.
The classification of the quantifiers into the classes in (87) have other benefits in
enabling us to identify and explain other linguistic phenomena (for example, negative
polarity items cannot appear in the scope of monotone increasing functors). Other
entailment patterns have also been identified and these, together with further general
properties of quantifiers, allow more universals to be stated. These universals and
constraint are somewhat technical in nature, and no more will be presented here, but
I hope that the short discussion in this section gives some idea of the fascinating and
fruitful ideas resulting from the study of generalised quantifiers.
6.7 Further Reading

There is a good deal written about quantification and the interpretation of noun
phrases and this section can do no more than indicate the more important texts on
particular topics. The classical quantifiers are given extensive treatment in many
introductory logic and linguistic textbooks of which the most accessible introduction
(which includes a discussion on translation) can be found in Guttenplan (1986:
166-246). The linguistic introductions to logic also discuss the classical quantifiers
in detail: see Allwood, Andersen and Dahl (1977: ch. 4), Partee, ter Meulen and Wall
(1990:137-154), McCawley (1981: ch. 4), Chierchiaand McConnell-Ginet (1990: ch.
3) and, for a more formal discussion, Dowty, Wall and Peters (1981: ch. 3). The
decomposition of quantified NPs using the ^-operator can be found in Montague
(1973) and Dowty, Wall and Peters (1981:104-111). Partee (1986) discusses the type
change from e to «e,t>,t>. There are a number of discussions about adjectives, of
which Klein (1980) and Siegel (1979) are the most accessible, while Kamp (1975),
Chierchia and McConnell-Ginet (1990: 370-377) and Keenan and Faltz (1985:
118-150; 308-333) require an acquaintance with intensional semantics. The literature
about definiteness and the interpretation of definite noun phrases is already vast and
still growing. The analysis of the in this chapter owes its origins to Russell (1905) and
Montague (1973) with Strawson (1950) providing the antithesis. See also Allwood,
Andersen and Dahl (1977: 148-155), McCawley (1981: 176-182) and the more
extended discussions in Hawkins (1978), Heim (1989), Kamp (1981), Ludlow and
Neale (1991) and the papers in Reuland and ter Meulen (1987). (Carlson (1989) and
Ojeda (1991) discuss the generic use of the definite article in English.) The issue of
quantifier scope ambiguities is addressed, inter alia, in Cooper (1983) and May
(1985). The literature on generalised quantifiers has grown steadily since the
publication of Barwise and Cooper (1984). Partee, ter Meulen and Wall (1990: ch. 14)
and Chierchia and McConnell-Ginet (1990: ch. 9) both provide introductions to the
topic, and more detailed discussions can be found in Keenan and Stavi (1986), Zwarts
(1983) and van Benthem (1986: chs. 1 &3). McCawley (1981 :ch. 14) discusses other
195
6 Quantification
types of quantifier and touches on mass terms and generics. Mass terms and plurals
are not discussed in this book, but the reader will find interesting, if difficult,
discussions of these topics in Link (1983), Hoeksema (1983) and L0nning (1987).
196
Inference
7.1 Making inferences

The semantic theory developed up to Chapter 6 has concentrated mainly on the
interpretation of sentences and phrases in isolation from each other, but one of the
criteria for assessing the adequacy of a semantic theory set out in Chapter 1 is that it
should account for the meaning relations that hold between different expressions in
a language. This means, amongst other things, that the semantic theory proposed here
ought to guarantee that, where reference and context are kept constant, the sentences
in (1 .b) and (1 .c) are paraphrases of (1 .a) while (1 .d) and (1 .e) are entailments of it
and (l.f) and (l.g) are contradictions of it.
(1) a. Jo stroked the cat and kicked the dog.
b. Jo kicked the dog and stroked the cat.
c. The cat was stroked by Jo and the dog was kicked by Jo.
d. Jo stroked the cat.
e. Someone kicked the dog.
f. The dog wasn't kicked.
g. No-one stroked anything.
The intuitively identified relations between the sentences in (1) derive from the
interpretations of the conjunction and, the negative not and the quantifier pronouns
no-one and someone. Such relations are generally referred to as logical entailments,
paraphrases or contradictions. (Note that these terms are used ambiguously between
the relation that holds amongst sentences, as here, and the product sentences
themselves, as in the first paragraph above.) Because of their logical nature, these
relations should be directly explained by the interpretation given to the logical
elements associated with the relevant grammatical expressions (&, ~ and 3) by the
model theory. Meaning relations of this sort are used to make inferences. An
inference is the recovery of information from a discourse which is implicitly or
explicitly presented as true by the speaker, but which is not stated directly. For
example, an interlocutor, say Bertie, may infer the sentence in (2.b) from the utterance
of (2.a) in a situation where Bertie knows that Chester did eat the cake.
(2) a. If Chester ate the cake, then he will probably die.
b. Chester will probably die.
More interesting chains of inference result from combining different bits of
information, expressed in sentences or formulae, to arrive at some conclusion. For
example, from the sentence in (3.a) it is possible to arrive at the inference in (3.f)
using the information expressed in the sentences in (3.b) to (3.e). In logical terms, (3)
has the structure of a valid inference pattern, since the truth of the conclusion in
(3.f) is guaranteed by the truth of all of the premisses in (3.a) to (3.e). We may indeed
think of an inference pattern of this sort as an extended entailment relation where the
set of premisses are treated as a compound sentence conjoined by and which directly
197
7 Inference
entails the conclusion. Such a pattern is called an argument by logicians, but the term
will not be used here in order to avoid a potentially confusing ambiguity with the
argument of a functor. In order to show the extent to which the semantic theory set up
so far is adequate, it is necessary to show the extent to which it can capture the
meaning relations that hold between sentences and account for intuitively valid
inference patterns like that in (3).
(3) a. The dog has eaten the cake.
b. The cake is poisonous.
c. If something poisonous is eaten, the entity that ate the poisonous substance
will become ill.
d. A dog is an animal.
e. If an animal is ill, a vet should be called.
f. Someone should call a vet.
Inference of the sort shown in (2) and (3) is called deductive inference because
the truth of the conclusion results from the logical interpretation of the premisses.
There are, however, inference processes that rely on non-logical relations between
expressions in a particular language. One such relation results from lexical meaning
and the lexical relations that hold in the vocabulary of some language. The term
lexical meaning is to be interpreted as concerning that part of the meaning of a
sentence that results, not from its logical structure (in terms of conjunctions and
quantifiers), but from the idiosyncratic meanings of its component words. A
distinction is thus being made between the meanings of grammatical or functional
words like a, every, and, i/and so on, and content lexemes that translate into constants
in the logical translation language. Expressions of the former sort are interpreted in
the same way with respect to all models, whilst the denotations of the latter
expressions may change from model to model. There are, however, relations that are
expected to hold between content expressions within particular models. For example,
in (4), (4.b) and (4.c) may be inferred from (4.a), the first due to the meaning of the
word kill and its relation to the word dead and the latter due to that of the words owl
and mouse and their relations to the words bird and mammal, respectively.
(4) a. An owl killed a mouse.
b. A mouse is dead.
c. A bird killed a mammal.
The inferences recovered from the meanings of words, however, are weaker than
those that derive from the interpretation of logical expressions in that they can more
readily be cancelled. This is most obvious in figurative or metaphorical speech where
properties usually associated with certain words are influenced by those associated
with other words in the sentence. The sentence in (5 .a), for example, is not
contradictory in the same way that (5.b) is, and conveys more obvious information,
despite the fact that the semi-formal sentences in (5.c), (5.d) and (5.e) describe
apparent entailments associated with the words idea and swallow that ought to license
the contradictory inference in (5.b). The position is taken here that lexical meaning
relations are not therefore absolute constraints on models, so not purely logical and
thus able to be overridden in certain circumstances.
198
Making inferences
(5) a. Jo swallowed the idea.

b. An idea is and is not abstract.
c. If x swallows y, then x is concrete.
d. If x is concrete, x is not abstract.
e. An idea is abstract.
Other types of inference exist, weaker than those derived through the meaning
relations of words and sentences which derive from presupposition and implicature.
The former has already been mentioned briefly with respect to definiteness and
uniqueness in Chapter 6 and Chapter 1. The difference between presupposition and
entailment is that presuppositions persist for both positive and negative sentences and
that the falsity of the presupposition does not so obviously falsify the original sentence
as does the falsity of an entailment. This is illustrated in (6) where the truth of (6.a)
presupposes that of (6.b) and where the falsity of (6.b) does not falsify (6.a) as does
the falsity of (6.c) (an entailment of (6.a)). This topic is a controversial and interesting
one, but does not directly concern us here and no more will be said about it.
(6) a. The King of France is not bald.
b. There exists an entity that is King of France.
c. There exists an entity that is not bald.
Another form of pragmatic inference is briefly discussed in Section 7.4 to
account for some of the apparent anomalies between the interpretations of certain
natural language constructions and their logical counterparts. This type of inference
results from the implicatures of an utterance. Implicature is probably the weakest form
of inference that can be made, since it holds only where there is no information to the
contrary and can be contradicted without contradicting the original sentence. For
example, the truth of (7.a) implicates the truth of (7.b), but this implicature can be
cancelled without contradiction as shown by the sentence in (7.c) which is not
obviously contradictory despite the fact that it denies the truth of (7.b) (although most
speakers of English might judge the former to be odd or paradoxical).
(7) a. Every boy in the room kissed Fiona.
b. Someone kissed Fiona.
c. Yes, that's true. Every boy in the room did kiss Fiona, but there were no
boys in the room!
The different sorts of inference mentioned above seem to form a hierarchy in
terms of how they relate to the sentence (or sentences) from which the inference is
made. Inference deriving from entailment (or paraphrase or contradiction) must
always hold, while that made through implicature need not hold. The inferences
determined by lexical meaning (and presupposition) fill intermediate slots in the
inference hierarchy shown in (8).
(8) Inference hierarchy: entailment > lexical meaning > implicature.
In this chapter, we will concentrate on deductive inference that can be made
using the logical operators and connectives that have been introduced in previous
chapters, showing how the interpretation given to them explains common inference
strategies. We will then take an all too brief look at a way of incorporating lexical
199
7 Inference
meaning into the theory and end with a short discussion of the way implicature can
account for some non-truth-functional aspects of the English counterparts of the
logical connectives &, v and —».
7.2 Logical deduction

Deductive inference derives from the primary meaning relations between sentences,
entailment, paraphrase and contradiction. By formally defining the latter, therefore,
we can show how adequately the model theory defined in Chapter 6 accounts for valid
inferences. Within truth-conditional semantics, the primary logical meaning relations
can be given general definitions in terms of the relation between the truth or falsity of
the proposition expressed by one sentence and that expressed by another.
One sentence is an entailment of another only if the truth of the former is
guaranteed by that of the latter. In other words, whenever (the proposition expressed
by) a sentence is true, (the propositions expressed by) its entailments are also true,
although the reverse need not be the case. These are, of course, just the truth-
conditions of material implication and so we can characterise entailment between two
sentences in terms of their combination using —>. Since entailment, like the other
logical meaning relations, holds in all situations, the material implication must be true
in all situations. This allows us to define entailment as holding between two sentences
when the formulae associated with each of the sentences conjoined by the material
implication sign, ->, make a tautology, a formula that is always true no matter what
the denotations of its component parts happen to be. Adopting the symbol I- to signify
the relation entails (so that Sj h S2 means sentence 57 entails sentence S2) provides the
definition of entailment in (9).
(9) Entailment: S, \- &> iff S,' -» S2' is a tautology.
To see how this definition works, we can take as an example the simplest
entailment in our fragment which results from an inference rule that logicians call
conjunction elimination. This rule allows the inference that each conjunct of a true
formula is also true. For example, in (10), the truth of both (lO.b) and (lO.c) may be
inferred from that of (lO.a) according to this rule.
(10) a. 3x [Vy[cake'(y) <-> x = y] & (eat'(x))(chester')] & sick'(chester').
b. 3x [Vy[cake'(y) <-> x = y] & (eat'(x))(chester')] .
c. sick'(Chester').
According to (9), if conjunction elimination is a valid rule, a sentence that translates
into a formula conjoined by & (e.g. (lO.a)) entails the sentences that translate into the
conjuncts (e.g. (lO.b) and (lO.c)), provided that the material implication formed from
the conjoined formula and each of its conjuncts is always true no matter what truth
values are associated with the antecedent or the consequent. In other words, the
sentence in (11.a) entails both (11 .b) and (1 l.c) if (and only if) (12.a) and (12.b) are
tautologies.
(11) a. Chester ate the cake and was sick.
b. Chester ate the cake.
c. Chester was sick.
200
Logical deduction
(12) a. (3x [Vy[cake'(y) <-> x = y] & eat'(x)(chester')] & sick*(Chester'))

-> 3x [Vy[cake'(y) <-> x = y] & eat'(x)(chester')].
b. (3x [Vy[cake'(y) <-> x = y] & eat'(x)(chester')] & sick'(Chester'))
—> sick'(Chester').
It is easy to show that both formulae in (12) are indeed tautologies by using the
truth table method introduced in Chapter 3. If we represent the formula denoted by
Chester ate the cake as p and that expressed by Chester was sick as q, we can see that
the formula (p & q) —> p is a tautology because the final implication always has the
value t (or 1) no matter what truth values are assigned to p and q as shown in (13).
Hence, according to (9), the entailments in (11) are valid and our theory successfully
guarantees that a sentence of the form S} and S2 entails both S} and S2 (and hence that
both these sentences may be inferred from the former).
(13)
p q p&q (p & q) -> p
t t t t
t f f t
f t f t
f f f t
The two other sentential meaning relations of paraphrase and contradiction can
also be defined in terms of tautologies that result from combining appropriate
formulae with particular connectives. Two sentences paraphrase each other if they
both have precisely the same truth-conditions. In other words, sentences that are
paraphrases of each other express propositions that are both true or both false (i.e.
logically equivalent) in all the same situations. Hence, we may show that two
sentences are paraphrases by showing that the formulae they express, when conjoined
by the equivalence connective, <->, make a tautology. This is expressed by the
definition in (14) where the symbol H h is used to signify the relation paraphrases.
(14) Paraphrase: S, HK &_ iff S,' <-> &' is a tautology.
We have already seen the equivalence that holds between ^-expressions applied
to an argument and their ^-converted equivalents in Chapter 5. By (14), this is
sufficient to make passive and active pairs paraphrases as, intuitively, they should be.
For example, the semantics of the lambda operator ensures that the formulae in (15.c)
and (15.d) are truth-conditionally equivalent (so that their conjunction by <-> is always
true) and, thus, by (14), the English sentences that they translate, (15.a) and (15.b),
respectively, are paraphrases as required.
(15) a. Chester was kicked by Jo.
b. Jo kicked Chester.
c. Xx [(kick'(x))(jo')](chester').
d. (kick' (Chester' ))(jo').
There is another simple logical paraphrase between sentences which was
mentioned in Chapter 3 and involves the truth-conditional equivalence between
201
7 Inference
positive formulae and their doubly negated counterparts. As we saw in Chapter 3, two
negatives in logical expressions are equivalent to a positive, so that ~(~p) is truth-
conditionally equivalent to p, where p is any formula. This is shown in the truth table
in (16) and so, according to (14), (17.a) and (17.b) are paraphrases because (17.c) is
a tautology.
(16)
p ~p ~(~p) p <-> ~(~p)
t f t t
f t f t
(17) a. It rained.
b. It is not the case that it didn't rain.
c. rain' <-> ~(~rain').
Because the definition of paraphrase in (15) utilises truth-conditional equivalence, it
also predicts that two paraphrases entail each other. This results from the fact that a
formula of the form p <-> q is truth-conditionally equivalent to one of the form
((P —* Q) & (Q —> P)) ( a s shown in the truth table in (43) of Chapter 3). By conjunction
elimination, therefore, we can infer that both p —> q and q —» p hold and so by the
definition of entailment in (9), paraphrases that translate into p and q are also
entailments of each other. Thus, the definition of paraphrase in (14) treats the relation
as equivalent to mutual entailment, as suggested in Chapter 1.
The final relation between sentences that we need to look at is contradiction.
Here, the truth of the proposition expressed by one sentence guarantees the falsity
of that expressed by another. There are a number of ways this can be defined using
the connectives, one of which is to use material implication and negation to bring out
the relation between contradiction and entailment. Essentially, the definition of
contradiction in (18), which uses the symbol t= to denote the relation contradicts,
defines one sentence as contradicting another if the latter entails the negation of the
former.
(18) Contradiction: Si •= S2 iff ST -» ~S 2 ' is a tautology.
From (18) it is easy to see that the negation of any entailment is a contradiction. Thus,
the sentence in (19.a) is contradicted by that in (19.b) which is the negation of an
entailment derived from conjunction elimination. This is guaranteed by the fact that
(19.c) is a tautology, which is shown by the truth table in (20) where p represents the
translation of Chester ate the cake and q that of Chester was sick.
(19) a. Chester ate the cake and was sick.
b. Chester wasn't sick.
c. (3x [Vy[cake'(y) <-> x = y] & eat'(x)(chester')] & sick'(Chester'))
-^ ~(~(sick'(Chester'))).
202
Logical deduction
(20)
p q p&q ~q ~(~q) ((p & q) -> ~(~q))
t t t f t t
t f f t f t
f t f f t t
f f f t f t
Having given a formal definition of the primary meaning relations between

sentences, we can proceed to an examination of some of the valid deductive inferences
that are captured by the semantic theory proposed in Chapters 2 to 6. Clearly, any
entailment or paraphrase of a sentence is a valid inference that can be made from that
sentence. But, as we have seen, inferences often operate over sets of sentences rather
than single ones. In other words, deductive inference normally takes the shape of an
inference pattern that contains a set of premisses and a conclusion. An inference
pattern is valid if, and only if, the truth of all its premisses guarantees the truth of the
conclusion. This is like entailment in an extended sense where the entailing sentence
is a conjunction of all the premisses and the conclusion is entailed by this conjunction.
An inference pattern may therefore be defined as valid, if the material implication
formed with the conjunction of the translations of its premisses and that of its
conclusion as the consequent is a tautology, as in (21).
(21) An inference pattern Si,...,Sn.i \- Sn is valid iff (S,' & .... & Sn.,') -» Sn' is a
tautology.
A simple inference pattern that is intuitively valid is shown in (22) where (22.a)
and (22.b) are the premisses and (22.c) is the conclusion.
(22) a. Fiona is either at home or in the library.
b. She isn't in the library.
c. Then she is at home.
According to (21), we can show that this is valid according to the model theory by
proving that a formula with the structure <(> —> \|/ is a tautology, where \\f is the
translation of (22.c) and ()> is the conjunction of the translations of (22.a) and (22.b).
If we let p stand for the formula translating Fiona was at home and q stand for that
translating Fiona was in the library, then the logical structure of the inference pattern
is ((p v q) & ~q) —> p. The truth table in (23) shows that this expression is
tautologous, thus proving that the inference pattern in (22) is actually valid according
to the semantic interpretation provided for or and not.
203
7 Inference
(23)
p q pvq ~q (p V q) & ~q ((p vq)&~q)-)p
t t t f f t
t f t t t t
f t t f f t
f f f t f t
7.2.1 Using the connectives

In this section, we look at some of the many valid inference patterns that can be
constructed using the propositional connectives. Only a few of these are discussed
here and the reader is referred to a good introduction to propositional logic for a
fuller discussion. Only the most commonly used patterns involving the expressions
and, or, not and //are discussed below.
We have already seen one valid inference pattern involving disjunction and
negation in (22) above. Another one, called disjunction elimination by logicians,
exploits the relation between disjunction and implication to arrive at a particular
conclusion. An inference pattern with the appropriate structure is shown in (24), where
(24.a), (24.b) and (24.c) are the premisses and (24.d) is the conclusion. The logical
structure of the premisses and conclusion in terms of propositional variables and
connectives is shown on the right hand side of the example where p stands for Jo
kicked Bertie, q for Jo kicked Ethel and r for Ethel hit Jo. The validity of this inference
pattern is shown by the truth table in (25) which is constructed from (24) along the
lines set out in (21). Since the complex formula on the right is always true no matter
what values are assigned to the basic formulae, the inference pattern is valid and the
conclusion in (24.d) may be inferred from the premisses, as required. (Note that the
formula ((p v q ) & ( p - > r ) & ( q - > r ) ) - > r is false if, and only if, all of the conjuncts
in the antecedent are true and the consequent is false. To show that it is true, therefore,
it is sufficient to show that one of the conjuncts is false or that the consequent is true.
Hence, the intermediate step of showing the truth value of the antecedent (p v q) &
(p —> r) & (q —> r) is omitted from (25).)
(24) a. Jo kicked Bertie or Ethel. pvq
b. If Jo kicked Bertie, then Ethel hit Jo. p-»r
c. If Jo kicked Ethel, then Ethel hit Jo. q-»r
d. Ethel hit Jo. r
Other important inference rules involve material implication. Given the truth of
an if... then sentence and its antecedent, we can confidently infer the truth of the
consequent, as illustrated in (26). This inference pattern is called modus ponens by
logicians and its validity is shown by the truth table in (27).
204
Logical deduction
(25)
p q r (pvq) p->r q ^ r ((pvq) &(p->r) &(q->r))->r
t t t t t t
t t f t f f t
t f t t t t
f t t t t t
t f f f t t
f t f t f t
f f t f t t t
f f f f t t t
(26) a. If Ethel was not at home, then she was in the pub.
b. Ethel was not at home. p
c. Ethel was in the pub. q
(27)
p q (p->q) (p->q) &p ((p -> q) & p) -> q)
t t t t t
t f f f t
f t t f t
f f t f t
Another inference rule involving material implication is called modus tollens

and is in effect the reverse of modus ponens. Given a conditional and the negation of
its consequent, it is possible to infer that the antecedent is also false. In other words,
given two sentences If SI then S2 and It is not the case that S2, we infer It is not the
case that 57. An example of this type of inference pattern is given in (28) where the
final conclusion in (28.d) relies on the equivalence between p and —p discussed
above.
(28) a. If Ethel was not at home, then she was in the pub. ~p -> q
b. Ethel was not in the pub. ~q
c. It is not the case that Ethel was not at home. —p
d. Ethel was at home. p
Exercise 7.1:
Prove that modus tollens is valid by constructing the appropriate truth table.
205
7 Inference
These inference pattern types can be put together to construct quite complex
chains of inference with new premisses being deduced from old ones to arrive at an
ultimate conclusion. In texts, however, the premisses and intermediate conclusions of
a chain of inference are often not expressed. This is illustrated in the short dialogue
between Bertie and Fiona in (29).
(29) a. Fiona: That's Ethel now.
b. Bertie: What is she doing?
c. Fiona: Well, she didn't kick the cat.
d. Bertie: OK, I'll slice the lemon.
The interesting thing about this exchange from our point of view is how Bertie's
response in (29.d) relates to Fiona's statement in (29.c). We can show how Bertie gets
to his conclusion by making explicit the unexpressed premisses that he uses with
Fiona's statement in (29.d) to come to intermediate conclusions that ultimately lead
to his assertion of (29.d). The full inference pattern in English is set out in (30). (30.a)
is the initial premiss asserted by Fiona. Bertie uses the (implicit) premiss in (30.b) to
arrive at the intermediate conclusion in (30.c) via modus tollens. The latter sentence
is, as we have seen, a paraphrase of (30.d) which can then be used to derive (30.f)
from the premiss in (30.e) via modus ponens. This conclusion can then be combined
with another conditional premiss in (30.g) to give the conjunction in (30.h). Applying
conjunction elimination to this (cf. (12) above) yields (30.i) which combines with the
final premiss in (30.j) to give the final conclusion in (30.k) again via modus ponens.
(30) a. Ethel didn't kick the cat.
b. If Ethel doesn't win a golf match, she kicks the cat.
c. It is not the case that Ethel didn't win the golf match.
d. Ethel won the golf match.
e. If Ethel won the golf match, she wants a drink.
f. Ethel wants a drink.
g. If Ethel wants a drink, we need gin and sliced lemon,
h. We need gin and sliced lemon.
i. We need sliced lemon.
j. If we need sliced lemon, I'll slice it.
k. I'll slice the lemon.
The logical structure of this inference pattern is set out in (31) where q stands for
Ethel kicked the cat, p for Ethel won the golf match, r for Ethel wants a drink, s for We
need gin, u for We need sliced lemon and v for I'll slice the lemon. The inference rule
that licenses the steps taken at each point in the inference pattern is indicated on the
right hand side and inasmuch as each of these small steps is valid, so the whole
inference pattern is validated by the semantic interpretation given to the connectives
in the model theory.
(31) a. ~q. premiss
b. ~p —> q. premiss
c. —p. deduction from a & b via modus tollens
d. p. deduction from c via negation elimination
e. p —» r. premiss
206
Logical deduction
f. r. deduction from d & e via modus ponens

g. r —» (s & u). premiss
h. s & u. deduction from f & g via modus ponens
i. u. deduction from h via conjunction elimination
j. u —> v. premiss
k. v. deduction from i & j via modus ponens
So far we have looked at the way inference patterns are constructed from
information gleaned from entailments and paraphrases, but conclusions can also be
drawn from the use of contradictions. The most obvious way that contradictions are
used in discourse is to challenge or disprove the truth of other assertions, as
exemplified in the simple exchange in (32).
(32) a. Jo: Fiona said she's working hard on her thesis today.
b. Bertie: But F ve just seen her in the pub.
c. Jo: Oh dear! She lied to me.
The reasoning that lies behind this exchange is fairly simple. Suppose that Fiona has
said to Jo that she is working hard on her thesis and so he asserts (32.a) to Bertie. On
standard assumptions that people generally speak the truth, both Jo and Bertie can
infer from this that Fiona is working hard on her thesis is true (let p represent the
formula that translates this sentence). Bertie, however, has seen Fiona in the pub and
asserts (32.b) (call the formula that translates Fiona is in the pub q). At this point an
unexpressed premiss comes into play to the effect that if someone is in the pub then
they are not working hard, from which we can get the formula q -> ~p. However, from
this and Bertie's assertion, we can infer ~p, via modus ponens. This, of course,
contradicts the earlier inference p. Although this does not contradict (32.a), since
Fiona did say that she was working hard, Jo can infer that she did not assert p
truthfully and his knowledge of lexical meaning allows him to infer that she lied.
Hence, while Bertie does not directly contradict Fiona's assertion, he provides
information that can be used to show that Fiona has been economical with the truth
with respect to her working hard on her thesis.
There is another use of contradictory information that involves a form of indirect
reasoning called a reductio ad absurdum argument (or negation introduction) by
logicians. In this form of reasoning, an assumption is made in the course of an
inference pattern which is not itself validated by the current premisses. This
assumption is then used to deduce logically that some other sentence is both true and
false. From this logically false conclusion, it follows that the assumption must be
false, since one cannot deduce a false conclusion from a true premiss. This form of
reasoning thus allows one to eliminate different possible explanations; any assumption
that leads to a contradiction must be false and so can be discarded. This may sound a
bit peculiar, but is exemplified in the dialogue in (33) which takes place between Jo
and Ethel after Jo has discovered that the poisoned cake has been eaten.
(33) a. Jo: The dog ate the poisoned cake.
b. Ethel: F ve just seen him playing with a ball.
c. Jo: Then who did eat the cake?
207
7 Inference
This exchange might seem peculiar because Jo asserts in (33.a) that a certain
entity ate the poisoned cake, but in (33.c) asks Ethel who that entity is. Ethel's
response to Jo's initial statement does not directly contradict Jo's statement in (33.a)
but she supplies information that when combined with certain unexpressed premisses
show that it must be false. Suppose that Jo finds the plate holding the poisoned cake
empty and so concludes that it has been eaten. In order to show that this is the case,
he must, by the interpretation given to such a sentence, find some entity that ate the
cake. In logical terms, he must find something that can be associated with the variable
z that satisfies the open formula 3x [Vy[cake'(y) & poisoned'(y) <-> JC = y] &
eat'(x)(z)]. He makes a first assumption that Chester is the culprit, but when Fiona
makes her assertion in (33.b), he brings to bear three background premisses that
show that his assumption must be false. The first is that if someone has eaten the
poisoned cake then they are dead; the second is that if someone is playing with a ball
then they are alive; and the third one derives from the meaning of the word alive to
the effect that if something is alive then it is not dead. From the second of the
implicational premisses and Ethel's assertion, we infer, via modus ponens, that
Chester is alive. From this conclusion and the third implicational premiss, modus
ponens again allows us to infer that Chester is not dead. However, from Jo's
assumption and the first of the additional premisses, we conclude that Chester is dead.
There is another deduction rule whose validity is intuitively obvious (and
provable) that allows us to construct a conjunction from two true formulae (this is
often called conjunction introduction). Since we have proved that Chester is dead
is true and that Chester is not dead is also true, we can infer that Chester is dead and
not dead is also true, but a formula of the form p & ~p can never be true. Hence, we
are forced to conclude that the original assumption, that Chester ate the cake, must be
false, since it was this that led to the contradiction. Hence, Jo can infer from Ethel's
statement that Chester didn't eat the cake so he is none the wiser as to who did and
responds with a question.
This somewhat convoluted chain of argument is spelled out in English in (34)
and the logical form of the inference pattern is shown in (35). Again, in the latter, the
rule that licenses each step is given on the right hand side. The sentence Chester ate
the poisoned cake is represented by p, Chester is dead is represented by q, Chester is
playing with a ball is represented by r and Chester is alive is represented by s.
(34) (Someone ate the poisoned cake.)
a. Chester ate the poisoned cake.
b. If Chester ate the poisoned cake, then he is dead.
c. Chester is dead.
d. If Chester is playing with a ball, then he is alive.
e. Chester is playing with a ball.
f. Chester is alive.
g. If Chester is alive, then he is not dead,
h. Chester is not dead.
i. Chester is dead and he isn't dead,
j. Chester didn't eat the poisoned cake.
208
Logical deduction
(35) a. P- assumption
b. p->q. premiss
c. q. deduction from a & b via modus ponens
d. r—»s. premiss
e. r. premiss
f. s. deduction from d & e via modus ponens
g. s —» ~q. premiss
h. ~q. deduction from f & g via modus ponens
i. q & ~q. deduction from c & f via conjunction introduction.
j- ~P- reductio ad absurdum
The reason that reductio ad absurdum arguments work can be seen by
considering the logical structure of the inference pattern in (35). From the assumption
of p, i.e. that Chester ate the poisoned cake, we come step by step to the contradictory
conclusion in (35.i), i.e. that Chester is and is not dead. As we have seen above, a
conclusion is truth-conditionally related to its premisses by material implication.
Hence, we can link the assumption p with the conclusion q & ~q in a conditional
expression as in (36.a). The creation of an implication combining an assumption with
a conclusion reached from that assumption by standard deductive rules is a valid rule
of logic, called conditional introduction. Because the consequent is a logical
contradiction, i.e. it can never be true, the conditional, by the semantics of —>, can only
be true if p is false. Hence, the generation of a contradiction from an assumption must
lead to the conclusion that the original assumption was false (a version of modus
tollens). This is clearly brought out in the conditional English sentence in (36.b) from
which it is intuitively necessary to conclude that Chester did not eat the poisoned
cake. Logically valid reductio ad absurdum arguments thus mirror certain human
deductive processes and our theory thus provides an explanation of these in formal
terms.
(36) a. p -> (q & ~q).
b. If Chester ate the poisoned cake, then Chester is and is not dead.
There are many other logical deductions utilising the connectives that follow
validly from the interpretation that has been given to them in earlier chapters.
However, the examples given here should be sufficient for the reader to see how they
work. Most inference patterns can be reduced to repeated steps involving the seven
rules introduced above, repeated as (37.a) to (37.g). (37.h) introduces another valid
inference pattern involving disjunction and negation which the reader can check is
indeed valid by showing that ((p v q) & ~p) -» q is a tautology.
(37) a. Conjunction elimination (CE): p & q i- p.
b. Conjunction introduction (CI): p, q H p & q.
c. Disjunction elimination (DEI): p v q, p -» r, q -» r h r.
d. Disjunction elimination 2 (DE2): p v q, ~p \- q.
e. Modus ponens (MP): p —> q, p \- q.
f. Modus tollens (MT): p -> q, ~q \- ~p.
g. Conditional introduction (Cndl): (p \- q) h p -> q.
h. Reduction ad absurdum (RA): (p h (q & ~q)) h ~p.
209
7 Inference
Exercise 7.2:
Using the rules of deduction in (37), show why the following deductions are valid.
Some ancillary premisses derived from the meanings of some of the words may need
to be invoked. Make sure that these are included in the structure of the inference
pattern.
i. Either Fiona or Bertie poisoned the cake. If Fiona poisoned the cake, then
she will have a bottle of poison in her bedroom and a book about unsolved
murders on her bedside table. Therefore, if Bertie didn't poison the cake,
then Fiona will have a book about unsolved murders on her bedside table.
ii. If Ethel lost the golf match, then she will be tired and unhappy and if she
is unhappy, there will be no gin left. But there is some gin left, so Ethel won
the golf match.
iii. If Jo likes Ethel, then he will give her a copy of Jane Eyre or a new bottle
of gin. If he gives her Jane Eyre, she will not be pleased and think he's a
creep, but if he gives her a new bottle of gin, she will be pleased but still
think he's a creep. So if Ethel doesn't think Jo is a creep, then he doesn't
like her!
7.2.2 Reasoning with quantifiers

At the end of Chapter 6, a number of entailments were mentioned that are associated
with different types of quantifier in terms of (left or right) monotonicity and also a
very general entailment, that of conservativity, which is believed by many formal
semanticists to be applicable to all quantifiers in every natural language. There are,
however, many other entailments from sentences containing noun phrases with
different quantifiers. In this section, however, we consider only deduction rules that
are associated with the classical quantifiers, V and 3.
Consider the inference pattern in (38) which is valid, according to (21), if, and
only if, the formula in (39) is a tautology.
(38) a. Every student is intelligent.
b. Fiona is a student.
c. Therefore, Fiona is intelligent.
(39) (Vx [student'(x) -> intelligent'(x)] & student'(fiona')) ->
intelligent' (fiona').
Because the formula in (39) is in the form of a material implication, it is false
only if the antecedent is true and the consequent is false. To prove that (39) is always
true, therefore, it suffices to show that this cannot ever be the case, i.e. that it is
impossible for every student to be intelligent and for Fiona to be a student but for
Fiona not to be intelligent. It is relatively easy to show that this is the case by using
a reductio ad absurdum argument, i.e. by showing that the assumption that (39) is false
leads to a contradiction. Assuming, then, that (39) is false, the antecedent formula VJC
[student'(x) -> intelligent'(x)] & student'(fiona') must be true. Since this is a
210
Logical deduction
conjunction, both student'(fiona') and VJC [student'(x) —»intelligent '(x)] must also be
true by conjunction elimination. For the latter formula to be true according to the
truth-conditions for universally quantified formulae discussed in the last chapter, the
open formula student'(x) -» intelligent'(x) must be satisfied by every possible
assignment of a value to x. One of these assignments associates x with the entity
denoted by fiona', i.e. WOMAN2, and satisfies student'(x) —> intelligent'(x) only if
it satisfies intelligent'(x) or fails to satisfy student'(x). Since we have assumed that
student'(fiona') is true, gwoMAN2* d o e s satisfy student'(x) and so must also satisfy
intelligent'(x). However, our original assumption that (39) is false requires
intelligent '(fiona') also to be false, but this leads to the contradiction that Fiona is both
intelligent and not intelligent. By the reductio ad absurdum argument, therefore, we
are forced to conclude that the assumption that (39) is false must itself be false. In
other words, (39) must always be true and so we have proved that (38.c) is validly
derived from (38.a) and (38.b). This type of inference, shown schematically in (40),
is called universal instantiation, because the premiss allows a variable bound by
the universal quantifier to be instantiated as any constant, a.
(40) Universal instantiation: Vx [P(x)] h P(a).
There is a parallel inference rule for existential formulae which is in a sense the
inverse of (40). Instead of instantiating a variable as a constant, the rule of existential
generalisation replaces a constant by a variable which is then bound by the existential
quantifier. This rule is stated in (41) and exemplified in (42). The premiss in (42.a)
licenses the conclusion in (42.b) because the formula in (42.c) is a tautology. Again,
this can be shown by assuming that it is false and showing that this assumption leads
to a contradiction. If (41. c) is false, then intelligent'(fiona') is true, and so the
assignment of the entity denoted by Fiona to x satisfies intelligent'(x). However, the
falsity of (41.c) requires 3x [intelligent'(x)] to be false so that there must be no value
assignment satisfying this open formula. We are therefore forced to conclude that
there is and is not a value assignment that satisfies intelligent'(x) which in turn entails
that (42.c) can never be false, thus proving that (41) is a valid deduction.
(41) Existential generalisation: P(a) \- 3x [P(x)].
(42) a. Fiona is intelligent.
b. Therefore, someone is intelligent.
c. intelligent'(fiona') —>3x [intelligent'(x)].
As an example of how the two deduction rules in (40) and (41) interact with each
other and with other rules of deduction, consider the inference pattern in (43). The
steps taken to get to the conclusion (43.d) from the premisses in (43.a) to (43.c) are
shown in (44). As before, the deduction rule used for each step appears on the right
hand side together with the number of the line on which relevant formulae appear. In
order to simplify matters, we assume that gin translates into an individual constant
and that alcoholic beverage is translated as the predicate a-b'.
(43) a. Ethel likes gin.
b. Gin is an alcoholic beverage.
c. All alcoholic beverages are bad.
d. Therefore, Ethel likes something that is bad.
211
7 Inference
(44) a. (like'(gin'))(ether). premiss

b. a-b'(gin'). premiss
c. Vx [a-b'(x) —> bad'(x)]. premiss
d. a-b'(gin') —> bad'(gin'). from c via universal instantiation
e. bad' (gin'). from b & d via modus ponens
f. (like'(gin'))(ethel') & bad'(gin').from a & e via conjunction introduction
g. 3x [(like'(x))(ethel') & bad'(x)]. from f via existential generalisation.
Further entailments and paraphrases involving the interaction of the existential
and universal quantifiers with the negation operator were mentioned in Chapter 6.
There it was noted that a sentence like (45.a) is a paraphrase of (45.b) and a sentence
like (45.c) is a paraphrase of both (45.d) and (45.e). In addition, each of the sentences
in (45.c), (45.d) and (45.e) all entail both (45.a) and (45.b).
(45) a. Someone didn't like Jo
b. Not everyone liked Jo.
c. Everyone didn't like Jo.
d. It is not the case that someone liked Jo.
e. No-one liked Jo.
According to the formal definitions of entailment and paraphrase in (9) and (14), we
can show that these entailment and paraphrase relations in (45) hold by showing that
their translations can be combined to form tautologies, using <-> and —> respectively.
The relevant formula to show the paraphrase relation between (45.a) and (45.b) is
shown in (46.a), that between (45.c), (45.d) and (45.e) appears in (46.b) and the
entailment relation is shown in (46.c).
(46) a. 3x [~(like'(jo'))(x)] <-> ~(Vx
b. Vx [~(like'(jo'))(x)] ^ ~(Bx
c. Vx [~(like'(jo'))(x)] -> ~(Vx
Again the proof that the formulae in (46) are tautologies can be made by showing that
they can never be false. As an example of the reasoning, let us take the formula in
(46.a) and assume it is false. For this to be so, either 3x [~(like'(jo'))(x)] is true and
~(VJC [(like'(jo'))(x)]) is false, or vice versa. Let us take the first possibility. For 3x
[~(like'(jo'))(x)] to be true, there must be some value assignment g' that satisfies
[~(like'(jo'))(x)]. By the semantics of negation, such a value assignment fails to
satisfy [(like '(jo '))(x)J. Hence, 3x [-(like '(jo '))(x)J is true only if there is some value
assigned to x that makes [(like '(jo f))(x)] false. For ~(\/x [(like '(jo '))(x)]) to be false,
on the other hand, VJC [(like '(jo f))(x)] must be true, which is so only if every value
assigned to x satisfies the propositional function [(like'(jo'))(x)]. However, we have
already seen that for the left hand side of the equivalence to be true there must be at
least one value of x that fails to satisfy this open formula, so that we arrive at a
contradiction: that there is and is not some value assignment that falsifies
[(like'(jo'))(x)], which shows that our first assumption is false.
Because there are two possibilities that guarantee the falsity of the equivalence
in (46.a), we have not yet shown that it can never be false. To complete the proof, we
212
Logical deduction
must show that the reverse situation, where 3x [-(like '(jo '))(x)J is false and ~(VJC
[(like '(jo *))(x)]) is true, also cannot occur. For the latter formula to be true, it must not
be the case that every value assigned to x satisfies (like '(jo '))(x), i.e. that there is some
value assignment that makes this propositional function false. For 3x [-(like '(jo '))(x)J
to be false, however, it must be the case that no value assigned to x satisfies
(like'(joy))(x). Again we arrive at the same contradiction as before: that there is some
value for x that does and does not satisfy the propositional function. Hence, we have
shown that neither of the component formulae in (46.a) can be true while the other is
false and so have confirmed the validity of the paraphrase.
Exercise 7.3:
Show that the formulae in (46.b) and (46.c) are tautologies by showing that the
assumption that they are false leads to a contradiction.
There is one final important point that needs to be made here with respect to the
entailments of quantified formulae. This concerns the difference between the universal
and existential quantifiers in terms of the existence in a model of entities that satisfy
the propositional functions that they combine with. As required by the truth-
conditions for 3 given in Chapter 6, existentially quantified formulae that are true
entail the existence in a model of at least one entity that has the property described,
and thus that something actually exists in the model. For example, if the formula
translating A lecturer screamed is true with respect to some model, then so are those
translating Some entity screamed, There was a lecturer and Some entity existed', i.e.
given the truth of (47.a) the formulae in (47.b) to (47.d) are also true.
(47) a. 3x [lecturer'(x) & scream'(x)].
b. 3x [scream'(x)].
c. 3x [lecturer'(x)].
d. 3x [exist'(x)].
The truth of (47.d) follows under the assumption that the extension of the
predicate exist' is always the set of all entities that exist in the model, i.e. A. If there
is some entity that satisfies any of (47.a) to (47.c), then this must be a member of A
and so must exist. The truth of (47 .b) and (47.c), on the other hand, follows from the
truth of (47.a) by a generalised version of the rule of conjunction elimination given
above. That this inference is valid is shown in (48) which uses a form of inference
sometimes called existential elimination. Because (48.a) is true (by assumption),
there must be some value assignment, g, that satisfies lecturer'(x) & scream'(x) This
allows us to infer the truth of a formula where x is replaced by an individual constant
that denotes whatever entity is assigned to x by g. We do not necessarily know the
identity of this individual, but just that there is some individual that screamed. Thus,
in (48.b) x is replaced by an arbitrary individual constant, symbolised as c, but not by
an actual individual constant like jo' or bertie'. From this step, the rest of the
deduction follows straightforwardly. Note that care must be taken in using this type
of deduction that inferences are not made about specific individuals. The extension of
the arbitrary constant c is not known, and so inferences with respect to particular
213
7 Inference
individuals are not licensed. Hence, we are not permitted to infer the truth of, say,
Fiona screamed from A lecturer screamed, only that someone or other screamed.
(48) a. 3x [lecturer'(x) & scream'(x)].
b. lecturer'(c) & scream'(c). existential elimination
c. lecturer'(c). conjunction elimination from b
d. 3x [lecturer'(x)]. existential generalisation from c
No existential commitment follows from the truth of a universally quantified
formula, however. It is not possible to infer from the truth of the formula translating
a sentence like Every lecturer screams either that something screams or that
something is a lecturer or indeed that anything exists at all. Thus, the truth of (49.a)
does not guarantee those of (49.b) to (49.d).
(49) a. Vx [lecturer'(x) -» scream'(x)].
b. 3x [lecturer'(x)].
c. 3x [scream' (x)].
d. 3x [exist' (x)].
The reason that these inferences cannot be made has to do with the implicational
nature of (49.a). Since an implication is true whenever its antecedent is false, the
formula in (49.a) is true if no value for x satisfies lecturer'(x), i.e. if there is nothing
in the extension of lecturer' in some model. Hence, we cannot infer (49.b), as it
cannot be guaranteed that there are any entities in the extension of lecturer'. Nor can
we infer (49.c), as (49.a) is true if both (49.c) and (49.b) are false, i.e. if there are no
lecturers and no screamers in the model. Furthermore, since there may be no entity
that satisfies lecturer '(x) or scream '(x), we cannot even infer that any entities exist at
all and so (49.d) also does not follow from the truth of (49.a). This has the
consequence that all the sentences in (50) express true propositions in our world (in
1991).
(50) a. Every blue sheep lives in the Highlands.
b. All pink elephants have green trunks.
c. Every Green Party member of the British parliament supports nuclear
power.
Although this may seem peculiar, it presents no very serious problems for a
theory that allows pragmatics to explain why sentences like (50.a) and (50.b) are
unlikely to be uttered as serious assertions outside of stories and jokes. More
importantly, we can use the lack of existential commitment in universally quantified
formulae to make intuitively valid inferences. Consider, for example, a situation in
which the sentence in (50.c) is uttered to someone in 1992 who has the background
knowledge that Green Party politicians do not support nuclear power. Assuming the
rough translations of (50.c) and the background knowledge as in (51.a) and (51.b),
where gp' stands for the translation of is a member of the Green Party, mp' that of is
a member of the British parliament and s' of supports nuclear power, the hearer can
make the inference in (51.c), that Green Party members of the British parliament do
and do not support nuclear power. Since the latter formula is contradictory (always
false whatever value is assigned to x), the hearer applies modus tollens to the
implication in (5 La) to derive (51.d), the conclusion that there are currently no Green
214
Lexical meaning
members of parliament. Hence, even the apparently bizarre lack of existential

commitment of universally quantified formulae helps to explain certain intuitively
valid inferences.
(51) a. Vx [(gp'(x) & mp'(x)) —> s'(x)]. premiss
b. Vx [gp'(x) -» ~(s'(x))]. premiss
c. Vx [(gp'(x) & mp'(x)) —> (s'(x) & ~s'(x))]. conjunction introduction
d. Vx [~(gp'(x) & mp'(x))]. modus tollens
7 3 Lexical meaning
The inferences discussed in Section 7.2 are explained with reference to the
interpretations given to logical operators, but there are valid inferences that are not
licensed by the model theory. For example, from the truth of (52.a) one can infer the
truth of the propositions expressed by each of the sentences in (52.b) to (52.d). Yet
there is nothing in the translations (and interpretations) of these sentences (shown in
(53)) that directly guarantees that these inferences hold in any model.
(52) a. Chester was killed by the poisoned cake.
b. Chester is dead.
c. Chester is not alive.
d. The poisoned cake caused Chester's death.
(53)a. 3x [Vy[(cake'(y) & poisoned'(y)) <-> x = y] & (kill'(chester'))«].
b. dead'(chester').
c. -(alive'(chester')).
d. 3x [Vy [(cake'(y) & poisoned'(y)) <-> x = y] & (cause'(die'(Chester')))(x)].
The inferences in (52) all depend, of course, on words in the sentences, not on
the logical structure of the sentences themselves. It is thus a property of the particular
English word kill that its extension relates to the extension of the English phrase cause
to die or that dead is the complementary of alive. Such relationships are not purely
logical in the way that, for example, no is logically related to not all. There is nothing
in the interpretation of the predicate kill' itself that requires all those things that are
killed to be in the extension of the predicate dead' in the same model, or that requires
nothing that is dead to be in the extension of the predicate alive'. Moreover, as we saw
in (5) above, inferences derived from lexical meaning are weaker than those derived
from the interpretations of logical operators, since the former can be suppressed to
give rise to non-literal interpretations, while the latter cannot without incoherence.
Content words are much more likely to undergo semantic change over time than
grammatical words. For example, the word selyin Middle English which gives us silly
in Modern English did not mean foolish or feeble minded, but rather pitiable or
deserving of compassion, a meaning still retained in parts of Scotland. The ancestor
of the Modern English word girl used to denote any human youth, but now, of course,
this has been restricted to female humans. Examples such as these are commonplace
in all languages whose earlier history is known.
Moreover, it is a commonplace observation that the vocabularies of different
natural languages divide the world in different ways. This means that the inferences
that can be derived from particular words in one language may be different from those
215
7 Inference
derived from a near equivalent in another. For example, the Modern German word
fahren implies motion by conveyance other than by foot, which is not the case for the
nearest equivalent in Modern English, i.e. the verb go. On the other hand, the nearest
equivalent to the English word granddaughter in Lakhota, an American Indian
language, is thakdza which has no implication of female gender and whose translation
into English would be grandchild. Thus, the German sentence in (54.a) implies that
in (54.b) while their English translations are not so related. On the other hand, the
English sentence in (54.c) implies that in (54.d) while the Lakhota sentence in (54.e)
has no such implication.
(54) a. Gestern ist Jo zum Kino gefahren.
yesterday is Jo to-the cinema gone
4
Jo went to the cinema yesterday.'
b. Jo ist zum Kino zu Fuss nicht gegangen.
Jo is to-the cinema on foot not gone
'Jo did not go to the cinema on foot/
c. My granddaughter is ill.
d. A female is ill.
e. mithakoza khuZe.
my-grandchild is-sick
'My grandchild is sick.'
Inferences derived from lexical meaning are thus here interpreted as language
particular and subject to semantic change (although there are those who would not
adopt this position). The model-theoretic interpretation of logical expressions,
however, captures general aspects of the meaning of grammatical/logical expressions
which are, by hypothesis, unchanging and constant across all natural languages all of
which have words, morphemes or other ways of indicating these fundamental
semantic concepts. There appears thus to be a considerable difference between the
meaning relations that result from the meaning of content words and those that result
from the interpretation of the logical operators. The latter have an invariant
interpretation that guides the interpretation of the whole system and thus have their
meaning encoded within the model theory itself. Lexical meaning, on the other hand,
being language particular, subject to change and weaker than logical meaning, must
be specified in some other way, outside the model theory.
Lexical implications of the sort noted above result from the sense of a lexeme
rather than its denotation. The denotation of a lexeme is its relation to things external
to the linguistic system. In the theory presented in earlier chapters, it is this aspect of
lexical meaning that is paramount and captured by associating each lexeme with the
entity, set of entities, etc., of which it is correct to predicate the lexeme. This is the
extension of the lexeme, which has already been contrasted with its intension and
which will be discussed further in Chapters 9 and 10. The sense of a lexeme, on the
other hand, has a number of different interpretations within the philosophical and
linguistic literature on linguistic meaning. Some philosophers, like Montague
following Frege and Carnap, interpret sense in the way that intension will be
interpreted in later chapters, as a logical function that determines the extension of an
expression (basic or derived) in different situations (see Chapters 9 and 10). Many
linguists, on the other hand, interpret sense in a relational way as concerned only with
216
Lexical meaning
the set of meaning relations lexemes bear to other lexemes (or expressions) within a
particular language. There seems some benefit in maintaining a distinction between
these two notions (see Chapter 10) and so it is the latter interpretation of sense that is
used here. In other words, the notion of sense is to be defined in terms of the different
of sense relations that hold between lexemes in a language.
Many different types of sense relations have been recognised between the words
and expressions in a language, some of which are central to the linguist's notion of
sense. One important sense relation is that of hyponymy where the sense of one
lexeme, the hyponym, is included in that of another, the superordinate, and the
extension of the former includes that of the latter (e.g. rose and flower). The relation
of synonymy is defined where two lexemes have the same sense and extension, and
so may be defined as mutually hyponymous. Total synonymy is rare (an example
may be quicksilver and mercury, although the former term is now archaic), but partial
synonymy occurs relatively frequently where two lexemes have almost the same sense
and almost the same extension (e.g. pullover and jumper or sweater in British
English). On the other hand, two lexemes are described as opposites if their
extensions are required to be distinct and their senses are contradictory with respect
to each other. There are a number of specific types of opposites. Antonymy (in the
restricted sense of this term) covers gradable opposites like hot/cold. Gradable
antonyms form instances of contraries (see below) and implicitly or explicitly invoke
a field over which the grading takes place, i.e. a standard of comparison. So, for
example, the adjective tall changes its extension according to the properties of the
entities of which it is predicated (e.g. tall gerbil vs. tall giraffe). Complementaries
(or binary antonyms) are all non-gradable opposites which, from the extensional
point of view, divide the universe of discourse into two disjoint sets (e.g. dead vs.
alive and abstract vs. concrete). The last opposite relation is that of the contraries
which involve the general opposition sense relation but without the negation of one
of the lexemes implying the truth of the other (thus excluding contradictories). We
can use this term to refer to non-binary contrasts found among co-hyponyms like
peony, rose, tulip, etc or ranks (e.g. Monday, Tuesday, Wednesday, etc.).
Finally, there are three more general sense relations that involve more complex
properties of the arguments of predicates. The first is that of the selectional
restrictions that predicates impose on their arguments. For example, the verb kick
requires an animate (and legged) subject and a concrete direct object while the verb
solve requires an animate and usually human subject and an abstract object (e.g. The
child solved the problem versus IThe rock solved the problem versus IThe student
solved the chair). Next, we have converses which are relational terms where the
relation denoted by one lexeme is reversed to give the relation denoted by another
(and vice versa). These often get called antonyms but they do not conform to the
general definition given above, since there need not be an oppositeness relation
involved. For example, sell is the converse of buy, but Bertie bought some gin from
Ethel does not imply that Bertie didn 't sell some gin to Ethel, though it does imply that
Ethel sold some gin to Bertie. Finally for our purposes, the meaning of lexemes may
be related by decomposition to the meanings of expressions involving fundamental
notions like causality (e.g. if x kills y then x causes y to die) or change of state (e.g.
soften and cause to become soft).
To define the sense of a word, it is sufficient to define the sense relations that it
217
7 Inference
bears to other expressions in the language, i.e. identifying its homonyms, hyponyms,
superordinates and opposites as well as any other selectional or decompositional
properties it may have. The important question is thus how to capture the sense
relations within truth-conditional semantics. The theory of semantics that has been
adopted here gives primacy to sentence meaning over lexical meaning, or, more
precisely, to the interpretation of the formulae that translate sentences rather than to
the interpretation of constants. Until now, the semantic contribution of words has been
restricted to the contribution they make to the truth-conditions of every sentence in
which they can appear. Since content lexemes are translated (and interpreted) as
constants in the logical representation of a sentence, this makes their extension their
primary contribution to the meaning of a formula. Given that, in truth-conditional
semantics, the central aspect of the meaning of an expression is taken to be its
denotation, one may think of sense relations as specifying relations between the
extensions of different lexemes and expressions. For example, tulip and rose are both
hyponyms of flower which means that every tulip and every rose is a flower or, more
formally, that the extensions of the predicates tulip' and rose' are subsets of the
extension offlower'. Furthermore, rose and tulip are opposites (incompatible co-
hyponyms) and so their extensions are disjoint. Thus, the sense of a lexeme may be
defined as the constraints it imposes on the extensions of other expressions with
respect to its own extension. The meaning of a lexeme in truth-conditional semantics
can thus be defined in terms of both its extension and its sense relations, as in (55).
(55) Lexical meaning: The meaning of a lexeme is the contribution it makes to
the truth or falsity of the proposition expressed by any sentence in which it
can appear together with the constraints it imposes on the extensions of
other expressions in those sentences.
Defining sense relations in terms of constraints on the denotations of expressions
allows them to be given a representation in LQ and thus a transparent interpretation
within the theory. It is common within lexical semantics to identify the sense relations
of particular lexemes by inserting a lexeme into a particular sentence frame and seeing
whether the sentence so formed appears to be tautologous. For example, to test for
hyponymy, the hyponymous lexeme is substituted for X and its superordinate for Y
in the sentence frame Every X is a Y. Thus, the tautologous feel of Every tulip is a
flower indicates the hyponymy between tulip andflower whereas the sentence No tulip
is a rose indicates the incompatibility between tulip and rose and so on. Such
sentences have a simple and direct translation into formulae in LQ as in (56) where
(56.a) requires the set of all tulips to be a subset of the set of all flowers and (56.b)
requires the set of tulips to have a null intersection with the set of roses.
(56) a. Vx [tulip' (x) -> flower' (x)].
b. Vx [tulip'(x) -> ~rose'(x)].
Formulae like those in (56) are called meaning postulates and have the form of
statements in the logical metalanguage that specifically relate formulae containing
different logical constants which translate particular lexemes and other expressions.
Meaning postulates all have the form of an implication where the antecedent consists
of a formula constructed from a single constant whose argument places are bound by
universal quantifiers. The consequent of the implicational statement is a more or less
218
Lexical meaning
complex formula consisting of related constants at least one of whose argument places
contains a bound variable shared with the antecedent and whose other argument places
are either existentially or universally bound. For example, the meaning postulates in
(57) capture the relations between the sentences in (52). (57.a) relates the extension
of kill' to that of cause to die; (57.b) allows the inference from Someone caused
something to occur to the fact that something did occur; (57.c) relates dying and
becoming dead and (57.d) relates the extensions of dead and alive. The steps used to
make the inferences in (52) using these meaning postulates are shown in (58). The
latter inference pattern assumes appropriate interpretations for tense, aspect and the
(actually problematic) expressions become' and cause'. Indeed, many decompositional
meaning postulates like those in (57.a) to (57.c) need to make reference to tense,
aspect and modality, matters that are the topics of the final chapters of the book.
(57) a. Vx [Vy [kill'(y)(x) -> (cause'(die'(y)))(x)]].
b. VP [Vx [Vy [(cause' (P(x)))(y) -> P(x)]]].
c. Vx [die'(x) —» become' (dead'(x))].
d. Vx [dead'(x) —> -alive' (x)].
(58) a. Bx [Vy [(cake'(y) & poisoned'(y)) <-> x = y] & (kill'(Chester'))(x)].
b. 3x [(kill'(Chester'))(x)]. conjunction elimination
c. 3x [(cause'(die'(Chester')))(x)]. modus ponens from b & (57.a)
d. die'(Chester'). modus ponens from c & (57.b)
e. become'(dead'(Chester')). modus ponens from d & (57.c)
f. become'(-alive'(Chester')). modus ponens from e & (.d)
Using meaning postulates, we can define the sense relations in terms of how they
relate the extensional denotations of lexemes and other expressions in the language.
Hyponymy, for example, is defined between two lexemes if there is a meaning
postulate connecting the constants that they translate into which has the structure
specified in (59.a), i.e. as an implicational statement with the argument place
universally quantified, as the examples in (59.b) to (59.d) show.
(59) a. Hyponymy: X is a hyponym of Y iff there is a meaning postulate relating
X' and Y' of the form: Vx [X'(x) -» Y'(x)] (the extension of X is a subset
of that of Y).
b. Vx [dog' (x) -» mammal' (x)].
c. Vx [mammal'(x) —> animal'(x)].
d. Vx [terrier'(x) -^ dog'(x)].
Because of the fact that two formulae p —» q and q —> r logically entail p —» r (as
readers can check for themselves), it follows that any sense relations defined for the
consequent are also defined for the antecedent. We can, therefore, construct further
meaning postulates for hyponyms from meaning postulates associated with their
superordinates like those in (60) which are derived from those in (59). If we then
define the sense of a lexeme to be the set of meaning postulates that have the constant
translating the lexeme in the antecedent of the implicational formula, the sense of a
superordinate is contained in the sense of the hyponym. Notice also that by this
definition, it turns out that a hyponym, X, with superordinate, Y, is also a hyponym
of all superordinates of Y.
219
7 Inference
(60) a. Vx [dog'(x) -> animal'(x)].

b. Vx [terrier' (x) —> mammal'(x)].
c. Vx [terrier' (x) —» animal' (x)].
Synonymy, as we have seen, is defined where two lexemes are mutually
hyponymous, in other words where their extensions (and senses) are required to be
identical, as shown in (61).
(61) a. Synonymy: X is a synonym of Y iff there is a meaning postulate relating
X' and Y' of the form: Vx [X'(x) <-> Y'(x)] (X and Y have identical
extensions),
b. Vx [pullover'(x) <-» sweater' (x)].
Unfortunately this definition does not allow for partial synonymy, since extensions
in the theory are discrete, i.e. there are no fuzzy edges where, for example, there may
be things that are almost pullovers but that are not sweaters. Since, as we have noted,
full synonymy is rarely, if ever, attested, this makes the definition in (61) less useful
than it might be. It is possible that a pragmatic explanation may be found for the
interpretation of synonymy as partial rather than full, but in the meantime, the effect
of partial synonymy may be incorporated by replacing the universal quantifier in (61)
by the quantifier most, so that we interpret synonymy as Most X are Fand Most Y are
X.
The definition of the general property of oppositeness has the form of the
meaning postulate in (56.b). Because p —> ~q has the same truth-conditions as q —> ~p
through modus tollens, we can infer from meaning postulates like that in (62.b) that
if the positive counterpart of the consequent is true then the antecedent cannot be.
Hence, from (62.b) we can infer a meaning postulate to the effect that no dogs are
cats, i.e. VJC [dog'(x) -> ~cat'(x)].
(62) a. Opposites: X is an opposite of Y iff there is a meaning postulate relating
X' and Y' of the form: Vx [X'(x) -> ~Y'(x)] (the extension of X is distinct
from that of Y).
b. Vx [cat'(x) -> ~dog'(x)].
All opposites must conform to the general definition in (62.a) but the different
types may distinguished by imposing further restrictions. For example,
complementaries are opposites where the negation of one also implies the positive of
the other, as defined and exemplified in (63).
(63) a. Complementaries: X is the complementary of Y iff there is a meaning
postulate relating X' and Y' of the form:
Vx [(X'(x) -> ~Y'(x)) & (~X'(x) -» Y'(x))] (the extensions of X and Y are
distinct and the complement of the extension of X is equal to the extension
ofY).
b. Vx [-concrete'(x) -» abstract' (x)].
c. Vx [-alive'(x) -» dead'(x)].
220
Lexical meaning
Again, we can infer that if X is the complementary of Y then Y is the complementary

of X. This follows from (63.a) by modus tollens, since ~Y'(x) implies —X'(x), i.e.
X'(x). This gives rise to the postulates in (64) paralleling those in (63.b) and (63.c).
(64) a. Vx [-abstract' (x) -> concrete' (x)].

b. Vx [~dead'(x) -> alive'(x)].
Antonyms are opposites restricted to some particular domain. For example, a big
mouse is not small for a mouse although it may be small for an elephant. Letting P
stand for some property, like being a mouse or an elephant, we can define antonymy
as in (65). Strictly speaking such meaning postulates go beyond the extensional
semantics so far developed, but they can be modelled in the semantic theory put
forward in Chapter 10 and they are included here for the sake of completeness. The
intended interpretation of (big '(P))(x) is that x is big for a P or x is a big P, whatever
constant P may be. (The meaning postulate in (65.c) results from (65.b) by modus
tollens.)
(65) a. Antonyms: X and Y are antonyms iff there is a meaning postulate relating
X' and Y' of the form: Vx [VP [(X'(P))(x) -> ~(Y'(P))(x)]] (the extensions
of X and Y are distinct for some given domain).
b. Vx [VP [(big'(P))(x) -> ~(small'(P))(x)]].
c. Vx [VP [(smaH'(P))(x) -> ~(big'(P))(x)]].
The definition of converses in (66.a) requires a schematic meaning postulate that
allows for predicates with different valencies. Many converses are also opposites as
illustrated in (67).
(66) a. Converses: X is the converse of Y iff there is a meaning postulate relating
X' and Y' of the form: Vx...Vy [X'(y)...(x) -> Y'(x)...(y)] (wherever
<ei,e2,...,en> is in the extension of X, <en,en.i,.-.,ei> is in the extension of Y).
b. Vx [Vy [Vz [(buy'(x)(y))(z) -> (sell'(z)(y))(x)]]].
c. Vx [Vy [Vz [(sell'(x)(y))(z) -> (buy'(z)(y))(x)]]].
(67) a. Vx [Vy [(left_of (x))(y) -> (right_of (y))(x)]]
b. Vx [Vy [(left.of (x))(y) -> ~(right_of «)(y)]]
c. Vx [Vy [(daughter'(x)Xy) -> (parent'(y))(x)]]
d. Vx [Vy [(daughter'(x))(y) -> -(parent'(x))(y)]]
Selectional restrictions and decomposition have similar definitions involving the
specification of further required properties of arguments. The latter differs in
containing information about causation, change of state and possibly other cognitively
important properties. An example of a simple selectional restrictions appears in (68.b)
and (68.c) and (68.d) show the decompositional properties of the verb melu where
melt" is the translation of the transitive verb and melf is that of the intransitive .
(68) a. Selectional restrictions/decomposition: A lexeme X imposes some
condition on the extension of one of its arguments.
b. Vx [sing'(x) -^ -abstract'(x)].
c. Vx [Vy [(melt"(y))(x) -K(cause'(melt'(y)))(x)]].
d. Vx [melt'(x) -»become'(-solid'(x))].
221
7 Inference
Exercise 7.4:
Write a set of meaning postulates that specify the sense relations that hold between
the sets of words that follow. Are there other sense relations that involve these words
and are all the sense relations noted as strong (or necessary) as each other?
i. run, move, walk, saunter, stroll.
ii. abstract, concrete, human, animate, inanimate, liquid, solid, think, flow,
interpret, crumble,
iii. horse, mare, filly, foal, offspring, child, mature, adult, human, animal,
mammal.
Meaning postulates are intended to impose constraints on the extensions of

certain expressions, and the sense relations defined above hold only where there is
such a constraint. It is not sufficient to define one lexeme a as the opposite of another,
p, merely because within some model Vx [ot'(x) —> ~|3'(x)] happens to be true. For
example, it is not the case that student is the opposite of dog, merely because in our
formal models no student is a dog (and vice versa). Nor is it the case that genius and
lecturer are synonyms merely because VJC [genius'(x) <-> lecturer*(x)] happens to be
true in M2. Rather, these lexemes are unrelated in terms of sense, because there are
no meaning postulates connecting them. The sense of a lexeme is defined exclusively
by the set of meaning postulates that specify the relationship of its extension with
those of other expressions within the language. This remains the case whatever the
truth of particular postulates within a particular model, but this brings up the problem
of the status of the meaning postulates with respect to different models or states of
affairs and to the theory as a whole.
There has been a tendency within the logical tradition, including Montague, to
interpret meaning postulates as constraints on admissible models. In other words,
meaning postulates restrict the set of all possible models (the denotation assignment
functions) which can be used to interpret a language with respect to the situation being
represented. Only those models are admissible in which the meaning postulates
associated with the set of constants are true. Under this interpretation, meaning
postulates are raised to the status of logical (or analytic) truths, since they are true of
all situations that can be described in some language. This means that inferences
derived from these have the same status as deductive inferences derived from the
interpretation of logical operators. As we have already seen, however, it is not always
the case that inferences derived from lexical meaning behave in the same way. It is
possible to suppress inferences derived from lexical meaning in a way that is not
possible with deductive inferences. When this happens, it gives rise either to a
figurative interpretation or to a generalisation of the applicability of a term. An
example of the latter appears in (69). The sentence in (69.a) is generally supposed to
capture the meaning of the noun bachelor, but the term may be applied to a woman,
as in (69.b), without causing the reader any difficulty in interpretation. This is so,
despite the fact that it seems to be equivalent to the less acceptable sentence in (69.c),
which states directly that the entity denoted by the subject noun phrase is both a
woman and a man (cf. the formula in (69.e)). The sentence in (69.b) would, in normal
222
Lexical meaning
circumstances, be interpreted as stating that the relevant woman is unmarried, i.e. the
maleness of bachelorhood is suppressed given the stated information that the subject
is a woman (and the complementary relation between man and woman with respect
to humans). The interpretation of (69.c), however, is less clear than this and could be
interpreted in a number of ways: that the person referred to behaves socially like an
unmarried man, that the person is a transvestite (i.e. looks like a woman) or that the
person is a transsexual (i.e. is genetically male). The interpretation of the logically
contradictory sentence in (69.d), on the other hand, which is derived from (69.c)
through the sense of man, is even less informative than the latter. It requires much
more contextual information for its interpretation and sounds much more bizarre in
isolation than (69.c) and has a very different status from (69.b). This is not explicable
if meaning postulates have the status of analytic truths, since all three example
sentences would be true in all the same circumstances and so convey the same
information.
(69) a. All bachelors are unmarried men.
b. That woman is a bachelor.
c. ?That woman is an unmarried man.
d. *That woman is not a woman and is unmarried.
e. 3y [woman'(y) & ~3x [(marry'(x))(y) & man'(y)]].
Instead of taking meaning postulates to be necessary conditions on models, we
can interpret them as independent statements (perhaps located in the lexicon) that
provide information about normal circumstances. Inferences made from them then
are assumed to hold where there is no information to the contrary. So, for example,
the statement in (69.a) about bachelor is assumed to be generally, rather than
universally, true. It holds unless there is a stronger reason to believe that part of the
information that can be deduced from the postulate cannot be true, because if it did it
would lead to a logical contradiction. In such a situation, only part of the information
encoded in the meaning postulate is inferred. It is as if the implicational connective
in formal meaning postulates were not interpreted exactly as -> but as a similar
connective, call it 1, for which the inference pattern structure in (70) is valid.
Applying this inference pattern to (69), we infer that Fiona is unmarried and the
information derived from (69.a) as to her being a man is dropped, because it is
explicitly stated that she is a woman.
(70) a. p i (q & r). MP
b. p. premiss
c. ~r. premiss
d. q. conclusion
The formal interpretation of a connective like 4- will not be further pursued as
there is much more to the retrieval of lexical inferences than just setting up meaning
postulates containing such an implicational connective. In particular, pragmatics
(mainly in the guise of cognitive relevance) must come into play in examples like
(69) to explain why, for example, it is a bachelor's maleness that is suppressed rather
than a woman's non-maleness (i.e. it is the inference derived from bachelor that is
altered, not that derived from woman). Furthermore, the notion of sense put forward
here allows a distinction to be made between sentences and situations in terms of how
223
7 Inference
few or how many lexical inferences need to be suppressed to avoid contradiction, but
this requires more semantic machinery than is yet available. (But see Chapters 9 and
10 for some further discussion of the matter.) For these reasons, a proper formalisation
of the way meaning postulates work will not be spelled out here. It is unfortunate that
while lexical semantics has, at least until very recently, been the primary focus of
linguistic semantics, a great deal of work remains to be done on this topic within
logical semantics. It does, however, seem a reasonable hypothesis that central aspects
of lexical meaning can be analysed in terms of the definition of sense relations as
constraints on the denotations of expressions within a particular language. It also
seems reasonable that such constraints cannot be stated as part of the model theory (as
conditions on admissible models), if their comparative weakness with respect to the
logical operators is to be captured. Thus, however the formal procedure for making
inferences from meaning postulates is to be set up, these two points must be
incorporated.
7.4 Non-truth-conditional aspects of the connectives

In this last section, we take a look at some aspects of the meaning of sentences
containing the conjunctions and, or and if ..then that are not explained by the
truth-conditional interpretation of their logical counterparts, &, v and —». This
discussion is by no means exhaustive and there are aspects of the meanings of the
connectives that are not discussed. The point of this section, however, is to indicate
how inference patterns can be constructed to account for certain aspects of meaning
not determined by truth-conditional interpretation by considering language use. The
success of this enterprise provides support for the hypothesis that truth-conditional
semantics can provide a sufficient basis for a general theory of meaning, as long as it
is supplemented by a rich enough theory of pragmatics.
7.4.1 And
First, let us look at the semantics of sentences containing and. As discussed in
Chapter 3, the truth-conditions of a formula containing the connective & are met if,
and only if, both conjunct formulae are true. This holds irrespective of the order in
which the conjunct formulae appear, because & is a commutative operator, as we have
already seen. The sentence in (71.a) is thus truth-conditionally equivalent to that in
(71.b).
(71) a. The skier fell over and broke her ankle,
b. The skier broke her ankle and fell over.
Most speakers of English, however, would interpret an utterance of (71 .a) as meaning
something different from an utterance of (71.b). This is because there is a tendency
to assume that conjuncts are causally or temporally related, if the events described
are such that they can be so related under normal assumptions. In (71 .a), for example,
the sentence The skier fell over comes before (she) broke her ankle and so the most
likely interpretation is that the falling preceded, and possibly caused, the breaking of
the ankle. This temporal/causal relation is reversed in (71.b) where the order of the
constituent clauses are reversed: the skier's breaking her leg precedes, and possibly
224
Non-truth-conditional aspects of the connectives
causes, her falling over. This difference between (71.a) and (71.b) could be taken
as an indication of the incorrectness of the semantic interpretation provided for the
conjunction and by the truth table for &. To establish whether or not this is so, we
have to ask ourselves a few questions: is the temporal/causal relation between two
conjoined clauses found between clauses in other contexts? If it is, can a unified
treatment of the two constructions be proposed without invoking the semantics of
the conjunction? We must also ask whether the causal/temporal implication always
holds of conjoined clauses and finally whether it is possible for two conjoined
clauses to translate into a true formula if the truth-conditions provided for and are
not met.
The answer to the first question is quite straightforward: exactly the same
temporal/causal effect is found when two or more sentences are strung together in
a text without any overt co-ordinating morpheme. Thus, the simple text in (72.a)
appears to be more coherent than that in (72.b) where the order of the sentences is
changed, despite the fact that they could both be said to describe the same situation.
(72) a. Fiona gave the cake to Bertie. Bertie gave the cake to the dog. The dog ate
the cake. The dog died. Bertie laughed.
b. Bertie laughed. The dog died. Fiona gave the cake to Bertie. The dog ate the
cake. Bertie gave the cake to the dog.
The peculiarity of the text in (72.b) may be, at least in part, ascribed to the fact that
texts of this sort are taken to describe a series of events occurring in the order in which
the sentences appear. Imposing such a temporal order on (72.b), however, makes it
difficult to construct a sensible causal chain out of the events and so it appears
incoherent. Temporal and causal order is thus imposed on a series of non-conjoined
sentences forming a text as well as on conjoined sentences. This strongly suggests
that this aspect of the meaning of conjunctions has nothing to do with the meaning
of the word and but with the way texts are constructed. This hypothesis is confirmed
by the fact that situations can be presented (and often are, in practice) with later
events being described before earlier ones.
Furthermore, the causal/temporal implications of and are not always in evidence.
For example, sentences like The skier broke her ankle and the instructor cracked his
ribs and Jo's aunt is insane and his grandfather was in jail for many years do not
imply that the situations described by each of the conjuncts is connected causally or
temporally. On the other hand, it is never possible for a conjoined sentence to
describe a true state of affairs when one of its component sentences describes
something actually false. So for example, it is perfectly possible for (71 .a) and (71 .b)
to describe the same temporal state of affairs, but it is not possible for such a
statement to be true, if, in fact, the skier did not break her ankle. Thus, the
temporal/causal effects associated with conjunction can be cancelled or weakened and
appear in other textual situations which do not contain and, whereas the truth of each
conjunct cannot be denied without denying the truth of the whole sentence. This is
evidence, therefore, that the truth-conditions for & provide the central core of the
meaning of and while temporal and causal effects are secondary and can be explained
by a theory of language use.
225
7 Inference
7.42 Or
A similar argument may be applied to one of the apparent oddities in the truth-
conditions associated with or via its translation as logical disjunction, as discussed
in Chapter 1. Because of the truth-conditions of v, a disjunctive formula may be true
if both of the disjuncts are true. However, in English, the use of a disjunctive sentence
often implies that only one of the disjuncts is true. It would, for example, be
extremely odd for someone to utter the sentences in (73) knowing that both disjuncts
are true.
(73) a. Fiona was a Professor of Linguistics or Dean of the Arts Faculty,
b. Bertie's jacket was in the bedroom or in the cloakroom.
Although the truth-conditional interpretation of or allows both sentences in (73)
to be true of situations in which Fiona was a professorial Dean of the Arts Faculty and
Bertie's jacket was both in the bedroom and in the cloak-room, these statements would
not usually be given such an interpretation. Very often disjunctive sentences have an
exclusive meaning, i.e. one which disallows situations where both disjuncts are true.
Hence, the sentence in (73.a) would normally be interpreted so that Fiona was either
a Professor of Linguistics or Dean of the Arts Faculty, but not both. To capture this
intuition, or could be interpreted, not as inclusive disjunction, but as exclusive
disjunction, ve,which has the truth-conditions in (74) and truth table in (75), repeated
from Chapter 3.
(74) Truth-conditions for exclusive disjunction: If <|) and \\f are formulae then
[0 v e \|/]M-s is 1 iff [()) & \j/]M'8 is 0 or [~ty & ~y] M * is 0.
(75)
p q PVe^
t t f
t f t
f t t
f f f
The problem with adopting exclusive disjunction as the truth-conditional meaning of

or is that while many uses of disjunction do imply that only one of the disjuncts
holds, there are some occasions where we do not want to preclude a situation in
which both component formulae are true. Consider, for example, the sentences in
(77) which we do not want to be falsified by situations where both disjuncts are true,
i.e. where there are students who have a private income and rich parents, or where
a job applicant has two Higher and five 'O' grade passes, or where the lecturer does
not want to rule out the possibility of a zealous student reading both articles.
(76) a. Students who have a private income or rich parents are better off at
University than those who do not.
b. Applicants for the job must have two Higher, or five Ordinary, grade passes.
c. You must read the article on presupposition or the article on implicature by
next Wednesday.
226
The sentences in (76) (and many others like them) show that exclusive
disjunction on its own makes too strong a claim about the truth-conditions of English
or, since inclusive readings of disjunctive sentences must be allowed for. This could
be taken as an indication that or is ambiguous between v and ve. While this is not an
unreasonable suggestion, it is probably incorrect. Gazdar (1979) presents logical
arguments against the ambiguity theory by showing that the use of exclusive
disjunction leads to bizarre logical consequences in negative sentences. For
example, under the ambiguity hypothesis the sentence in (77.a) has a propositional
structure that may be represented as either (77.b) or (77.c).
(77) a. Edinburgh isn't boring or provincial.
b. ~(pvq).
c. ~(p ve q).
The problem with (77.c) is that it can be shown to be truth-conditionally
equivalent to the formula in (78.a) which, by the definition of paraphrase in (14),
entails that (77.a) is a paraphrase of (78.b). Intuitively, however, these two sentences
are not paraphrases, thus indicating that the exclusive reading of (77 .a) is not
acceptable. Since there appears to be no reason why the ambiguity should fail in this
example, the implication is that or is never ambiguous.
(78) a. ((~p&~q)v(p&q)).
b. Either Edinburgh isn't boring and isn't provincial, or it's both.
A less technical way of showing that the ambiguity theory is not the best account
of the semantics of English or is to show, as we did with and, that the apparent
differences can be traced to other factors. Because of sentences like those in (76), we
cannot take exclusive disjunction as basic because otherwise the formulae that
translate such examples would be false where both disjuncts are true; an undesirable
consequence, as we have seen. If we assume that inclusive disjunction is basic,
however, we can show that the exclusivity effects result from non-truth-conditional
aspects of meaning. There are two major sources for exclusive readings of
disjunctions. In one type of disjunction, exemplified in (73.b), the disjunction is
interpreted exclusively because of our knowledge of the way the (real) world is.
Knowing that a jacket is an entity and rooms are locations, we are forced to give the
sentence an exclusive reading, otherwise it could be the case that the jacket in
question was in two different locations at once, something that our knowledge of the
world tells us is impossible. A sentence like Bertie ys jacket is in the bedroom or on the
floor, however, can be given an inclusive reading, because the properties of being on
the floor and being in a bedroom are not mutually exclusive. Hence, there are no real
world constraints on interpreting the disjunction inclusively and the exclusive reading
is not forced.
The exclusive reading of the sentence in (73.a), however, cannot be accounted
for in precisely the same way, since it is perfectly possible for someone to be a
Professor of Linguistics and a Dean of a Faculty at the same time. To account for
the preferred reading here, we must turn, as we did with and, to a theory of language
use. As discussed in the introduction, Grice (1975) suggests that there are a
number of rules that speakers adhere to when engaging in conversation. In
conversations, participants are assumed to be being co-operative and are sticking to
227
7 Inference
certain principles governing conversational behaviour. One of the maxims can be

used to account for the oddity of someone uttering a sentence like (73.a), when, in
fact, they know perfectly well that Fiona is both Professor of Linguistics and
Dean of the Faculty. The relevant maxim is repeated in (79):
(79) Maxim of quantity: Make your exchange as informative as is required.
Amongst other things, the maxim of quantity requires speakers to use a conjunction
if they believe that two sentences express true propositions or to state that two
disjoined sentences may both be true. In a situation in which speakers conceive
that it may be possible that both component formulae expressed in (73.a) may be true,
they ought to say something like Fiona is Professor of Linguistics or Dean of the
Faculty or both. By using disjunction, rather than conjunction, therefore, a
co-operative speaker implies that they do not think that Fiona is both Professor and
Dean. Hence, from the use of the or, there is an implicature that the speaker believes
that it is not the case that both disjuncts are true. This gives rise to the effect of
exclusive disjunction, without resorting to an interpretation of or as ve. Furthermore,
the implication of exclusivity can be cancelled, allowing (73.a) to be true even if both
disjuncts are true. After all, someone who uttered this sentence would be somewhat
aggrieved at being accused of lying if it so happened that Fiona was, in fact, both
Professor and Dean. This would not be explained if the exclusive reading of this
sentence was part of the truth-conditions of or and not an implicature derived from
conversational use. The English disjunction morpheme or should not, therefore, be
treated as a homonym, but non-ambiguously as inclusive disjunction. Its interpretation
in particular contexts may giveriseto exclusive interpretations by the meanings of its
component formulae and knowledge of the world or through the operation of
pragmatic principles like the maxim of quantity. A theory of language use again
can be shown to interact with basic principles of semantic interpretation to give the
desired results.
Before going on to look at conditionals, there is one entailment relation
involving disjunction that is problematic and that should be mentioned here. The
truth-conditions for v guarantee that a complex formula p v q is true if p is true,
irrespective of the truth value of the other component formula. This has the effect of
making the formula p -»(p v q) a tautology. According to the definition of entailment
in (9), this means that we predict that a true sentence Sj entails Sj or S2 where S2 is
any other sentence at all. In other words, any sentence that expresses a true proposition
entails a disjunctive sentence containing it. But this is bizarre, as the examples in (80)
show.
(80) a. Grass is green. Therefore, grass is green or violets are blue.
b. Chester ate the cake. Therefore, Chester ate the cake or Ethel was crazy.
The oddity of these examples, however, can again be explained with reference to
pragmatic factors. Indeed, we may again appeal to the maxim of quantity. Clearly, if
it is the truth of a single proposition that needs to be conveyed, then the maxim of
quantity requires the utterance only of the sentence that expresses that proposition.
Any addition, like an extra disjunction, is extra unnecessary information and so a
flouting of the conversational maxim.
228
7.4.3 If
There are a number of peculiar things about the interpretation of conditional
sentences as logical material implication. One of these is the peculiarity of asserting
conditional sentences when there appears to be no apparent causal relation between
the antecedent and the consequent. Compare, for example, the sentences in (81).
(81) a. If the cake was poisoned, then the dog will die.
b. If the cake was poisoned, then the dog will eat it.
c. If the cake was poisoned, then the dog won't be hungry.
d. If the cake was poisoned, then the dog was called Chester.
e. If the cake was poisoned, then the dog is a space alien from Alpha Centauri.
The acceptability of the sentences in (81) declines from (81 .a) to (81 .e) as it becomes
more difficult to connect the situation described by the antecedent with that described
by the consequent. This feeling of connectivity between the two sentences is much
stronger with conditionals than with conjunctions and is not so easily accounted for
(particularly as material implication is not commutative, i.e. p —> q is not truth-
conditionally equivalent to q —> p). However, we can again provide a pragmatic
explanation for this phenomenon using Gricean principles of conversation. The
explanation utilises the maxim of quantity in (79) and the maxim of quality, repeated
in (82).
(82 ) Maxim of quality: Do not say what you believe to be false and do not say
that for which you lack adequate evidence.
To see how these maxims help explain the gradation of acceptability in (81),
consider what one is trying to convey in asserting a conditional sentence. In general,
the utterance of a conditional sentence asserts that, as far as the speaker is aware, it
is not true that the antecedent is true and the consequent is false simultaneously, i.e.
the maxim of quality is adhered to by avoiding the assertion of something known to
be false. For listeners to believe such a conditional, they should be able to pick out
a connection between the two clauses that enables them to be certain that their
truth-conditions are met. If there is no such connection then they could believe the
formula expressed by the conditional only by having reasonable grounds for
believing that the antecedent is false or the consequent true. By the maxim of
quantity, however, the speaker is then revealed as being un-co-operative because he
or she should either have simply stated the consequent or denied the truth of the
antecedent. The conditional says too much, as it were. Thus, the assertion of a
conditional is only felicitous if there is some connection between the antecedent and
the consequent and if the speaker believes the conditional to be true.
The assertion of conditionals containing apparently unconnected component
sentences, however, can sometimes be used felicitously. The utterance of the sentence
in (81.e) may be felt to be more acceptable than that of the sentences in (81.c) and
(81.d). The reason for this can be found in the interaction of typical beliefs held by
interlocutors and modus tollens. From the truth-conditions of —>, this sentence is true
if (and only if) the dog is a space alien from Alpha Centauri or the cake wasn't
poisoned. Someone hearing an utterance of (81.e) is unlikely in ordinary
229
7 Inference
circumstances to entertain the possibility that there are dogs which are space aliens
from Alpha Centauri. If hearers have no other grounds for assuming that speakers are
flouting the maxim of quantity by uttering something they know is untrue, the former
must, by modus tollens, assume that the antecedent is false. Thus, anyone who utters
the sentence in (81 .e) is normally assumed to be conveying that they believe that the
cake is not poisonous. In a similar fashion, it is possible to convey that a consequent
really is true by having as an antecedent a sentence that expresses a well known
truth, since, for the conditional to be true, the consequent must also be true. For
example, the sentence If two plus two equals four, then my client is innocent strongly
conveys that my client is innocent (example taken from McCawley (1981: 220)).
Hence, Grice's conversational maxims, together with the truth-conditional
interpretation we have provided, can be used to account for a number of peculiarities
in the interpretation of conditionals.
That a pragmatic explanation of these facts is feasible is strengthened by the
fact that the same connectivity is felt between paraphrases of conditionals that use
a disjunction. A formula containing —> is truth-conditionally equivalent to a
disjunction where the antecedent formula is negated, as can be checked by showing
that (p -> q) <-» (~(p) v q) is a tautology. Hence, our theory predicts that the sentences
in (83) are paraphrases of those in (81).
(83) a. Either the cake wasn't poisoned or the dog will die.
b. Either the cake wasn't poisoned or the dog will eat it.
c. Either the cake wasn't poisoned or the dog won't be thirsty.
d. Either the cake wasn't poisoned or the dog is called Chester.
e. Either the cake wasn't poisoned or the dog is a space alien from Alpha
Centauri.
The acceptability of the sentences in (83) declines in precisely the same way as their
equivalents in (81). Hence, we can conclude that the relatedness between
antecedent and consequent in a conditional does not result from the semantics of if
and we are not in a position to reject the truth-conditional interpretation of natural
language conditionals as material implication.
Unfortunately, while a theory of language use can be used to account for some
of the vagaries of the use of material implication as the interpretation of English if...
then, it is hard to see how it can account for all of them. Apart from the existence of
counterfactual conditionals like If Chester had not eaten the poisoned cake, he
would have lived and If I were rich, I would spend every day gardening which require
something more than simple material implication for their interpretation, the very
equivalence between material implication and disjunction noted above argues against
the correctness of this analysis. Thus, while the sentences in (83) seem to be
reasonable paraphrases of those in (81), there are many pairs of sentences with the
same structure that are not paraphrases of each other. Consider, for example, the
sentences in (84) (taken from Gazdar (1979:83)) and (85) (taken from McCawley
(1981:222)). Neither of these pairs of sentences are paraphrases, despite the fact that
the use of -> to translate //"predicts that they are.
(84) a. If you have your ankle removed your foot will drop off.
b. Either your foot will drop off or you will not have your ankle removed.
230
Further reading
(85) a. If you come a step closer, I'll scream.

b. Either you won't come a step closer or I'll scream.
It is considerations like these, rather than the problem of conditionals with false
antecedents, that militate against the use of material implication to translate
conditional sentences in natural languages. Many suggestions have been put forward
for the analysis of such sentences and all go well beyond ascertaining the truth of
conditionals over a single state of affairs. That is to say, the truth of a conditional like
If it doesn't rain, then there is drought depends not just on whether it is not raining
and there is drought here and now, but on whether in all or the majority of cases in
which it doesn't rain, there is drought. Such approaches take us beyond the
extensional semantics that has been developed in Chapters 2 to 6 into the realm of
intensionality, the topic of the final chapters of this book. Conditionals, however, will
not be further discussed, as the topic is extremely controversial and requires more
space than can be given here. The reader is referred to the books and articles
mentioned in the next section for more information, but these should not be tackled
until the reader has discovered more about intensionality.
7.5 Further Reading

For a general discussion of inference, see Sperber and Wilson (1986: ch. 2). Logical
deduction figures largely in textbooks on logic and the rules mentioned in this chapter
(and many others) can be found by looking at any reasonable introductory book on
predicate and propositional logic. As before, the most accessible account is given in
Guttenplan (1986:124-165), which deals with the inferences that are generated by the
propositional operators, and pp. 245-276, which deals with the quantifiers. All wood,
Andersen and Dahl (1977: ch. 6) provides a concise summary, as does Partee, ter
Meulen and Wall (1990:115-136; 154-180), while Martin (1987: ch. 5) gives a rather
more formal account of logical deduction. For a detailed discussion of lexical meaning
in general linguistic semantics, see Cruse (1986) and also the discussions in Lyons
(1977: 250-335) and Lyons (1981: chs. 2 to 4). Pulman (1983) gives an interesting
discussion of the possibility of constructing a theory of lexical meaning, taking up the
challenge of the philosophers W. V. O. Quine (see Quine (I960)) and H. Putnam (see
Putnam (1975)). There is not a great deal of literature specifically about lexical
meaning within formal semantics: there is a slightly dated discussion in Kempson
(1977: ch. 6) and a more sophisticated one in Chierchia and McConnell-Ginet (1990:
ch. 8). Dowty (1979: 193-321) provides a lengthy discussion of the topic with
particular reference to grammatical aspect. Thematic (participant) roles are given an
analysis in terms of lexical entailments defined by meaning postulates in Dowty
(1989) and Dowty and Ladusaw (1988). See also the articles in Eikmeyer (1981) and
Ballmer and Pinkal (1983). The non-truth-functional aspects of the propositional
connectives have received a considerable amount of attention since Grice (1975;
1978): Levinson (1983: ch.3) discusses the topic, as do McCawley (1981: 215- 233)
and Chierchia and McConnell-Ginet (1990: 187-203). Gazdar (1979: chs. 3 & 4)
provides a formalisation of the notion of implicature and a discussion of the
truth-functional connectives. These problems have been re-addressed within the
literature on Relevance Theory: see, especially, Blakemore (1987; 1989) and, for a
231
7 Inference
critique of some aspects of the Gricean approach, Carston (1988). The literature on
conditionals is predictably large. Gazdar (1979: 83-87) provides semantic reasons for
rejecting material implication as a reasonable interpretation of the meaning of if
McCawley (1981: 296-326) discusses modalised implication and counterfactual
conditionals, the latter based on the seminal discussion in Lewis (1973). Van Benthem
(1986: ch. 4) provides an account of conditionals as generalised quantifiers, while the
papers in Traugott, ter Meulen, Reilly and Ferguson (1986) provide a good overview
of the main issues for formal and general linguistic semantics and language typology.
232
8 Time, Tense and Aspect
8.1 Temporal contingency

The grammar fragment developed in previous chapters generates sentences only in the
past tense form. This has been a matter of expediency to allow more natural English
to be used for the examples. However, the temporal properties of the sentences
implied by the use of this tense have, in fact, been completely absent from their
interpretation. The models we have been working with only contain a set of basic
entities, A, and a function F which assigns an extension to each lexeme in the
language. The denotations assigned by the latter are, however, static and no notion
of change or development is (or could be) incorporated. This means that really the
model theory treats all formulas as universal truths or universal falsehoods, as if
they were all of the same sort as sentences like e - me2, All humans are mortal, No
bachelors are unmarried, The square root of nine is seventeen, and so on. The
propositions expressed by such sentences have the same truth value at all times and
so may be thought of as 'timeless'. Most of the sentences generated by the grammar
fragment, however, translate into formulae that could vary in truth value according to
time and place, and other contextual factors. For example, A lecturer screamed may
be true if uttered today but false if uttered the day before yesterday. Such sentences
do not denote universal truths or falsehoods but contingent ones, ones whose truth
depends on what is happening or has happened at a particular time. Most, if not all,
languages have ways to refer to situations that are happening now or that have
happened in the past or that are going to happen in the future. Furthermore, many
languages can present situations as ongoing or complete (e.g. Fiona was singing a
song versus Fiona sang a song) and the duration or time of occurrence of events can
be specified directly within sentences (e.g. Fiona sang for five minutes yesterday
afternoon).
In this chapter, we are concerned with defining a theory that enables the
difference between timeless and temporally bound sentences to be captured. To do this
we will sketch a fairly simple theory of time and use this to provide an account of the
way English may refer to present, past or future events (tense) and how such events
are presented as ongoing or complete (aspect). The analysis concentrates on basic
tense and aspect, and no attempt is made to incorporate temporal adjuncts like the
adverbial phrases yesterday, on Tuesday, etc. or sentential adjuncts like when clauses.
8.2 Time
Before tense or aspect can be incorporated into the theory, the concept of time needs
to be introduced. According to one view, time consists wholly of a series of moments
which are all ordered with respect to each other so that each moment either precedes
or follows all the others. This captures the intuition that time is directional, irreversible
and uniform. This can be formally defined by a set, T, of moments of time and a
relation is earlier than or precedes symbolised as < which has the properties in (1).
233
Time, tense and aspect
(1) For all t, and tj e T:
a. ~(t,<t,).
b. ((t,<t J )&(t J <t k ))-»(t,<t k ).
c. (t i <t J )^~(t J <t l ).
d. (t, < tj) v (tj < t.) v (t, = tj).
The condition in (La) disallows any moment of time from preceding itself, thus
making < irreflexive; (l.b) imposes transitivity on the relation by requiring any
moment of time that precedes another to precede all other moments that the latter
precedes; and the third condition ensures that the relation < is asymmetric by
disallowing moments of time from both preceding each other. The final condition in
(1 .d) makes the relation dense by ensuring that there are no moments of time outside
the series T and that every moment precedes or follows every other moment except
itself. Given these conditions, one may think of T as the set of natural numbers and
< as the relation is less than. This concept of time may also be represented by the time
line in (2) where the line represents moments of time ordered from the past into the
future as indicated by the arrow and t is a particular moment on this series.
(2)
8.2.1 Intervals of time

Moments of time seem adequate for assessing the truth of formulae containing
expressions referring to instantaneous events like sneeze', but there are many activities
like writing a book or painting a picture, which occur over a period of time during
which it may not be true that for all moments in that period someone is actually in the
process of writing or painting. Thus, while it is true that I am currently writing this
textbook in August 1991, it is not true that I was actually writing three hours ago
(since I was still in bed) and yet one of the interpretations of the sentence Ronnie is
writing a book on semantics was as much true three hours ago as it is now.
Furthermore, events may be presented as occurring during the period when some other
event is occurring. For example, I may describe my current situation with the sentence
I am eating my breakfast while writing this sentence, presenting two different events
as occurring at the same time. To be able to account for situations such as book-
writing and other non-instantaneous events and to allow for events to be contained in
or overlap others, the truth of formulae may be assessed not at moments of time but
at intervals of time. Thus, the proposition expressed by Ronnie wrote a book is
assessed as being true if, over a period of time that has now finished, Ronnie was
writing a (particular) book and that book is now complete. This sentence does not,
however, require Ronnie to begin and complete the writing of a book at some
particular moment in time, nor that it be true that Ronnie wrote at every moment of
the time that he took to complete the book. In interval semantics, moments of time
are primitive, and intervals of time are defined in terms of these, as a continuous set
234
Time
of moments of time ordered by the precedence relation, <. The set of all possible
intervals of time is thus defined as in (3) as the set of all continuous subsets, i, of a set
of moments of time, T, ordered by the precedence relation, <
(3) I = {i c T I ((t, € i) & (tk e i) & (t, < tj < t*)) -> (tj e i)}.
Intervals of time may overlap with, be contained in, or precede, other intervals.
Two intervals i and j overlap (written ioj) if they share moments of time; one interval,
i, is contained in another, j , (written i c j ) if every moment in i is also in j ; and one
interval, i, precedes another, j (written i < j), if every moment in i precedes every
moment in j . These definitions are summed up in (4) and a schematic example
illustrating the concepts is given in (5). In the latter, the four lines under the time line
marked i, j , k and 1 are all intervals of time such that k overlaps with i and j , 1 is a
subinterval of (is contained in) j , and i and j are independent of each other, except
that i precedes j (and k precedes 1), as summarised in (6). (By the definition in (4.c)
i does not precede k and k does not precede j , because they overlap.)
(4) For all temporal intervals i and j in I:
a. i o j if, and only if, i n j * 0 .
b. i c j , if, and only if, i n j = j .
c. i < j , if, and only if, Vte i [Vt'ej [t < t']].
(5)
(6) a. i<j,k<l.
b. iokjoljok.
c. lcj.
For notational convenience, an interval of time is represented, not by using set
notation, but by enclosing the endpoints of the interval between square brackets.
Thus, the interval symbolised as [t^tg] is the set of all moments of time between t4 and
t8, i.e. {t4,t5,t6,t7,t8}. An interval of time that is no more than a moment is represented
within interval semantics as a unit set of times, i.e. as an interval consisting only of
a single moment of time, {t}, and is indicated by enclosing the moment of time
between square brackets. For example, [t7] is the interval of time consisting only of
the moment t7, i.e. {t7}. These definitions are given in (7).
(7) a. Interval of time: [t to tj =def {11 U < t < t 2 }.
b. Moment of time: [t] =ber {t}.
235
8.2.2 Temporal models

In order to incorporate time into our semantic theory, the models with respect to
which the truth of a formula must be assessed must be extended to the ordered
quadruple in (8).
(8) Temporal model: «A,F,T,<».
As before, A is a set of entities specifying the ontology of the model and T is a set of
moments of time ordered by the dense linear ordering <, as defined in (1). Intervals
of time as such are not part of the model directly as they are not primitive but derived
from the set of moments of time, T, and the precedence relation, <, in the way
specified in (4). As before, F is the function that assigns denotations to lexemes in the
model, but now, of course, it must do so at particular temporal intervals. F may,
therefore, be defined as a function from lexemes to sets of ordered pairs of times and
extensions, <i,A>, where i is a temporal interval and A (Delta) is the extensional
denotation of a constant at the time i. The models defined in this book are all complete
in the sense that it is always possible to determine the extension of any well-formed
expression in the object language. It is therefore necessary to ensure that there are no
times for which the extension of some expression is not specified. In other words, F
assigns denotations to basic expressions in such a way that the extension of the latter
can be ascertained for every interval defined on T so that there are no 'holes' in the
model where the extension of a particular constant is unspecified at some particular
time. This is done by requiring that the union of all temporal intervals in the range of
F(a) for each constant a be equal to the largest possible interval, i.e. the set of times
T. This strong condition on (complete) temporal models is formally written as the
statement in (9) where the symbol U indicates the union of all sets i where i is the
interval of time that forms the first element in an ordered pair in F(oc).
(9) U<i,A> e F(cc) = T, for all constants a in the object language.
As an example of a temporal model, let us consider a situation in which the four
humans of our previous models, Jo, Fiona, Bertie and Ethel, previously unacquainted,
meet for the first time. On first impressions, both Fiona and Ethel like Jo and Jo likes
both of the others. Nobody likes Bertie at first and he likes no-one else. After a time,
however, Fiona decides she doesn't like Jo, but likes Ethel instead. At the same time,
Ethel decides she does like Bertie. After another period of time, Ethel too decides she
doesn't like Jo but does like Fiona. Bertie also comes to like Ethel. This situation can
be formally represented in a model, call it M8, containing four entities MANi (Jo),
MAN2 (Bertie), WOMANt (Ethel) and WOMAN2 (Fiona) and nine moments of time,
ti to U (a ludicrously small timescale in reality but sufficient to illustrate the idea). The
denotation assignment function, F8, models the above situation by assigning the
denotation to like' in (10), where there is no moment of time in which the lexeme does
not have a specified extension (even if this is null for some moments). We assume
here, and elsewhere, that individual constants denote the same entity at all times in a
model so that F8 assigns to each individual constant, jo', flona', bertie' and etheV, a
single member of A for the largest interval in I, i.e. T itself.
236
Time
(10) a. F8(jo') =<[ti,t,],MAN,>.

b. F8(fiona') =<[t1,t9],WOMAN2>.
c. F8(bertie') =<[tl9t,],MAN2>.
d. F8(ether) =<[t,,«,WOMAN,>.
e. F 8 (like')={<[t,],0>,
<[t a ,y,{<MAN 1> WOMAN 1 >,<MAN 1 ,WOMAN 2 >,
<WOMAN1,MAN1>, <WOMAN2,MAN1>}>,
<[t4,t6],{<MAN1,WOMAN1>, <MAN1,WOMAN2>, <WOMAN1,MAN1>,
<WOMAN 2 ,WOMAN 1 >,<WOMAN 1 ,MAN 2 >}>,
<[t7,t9],{<MAN1,WOMAN1>,<MAN1,WOMAN2>,
<WOMAN1,WOMAN2>,<WOMAN2,WOMAN1>,
<WOMAN1,MAN2>, <MAN 2 ,WOMAN 1 >}>}.
The situation specified in (10) can be represented diagrammatically as (11), where
each line bounded by square brackets under the time line indicates the interval during
which one person likes another.
(11)
H h H h
0
<WOMAN2,MAN1>
<WOMAN1,MAN1>
<WOMAN2,WOMAN1>, <WOMAN1,MAN2>
<WOMAN1,WOMAN2>, <MAN2,WOMAN1>
•
, <MAN1,WOMAN2>
To ascertain the truth of temporal formulae, however, it is necessary to know not

only the model, but also the time at which the formula is being assessed. For example,
the formula (like'(jo'))(fiona') is true with respect to (10) at the interval [t3] (a
moment of time), but false at [t 7 ], while (like'(fiona'))(jo') is true at all intervals
except [ t j . Hence, all logical expressions must be interpreted with respect not only to
a model, M, and an assignment of values to variables, g, but also an interval of time,
i. The extension of an expression, oc, with respect to M, g and i is symbolised as
To interpret temporal formulae, the base clause of the model theory must be
altered to identify the meaning of a constant at a particular time with its extension at
237
that time as specified by the denotation assignment function of the model (12.a).
Because the value of F applied to a constant a is a set of ordered pairs, F(oc) may be
interpreted as a function from intervals to extensions. Thus, where <i,A> is in F(oc),
F(oc)(i) is the extension, A, of a at interval i. For example, ¥%(like')([t2M) is the set
{<MAN,,WOMAN,>, <MAN1,WOMAN2>, <WOMAN1,MAN1>,
<WOMAN 2 ,MANi>} (according to (8.e), above). The rule for interpreting
functor/argument structures at a particular time is then given by the rule of functional
application relativised to temporal intervals. Thus, the interpretation of an expression
f(a) with respect to a model, a variable assignment and a particular interval of time,
is determined by the application of the extension of the functor,/, at that time to that
of the argument at the same time, cf. (12.b).
(12) Given a model M = «A,F,<,T»:
a. If a is a constant, then [oc]M& = F(a)(i).
b. If f(a) is an expression in LQ, then [f(a)]M^ = [f] M ^ ([a]M*9.
Given the revised model theory in (12), to obtain the interpretation of the
expression (like'(jo'))(fionay) at the interval [t2,t4], one must apply the function
[like'(jo')]mâ^ to the argument [fiona*]™*^2™. Since fwna' is a constant in the
logic, [fiona']m*<[t2M] is given by FJfiona') applied to the interval, [t2,t4]. Under our
assumption that individual constants like fwna' extensionally denote the same entity
at all times, Fs(fiona')([t2M) is the same as F*(fwna')([UMX i.e. WOMAN 2 . The
functor expression, on the other hand, like'(jo'), is not basic, and so its extension at
the interval [t^tj is computed by a second application of (12.b) from the extension of
like' at this interval applied to that of jo' in M8 at the same time. Since Jo translates
as an individual constant, [/V]*18^12'41 is F8(/o')([t1,t9]), i.e. MANX. The function
denoted by [like 'JMSAIQÎ [S given by F8(//Jte')([t2,t4]) which, as we saw above, is that
function in({0,l}A)Athat defines the set {<MAN 1 ,WOMAN 1 >,<MAN 1 ,WOMAN 2 >,
<WOMAN 1 ,MAN 1 >, <WOMAN 2 ,MANi>}. Applying the appropriate function to the
entity denoted by jo', i.e. MANi, gives the characteristic function of the set
{WOMANi,WOMAN2), which is thus the value of [like '(jo ')]*»*#&*. The application
of this function to the extension of fwna' yields the value 1, since WOMAN 2 is in the
set denoted by the predicate at that time. Hence, the formula (like '(jo '))(fiona') is true
with respect to M8 at the interval [t 2 ,tj. (13) summarises the steps taken to arrive at
this interpretation.
(13) a. [(like'(jo'))(fiona')] M w 2t4 J is 1 iff [like'(jo')]M8-8[t2t4] ([fiona'] M8 ^ t2t4 0 is 1.

b. [like'(jo')] M8 ^ t2t41 ([fiona'] M8 ^ t4 0 is 1 iff
([like']M8'8-tt2-t4i ([jo>]M8'8tt2t40)([fiona']M8'8it2t40 is 1.
c. [fiona']M8'8tt2t4] is F8(fiona')([t2,t4]) is WOMAN 2 .
d. [jo']M8'8lt2t4] is F8(jo')([t2,t4]) is MAN,.
e. [like'] M8 '8[ t2t4 i is F 8 ( l i k e ' ) ( [ t 2 , t 4 ] ) , the function defining
{<MAN b WOMAN 1 >,<MAN 1 ,WOMAN 2 >, <WOMAN^MAN^,
<WOMAN 2 ,MAN,>} (call this K).
f. K ( M A N 0 ( WOMAN,) = 1.
g. So [(like>(jo'))(fiona')]M8-8ft2t4]is 1.
238
Time
Exercise 8.1:
The verb like is a stative verb so that its extension at some interval of time remains
the same for every subinterval of that interval. Assuming this to be so, work out the
extensions of the following expressions at the intervals indicated with respect to the
model in (10).
i. (like'(ether))(fiona') at
ii. (like'(fiona'))(jo') at [tj.
Ascertaining the truth or falsity of formulae with respect to temporal intervals

that correspond to the times specified by the denotation assignment F is given directly
by the theory of interpretation in (12), but problems arise in interpreting formulae with
respect to intervals that are not specified by F. In the previous exercise, we assumed
that, because like is stative, if x likes y at an interval i, then x likes y at all subintervals
of i. Without any contextual information to the contrary, this seems intuitively correct
and solves the problem of ascertaining the extension of certain types of constant (but
not all; see section 8.3.2) at subintervals of the intervals specified by F.
A more intractable problem arises, however, when we consider the truth of a
formula with respect to an interval of time which spans a change in the extension of
a constant in the formula being interpreted. To illustrate the problem consider the
interpretation of the formula (like'(jo'))(fiona') at the interval [t4,ts]. According to
(10), this formula is true at moment [t4] but false at moment [t5]. What then is the truth
value of [(Uke'(jo'))(fiona')]m*itW} This question requires an answer to the more
general question: what exactly is meant by the expression 'true at an interval of time' ?
Clearly if we mean 'true at some time during an interval', then the example formula
is true with respect to the little model given above, since there is a subinterval ([t4])
at which (like '(jo '))(fiona') is true. On the other hand, if it means 'true throughout an
interval of time', then, of course, [(like'(jof))(fionaf)]M**[t4t5] is false, because there is
a subinterval ([t5]) at which the formula is false. Although this is not a problem that
arises directly in this chapter, because of the way the truth of formulae is assessed
with respect to arbitrary intervals of time, a grammar fragment that analyses temporal
adverbials like on Tuesday, last week, two years ago, needs to have this problem
solved in order that sentences like Fiona sang last Tuesday may be interpretable in
a situation in which there are periods on the appropriate day in which Fiona did indeed
sing, but other periods in which she didn't. Furthermore, a formula should in principle
be interpretable with respect to any interval in the model, and so some solution must
be provided whereby the extensions of expressions can be ascertained at all temporal
intervals irrespective of whether or not those intervals are directly mentioned by the
denotation assignment function of the model in question.
The solution that is adopted here takes the notion of truth at an interval to be
equivalent to truth at some time during an interval. The reason for this decision can
be seen from a consideration of the truth-conditions of sentences like the ones in (14).
239
(14) a. Jo kicked the dog yesterday.

b. Fiona sang the song last Tuesday.
c. Bertie will paint a picture next year.
d. Ethel knew how to play golf last year.
These sentences all express true propositions, if the event described by the verb and
its arguments occurred at some time during the interval described by the adverbial
expression. For example, the sentence in (14.b) translates into a formula that is true,
if the person called Fiona sang the (contextually salient) song at some time on the
Tuesday before the time of utterance. It does not, however, require that Fiona sang the
song all throughout the day in question. The sentence in (14.d) appears to imply that
Ethel knew how to play golf throughout the whole of the year preceding the one in
which the sentence is uttered, but it would not be falsified if, for example, she knew
how to play golf from January to September of that year and then was in an accident
that made her forget that there exists a game called golf, let alone how to play it.
In general, therefore, we can take truth at an interval to be defined according to
truth at some period in that interval. Formally, this can be defined by taking the
denotation of a constant, a, at an interval i to be determined by the union of the
extensions of a at all intervals that overlap with i, as specified by the denotation
assignment function, F. For example, [like']m>s[t4t5] is given by F9(like')Q) for all
intervals j that overlap with [t4,t5]. According to the model, Mg, there are two such
intervals in the domain of F: [t2,t4] and [tS9U\ (see (lO.c)). In the former interval like'
denotes the set {<MAN1,WOMAN1>, <MAN1,WOMAN2>, <WOMAN1,MAN1>,
<WOMAN 2 ,MAN 1 >}, whilst in the latter it denotes the set {<MAN1,WOMAN1>,
<MAN 1 ,WOMAN 2 >, <WOMAN1,MAN1>, <WOMAN 2 ,WOMAN 1 >,
<WOMAN1,MAN2>}. The union of these two sets is {<MANi,WOMAN!>,
<MAN 1 ,WOMAN 2 >, <WOMAN 1 ,MAN 1 >, <WOMAN 2 ,MAN 1 >,
<WOMAN!,MAN2>, <WOMAN2,WOMAN1>} which provides the extension of like'
at the interval [t4,t5]. The formula (like '(jo y))(fwna') is thus true with respect to model
M« and interval of time [t4,t5].
The first clause in the model theory must now be revised as in (15) which also
contains a restriction on the overlapping intervals that are relevant to the
interpretation. According to the definitions in (4.a) and (4.b), above, proper
subintervals also overlap with their containing intervals. Although the extensions of
stative predicates at some particular time are maintained for all subintervals of that
time, this property is not shared by all predicates. For example, a sentence like Bertie
painted a picture last week should not be interpreted as expressing a true proposition
with respect to a situation where Bertie painted a picture over a two week interval
containing last week (see section 8.3.2 for more discussion of these sorts of predicate).
Hence, (14) excludes from the definition of the extension of a constant at some time,
i, the extensions of that constant at other times which properly contain i. However, if
i is in the domain of F then the definition in (15) reduces to that in (12.a), under the
assumption that F does not include overlapping intervals for individual constants.
(15) If a is a constant, then [a]M-^ is UA, where F(oc)(j) = A for all ioj where i is
not a proper subinterval of j .
240
Tense
The rest of the model theory, covering the interpretation of the propositional
connectives, the lambda operator and the quantifiers, needs also to be revised in the
light of the introduction of time. These are all straightforward revisions, and a fuller
discussion of the model theory is postponed until Chapter 10.
8.3 Tense
Given the theory of time and temporal models presented in the last section, the
interpretation of simple tensed formulae is fairly straightforward and is the topic of
this section. In the following subsections, a general theory of tense will be presented,
providing the basic interpretations of Present tense, Past tense, and Future tense
and discussing certain combinations and entailments of these. Many languages do not
have morphological tense systems and very few have a three-way tense system, but
all languages seem to have ways of referring to times other than the time of utterance.
The theory put forward in this chapter is a general one, in the sense that it does not
claim to analyse specific tense systems, but provides the machinery to talk about
present, past and future time reference. It should be emphasised that as we move away
from a purely extensional semantics like that developed in Chapters 2 to 6, the amount
of controversy and debate increases. Readers are asked to bear in mind, therefore, that
what is presented here is no more than an outline of what is acknowledged to be an as
yet incomplete theory in a fairly rudimentary state of development. Readers are
invited to pursue their own ideas and are encouraged to compare the truth-conditions
suggested for the interpretation of different tenses with their intuitions about their own
language or their knowledge of languages other than English.
8.3.1 Past, present and future

Tense is a deictic category that locates in time the state of affairs described by a
sentence. This location is not precise, but relative to some other time, a temporal
reference point, the most important of which is the time of utterance, the interval
(usually assumed to be a moment) in which a sentence is uttered by some speaker. It
is with respect to this time that present, past and future for the speaker are computed.
Thus, for those languages that have it, a declarative sentence in the present tense can
be used to assert that the event described by the sentence occurs at the time the
sentence is uttered, i.e. the time of the event is located at the time of utterance. A
declarative sentence in the past tense, on the other hand, can be used to assert that
the event occurred prior to the time of utterance and one in the future tense, that the
event will occur after the time of utterance. Although in a discourse the time of
utterance is always the main reference time, it is possible for other times to act as the
principal reference time for the interpretation of a tensed sentence, particularly for
multiply tensed sentences (see Section 8.2.2) or embedded clauses. Hence, the
definitions given below refer only to principal reference times rather than the time of
utterance, in order to maintain generality.
Since tense locates situations in time that are described by a sentence, the tenses
can be represented in logic as operators over formulae. Thus, temporal logics contain
a set of tense operators which, being expressions of type <t,t>, combine with
formulae to yield tensed formulae. The introduction of these operators into the
241
language LQ gives the tensed logical language, Lr. The present tense operator is
symbolised as Pres and the representation of a sentence like Fiona sings in Lr is Pres
(sing '(fiona')), and that of Every student sings is Pres (\/x [student '(x) —> sing '(x)]).
As mentioned above, the use of the present tense may indicate that the truth or
falsity of a formula is to be assessed with respect to the primary reference time,
usually the time of utterance. Thus, on one interpretation, the proposition expressed
by the sentence Fiona sings is true if there is someone called Fiona who is singing at
the time the sentence is uttered. In other words, the present tense locates the time of
the event or situation described by a sentence at the time of utterance and so the
interpretation of the LT representation ofFiona sings is identical to [sing '(fiona 'j] M8 «•»,
where i is the time of utterance. Thus, the interpretation of formulae containing the
present tense operator at some time i is equated with the interpretation of the formula
without the operator at the same interval of time. The truth-conditions of such
formulae are spelled out formally in (16) and represented diagrammatically in (17)
where i is the reference time placed on the time line which clearly shows that the time
at which the embedded formula ty is to be interpreted is the same as the principal
reference time.
(16) Present tense: [Pres (<!>)]**** is 1 iff [$]*** is 1.

(17)
i
Pres(<|>)
The fact that present tense formulae at a particular time are interpreted
identically to the corresponding untensed formula at that time leads many logicians
to omit the present tense operator altogether. However, many languages
morphosyntactically mark the present tense and many distinguish tensed from
untensed sentences. The linguistic evidence thus points to the need for a present tense
operator, as well as past or future tense ones, even though this may be semantically
unnecessary, according to the interpretation given in (16). In a discourse, however, it
is possible that the use of tense introduces particular reference times that can be
subsequently referred to by pronominal expressions (like then, from that time on, etc.).
Untensed sentences, on the other hand, do not seem to have this property. A more
sophisticated theory that considers the needs of discourses, rather than isolated
sentences, would thus very likely make use of a present tense marker and so a present
tense operator is adopted here, despite the fact that it is truth-conditionally otiose.
To capture the force of the past tense, an operator Past is used which gives rise
to formulae like the ones in (18) which represent the formulae expressed by the
English sentences in (19), respectively.
242
Tense
(18) a. Past (sing'(fiona')).

b. Past (3x [student'(x) & (like'(jo'))(x)]).
c. Vx [adult' (x) -> Past (child'(x))].
(19) a. Fiona sang.
b. A student liked Fiona.
c. Everyone who is an adult was a child. (Every adult was a child.)
The interpretation of past tense formulae locates the time of the event described by
the predicate at a time prior to the principal reference time, so that the truth or falsity
of a formula like Past (sing '(fiona')) is determined by whether there is some interval
of time before the moment of utterance at which the formula sing'(fiona') is true. If
there is, then Past (sing'(fiona')) is true, if not, then it is false. (20) gives the formal
definition of the interpretation of formulae in the scope of the past tense operator
while (21) represents this diagrammatically and shows clearly the 'pastness' of the
situation with respect to the principal reference time.
(20) Past tense: [Past (ft)]*** is 1 iff there is a time j such that j < i and [c()]M*)
isl.
(21)
Past(<())
In order to see how past tense formulae are interpreted with respect to a
particular model, let us adopt the simple model, M8, presented in the last section
augmented with the denotations of sing'y student' and happy' in (22).
(22) a. F8(sing') = {<Hl9U\9{WOMAN2}>,
<[t3,t8],{MAN1,WOMAN1,WOMAN2}>,
<[Ul {MAN2,MAN! ,WOMAN2, WOMAN! }>}.
b. F8(student') =
c. F 8 (happy') = {<[t,,t2],{MAN1, WOMAN, }>,

<[t3,t6],{MAN1}>,
<[t 8 ,tj, {MAN2,WOMAN2, WOMAN2} >}.

If the sentence Fiona sang was uttered at the moment [t8], the formula that represents
it, i.e. Past (sing'(fiona')), is true only if there is an interval of time i prior to [t«] in
which sing'(fiona') is true, i.e. if WOMAN2 (the entity denoted by fiona') is in
[sing ']M**\ Clearly, there are a number of intervals at which this is so according to F8,
[tjjtg] being one of these (cf. (22.a)). Hence, Past (sing'(fiona')) is true with respect
to M« and time [ t j . On the other hand, the formula is not true at the interval [tj, since
243
there is no prior interval in which Fiona sings.

(23) presents the formal procedure for verifying the more complex formula Past
(VJC [student'(x) -> sing'(x)]) (the translation of the sentence Every student sang) with
respect to M 8 and time of utterance [ t j . The complexity of this example is due entirely
to its rigour and nothing is introduced that was not in the discussion of the past given
above. In (23) and the exercise that follows, it is assumed that the properties of
singing, being a student, being a lecturer or being happy are all stative so that if
someone has one of these properties at some interval of time, then they have it for all
subintervals of that time.
(23) a. [Past (Vx [student'(x) -^ sing'(x)])]M8'8.r*] is 1 iff there is an interval of time
j such that j < [t6] and [Vx [student* (x) —»sing'(x)]]M8«8o is 1.
b. Let j be [ M J , then [Vx [student* (x) -» sing'(x)]]M8'8«3.<4] i s 1, iff for every
value assignment g' like g but for the value of x, [student'(x) —>
sing'(x)]M8'8'<tt3-t4J is 1.
c. Let g' = gMANi/x? t n en [student'(x) -> sing'(x)]M8'8MAN1/xft3t4] is 1 iff
[student']M8'«MAN1/Xtt3t4i is 0 or [sing']M88MAN1/xft3^ is 1.
d. [student'(x)]M8-8MAN1/xtt3t4i i s 1 since MAN! e F8(student')([t2,t9]), student' is
stative and [t3,t4] £ [t 2 ,tj.
e. [sing'(x)]M8'8MAN1/x[t3t4iis 1 since M A ^ € F8(sing')([t3,t8]), sing' is stative and
[t3,t4] C [ta,t8].
f. Hence, [student'(x) -> sing'(x)]M8-8MAN1/xft3t4] is 1.
g. Let g' = gMAN2/x> t hen [student'(x) -> sing'(x)]M8-8MAN2/x<[t3>t4] is 1 because
[student'(x)]M8-8MAN2/xit3t4i i s 0 (as MAN2 g F8(student')([t3,t4])).
h. Let g' = gwoMANi/x^ [student'(x) -> sing'(x)]M88WOMAN1/x'[t3t4i is 1 because
[student'(x)]M8-«WOMAN1/x'ft3t4] is 0 (as WOMAN, £ F8(student')([t3,t4])).
i. Let g' = gwoMAN2/X) [student'(x) -> sing'(x)]M8'8woMAN2/x,[t3,t4] i s i , because
[student'(x)]M8-8woMAN2/x,[t3,t4] i s i ( W O M A N 2 G F 8 (student')([t3,t 4 ]) and
[sing'(x)]M88woMAN2/x,[t3.t4] IS \ ( W O M A N 2 G F 8 (sing')([t 3 ,t 4 ])).
j. Hence, since there are no other g', [Vx [student'(x) —> sing'(x)]]M8'8it3t4i is
1.
k. Hence, since [t3,t4] < [t 6 ], [Past (Vx [student'(x) -> sing'(x)]]M8-8ft6] i s l.
Exercise 8.2:
For each of the following formulae, give an English sentence which could express it
and interpret it formally (as in (23)) with respect to the model M8 and the time of
utterance shown.
i. Past (~(3x [student'(x) & happy'(x)])) at [t 5 ].
ii. Pres (Vx [student'(x) & Past (happy'(x))]) at [ t j .
iii. Past (sing'(jo') & (like'(jo'))(fiona')) at [t 7 ].
Very few languages have a morphological future tense and the interpretation of
sentences that refer to future time is not straightforward because modality is usually
involved. However, to represent simple future reference in LT, there is another
operator, Fut, which gives rise to formulae such as Fut (sing '(fiona')) which can be
244
Tense
used to translate English sentences with the auxiliary will as in Fiona will sing. The
interpretation of future tensed formulae proceeds in the same way as for past tense
ones, except that we look for a time following the time of utterance at which the
simple untensed formula is true. This gives the truth-conditions in (24) which again
can be diagrammatically represented by the time line in (25).
(24) Future tense: [Fut (<t>)]Mgi is 1 iff there is a time j such that i < j and [<t>]M*>J
isl.
(25)
As an example, consider the interpretation of Fut (sing '(fwna')) with respect to

Mg and the time of utterance [t 3 ]. This is true if, and only if, there is an interval of time
following [t3] where Fiona actually sings. Since [sing ']M8,g.[t5,t7] contains WOMAN 2 , the
entity denoted by fwna', and the time of utterance, [t 3 ], precedes the interval [t s ,t 7 ], the
formula is true. (26) gives a further illustration of the semantics of Fut, and shows the
formal interpretation of the formula Fut (3x [student'(x) & happy'(x)]), one of the
translations of A student will be happy, with respect to M 8 and the time of utterance
(26) a. [Fut (3x [student' (x) & happyXx)])]M8-«^ is 1 iff there is atime j , where [t6]
< j , such that [3x [student'(x) & happy'(X)]] M8 «J is 1.
b. Let j = [U,Ul Then 0 x [student'(x) & h a p p y ' ( x ) ] ] " 8 * ^ is 1 iff there is a
value assignment g' exactly like g except for the value assigned to x such
that [student'(x) & happy'(x)]**8*'*8-®] is 1.
c. Let g' = gwoMAN2/x T n e n [student'(x) & happy'(x)]M8-8WOMAN2/x-£t8t9i is 1 iff
[student'(x)]M8-8WOMAN2/xit8t9] is 1 and [happy'(x)]M8-8WOMAN2/*[t8t9J is 1.
d. Since WOMAN 2 e [student']M8-8WOMAN2/Xft8t9\ [student'(x)] m ^ O M ™ww [s
1.
e. Since WOMAN 2 e [happy']M8-8WOMAN2/x-tt8t9J, [student'(x)]M8-8WOMAN2/xit8t9]is 1.
f. Hence, [student'(x) & happy'(x)]M88'[t8t9i is 1.
g. Hence, [3x [student'(x) & happy'(x)]] M8 ^ t8t9 i is 1.
h. Hence, since [U] < [t8,t9], [Fut (3x [student'(x) & happy'(x)])]** 8 *^ is 1.
245
Exercise 8.3:
What are the truth-conditions of the following future formulae? What English
sentences might these translate and what do they reveal about the interaction of the
operators Fut, ~ and 3?
i. ~(Fut (3x [student'(x) & happy'(x)])).
ii. ~(Fut (3x [student'(x) & happy'(x)])).
iii. Bx [Fut (-(student'(x) & happy'(x)))].
iv. 3x [~(Fut (student'(x) & happy'(x)))].
8.3.2 Complex tenses

Because tenses are represented as operators over formulae in LT, it is possible to
iterate them, just as it is possible to have more than one quantifier in front of an open
formula. Thus, according to the syntax of Lr all of the formulae in (27) are
well-formed.
(27) a. Fut (Fut (sing'(fiona'))).
b. Past (Past (sing'(fiona'))).
c. Fut (Past (sing' (fiona'))).
d. Past (Fut (sing'(fiona'))).
e. Past (Fut (Past (Past (Fut (Fut (sing'(fiona'))))))).
The iteration of tenses in this way is a mixed blessing. Expressions like (27 .e) are
likely to translate no natural sentence in any human language, although they can be
imitated using very formal language. For example, the expression in (27.e) can be
represented in formal English by repeating the phrases it was the case that and it will
be the case that, i.e. by the sentence // was the case that it will be the case that it was
the case that it was the case that it will be the case that Fiona will sing. However,
such expressions are unlikely ever to be uttered in any actual situation and, more
importantly, finding a language that has a single tense in its inventory that would
reflect this temporal contour is highly improbable. However, the fact that there is a
sentence of English that can be translated into the formula in (27.e), no matter in how
stilted a fashion, indicates that the explanation for its unnaturalness in terms of human
language lies elsewhere than in semantics. We will not attempt to provide a pragmatic
account of the improbability of a sentence expressing the formula in (27.e), but will
assume that one can be constructed using Grice's maxims and so accept that tenses
can be iterated.
Double combinations of tenses are, however, needed to provide an account of
part of the tense/aspect systems of a number of languages, including English. The
double past tense formula in (27.b), for example, reflects the temporal information
contained in the English pluperfect. Consider the sentence in (28) which expresses a
true formula if, on the day before the time of utterance, there is a time before that at
which Fiona indeed sang the song.
(28) By yesterday, Fiona had sung the song.
246
Tense
Thus, two reference times, other than the time of utterance, are required to interpret
such a sentence: yesterday, on which the formula Past (sing '(fiona')) is true, and some
time before it when sing'(fwna') is true. These are the truth-conditions of the double
past tense in (27.b), so that the pluperfect tense can be represented by the diagram in
(29) where, as before, i is the time of utterance and j and k are two temporal intervals
standing for the referent of yesterday and the time at which Fiona did the singing,
respectively.
(29) Pluperfect:
Past(<(>) Past(Past((|>))
Double futures also underlie the temporal interpretation of certain constructions

in human languages. Although English has no proper double future tense, the sense
can be captured by using the construction will be going to as in (30).
(30) Fiona will be going to sing my favourite song (when she has learnt it).
Although the main clause of this sentence also conveys the intention of the entity
denoted by the subject, the temporal information it conveys is that expressed by the
formula in (27.a). Again there are two reference points in addition to the time of
utterance: the time when she has learnt the song and the time after this when she
actually sings it. Hence, the situation can be represented by the diagram in (31) where
j represents the time Fiona learns the song and k when she sings it. The diagram of the
truth-conditions associated with the formula in (27.a) shows only the temporal relation
between the events described by the sentence in (30). It gives no indication of the
modality or the aspect of the English sentence and thus does not provide a complete
explication of the meaning of that sentence.
(31)
Fut(Fut(<)))) Fut(<|))
247
Exercise 8.4:
Show that the diagrams in (29) and (31) represent the truth-conditions for the
pluperfect and the be going to construction in English, by working out formally the
truth or falsity of the formulae in (27.a) and (27.b), with respect to M8 and the
moments of time [t3] and [ t j , respectively.
The two diagrams in (29) and (31) reveal an entailment that each of the formulae
in (27.a) and (27.b) have. That is, if the proposition expressed by Fiona had sung the
song is true at the time of utterance then so is that expressed by Fiona sang the song
(32.b). Similarly if Fiona will be going to sing the song expresses a true proposition
at a particular time, then so does Fiona will sing the song (32.a).
(32) a. Fut (Fut (((>)) -> Fut (<)>).
b. Past (Past (<|>)) -> Past ($).
The mixed combinations of tenses in (27 .c) and (27.d) also figure in English. The
formula in (27.c) is expressed by the sentence in the future perfect given in (33)
(ignoring the adverbial by midnight and the direct object for simplicity).
(33) Fiona will have sung my favourite song by midnight.
This expresses a true proposition only if there is a future time (before midnight) in
which the proposition expressed by the sentence Fiona sang myfavourite songis true.
This situation can be shown schematically by the diagram in (34), which, as before,
utilises three reference times including the time of utterance.
(34) Future perfect:
Fut(Past(0)) Past((|>)
The semantics of the operators Past and Fut do not require the reference point k in
(34) to be in the future with respect to i. The interval j must be in the future with
respect to i, but the formula Past (§) is true with respect to j if (and only if) k is before
j , i being irrelevant to its interpretation. Hence, $ could be true in the future, past or
present with respect to i. The implication in (35) is thus a tautology on this analysis.
248
Tense
(35) Fut (Past (<|>)) -> (Past ((|)) v Pres (<|>) v Fut (()>)).
This interpretation of the future perfect is, therefore, rather weak because it does
not determine when the event in question is likely to happen with respect to the time
of utterance. On the other hand, the use of this tense apparently conveys the
information that the simple formula, <> | , is true with respect to i, so that one might
want to extend the theory to make the time of utterance continue to be relevant at all
subsequent evaluations of the truth of the embedded formulae at particular times. For
example, one might evaluate Fut (Past (§)) as being true at i only if Past (§) is true at
an interval j following i and(|) is true at an interval k also following i (and, of course,
preceding j). This would make any future perfect formula true only in the situation
diagrammed in (34). However, this seems an unnecessary move. In the first place, we
can appeal to the Maxim of Quantity to account for the non-use of the future perfect
where it is known whether Past (§) or Pres (§) is true. This is because in normal
conversation we state what we believe to be true and no more or less. If we know that
something has happened or is happening, then that is what we should say, by using the
past or present. If we know that something has not happened in the past or is not
happening in the present, then we must use a disjunction like Fiona is singing the song
now or will sing it soon or Fiona has already sung the song or will sing it soon. Only
if we know that something did not happen in the past and is not happening now, and
that this something is to happen by a certain time, would we use the future perfect.
Hence, we can account for the implicature of the futurity of <|) in Fut (Past (§)), using
pragmatic principles.
Furthermore, it seems that the apparent futurity of k in (34) with respect to i is
not contradicted by situations in which k is equal or prior to i. For example, (36) gives
four possible responses to the sentence in (33). Only the response in (36.d) seems to
deny the truth of (33), while the others merely specify the time at which Fiona sings
the song. Hence, the apparent weakness of the future perfect on this interpretation
appears to be borne out.
(36) a. She did, two hours ago! You must have missed it.
b. You'd better be quick; she's singing it now.
c. She's going to sing it in a quarter of an hour.
d. No, she won't. She's never sung that song and never will. She hates it!
Similar considerations apply as well to formulae with the structure in (27.d),
where future formulae are evaluated in the past as shown in the diagram in (39). Again
k may be before, after or identical to i, the implication in (37) again turning out to be
a tautology. There is no recognised 'tense' in English that captures this temporal
relationship, but the past tense of be going to as in (38) seems to come closest to it.
(37) Past (Fut (()))) -> (Past (<|>) v ty v Fut (<())).
(38) Fiona was going to sing my favourite song by midnight.

This sentence can be used to imply that the speaker believes that Fiona has not sung,
is not singing and will not sing my favourite song. However, such an assumption is
cancellable as is shown by the fact that the replies in (36.a) to (36.c) contradict the
assumption, while not contradicting what is said by the sentence in (38). Hence, we
249
may take the speaker's belief as an implicature and not an entailment of the sentences
and maintain the truth-conditional analysis captured in the (partial) translation in
(27.d).
(39)
Fut((J>) Past(Fut(<!>))
*Exercise 8.5:
Discuss the truth-conditions and provide an appropriate logical translation of
sentences containing the adverbs always and never like those in (a) to (d), below.
What problems arise with the interaction of the scope of the different operators in
these translations?
i. Jo was always a fool,
ii. Jo will always be a fool,
iii. Jo was never clever,
iv. Jo will never be clever.
The theory of tense presented above is idealised with respect to actual linguistic
data. The specifics of grammatical tense in particular languages often have different
interpretations from those presented in Section 8.2.1. For example, in English, the
grammatical tense that covers the present (the so-called present tense) also allows a
future interpretation, although it does not allow a past tense reading (without
pragmatic implicatures), as shown in (40).
(40) a. Jo goes home today.
b. Jo goes home tomorrow.
c. ?Jo goes home yesterday.
Furthermore, the semantic future tense is encoded using a periphrastic construction
involving the auxiliary verb will/shall in English which emphasises the apparent
modality of this tense. Thus, English may be said to distinguish semantic past and
non-past and to lack a true future tense. The theory presented above, however, is not
intended as a complete theory of the tense system of English (or any other language),
but as the basis of such a theory. Clearly, in analysing the tense system of any
250
Simple aspect
human language, one needs a theory of time and some way of expressing past, present
and future relative to some point of reference. This has been done and been shown to
have certain interesting properties that hold, or appear to hold, of English. The theory
provides a reasonable base for an analysis of the specifics of a particular system of
tense.
8.4 Simple aspect

The previous section deals with the deictic category of tense, which locates an event
in time with respect to other times, in particular the time of utterance. Temporal
relations may, however, also be presented according to the perceived internal
properties of the event or situation being described. For example, an activity can be
presented as continuing at some interval of time without indicating how long the
activity has lasted or will last in relation to that time, as illustrated by the sentences
in (41). On the other hand, an event may be presented as complete or undifferentiated,
as illustrated in (42).
(41) a. Fiona is singing the song.
b. Fiona was singing the song.
c. Fiona will be singing the song.
(42) a. Fiona sings the song.
b. Fiona sang the song.
c. Fiona will sing the song.
The examples in (41) are said to be in the imperfective aspect whilst those in (42) are
in the perfective aspect. Other verbal aspects may be defined as subtypes of this basic
distinction. For example, one can further subdivide the imperfective into habitual
(43.a) and continuous (43.b).
(43) a. Jo used to kick the cat.
b. Jo was kicking the cat.
Finer grained distinctions may be made by dividing the continuous aspect into
progressive and non-progressive types or recognising punctual versus durative
events or a difference between events that involve the bringing into existence of some
entity (e.g. paint a picture) and those that just involve an activity of some sort (e.g.
walk), and so on. It is not the intention of this section, however, to discuss the category
of aspect in all its manifestations (this would require a whole volume to itself).
Instead, a basic definition of the perfective/imperfective distinction and some
characterisation of some of the different types of event denoted by different verbs is
given in terms of temporal intervals.
8.4.1 Perfective and imperfective

Despite the controversy that surrounds almost any formal definition of the semantics
of the different aspects, the use of temporal intervals allows us to give at least a
rudimentary definition of the semantics of the basic perfective/imperfective
distinction. In this section, I present definitions of these that follow Bennett and Partee
251
(1978) and Dowty (1979). The analysis presented here is, however, far from complete
and, for fuller treatments, the works cited at the end of this chapter should be
consulted.
The perfective aspect, as mentioned above, presents a situation as temporally
undifferentiated, but complete, and ignores any internal structure it may have. For
example, the sentence Jo kicked Chester presents the event of Jo kicking Chester as
a complete unit without indicating how long or how often it occurred. Hence, to know
the truth value of the formula that translates this sentence, all we need to know is
whether the event occurred or not (at some time in the past). Truth-conditionally,
therefore, the perfective can be interpreted in the same way as we have been
interpreting all formulae in this chapter, i.e. directly with respect to a model, a variable
assignment and a time, the latter being the time of the event. The perfective, then, may
be (and often is) treated as the semantically neutral aspect, without any need for an
operator (like the tense operators, Pres, Past and Fut). Any sentence in the perfective
aspect, therefore, translates directly into L T and is interpreted using rules already
given. Thus, the sentence Jo kicked Chester translates (after lambda conversion) into
Past ((kick'(chester''))(jo')), a formula that is true if, and only if, Jo actually kicks
Chester at some time prior to the time of utterance.
It is possible, however, that, like the present tense, the perfective should also be
signified by the use of an operator, particularly as there are languages that have
morphosyntactically marked perfective aspects and others may combine perfective
and imperfective aspects. Furthermore, it could be argued that the completeness
property of the perfective should be represented in its semantics. This may be done
by requiring the temporal interval at which the truth of a perfective formula is assessed
to contain both initial and final moments of the event being described by the sentence.
This would mean that the proposition expressed by a sentence like Jo kicks Chester
would be true only if the event of Jo's kicking Chester is begun and finished at the
time of utterance. Furthermore, by the semantics of the past tense given above, a
sentence like Ethel played a game of golf would express a true proposition only if
Ethel actually completed the game of golf at some point before the time of utterance.
It would not be true, however, if, for example, she was still playing the last hole when
the sentence was uttered. In the future tense, such an analysis would entail that the
initial moment of the event being described must follow the time of utterance. So, for
example, the sentence Fiona will sing my favourite song would express a true
proposition only if Fiona had not already begun singing my favourite song at the time
of utterance.
Given these informal descriptions of the truth-conditions of past, present and
future perfective sentences, we may introduce a perfective aspect operator, Perf,
whose formal interpretation is given in (44).
(44) The perfective: [Perf (<J>)]M-8J is 1 iff [c()]M•»•* is 1 and i contains initial and
final points if $ inherently contains these.
Initial and final points are taken to be initial and final subintervals (as defined in (45))
which are no more than a moment (i.e. intervals consisting only of a single point in
time). The requirement for initial and final points applies only to predicates that
inherently contain them. This is to accommodate stative predicates like be good or like
which do not necessarily have beginning or end points. For example, the sentence
252
Simple aspect
Bertie is good expresses a true proposition, if Bertie is actually good at the moment
of utterance. Its truth does not, however, depend on Bertie* s ceasing to be good at this
point. The proposition is true, even if Bertie continues to be good until the end of his
life. Consequently, bounding points are only taken into account in situations (like
playing a game of golf or singing a song) where these are inherently required by the
meaning of the predicate. For those events, like being good, which do not necessarily
have beginning or end points, the perfective reduces to the simple assessment of the
truth of the formula at an interval. We will return to the inherent aspectual properties
of predicates in the next section.
(45) a. Initial subinterval: i is a initial subinterval of j , iff i is a subinterval of j and
there is no t € j - i and t' e i such that t < t \
b. Final subinterval: i is a final subinterval of j iff i is a subinterval of j and
there is no t e j - i and t' e i such that t' < t.
The definition of the perfective in (44) gives rise to the truth-conditional
interpretation in (46) of the formula Past (Perf(3x [book '(x) & write '(x)(jo')])) which
translates the sentence Jo wrote a book. This formal description of the truth-
conditions of the formula translating this sentence, given in (46), guarantees that it
is true only in circumstances in which Jo completed a book some time prior to the time
of utterance. It would not, however, be true if, for example, Jo had begun a book, but
not actually finished it, i.e. if the end point of the action is not before the time of
utterance.
(46) a. [Past (Perf (3x [3x [book'(x) & (write\x))(jo')]]))] *«•** is 1, iff there is an
interval j such that j < i and [Perf (3x [book'(x) & (write'(x))(jo')])]M8-8<J is
1.
b. [Perf (3x [book' (x)&( write' (x))(jo')])]M8-8J is l,iff j includes beginning and
final points and [3x [book'(x) & (write'(x))(jo')]]M8«J is 1.
c. [3x [book'(x) & (write'(x))(jo')]]M8«J is 1 iff there is a value assignment g'
exactly like g but perhaps for the value of x, such that [book'(x)]M8«'J is 1
and [(write'(x))(jo')]M8-8'Jis 1.
While the perfective describes an event as complete, the imperfective aspect
describes it as on-going or incomplete. To represent this aspect, we introduce the
semantic operator, Impf, which is interpreted as requiring the event being described
by the formula in its scope to be true of an interval of time properly containing the
time at which the imperfective formula is being assessed. So, for example, the
sentence Jo is kicking Chester is translated into the formula Pres (Impf
((kick''(chester '))(jo'))) which is true with respect to an interval i, only if there is a
larger interval, j , containing i, such that (kick'(chesterf))(jo') is true at j . In plainer
English, this means that the event described by an imperfective formula must go on
around the time with respect to which it is to be interpreted. The formal definition of
this appears in (47).
(47) The Imperfective: [Impf (<|>)]M.8J is 1 iff there is an interval j such that i is
a proper subinterval of j and [§]M&i is 1.
253
To see the effects of this definition, consider the interpretation of the more complex
sentence in (48). This expresses a true proposition if there is some time prior to the
time of utterance at which someone called Jo enters and this event occurs in an
interval, longer than that in which Jo enters, when someone called Fiona sings.
(48) Jo entered, while Fiona was singing.
If the model M8 is extended to include denotations for the verb enter as in (49) and
while is translated simply as a form of conjunction, then the truth of the formula
translating (48) uttered at interval [t5] with respect to M8 is determined as in (50).
(49) a. F8(enter')([t2,t3]) = {MAN1? WOMAN,}.
b. F8(enter')([t1]) = F8(enter')([t3,t9]) = 0 .
(50) a. [Past (Perf (enter'(jo')) & Impf (sing'(fiona')))]M8-8<t5J is 1 iff there is an
interval, j , such that j < [t5] and [Perf (enter'(jo')) & Impf
(sing'(fiona'))]M8-8Jis 1.
b. Let j = [t2,t3], then [Perf (enter'(jo')) & Impf (sing'(fiona'))] MS^CO] [S I iff
[Perf (enter*(jo'))]M8'gtt2t3J is 1 and [Impf (sing'(fiona'))]M8*112-01 is 1.
c. [Perf (enter'(jo'))]^-*^'13! is 1 iff [t2,t3] contains initial and final points for
[enter'(jo')]M88it2t3i and [enter*(jo*)]M8*[tW3i is 1. Since [t2,t3] does contain
beginning and end points for [enter'(jo')]M8stt2t3i and [enter' CJo*)]M8*#It2*t311S
1 (both from F8(enter')([t2,t3]) in (49.a)), [Perf (enter'Go'))] ******* is 1.
d. [Impf (sing' (fiona'))] M8<g ft2«] is 1 iff there is an interval of time, k, such that
[t2,t3] is a proper subinterval of k and [sing'(fiona')]m** is 1.
e. Let k = [ti,tj.
f. F8(sing')([t!,t2]) and F8(sing')([t3,t8]) both contain WOMAN2, the value of
[fiona' ]M8,g.[ti,t9] ^ d [sing']M88ttlt8J also contains WOMAN2, since the interval
[ti,t8] overlaps with both of these intervals. Because sing is an activity verb
we can infer that [sing'(fiona')]1*8*1 is the same for all subintervals 1 of [t^tg]
(see next section). Hence, [sing'(fiona')]M8«ftlt5i is 1.
g. Since [t2,t3] is a proper subinterval of [UM and [sing'(fiona')]M8«[tlt5i is 1,
[Impf (sing'(fiona'))]™*<w is 1.
h. Since [Perf (enter'(jo'))]MS*^]is 1 and [Impf (sing'(fiona'))]M8*it2t3i is 1 and
[t2,t3] < [tj], [Past (Perf (enter'(jo')) & Impf (sing'(fiona')))] m *w is 1.
We can represent this situation more succinctly using a timeline. The diagram
in (51) shows the time of utterance, [ts], as i, the time of Jo's entry, [ t ^ ] as j and the
period of Fiona's singing, [ti,tj, as k. It is not necessary that k, the period of Fiona's
singing, be prior to the time of utterance, merely that j should be. This is generally
true of all imperfective formulae: the tense requirements need only be met by the
proper subinterval at which the truth of the formula is assessed, not by the wider
interval at which the base formula is true. Intuitively this is the correct result, as we
do not want to say that the proposition expressed by (48) is false if Fiona is still
actually singing at the time of utterance.
254
Simple aspect
(51)
Although the truth-conditions given to the imperfective in (47) properly locate

the principal reference time inside the time over which some event is occurring, it
suffers from a serious deficiency. The problem is that if an imperfective sentence like
Fiona was singing my favourite song expresses a true proposition, according to (47),
then there must be some interval of time at which the perfective aspect of the sentence
Fiona sang my favourite song is also true with respect to some reference time. The
entailment relation in (52) is valid with respect to the model theory containing (47),
as can be checked by showing that no situation can be constructed in which, for
example, Impf (Past (singf(fwnaf))) is true at some time i, but there is no other
interval j in the model at which Perf (Past (singy(fiona'))) is true.
(52) If [Impf (c|))]M** is 1, then [Perf (<fr)]M*i is 1 for some interval j .
However, although it may well be the case that the conditional in (52) holds for many
situations, it does not always do so, as can be seen by considering the sentences in
(53), where the truth of (53.b) does not necessarily follow from that of (53.a). It is
possible, for example, that Fiona has never sung my favourite song before, but was
singing it when she was electrocuted and thus never actually completed it. In such a
situation, because Fiona did not complete the singing of the song, the formula Perf
((sing,'(a))(flona')) (where a stands for the translation of the song for convenience)
is not true of any interval of time.
(53) a. Fiona was singing my favourite song, when she was electrocuted by a faulty
microphone,
b. Fiona once sang my favourite song.
The reason (52) is valid in LT is because the imperfective is defined as occurring
if (and only if) the primary reference time is a proper subinterval of an interval of time
over which some event took place. For lmpf(§) to be true at i, therefore, there must
be an interval j properly containing i, at which <j) is true. This problem is often referred
to as the imperfective paradox and has led to a considerable debate about the proper
meaning of the imperfective aspect and to suggestions for revising (47) to get around
the problem. Discussion of alternative suggestions is not possible here, but I have
drawn the reader's attention to the problem to show that analyses that appear to
account for some part of the meaning of some grammatical construction may have
unforeseen consequences. The advantage of formalisation is that such consequences
can be made plain and steps taken to rectify the problem, while trying to maintain the
original insight. Without formalisation, it is possible for undesirable consequences to
remain hidden and to lead to further problems elsewhere in the theory.
255
Exercise 8.6:
Let us add the following to the model M8, given above:
a. two entities S O N d and SONG2 are added to A.
b. F is extended as follows:
F 8 (song') = {<[t 1 ,t 9 ],{SONG 1 ,SONG 2 }>},
F8(singt') = {<[t1,t2],{<WOMAN2,SONG1>}>,
<[t3,t5],{<MAN1,SONG1>, <WOMAN2,SONG1>, <WOMAN1,SONG1>}>,
<[Ws],{<MAN1,SONG2>, <WOMAN2,SONG2>, <WOMAN1,SONG1>}>,
<[t9],{<MAN1,SONG2>, <WOMAN 2 3ONG 2 >, <WOMAN1,SONG1>,
<MAN2,SONG!>}>}
Assume that sing/ (indicating the transitive use of the verb sing) inherently contains
initial and final points and that these are provided by the intervals specified by F 8 , but
that sing' (indicating intransitive use) does not. Work out the truth values of the
following formulae with respect to this model and a moment of utterance [t 5 ].
i. Past (Perf (3x [(singt'(x))(jo')])).
ii. Fut (Perf (3x [(singt'(x))(ethel')])).
iii. Fut (Perf (3x [sing'(x)])).
iv. Past (Perf (~(3x [sing'(x)]))).
v. Pres(Impf(sing'(jo'))).
vi. Fut(Impf(3y [3x [
8.4.2 States and actions

One of the reasons why studies of aspect (both formal and informal) become so
complex is that the semantics of individual words, or groups of words, within a
sentence contributes almost as much to the aspectual interpretation of that sentence
as any morphosyntactic marking of grammatical aspect. We have already come across
this in our discussion of the perfective, where the notion of beginning and end points
was required to account for the completeness property for predicates that have
inherent boundaries, like build a garage, etc. We have also seen that the extensions
of stative predicates (like like, know, be good) at some interval i are constant at all
subintervals of i. The study of the different aspectual classes of predicates has a long
history, going back at least to Aristotle in the fourth century BC. Aristotle's original
classification has been the source of revision and debate by philosophers ever since
and now the subject has been taken up, with somewhat different emphasis, by
linguists. The literature on this complex topic is vast and we can do very little in this
book but look briefly at one particular classification, that of Zeno Vendler, and some
of the entailments associated with the classes he proposed. These properties are not
strictly logical in the sense that the semantic theory can derive them as theorems.
Rather they have to do with the lexical meanings of expressions and so must be stated
separately from the model theory itself. As we saw in Chapter 7, this may be done by
the use of meaning postulates which specify the idiosyncratic semantic properties of
predicates as part of the lexicon of some language.
256
Simple aspect
The basic distinction that can be made between predicates is that between states
(or stative predicates) and actions. Syntactically these two classes can be
distinguished (in English, at any rate; these syntactic properties do not necessarily
carry over into other languages) by the facts that stative predicates do not easily go
into the progressive (54.a) unlike actions (54.b); imperatives usually have an action
reading, even where stative verbs are used (55); and the pseudo-cleft construction is
more natural with action verbs than stative ones (56).
(54) a. ?Ethel is knowing the answer,
b. Ethel is playing golf.
(55) a. ?Know the answer!
b. Play golf!
(56) a. ?What Ethel did was know the answer,
b. What Ethel did was play golf.
We have already seen the primary semantic property of statives: the extensions of
such predicates at one interval remain the same for all subintervals, including
moments. We can capture this property by the meaning postulate in (57) which
guarantees that sentences like that in (58) sound tautologous.
(57) If f is a stative predicate, then if [f(a)]M*s is true, then so is [f(a)]M«J for all
subintervals j of i.
(58) If Jo was tall for his age between the ages of two and four years old, then
Jo was tall for his age at the age of three.
Actions are subdivided by Vendler into three categories: activities,
accomplishments and achievements. The first group includes predicates denoting
general activities like running, walking, driving a car, singing, etc. These differ from
achievements (e.g. recognise, find, lose, die, etc.) and accomplishments (e.g. sing a
song, paint a picture, write a book, etc.) in having no inherent boundaries, i.e. no
necessary initial and final points to the events they describe. There is nothing inherent
in the meaning of the word sing that requires the action to have a beginning and an
end. After all the sentence The gods are singing could express a proposition that is
true at all times, past, present and future.
The main entailment of activity predicates involves a relation between their
imperfective and perfective uses. Thus, if the imperfective use of an activity predicate
is true, then its perfective (in the same tense) is also true. For example, if the dog was
howling is true now, then the dog howled is also true. This is captured by the meaning
postulate in (59). Because activities do not have inherent boundaries the perfective
formula reduces truth-conditionally to the simple base formula, i.e. [Perf (<|>)]M'84 =
[c|)]M,8,i for a n formulae, <|), containing activity predicates.
(59) If f is an activity predicate, then if [Impf (f(a))]M<& is true, then so is [Perf
(f(a))]M*o
A further entailment involves the truth of formulae containing activity predicates at
subintervals. In general, it seems to be the case that if Jo walked for an hour then he
walked for all periods in that hour, apart from individual moments of that hour
257
(walking seems to have to be defined over periods greater than a moment, since a
snapshot of someone walking shows no movement by the walker). We can symbolise
the durational part of the activity asfor'(NPtemp) (a predicate modifier) where NPtemp
stands for some temporal NP (which we may allow directly to denote intervals of
time, although this poses certain problems). Thus, we arrive at the meaning postulate
in (60).
(60) If f is an activity predicate, then if [for'(NPtemp)(f(a))]M-8i is 1 where
[NPtemp]^1 = j , then so is [f(a)]M<8\ for all subintervals k of j that are larger
than a moment.
Finally, we must look at achievements and accomplishments. In the first place,
the meaning postulates in (59) and (60) do not hold of these. One cannot infer from
the truth of Fiona was singing my favourite song, that Fiona sang my favourite song
is true: she may have been interrupted somehow and never finished the song. Nor can
one assume that if Bertie wrote a book for an hour then he wrote a book at any time
in that hour. Achievements differ from accomplishments in that they appear to be
momentaneous events. Hence, durational phrases sound peculiar, if not totally
ungrammatical, as witness the achievement predicates in (61) versus the
accomplishments in (62).
(61) a. ?Jo recognised the dog for a few minutes,
b. ?The dog died for a few hours.
(62) a. Jo sketched the dog for a few minutes,
b. The dog barked for a few hours.
Furthermore, accomplishments which are complete in a particular period enable
us to infer that the imperfective form of the accomplishment is true for all subintervals
of that period, excluding final subintervals. Thus, if in'(NPlemp) is the logical
representation of phrases such as in an hour, the meaning postulate in (63) guarantees
the truth of (64) in all normal situations.
(63) If f is an accomplishment predicate, then if [in'(NPtemp)(Perf (fla)))]***1 is
true, where [NPtemp]1^1 = j , then [Impf (f(a))]M'*k is also true for all
subintervals k of j excluding final subintervals.
(64) If Jo painted a picture in an hour, then Jo was painting a picture during that
hour.
There is a lot more that can be said on this subject and there are more subtle
differences between the different classes of predicates that can be teased out, but this
brief sketch should give some idea of the range of variation in entailments between
them and how these may be captured formally in the semantic theory described in this
book.
258
Scope ambiguities
8.5 Scope ambiguities

So far we have been looking at tense and aspect from the purely semantic point of
view. Translations of English sentences have been given without any attempt to
specify the rules for deriving them in a compositional manner and reliance has been
placed on the reader's ability to work out the relation between the sentences in the
object language and the corresponding formulae in LT. This relation should be clear
enough, but specifying the translation rules from English into LT depends so much on
the syntactic theory that one adopts that it seems sensible at this point to dispense with
our previous practice of constructing complete and precise grammar fragments. This
should not give the impression that a translation procedure cannot be given, but
different theoretical frameworks encounter different problems and have different
advantages for providing a precise translation procedure. Hence, a procedure worked
out for one syntactic theory is not necessarily going to provide insights into the
procedure for another. Although no attempt will be made to give a comprehensive
translation algorithm for all the constructions discussed above, this chapter ends with
a discussion of one problem that arises with respect to the relative scopes of the tense
(and aspect) operators and any quantifiers contained in subject phrases. This problem
is analogous to that of quantifier scope ambiguities discussed in Chapter 6 and will
have to be solved by the translation procedure for any theory.
To see the problem in more detail, let us update the grammar fragment of
Chapter 6 to incorporate simple present and past tense verbs (i.e. ignoring entirely
future tense and aspect). Using a binary feature PAST, the basic sentence rule is
revised to become (65.a) and all the VP rules given in previous chapters are now
assumed to allow a verb to be marked + or - PAST, according to whether the VP node
that dominates it is + or -. There are two translations (65.a) depending on the value
of the feature [PAST] which are given in (65.b) and (65.c). (Although this seems to
be a violation of the rule-to-rule hypothesis, it is not, if different feature values in a
syntactic rule are defined as givingriseto different rules.) The effect of the translation
in (65.b) is to accommodate the present and future readings of the English present
tense as illustrated in (40), above. Given these rules, a sentence like Every student
passed an examination receives the partial analysis in (66). After applying lambda
reduction to the translation in (66.c) we get the formula in (67) which has the
truth-conditions in (68). These guarantee the truth of (67) only in situations where all
students at some time prior to the time of utterance pass an examination at that same
time.
(65) a. S[oPAST] -> NP VP[ocPAST].
b. S[-PAST] => [Pres (NP'(VP)') v Fut (NP'(VP'))].
c. S[+PAST] =f> [Past (NP'(VP'))].
(66) a. passed an examination
=> Xy [Bz [examination'(z) & (pass'(z))(y)]].
b. every student
=> XQ [Vx [student'(x) -> Q(x)]].
c. every student passed an examination
=> Past(AQ [Vx [student'(x) -> Q(x)]]
(Xy [3z [examination'(z) & (pass'(z))(y)]])).
259
(67) Past (Vx [student'(x) -> 3z [examination'(z) & (pass'(z))(x)]]).

(68) a. [Past (Vx [student'(x) -> 3z [examination'(z) & (pass'(z))(x)]])]M88-i is 1
iff there is a time j < i such that [Vx [student'(x) —> 3z [examination'(z) &
(pass'(z))(x)]]] M ^Jisl.
b. [Vx [student'(x) -> 3z [examination'(z) & (pass'(z))(x)]]]M8-gi is 1 iff for
every substitution a e A for x [student'(x) -> 3z [examination'(z) &
(pass'(z))(x)]]M8-8a/xoisl.
c. [student'(x) -> 3z [examination'(z) & (pass'(z))(x)]]M8'«a/x,j is 1 iff either
[studentXx)]* 18-^ is 0 or [3z [examination'(z) & ( p a s s ' ^ X x ) ] ] * 1 8 ^ is 1.
d. [3z [examination'(z) & (pass'(z))(x)]]M8«a/xJ is 1 iff there is a substitution b
e A for z such that [examination'(z) & (pass'(z))(x)]M88[a/xib/zo is 1.
e. [examination'(z) & (pass'(z))(x)]M8 ^ b / z o ls \ iff [examination'(z)]* 18 ^ 1 ^
is 1 and [(pass'(z))(x)]M8'8fa/xib/zo i s 1.
The crucial point for our purposes is seen in (68.c) which shows that on this
reading the entities passing the examination are also students at the time of passing.
This is because student'(x) and 3z [examination y(z) & (pass '(z))(x)] are both assessed
with respect to the interval j , in the past with respect to the time of utterance.
However, there is another possible interpretation of this sentence where the entities
passing the examination in question are not necessarily students at this past time, but
are students at the time of utterance. For example, if the examination was an entrance
exam to a university, the people passing it would in general not be students at the time
of passing. To get this second reading, we can use the rule of quantifying in
introduced to handle quantifier scope ambiguities in Chapter 6. Using this rule to
postpone the combination of the translation of the subject phrase until after the tense
operator has been introduced yields the translation for Every student passed an
examination in (69.a) which reduces to (69.b). (69.a) is derived by supplying A,P[P(y)]
as the translation of the subject, then applying the tensed sentence rule, followed by
quantifying in the translation of every student, as readers can check for themselves.
(69) a. XQ [Vx [student' (x) -> Q(x)]]
(ky [Past (XPP(y)(Xx,[3z [examination'(z) & (pass'(z))(x,)]]))]).
b. Vx [student'(x) -> Past (3z [examination'(z) & (pass'(z))(x)])].
The truth-conditions of (69.b) are computed as in (70) with the crucial
information appearing in (7O.a and b). These show that we are assessing the truth of
student'(x) at the time of utterance, i, while the truth of 3z [examination'(z) &
(pass'(z))(x)] is assessed at the past interval, j .
(70) a. [Vx [student'(x)-> Past (3z [examination'(z) & (pass'(z))(x)])]]M8'8jis 1
for every substitution a e A for x [student'(x)]* 18 -^ is 0 or [Past (3z
[examination'(z) & (pass'(z))(x)])]M8i8a/xi is 1.
b. [Past (3z [examination'(z) & (pass'(z))(x)])]M8-8a/xi is 1 iff there is a time j
prior to i such that [3z [examination'(z) & (pass'(z))(x)]]M88a/xJ is 1.
c. [3z [examination'(z) & (pass'(z))(x)]]M88a/xJ is 1 iff there is a substitution b
G A for z such that [examination'(z) & (pass'(z))(x)]M88fa/x^b/zJ is 1.
d. [examination'(z) & (pass'(z))(x)]M8•giî^.j is 1 iff [examination'(
is 1 and [(pass'(z))(x)]M8â/xJb/zJ is 1.
260
Scope ambiguities
Thus, we find further scope interactions between logical operators, quantifiers

and tenses, which do appear to mirror certain genuine ambiguities in English
sentences. Ignoring any aspectual operators, the sentence every student passed an
examination turns out to be at least six ways ambiguous, according to the different
scopes of the operators Past, 3 and V. Two of these have already been discussed. The
others are shown in (71).
(71) a. Past (3z [examination'(z) & Vx [student'(x) -> (pass'(z))(x)]]).
b. 3z [examination'(z) & Past (Vx [student'(x) -> (pass'(z))(x)])].
c. 3z [examination'(z) & Vx [student'(x) —> Past ((pass'(z))(x))]].
d. Vx [student'(x) —> 3z [examination'(z) & Past ((pass'(z))(x))]].
Exercise 8.7:
Work through the truth-conditions of the formulae in (71) to show how they differ in
meaning. Are all of these readings intuitively plausible as interpretations of the
sentence Every student passed an examination?
The proliferation of scope ambiguities of this sort gives rise to a number of

questions that need consideration. In particular, it is necessary to ask whether all the
ambiguities predicted by the free application of the rule of quantifying in are plausible
for some particular sentence. As scope ambiguities involve all the logical operators,
readings involving scope differences between tenses and aspects also need to be
considered. If certain readings do not seem to represent genuine readings of a
particular sentence or set of sentences, then the limits to ambiguity need investigation
and some motivated constraints need to be put on the operation of the rule of
quantifying in to allow only those ambiguities that are felt to be valid and to reject all
others. It is not my intention to pursue these weighty matters here, but the reader
should be aware that much more research needs to be carried out into the domain of
scope ambiguity, before we can be confident that the treatment of tenses and aspects
as logical operators is the correct one.
Furthermore, one has to ask whether the rule-to-rule hypothesis can be
maintained, given that any construction containing more than one logical operator has
more than one output for any single input. Whether or not the rule-to-rule hypothesis
should be abandoned depends on whether one's syntactic theory can provide
alternative syntactic analyses, with the same output, mirroring the different semantic
readings. Again, this topic will not be pursued here, bound up as it is with theory
particular considerations. Scope ambiguities do, however, cast doubt on the validity
of the rule-to-rule hypothesis (although not the principle of compositionality) as
holding for the full range of semantic phenomena found in human languages.
261
8.6 Further reading

General linguistic discussions of tense and aspect can be found in Comrie (1985),
Comrie (1976), Lyons (1977: 677-690; 703-718), and Dahl (1985) which all have
good bibliographies. For discussions of tense logic with respect to moments of time
rather than temporal intervals see, inter alia, Lyons (1977: 809-823), McCawley
(1981: 340-359), Reichenbach (1947), Prior (1967), Rescher and Urquhart (1971),
Gunthner (1976), Dowty, Wall and Peters (1981: 112-121), and Chierchia and
McConnell-Ginet (1990: 223-1234). The suggestion that temporal intervals are more
useful for the analysis of natural languages is proposed in Bennett and Partee (1972)
and developed subsequently in Bennett (1977), Dowty (1979: ch. 3), Aquist (1976),
Baiierle (1979), Cresswell (1985a: ch. 3), and many others. The classification of verbs
in terms of their aspectual properties is the subject of many books and articles, among
which see especially Vendler (1967), Dowty (1979: ch. 2), Mourelatos (1978) and
Verkuyl (1972; 1989). Discussions of tense and aspect can be found in many
collections of papers, among which see, for example, Rohrer (1980) and Tedeschi and
Zaenen (1981). A critique of interval semantics can be found in, for example, Tichy
(1985) (but see also the reply in Cresswell (1985a: 85-95))). A considerable amount
of discussion of aspect currently takes place with reference to the notion of events
which are considered to be ontological primitives like the individual entities
encountered in this book. Event theory owes its current popularity to Davidson
(1967) (and see also the other articles in Davidson (1980)); some examples of this
approach can be found in Dowty (1989), Bach (1986), ter Meulen (1983; 1984) and
Parsons (1989) which gives a good discussion of the Imperfective Paradox and
presents a solution.
262
Possible Worlds
9.1 Where entailments fail

In previous chapters (particularly Chapter 7), we looked at certain types of entailment
relations that are guaranteed by the theory of interpretation set out in the earlier part
of this book. Certain contexts exist, however, where expected entailments do not hold.
Consider, for example, the inference pattern in (1).
(1) a. The Morning Star is the planet Venus.
b. The Evening Star is the Morning Star.
c. Therefore, the Evening Star is the planet Venus.
The validity of this inference pattern illustrates a general rule that holds in the
extensional semantic theory developed in Chapters 2 to 6 of this book. This rule is
called Leibniz's Law or the Law of Substitution and it allows the substitution of
extensionally equivalent expressions for one another in a formula while maintaining
the truth value of the original formula. Thus, in (1), since the Morning Star and the
Evening Star denote the same entity, the latter expression may be substituted for the
former in the first premiss to give the conclusion. Indeed, because all three terms in
(1) have the same extension all of them may be substituted for each other salva
veritate (the Latin phrase used by Leibniz meaning 'with truth unchanged'). The Law
of Substitution can be formally defined as in (2) which, in words, says that if an
expression a is extensionally equivalent to another expression b, then a formula (>| is
truth-conditionally equivalent to the formula formed from <j) by substituting an
instance of b for every instance of a.
(2) (a = W->[4><->4>*].
That substitution of this sort does really hold in the semantic theory (without tense)
developed up to Chapter 7 can be shown by trying to show that <)| could be true while
<|>b* is false for any extensionally equivalent expressions a and b. No formal proof will
be attempted here, but intuitively it should be fairly clear to the reader that the formula
in (2) does indeed hold, because truth-conditions are defined in a model in terms of
entities and functions over entities and not in terms of the names of entities or
functions. Hence, the truth of a formula depends on the extension of an expression in
a model and not on the expressions themselves. If two expressions, say morning-staf
and evening-star \ denote the same thing, say the entity VENUS, then the truth of a
formula containing either of these expressions depends only on whether the entity
denoted, i.e. VENUS, is mapped onto 1 by the function denoted by some predicate,
e.g. [Xx [x = venus']]M s (the set of all entities identical with Venus).
The Law of Substitution in (2) does not, however, hold in all linguistic contexts.
Consider the inference patterns in (3), (4) and (5).
263
9 Possible worlds
(3) a. Necessarily, the Morning Star is the Morning Star.
b. The Morning Star is the Evening Star.
c. Therefore, necessarily, the Morning Star is the Evening Star.
(4) a. Bertie believes the Morning Star is the Planet Venus.
c. Therefore, Bertie believes that the Evening Star is the planet Venus.
(5) a. Bertie wants to look at the Morning Star.
c. Therefore, Bertie wants to look at the Evening Star.
None of these inference patterns is intuitively valid, despite the fact that they involve
the substitution of extensionally equivalent expressions in a formula. The inference
pattern in (3) fails because, while it is (logically) necessary that something is identical
to itself, it is not necessary for two different names to have the same extension. We
can imagine a possible state-of-affairs in which the Morning Star is the planet Venus,
but the Evening Star is, for example, the planet Jupiter. The fact that the proposition
expressed by the sentence The Morning Star is the Evening Star is contingently true
is not sufficient grounds for asserting that it is necessarily true. Contingent formulae
are those whose truth value could be otherwise than they are whilst non-contingent
formulae are either always true or always false and whose truth or falsity is
determined by their logical structure.
The invalidity of the inference patterns in (4) and (5) depends on the attitudes and
beliefs of the referent of the subject of the main clause. The person called Bertie could
well believe that the Morning Star is the planet Venus, but not believe that the
Evening Star and the Morning Star are the same thing. Thus, while the second premiss
in (4) is true, it is true outside Bertie*s beliefs and so cannot be used to construct new
statements about what Bertie believes from the original premise. In Bertie's system
of beliefs, the Evening Star could be the planet Jupiter and not Venus at all. Similar
reasoning argues against the validity of (5). Here, Bertie may easily want to see the
Morning Star and so be intending to get up at dawn, but this is not the same as
wanting to see the Evening Star at dusk. He may already have seen the Evening Star
and merely wants to catch a sight of its manifestation as the Morning Star. Notice that
here the failure in substitution in (5) may also persist even if Bertie knows that the
Evening Star and the Morning Star are extensionally equivalent. What is important is
a particular manifestation of some entity, the planet Venus.
Sentences containing verbs like want and believe exhibit another interesting
property. Consider the sentences in (6).
(6) a. Jo believes that the Ladies' Scottish Golf Champion is rich,
b. Jo believes that Ethel is poor.
In general, it is, of course, possible for these two sentences to express true
propositions at the same time. If, however, it is true that Ethel is the Ladies' Scottish
Golf Champion, it might seem that Jo has contradictory beliefs, believing of one and
the same person that she isrichand that she is poor. However, the contingent fact that
Ethel is a golf champion does not necessarily affect Jo's beliefs and we cannot
substitute the name Ethel for the subject noun phrase in (6.a) (or vice versa) in order
264
Where entailments fail
to infer the truth of the contradictory sentence (via conjunction introduction and the
opposite relation between rich and poor) Jo believes that Ethel is rich and Jo believes
that Ethel is not rich. If Jo does not know (or believe) that Ethel is the Ladies' Scottish
Golf Champion, then (6.a) means that Jo has a belief about someone who carries the
title of Ladies' Scottish Golf Champion without knowing (or perhaps caring) who that
person is. In other words, (6.a) may be a reflection of Jo's pre-conceived ideas about
those who win golfing tournaments and have nothing to do with the actual individuals
concerned. Such belief is often said to be de dicto, a belief 'about what is said or
mentioned'. In such contexts, the referents of noun phrases within the sentential
complement of a verb like believe are, one might say, within the subject's 'belief
worlds' and do not refer necessarily to entities outside those worlds.
There is a second way that (6.a) may be interpreted, however. The assertion of
such a sentence, instead of being interpreted as a simple statement about Jo's beliefs,
may be taken as asserting of a certain individual that Jo has a particular belief about
them. In other words, the speaker may refer to the particular individual that is denoted
by the expression the Ladies' Scottish Golf Champion, who may not be known to Jo,
the subject of the belief. In these cases, the belief is said to be de re, belief 'about the
thing' referred to. In such cases, the reference of a noun phrase in the sentential
complement of believe is to an entity outside the subject's beliefs, something 'in the
real world'. Furthermore, because de re reference is outside the subject's beliefs,
substitution of extensional equivalents is valid, but, for the same reason, this does not
lead to the conclusion that Jo has contradictory beliefs (see below).
This ambiguity between de re and de dicto reference is not confined to belief
contexts, but may be found after verbs which denote mental states, acts and attitudes
(a subclass of which, including believe, doubt, consider, are traditionally called by
philosophers verbs of propositional attitude). For example, the sentences in (7)
exhibit the same ambiguity as those in (6).
(7) a. Jo does not want to meet Ethel.
b. Jo wants to meet the Ladies' Scottish Golf Champion.
In the de dicto sense, if Ethel and the Ladies' Scottish Golf Champion are one and the
same, these sentences are not contradictory. Jo may very well want to meet the Ladies'
Scottish Golf Champion (perhaps under the illusion that she is a rich lady of leisure)
but definitely not want to meet Ethel, his secretary, whose filing system he has ruined.
On ade re reading, on the other hand, the sentences in (7) are contradictory, because
the speaker is asserting of some individual (i.e. Ethel) that Jo does and does not want
to meet her. People hearing a sentence like (7.b) meant in a de dicto sense might
interpret it de re. So, for example, Fiona might utter (7.b) to Bertie and the latter might
reply Oh, he wants to meet Ethel, i.e. he supplies the actual referent of the phrase the
Ladies' Scottish Golf Champion, even though Jo may still not want to meet Ethel. We
will return to this topic in Chapter 10.
There is one further problem with the verbs we have been discussing with respect
to Leibniz' Law. The way this is defined in (3), above, allows for the substitution, not
just of entity-denoting expressions, but of any two extensionally equivalent
expressions of any type. By this law, therefore, if someone believes any true formula
at all, then they should also believe every other true formula. Thus, if someone
believes that grass is green then they must, by Leibniz' Law, also believe that the
265
9 Possible worlds
Morning Star is the planet Venus, that 29 is the square root of 841 and that there is a
pink water lily in flower in my garden pond! This patent absurdity follows from the
fact every true formula denotes the value 1, so that any sentence that translates into
a formula with this value may be substituted for any other sentence with the same
value. This is clearly not true and verbs like believe are not truth-functional in their
complement position, because the truth of the whole proposition expressed by the
sentence does not depend on the truth (or falsity) of the proposition expressed by the
embedded sentence.
It is not only the law of substitution that is problematic in the above contexts.
Scope effects involving the logical quantifiers may also fail to hold. Consider the
inference pattern in (8).
(8) a. Jo met a secretary.
b. Therefore, there exists a secretary that Jo met.
As we saw in Chapter 6, the decomposition rule for transitive verbs ensures that any
quantifier in an object noun phrase has scope over the verb. Thus, in LQ, (9.a) provides
the translation of the premiss in (8) which is, of course, also the translation of the
conclusion of the inference pattern, hence ensuring that the truth of the premiss
guarantees the truth of the conclusion. From conjunction elimination applied to (9.a),
we can also conclude that a secretary exists (9.b) and that someone exists whom Jo
met (9.c). In other words, the interpretation given to the translation of the premiss in
(8) ensures that there exists someone specific whom Jo met, provided that the premiss
is true.
(9) a. 3x [(meet' (x))(jo') & secretary' (x)].
b. 3x [secretary'(x)].
c. 3x
After verbs of needing, wanting, seeking, etc., however, this type of entailment
does not necessarily follow. Thus, in (10), none of (lO.b) to (lO.d) necessarily follow
from(lO.a).
(10) a. Jo wanted to meet a secretary.
b. There exists a secretary that Jo met.
c. There exists a secretary.
d. There is someone whom Jo met.
If Jo met a secretary, then there is a specific secretary that Jo met, but if he wanted
to meet a secretary, then it does not follow that there is a specific secretary that he
wanted to meet. He may have in mind a particular individual secretary, but not
necessarily. He may just want to meet someone who is a secretary, without caring
which particular one. Indeed, (lO.a) could be true even if no secretaries exist, a
situation that is made more obvious in the sentence Jo wants to meet a unicorn which
can clearly express a true proposition without there being any unicorns for someone
to meet. Thus, certain verbs allow a non-specific reading for their objects (including
objects of subordinate non-finite verbs), because there is no particular entity with the
property defined in the noun phrase that is the object of the main predicate. In effect,
then, these verbs may prevent an existential quantifier in an object phrase from having
scope over the whole sentence. Again there is ambiguity, because a specific reading
266
Intension and extension
of the complement is possible. For example, Jo may have in mind a particular

secretary who has been recommended to him as being particularly efficient, although
he does not know exactly who this is. In this case, not every secretary will satisfy Jo's
wanting to meet them, but only the specific one who has been recommended for their
efficiency.
9.2 Intension and extension

The contexts in which straightforward entailments fail are often referred to as
referentially opaque (or oblique) contexts, opposed to the referentially transparent
contexts dealt with in Chapters 2 to 6. In order to make our semantics more adequate
than it obviously is, we need to find some way of being able to distinguish between
the objects of verbs like want, lookfor and those of verbs like meet, find. We also need
to decide what complement clauses should denote, since, as we have seen, they cannot
just denote truth values, but must still be associated with truth or falsity. The
philosopher Gottlob Frege was the first to tackle this problem directly. He proposed
that a distinction should be made between what he called the reference (German
Bedeutung) of an expression and its sense (German Sinn). The reference of an
expression is what we have been calling its extension, that is the entity or function that
an expression refers to in any model. Its sense, on the other hand, is more abstract,
more like the 'meaning' of the expression. We may think of the sense of an expression
as being the thing that each of the elements in the reference of the expression have in
common that cause them to be identified as such. As an illustration, the reference of
the English word red is, according to Frege, the extension as defined in previous
chapters, i.e. the set of all red entities. Its sense, however, is the property of being red,
the 'redness' of all these entities abstracted away from the individuals themselves. In
order to avoid confusion with the notion of reference as the act of picking out entities
in the world and that of sense as the meaning relation between expressions in a
particular language (see Chapter 7 and Lyons (1977:177ff)X it is better to follow the
terminology of Carnap (1947) and use, as we have been doing, the term extension for
Frege's Bedeutung and intension for Sinn.
Some may think of intensions as psychological concepts, but here the view is
taken that they are external to human beings (in the world, so to speak) and so publicly
accessible in some way. In either case, both intension and extension are intended to
refer to things in the world and so constitute different aspects of the denotation of an
expression, according to the definition of this concept in Chapter 1. This distinguishes
both terms from sense, which is, from the linguistic point of view, language internal
and defines the meaning relations between words and expressions of a particular
language, as we saw in Chapter 7. We may thus picture the 'meaning' of an expression
as minimally constructed from its sense and its denotation, as shown in the diagram
in (11). (The role of sense in this system is briefly discussed in the next chapter.)
267
9 Possible worlds
(11)
Meaning
Sense
Extension Intension
Making a distinction between extensional and intensional denotation allows a

distinction to be made between referentially transparent and referentially opaque
contexts. In the former case, a predicate is interpreted as a function over the
extensional type of its argument, in the way familiar from previous chapters.
Predicates that create opaque contexts, however, can be interpreted as denoting
functions that take the intensional type of their arguments as their domain. Thus, while
the verb meet denotes a relation between two entities, the verb look for denotes a
relation between an entity, the searcher, and the intension (or concept) of another. In
this way, we can distinguish the meanings of the sentences Bertie is looking for an elf
and Bertie is looking for a hobbit. The expressions elf and hobbit, we assume,
extensionally denote the null set in the actual world, and hence the noun phrases an
elf and a hobbit also both denote the null set (since no set has a non-null intersection
with 0). Their intensions, on the other hand, differ, since the property that identifies
an entity as an elf differs from that which identifies a hobbit, even though there are no
actual elves or hobbits picked out in the real world and so if Bertie actually found one
he would be able to identify it as such. Similarly, verbs like believe do not denote
relations between entities and truth values, the extensions of sentences, but relations
between entities and the intensions of sentences, which may be informally interpreted
as the propositions they express. To believe that the Morning Star is the planet Venus
is, therefore, not the same as believing that the Evening Star is the planet Venus
because the propositions expressed by these two sentences differs. In the first case, the
proposition expressed by the sentence is true if something called the Morning Star
(which is visible at dawn) is the same as the planet Venus whilst the second is true
if something called the Evening Star (which is visible at dusk) is the same entity as the
planet Venus. Because the intensions of these sentences differ, Jo can 'stand in a belief
relation' to one and not the other without contradiction, even if it is contingently true
that the Evening Star and the Morning Star extensionally denote the same entity.
Furthermore, by appealing to the notion of intension, we can salvage the
Principle of Compositionality. If the denotation of an expression can only be
extensional, then the examples in Section 9.1 constitute a direct refutation of this
principle, because the meaning of the whole does not rely on the extensions of the
complements of the main verbs. However, for the purposes of compositionality, we
can equate 'meaning' with intension and redefine the principle as in (12).
(12) Principle of Intensional Compositionality: The intension of a complex
expression is a function of the intensions of its parts.
268
Introducing other worlds
Frege attributed another important property to his concept of intension (Sinn) which
is stated as the Principle ofIntensional Reference in (13). This captures the traditional
principle that intension (the defining property of some class) determines extension (the
membership of a class).
(13) Principle of Intensional Reference: There is a relation between intensions
and extensions such that each intension determines a unique extension.
The idea that an intension determines a unique extension for some expression allows
the concept to be defined in terms of a function. The intension of an expression is the
property that unifies all the members of its extension, and so it enables the
identification of the extension to take place. Hence, an intension may be thought of
as a function that picks out an extension for any possible state-of-affairs. For example,
the intension of the word red can be considered to be the function that identifies the
set of red entities in any situation, i.e. a function from situations to extensions.
Before we go on to look at intensionality in more detail in Chapter 10, however,
we will first extend our semantic analysis to cover the simple modal adverbs
necessarily and possibly. The introduction of modality leads to an introduction into
the model of a set of possible worlds and lays the foundations for a discussion of
Montague's formal approach to intensionality and intensional contexts.
9.3 Introducing other worlds

Human beings are capable of imagining situations that are not actual and all languages
have the ability to refer to such states-of-affairs. Thus, while none of the sentences in
(14) (presumably) express true propositions on 12 August 1990, we can interpret them
well enough and imagine states-of-affairs of which they are true. Furthermore, as (15)
shows, English has many ways of presenting non-actual situations for the purposes of
conveying information. (15.a) contains a modal verb indicating the possibility of the
truth of the situation described by the non-modal counterpart of the sentence, (15.b)
illustrates a (slightly archaic) construction that presents the information of its modified
sentence as untrue but desirable while (15.c) exemplifies conditionals whose
component clauses are presumed to be untrue. Like the other constructions in (15),
such counterfactual conditionals rely on our ability to entertain the idea that
situations may be different from what they actually are (or are assumed to be).
(14) a. Richard Montague was thefirstperson ever to win a Nobel Prize for Logic.
b. Hitler's postwar pact with Japan came unstuck when Hirohito invaded the
USA, Germany's closest ally.
c. A team of robots from Mars began to clean up Earth's poisoned atmosphere.
(15) a. The depletion of the Ozone Layer might be caused by the excessive use of
aerosol sprays.
b. Would that I had never agreed to write this book!
c. If Richard Montague had been a linguist, he would never have written PTQ.
It is in the notion of alternative situations that the formalisation of the concept
of intension lies. This concept has already been used in this book in the representation
269
9 Possible worlds
of the meanings of the propositional connectives. As we saw in Chapter 3, the truth-

conditional meanings of the connectives, &, v, ~, -», <-», can be represented by truth
tables, each line of which represents a different situation with regard to the truth or
falsity of one or more formulae and provides the truth value in that situation of a
complex formula formed from these using a connective. Logicians tend to talk, not of
different situations, but of possible worlds, a term introduced by Leibniz. Just as we
may think of our own world, the actual world (the 'best of all possible worlds'
according to Leibniz) as a thing in its own right, so we may think of all possible
worlds in the same way (although possibly not as existing). In constructing intensional
models, therefore, we may introduce another set of ontological primitives, a set, W,
of worlds alongside the set, A, of individual entities and the set, T, of times, in a
temporal semantics. While Occam's Razor states that the postulation of different
constructs should be avoided, it crucially states that entities should not be posited
beyond need (entia non sunt multiplicanda praeter necessitatem). Since we need to
account for the existence of opaque contexts like those discussed in Section 9.1, and
since a purely extensional semantics using only individuals (and times) cannot explain
them, we clearly need something else to provide a semantic account of these
expressions. Unlike the postulation of a set of primitive times in temporal semantics,
however, the adoption of a set of primitive worlds is somewhat mysterious and less
intuitively appealing. Indeed, the ontological status of possible worlds has been the
subject of heated philosophical debate, but it is not my intention to go into this. If we
assume that possible worlds are theoretical constructs (whose ontological status is
similar to other mathematical constructs), then what we, as linguists, need to do is to
assess their usefulness in achieving the general aims of semantic theory that were set
out in Chapter 1. If the use of possible worlds enables an account of opaque contexts
to be given in a clear and precise fashion and helps to capture the meaning relations
that hold between different opaque sentences, then their adoption into the theory is
justified. Of course, if another way of accounting for intensional phenomena is found
that does not require the use of mysterious things like possible worlds, then, by
Occam's Razor, we should dispense with their use. However, although alternative
theories of intensionality have been suggested, none appear so far to have the scope
and simplicity of the possible worlds approach and so this is the one that will be
adopted here to provide an account of modality (and other intensional phenomena).
9.3.1 Simple modality

Some examples of modal sentences containing the modal adverbs necessarily and
possibly and the modal verbs may and must are given in (16).
(16) a. Every proposition is necessarily either true or false, but not both.
b. Bertie possibly knows that the Morning Star is the Evening Star.
c. Every proposition must be either true or false, but not both.
d. Bertie may know the Morning Star is the Evening Star.
There are a number of different types of modality and so there are a number of
different ways in which the necessity and possibility in the sentences in (16) may be
taken. Some different types of modality will be discussed in the next section, but here
we will concentrate only on logical necessity and possibility: that is to say, necessity
270
and possibility that have to do with truth or falsity within a logical system. We may
say that a formula is necessarily true (or false) in a logical sense (or alethic sense
from the Greek aletheia meaning 'truth'), if the system of logic used ensures that a
modal formula has to be true (or false), whatever model it is being interpreted with
respect to. On the other hand, possibly true (or false) formulae are just those formulae
that the logic does not define as necessarily false (or true). Thus, we may follow
tradition and divide up the set of all formulae not only into true and false but also into
contingent and non-contingent as represented in the diagram in (17) where the box
(17) Contingent and non-contingent propositions

true false
non-contingent contingent non-contingent
represents the set of all formulae.

The set of necessarily true formulae thus consists of all the non-contingent truths
while the set of necessarily false formulae consists of the non-contingent falsehoods.
Possibly true formulae are all those formulae that are not non-contingently false, i.e.
all non-contingently true formulae plus all the contingently true and false ones.
Formulae are possibly false if they are non-contingently false or contingently true or
false. This is shown in the diagram in (18) on the following page. Under this
interpretation of necessity and possibility, we may say that sentences like those in
(16.a) and (16.c) express true propositions if (and only if) the formula translating the
sentence Every proposition is either true or false, but not both is amongst the set of
non-contingently true formulae. On the other hand sentences like (16.b) and (16.d)
express true propositions if (and only if) the sentence Bertie knows that the Morning
Star is the Evening Star translates into a formula that is not amongst those that are
necessarily false.
271
9 Possible worlds
(18) Necessary and possible propositions
necessarily true possibly true
possibly false necessarily false
The fact that the interpretation of sentences expressing possibility and necessity
may be defined in terms of their non-modal counterparts implies that these two
concepts may be represented as operators like the tense and aspect operators of
Chapter 8, an analysis that also reflects the fact that the adverbs necessarily and
possibly axe sentential modifiers. We may therefore introduce two logical operators:
• translating necessarily or it is necessary that and o translating possibly or it is
possible that. We may also associate English sentences containing the modal verb
may with formulae containing o and those containing the verb must with formulae
containing • (but see Chapter 10).
Possible worlds can be used to provide a formal definition of the semantics of
these two operators. To do this, a set W of possible worlds is introduced into models,
which (if tense is ignored for the moment) thus consist of an ordered triple « A,F,W».
A is a set of entities, as before, W is a set of possible worlds and the denotation
assignment function, F, assigns an extension to each constant in every world in W. An
expression is thus interpreted with respect to a model, M, an assignment of values to
variables, g, and a possible world, wn (just as in a semantics with tense an
interpretation refers to an interval of time). Thus, if ty is a formula, then its
interpretation with respect to M, g and wn, i.e. [<|>]M*™, is the truth value of (() at w n . If
<(> is contingent, then the value of [(|)]M gwn may be different from [((>]M,g,wm where wm is
some member of W other than W,,.
We will not go into details of the translation and interpretation procedure for
formulae in models with possible worlds at this point, but in order to illustrate how
this is done let us consider the interpretation of the formula laugh '(fiona') with respect
to a very simple model, M9. In this model, there are five entities, two men, two women
and a dog which are associated with the same logical name in each of the four worlds
in the model, i.e. jo\ bertie\ ethel\ fiona' and Chester\ respectively. The only
predicate we are interested in is laugh' and both Jo and Fiona are laughing in one of
the worlds (wO, only Chester is laughing in another (w2), no-one is laughing in a third
(w3) and everyone is laughing in the last (w4). This model is formally set out in (19).
272
(19) M9 = «{MAN1, MAN2, WOMANX, WOMAN2, DOG},

F9(jo') = {<w1,MAN1>, <w2,MAN1>, <w3,MAN1>, <w4,MAN1>},
F9(bertie') = {<W!,MAN2>, <w2,MAN2>, <W3,MAN2>, <w4,MAN2>},
F9(ethel') = {<w1,WOMAN1>, <w2,WOMANt>, <w3,WOMAN1>,
<w 4 ,WOMAN 1 >},
F9(fiona') = { ^ . W O M A N ^ , <w2,WOMAN2>, <w3,WOMAN2>,
<w 4 ,WOMAN 2 >},
F9(chester') = {<w t ,DOG>, <w2,DOG>,< w3,DOG>, <w 4 ,DOG>},
F 9 (laugh') = {<w1,{MAN1,WOMAN2}>, <w 2 ,{DOG}>, <w 3 ,0>,
<w4, {MANXMAN* WOM AN,,WOM AN2 ,DOG} >},
{W!,W 2 ,W 3 ,W 4 }»
As mentioned above, formulae are now interpreted with respect to a particular

world, the reference world, just as they are to particular times in a temporal model.
(We could also, if necessary, select a particular world, say Wi, to represent the actual
world which may function as the base reference point in the same way as the moment
of utterance in Chapter 8.) Constants have the extension they are assigned by F at the
reference world and the values of function-argument structures are computed from the
extension of the function and applied to that of the argument at the reference world.
Thus, to compute the value of laugh'(fwna') at the world w b i.e.
[laugh '(fiona ^JM^W^ w e take the value of laugh' with respect to M 9 and W! and apply
it to the value of fiona' with respect to the same model and world, i.e.
[laugh ']M9>&<wl(\fiona *]M9'8-wl). Because laugh' is a constant, [laugh ']M9'g-wl is determined
by taking the value of F9( laugh') at W! which can be written as F9(laugh ')(wi), because
the value of F(ot) is a set of ordered pairs of worlds and extensions, i.e. a function
from the former to the latter. The value of ¥9(laugh')(wi) is, according to (19), the
(characteristic function of the) set {MAN l9 WOMAN 2 }. Since [fiona']M9&"\
V9(fiona')(yri\ is WOMAN, and WOMAN, is a member of the set {MAN1?
WOMANJ, the formula laugh '(fiona') is true with respect to M 9 and w2. On the other
hand, this formula is not true in worlds w2 and w3, because WOMAN 2 is not in the
extension of laugh' in either of these worlds, as can be seen from the model in (19).
Thus, a model containing possible worlds can provide characterisations of different
situations, in some of which Fiona is laughing and in some of which she isn't.
Now we return to the interpretation of modal formulae. Because non-contingently
true (i.e. necessarily true) formulae are true by virtue of the logical system used, they
are true in every (logically) possible world, just as necessarily false formulae are false
in every (logically) possible world. We can therefore define the truth-conditions of a
formula containing the operator Dby looking at the truth value of the formula without
the operator in every world. Only if the latter is true in every world is the modal
formula with • be true. Hence, the truth-conditions of EH are defined as in (20).
(20) | is a formula then [D(|)]M<8'wn is 1 iff [§]M* wm is 1 for all wm in W.
If <>
For example, given (20), the formula D( laugh'(fwna')) is true in world, w h if (and
only if) laugh '(fiona') is true in all worlds in the model M9. Since there is at least one
world (in fact, there are two) in which Fiona is not laughing, [•(laugh '(fiona f))]^î
is false. On the other hand, the disjunction of this formula with its negation,
Dflaugh'(fwna') v -laugh'(fwna')) is true with respect to M9, g and w, because
273
9 Possible worlds
laugh'(fiona') v -laugh'(fiona') is true in every world (as can be shown by looking

at the truth table for/? v -p which yields the value 1 for all possible values of/?).
Logically possibly true formulae, as we have seen, consist of all the necessarily
true formulae plus all other contingent formulae, whether true or false. In fact, the
possibly true formulae only exclude those that are necessarily false, i.e. those that can
never be true. Translating this into a possible worlds context we can define a formula
<x|) as being true if, and only if, it is not false in every possible world, i.e. that it is true
in some possible world. Therefore, we define the truth-conditions for possibility as in
(21).
(21) If § is a formula, then [o<|)]M'8'wn is 1 iff [c|)]M-8'wm is 1 for some wm in W.
Given this definition, the formula o(-laugh'(fiona')) translating the sentence It is
possible that Fiona doesn 't laugh is true with respect to M 9 and Wi, because there is
a world, w2, in which -laugh'{fiona') is true. On the other hand, the conjunction of
this formula with its positive counterpart, o(-laugh '(fiona') & laugh '(fiona')), is false
because there is no world in which (-laugh '(fiona') & laugh '(fiona')) is (or could be)
true.
Exercise 9.1:
Which of the following formulae are true in world Wi with respect to the model M9?
i. D~(laugh' (Chester')).
ii. o(~(laugh'(chester')) —» laugh'(jo')).
iii. -(Dlaugh'(chester')).
iv. ~(o(laugh'(jo') -> -(laugh'(jo')))).
v. o~(laugh' (fiona')).
vi. ~(olaugh' (fiona')).
Do the truth values of these expressions change depending on which world acts as the
reference world? Explain your answer.
There are a number of important entailment rules that are valid in this semantics
for modal formulae that also appear to be intuitively valid for the English sentences
they are intended to translate. In the first place, the semantics for • and o involve
quantification over possible worlds and so the entailment rules defined for the logical
quantifiers have counterparts in this modal logic. Necessity is interpreted in terms of
universal quantification over worlds, and a version of the rule of universal
instantiation, given in Chapter 7, is valid for all worlds in all models. This rule is
given in (22.a) and allows one to infer the truth of a formula $ from the truth of the
formula Q(). If we treat formulae as denoting functions from possible worlds to truth
values, then the validity of (22.a) follows from that of the corresponding first order
quantified formula in (22.b) for all worlds wm. It should, however, be clear without
(22.b) that if some formula is necessarily true (true in every possible world), then it
is true in any named world.
274
(22) a. Q|>-><|>.
b. V w [<Kw)] - > (Kw m ).
Logical possibility involves existential quantification over worlds and so the rule in
(23.a) is valid for all models and worlds. Again this may be seen as a version of
existential generalisation indicated in the formula in (23.b), where wm is any named
world. Intuitively, this too is valid, because if something is true at some named world,
then it must be possibly true (nothing actually true can be necessarily false).
(23) a. <J> —> <><>.
b. (t>(wm) - > 3w [<Kw)].
Other rules that are valid in the current logic are shown in (24) to (27) where
each formula is followed by examples in English plus their schematic translations to
illustrate their intuitive validity. (24) and (25) show the interdefinability of o and D,
while (26) shows that impossible formulae are those that are necessarily false and
(27) shows that possibly false formulae are those that are not necessarily true.
(24) a. O|)<-> ~o~<|>.
b. Fiona must be laughing or not laughing,
•(laugh'(fiona') v -laugh'(fiona')).
c. It is not possible that Fiona isn't laughing or not laughing.
~o~(laugh'(fiona') v -laugh'(fiona')).
(25) a. <x|> <r>
b. Fiona may be laughing,
olaugh'(fiona').
c. It is not the case that Fiona is necessarily not laughing,
-•-laugh'(fiona').
(26) a. ~O(() <->
b. It is not the case that Fiona may be laughing,
-olaugh'(fiona').
c. It must be the case that Fiona is not laughing,
•-laugh'(fiona').
(27) a. o~<J) <-> -Of).
b. It may be the case that Fiona isn't laughing.
o~laugh'(fiona').
c. It is not the case that Fiona must be laughing,
-•laugh'(fiona').
Exercise 9.2:
Explain why the semantics given for • and o guarantees that the rules in (24) to (27)
are valid.
275
9 Possible worlds
The interpretation of necessity proposed above also has the desired property that
it guarantees that the substitution of extensional equivalents is not valid in this
context. Recall the invalid inference pattern in (3), repeated as (28). A schematic
translation of this inference pattern appears in (29) where ms' is used for the
translation of the Morning Star and es' for that of the Evening Star.
(28) a. Necessarily, the Morning Star is the Morning Star.
c. Therefore, necessarily, the Morning Star is the Evening Star.
(29) a. D(ms'=ms').
b. ms' = e s \
c. Q m s ' = es').
The inference pattern set out above is invalid in the terms of modal logic because the
first premiss and the conclusion have different truth-conditions. (29.a) is true, as we
have seen, relative to a model M and world wn if, and only if, ms' = ms' is true in
every possible world in M. Since every constant is, by the definition of identity,
necessarily identical to itself, this formula is true of any set of possible worlds. (29.b),
on the other hand, is true if and only if two different constants happen to extensionally
denote the same entity. Although the premiss in (29.b) asserts that it is true that two
different constants do denote the same entity, it is true non-modally, i.e. only with
respect to some specified world (the actual one). Since it is not a logical impossibility
that two different constants extensionally denote different things, there must be some
logically possible world in which (29.b) is false. Hence, (29.c) must be false in a
model that contains all logically possible worlds, i.e. in any model likely to provide
an adequate interpretation for any natural language. The truth of (29.c) thus does not
follow from the truth of the premises in (29.a) and (29.b).
The putative inference pattern in (28) is set out in (30) in terms of the
truth-conditions of the expressions in (29). From this it is easy to see that (29.a) and
(29.c) have different truth-conditions and that therefore (28.c) does not follow from
(28.a) and (28.b). We have, therefore, provided a semantics that ensures that
extensional substitution into one of the opaque contexts noted in Section 9.1 is not
valid.
(30) a. [Qns' = ms'] M 8 wn is 1 iff [ms' = ms']M'8iWm in all wm e W, i.e. if [ms']M'8'wm
is identical to [ms']M«wm in every possible world.
b. [ms' = es'] M * wn is 1 iff [ms'] M * wn is identical to [es'] M * wn .
c. [dies' = ms']M'8'wn is 1 iff [es' = ms'] M * wm in all wm e W, i.e. if [ms'] M * wm is
identical to [es']Ms-wm in every possible world.
9.3.2 Accessible worlds

In models more satisfactory than the simplistic M 9 , the set of worlds would include
all the logically possible worlds, i.e. a set of worlds in which each non-contradictory
formula is true in at least one world. This has the consequence that every formula that
is not a logical falsehood is possibly true and thus every sentence in (31) is associated
with a true formula in some appropriate model.
276
(31) a. The Conservative Party might not win a fifth term of office in 1997.
b. Rabbits may be robots from Mars.
c. Snow may be a black sticky substance, hot to touch.
d. A tree may be an air-breathing quantifier that lives at the bottom of the sea
on a diet of logic textbooks.
Although all the sentences in (31) describe situations that are logically possible,
not all of them are as likely as the others. For example, (31 .a) is much more likely than
(31.b) and (31.d) is completely unlikely. Indeed, in the latter case the English word
tree appears to have lost all of the properties generally associated with trees in the
actual world and one could question whether a speaker asserting (31.d) as a true
statement in some possible world was actually speaking English. Because of such
differences in likelihood, the notion of possibility (and necessity) might be taken to
be relative to a particular world and not absolute for all worlds. This notion of the
relativity of modality can be defined in terms of a relation, R, of accessibility (or
alternativeness) between possible worlds. A world wm is thus R-accessible from
another world wn, if, and only if, wnRwm is true (i.e. if wn stands in relation R to w,n)
in which case wm is a possible alternative to wn. The relation, R, thus imposes
structure on the set of possible worlds, as the precedence relation, <, imposes structure
on the set of possible times, T, in temporal models. The relations R and <, however,
generally have different properties from < For example, unlike <, R is not usually
transitive so that if wJRWm and w,nRwo it is not necessarily the case that WnRwo. R is,
however, often reflexive, so that WnRwn holds. The actual properties of R depend on
the type of accessibility relation being modelled and thus no independent restrictions
are put upon its definition.
The introduction of accessibility relations into the model allows possibility and
necessity to be redefined in such a way as to refer only to worlds that are accessible
to the reference world and not to every world in the model. This relativisation of
modality to certain sets of worlds is reminiscent of the restriction of the domain of
quantification to pragmatically determined context sets discussed in Chapter 6 with
respect to definiteness, except that Ris assumed to be given in the model itself, rather
than by the pragmatics. The truth-conditions for o and • relativised to accessible
worlds are given in (32) where models contain a set of worlds plus an accessibility
relation as well as a set of entities and a denotation assignment function, i.e. M =
«A,F,W,R».
(32) a. If <(> is a formula, then [O|)]M'8<wn is 1 iff [ty]**™ = 1 for all wm e W where
WnRw m .
b. If <}> is a formula, then [oc()]M'8'wn is 1 iff [<()]M'g'wm = 1 for some w m 6 W where
There are many different sorts of accessibility relation that may be defined over the
set of possible worlds and these may yield different interpretations for modal
formulae. Where logical necessity and possibility are involved, R relates all worlds
that conform to the system of logic chosen. By (32), a formula that is logically
necessarily true is one which is true in all logical worlds, but not in worlds where the
logical structure breaks down. Whether the set of possible worlds should include
worlds that do not conform to logical rules of inference like conjunction elimination
277
9 Possible worlds
is a matter of debate, but there are more interesting accessibility relations that do
properly discriminate between logically possible worlds.
One such relation is that defined by the sense relations amongst words, such as
those discussed in Chapter 7. We saw there that word meaning may be partly
accounted for in terms of meaning postulates that relate the extensions of different
expressions within a particular language. In a semantic theory that allows for the
possibility of alternative states of affairs, the status of meaning postulates like those
in (33) is unclear. Do they constrain the extensions of predicates in all permissible
models or only in the actual world in particular models? We do not want to require all
possible worlds to conform to all sense relations (thus elevating meaning postulates
to the status of logical constraints) because we can, for example, happily accept
talking trains and excitable cars within the space of a children's story or an adventure
film. Nor do we want to restrict sense relations to holding only in the actual world,
since we could not then explain why a sentence like Colourless green ideas may sleep
furiously seems to describe an empirically impossible situation. If meaning postulates
are interpreted as defining accessibility relations between worlds, however, then the
anomaly associated with sentences like that just quoted or the peculiar sentence in
(31 .d) may be accounted for. For example, if a definition of the sense relations of
English includes the statements in (33) and we interpret R to be Rs, an accessibility
relation that relates worlds that conforms to these sense relations, then (31 .d) turns out
to be false according to the definition of possibility in (32.b). Any world in which this
sentence expresses a true proposition fails to conform to the constraints in (33),
because it requires some entity to be both abstract and not abstract, a logical
impossibility. Hence, such worlds are not accessible to the actual world and so A tree
may be an air-breathing quantifier that lives at the bottom of the sea on a diet of logic
textbooks is true in no world, wm, that is Rs-accessible to the reference world, i.e.
where wJRWn, does not hold.
(33) a. Vx [tree'(x) -* -abstract'(x)].

b. Vx [quantifier' (x) -^ abstract' (x)].
We may conceive of sense, according to the definitions of Chapter 7, as defining
a complex of accessibility relations that relate worlds according to how they conform
to the different sets of meaning postulates associated with different lexical items. The
sense of an expression, therefore, structures the set of worlds and makes it possible to
differentiate those expressions that have the same intension. (This is discussed further
in Chapter 10.) In this way, it is possible to give a more subtle account of the meaning
of words and the sentences that contain them, one that may provide a semantic theory
of figurative language in terms of possible worlds. Be that as it may, interpreting sense
relations in terms of meaning postulates that define accessibility relations enables the
differentiation of likely from unlikely possibilities.
278
*Exercise 9.3:
Consider a language with only two (one-place) predicate constants, P and Q, and two
individual constants, a and b, and a model consisting of only two elements, d and e2,
a denotation assignment function, F and a set of possible worlds W.
1. Define the set of all logically possible worlds, W, by listing the possible
combinations of extensions of P and Q in this model (i.e. by defining F).
How many worlds are there?
2. Which worlds are accessible to each other, if we add the accessibility
relation Rx to the model which is defined to hold between worlds that
conform to the meaning postulate Vx [P(x) —> Q(x)]? How many worlds are
no longer accessible?
3. What are the truth values of the following expressions with respect to this
model, assuming Ri accessibility?
i. D(~P(b)v Q(b)).
ii. «P(a) & ~Q(a)).
iii. ~O(~P(a) v ~Q(a)).
iv. ~D(P(a)vQ(a)).
While the accessibility relation on worlds imposed by sense relations may be

considered central to the meaning of non-figurative English expressions, other notions
of accessibility can be used to account for other sorts of modal relations that are
central to what one knows or what one's obligations are. In addition to the alethic
modality discussed in the last section, philosophers also recognise epistemic and
deontic modalities, amongst others. The former expression, derived from the Classical
Greek word for 'knowledge', defines necessity and possibility in terms of states of
knowledge while the second, a word derived from the Greek word meaning 'to be
binding', has to do with permission and obligation.
The sentences in (34) would generally be construed epistemically, since it is
likely that (34.a) is not logically true (we can conceive of worlds where room
temperature is below the freezing point or above the boiling point of water) and (34.b)
conveys little information from the logical point of view. Hence, the proposition
expressed by (34.a) is true if (and only if) water is wet at room temperature in every
world consistent with what is (objectively) known about the temperature of rooms and
about the properties of water. Similarly, (34.b) expresses a true proposition if and only
if in some world consistent with what is known about cold climates, precipitation and
snow, precipitation does indeed fall as snow in cold climates. The truth of the
propositions expressed by these sentences thus does not depend on worlds in which,
for example, rooms generally have a temperature of -50°C or in which water is a solid
at +20oC.
(34) a. At room temperature, water must be wet.
b. In cold climates, precipitation may fall as snow.
279
9 Possible worlds
Epistemic modality may be interpreted as truth with respect, not to worlds

consistent with what is generally known, but to worlds consistent with what an
individual, speaker or hearer, knows. Consider the two sentences in (35).
(35) a. Fiona must be drunk.
b. Bertie may sing at the party.
In terms of what is objectively known, the proposition expressed by (35.a) is
(probably) not true, since Fiona is drunk (at some particular time) is unlikely to be
true in every world consistent with objective knowledge. However, it can be
interpreted in such a way that it is true, provided that we take into consideration the
knowledge of the person who utters the sentence. For example, according to what a
speaker knows, it may indeed be necessarily true that Fiona is drunk, because she has
drunk six glasses of wine and does not usually drink alcohol. In other words, Fiona
must be drunk is true epistemically, if Fiona is drunk is true in every world consistent
with what is known by the speaker. Similarly, a speaker-oriented epistemic
interpretation of (35.b) requires there to be some world consistent with what is known
by the speaker in which Bertie sings at the relevant party.
The two sentences in (35) are ambiguous not only according to the scope of
knowledge assumed, but also according to the sort of modality intended. On a deontic
interpretation, an utterance of (35.b) asserts that Bertie is being allowed to sing at the
party while an utterance of (35.a) asserts that Fiona is obliged or required to be drunk
(e.g. because she is acting in a play that requires this behaviour or is taking part in
some initiation ceremony). More formally, (35.b) is true if (and only if) there is a
world that is deontically accessible from the actual world where Bertie sings at the
party is true and (35.a) is true if in every world deontically accessible from ours Fiona
is indeed drunk. Exactly what deontic (or indeed epistemic) accessibility involves is
a matter for the philosopher rather than the linguist, but the fact that different
accessibility relations may alter the truth values of modal sentences allows us to begin
to account for the different interpretations of sentences like those in (35).
Thus, the truth or falsity of modal statements depends on the different
accessibility relations holding between different worlds. These accessibility relations
have an effect, not only on the truth value of a formula translating a particular
sentence, but also on the inferences that can be drawn within the modal system. For
example, the entailment rule in (22.a), which allows the inference that 0 is true
whenever D<|) is true, is valid only if R is reflexive (i.e. where for all worlds, w,,,
wnRw is true). As pointed out in McCawley (1981:277), in certain interpretations of
moral necessity, where R is interpreted something like is morally better, or more
desirable, than, R is not reflexive. In such a modal system, we cannot infer from the
truth of All children must be good that All children are in fact good, because the
former sentence only requires the latter to be true in all morally more desirable worlds,
of which the actual world is not necessarily one.
The definition of such relations is, therefore, very important for the interpretation
of modal sentences and it is likely that an adequate semantic theory needs to contain
many accessibility relations, providing a rich and complex structure to the worlds
contained by any model. Context is then required for the hearer to identify the
particular relation intended by the speaker. Despite the importance of the notion of
accessibility, nothing further will be said about such relations. The reason for this is
280
Further reading
that, while particular interpretations and inferences are affected by the particular
relation supposed, the general theory is not affected by which accessibility relation is
assumed. Therefore, in the next chapter, it will be generally assumed that R defines
logical accessibility, unless otherwise stated.
9.4 Further reading

Referentially opaque constructions are discussed in Lyons (1981: 157- 167),
McCawley (1981:326-340), Chierchiaand McConnell-Ginet (1990:204-208), Partee,
ter Meulen and Wall (1990: 403-414) and Dowty, Wall and Peters (1981: 141-146).
Frege's discussion on Sinn and Bedeutung can be found in Frege (1892) and Dummett
(1981: chs. 5 & 6) gives a useful critical (philosophical) discussion of this. The notion
of possible world is discussed at length in, for example, Lewis (1981), and, for a
different point of view, Bradley & Swartz (1979). A full logical exposition of modal
logic can be found in Hughes and Cresswell (1968), but the subject is given a simple
introduction in Guttenplan (1986: ch. 16), McCawley (1981: 273-296) and Allwood,
Andersen and Dahl (1977: 108-110). The logic is discussed in the terms used in this
chapter in Dowty, Wall and Peters (1981: 121-131), and Chierchia and
McConnell-Ginet (1990:213-221). Different types of modality are discussed in Lyons
(1977: ch. 7), Lyons (1981: ch. 10) and accessibility relations are discussed in
McCawley (1981:273 ff. & ch. 11), Partee, ter Meulen and Wall (1990:414-423) and
Martin (1987: 281-295).
281
10 Intensional Semantics
10.1 Modelling intensions

In Chapter 9, an interpretation of the modal adverbs necessarily and possibly was
presented in terms of possible worlds, but the other contexts in which extensional
entailments fail were not discussed. As proposed in Section 9.2, the general solution
to the problem of referentially opaque contexts lies in the concept of intensionality,
but the interpretation of modality given above was not stated in terms of this concept
and it has not been made clear how possible worlds enable a formal definition of
intension to be made. Let us now remedy the situation and provide a general semantic
theory for opaque contexts, thus completing our survey of formal semantic theory.
The definitions for the interpretation of modal formulae given in Chapter 9
embody the idea that formulae may be true in some worlds but not in others, i.e. that
the extensions of formulae (i.e. truth values) may vary from world to world.
Furthermore, it was suggested at the end of Section 9.2 that an intension is something
that picks out the extension of an expression in any state of affairs. The intension of
a formula may thus be defined as something that specifies its truth value in every state
of affairs. Equating states-of-affairs with possible worlds, we interpret the intensions
of formulae as functions that map possible worlds onto truth values: functions that
map a possible world onto 1 if the formula is true in that world and onto 0, otherwise.
For example, the truth values of the formula laugh '(fwna') with respect to the model
M 9 differ according to the reference world. As can be seen from the specification of
this model in (19) of Chapter 9, this formula is true in worlds Wi and w4 but false in
worlds w2 and w3. The intension of laugh '(fiona') in M 9 is thus the function that maps
Wi and w4 onto 1 and w2 and w3 onto 0, as shown in (1).
(1) Intension of laugh'(fiona') in M9
While the truth value of two formulae may be the same in some world, their intensions
may differ. For example, both laugh '(fiona') and -(laugh'(Chester')) are true in world
Wi with respect to M9. However, the intension of the latter is the function that maps
Wi and w 3 onto 1 and w2 and w4 onto 0, since in the latter worlds Chester does not
laugh, as shown in (2.a). Furthermore, the formula laugh '(jo') v -laugh '(jo') although
again true in w h differs in intension from both the previously considered formulae,
since its intension is that function which maps every possible world onto 1, cf. (2.b).
282
Modelling intensions
(2) a. Intension of -(laugh'(Chester')) in M 9
b. Intension of laugh '(jo') v -laugh '(jo') in M9
As the diagrams in (1) and (2) show, interpreting the intensions of formulae as
functions from possible worlds to truth values allows semantic differentiation between
formulae that happen to have the same extension in some world. Although there are
some formulae that have the same intensions in M 9 (compare, for example, the
intension of laugh '(jo') with (1)), but which are not true paraphrases, this is an artefact
of the small scale of the model and not of the theory itself. In principle, if a full set of
logically possible worlds is incorporated into the model, it is possible using this
method to differentiate the intensions of all contingently true formulae.
Intensions cannot, however, be defined solely with respect to possible worlds.
As we saw in the last chapter, the concept of time is necessary to account for tense and
aspect distinctions. Since natural languages may combine tense and modality in
particular sentences, as, for example, in Jo possibly laughed yesterday and When she
has finished her degree, Fiona may apply for a job as a gardener, an adequate
semantics for natural languages must also combine elements of modal and temporal
interpretation. Models should, therefore, contain both times and possible worlds, as
well as entities. Furthermore, because temporal models allow the extensions of
expressions to change from interval to interval in the same way that modal models
allow them to vary from world to world, our semantic models must allow extensions
to vary along two dimensions: temporal interval and possible world. Thus, we may
think of extensions as being defined for a particular world at a particular time. In other
words, extensions are defined for a particular co-ordinate or index consisting of a
possible world, wn, and an interval of time, i, i.e. with respect to an ordered pair
<wn,i>. If one of the co-ordinates with respect to which an expression is interpreted
changes then the extension of that expression may change. Hence, [oc]Mgwni, the
extension of an expression a at the index <w m i> with respect to model M and variable
assignment g, may differ from [oc]M-8-wnJ or [a]M.g.wm.« where wn differs from wm and i
differs from j .
283
10 Intensional semantics
As an illustration, consider the diagram in (3) which shows the changing

extensions of the predicate laugh* in four worlds across four moments of time, to give
part of the intensional model Mi0. In (3), a line indicates an interval of time and the
set below the line shows the extension of laugh' at that interval. The horizontal axis
presents the different extensions for laugh' across the four moments of time within the
same possible world while the vertical axis traces its extensions at the same time in
different worlds.
(3)
tl t2 t3 t4
wl ** • ^
{MANX}
w2 -^
{DOG} 0
w3 -*
0 {WOMAN 2 } {WOMAN 1 ,WOMAN 2 }
w4 *< •
{MAN^MAN^WOMAN^WOMAN^DOG}
If intensions are defined as functions from states of affairs to extensions, then in

models with times and worlds, they must have the set of all world-time co-ordinates
(or indices) as their domain. Thus, the intension of the formula laugh'(fiona') with
respect to Mi0 is a function from pairs consisting of a world and a time to the truth
values, 0 and 1. Because laugh is an activity verb, its extension holds for all subsets
of an interval over which it is defined, as we saw in Chapter 8. Hence, we can define
the intension of laugh '(fiona') in terms of its truth values at each ordered pair of world
plus moment of time, as is done in (4), i.e. as a function from world/time pairs to truth
values. The intension of an expression, a, with respect to a model, M, and an
assignment of values to variables, g, is written as 3[oc]M«, so 3 [laugh'(fiona')]M1°s
represents the intension of the formula laugh'(fiona') with respect to M10 and g.
284
(4) sflaugh' (fiona')] M10'«
Exercise 10.1:
Diagram the intensions of the following formulae with respect to M10 (assuming that
jo\ fiona' and Chester' have the same extensions in all worlds and times, i.e. MANi,
WOMAN2 and DOG, respectively):
i. laugh'(jo').
ii. laugh'(chester') v laugh'(jo'),
iii. laugh'(fiona') & -laugh'(fiona').
iv. laugh'(fiona') —»laugh'(jo').
It is possible to add more co-ordinates to the indices in terms of which intensions

are defined. For example, it is often the case that locations are relevant to the
interpretation of sentences. Thus, if the sentence // is raining is uttered, it is usually
interpreted as an assertion that it is raining at the place in which the speaker is situated
(at the present time in the actual world), not that it is raining everywhere. Hence, we
might expand our semantic space to include locations and introduce into our models
a set of places together with some relation of proximity. Furthermore, the fact that /
and you are indexical (or deictic) expressions, like here and there or tense, indicates
that co-ordinates for speaker and hearer should be added to the model. Indeed, as
285
Lewis (1972) points out, there are many other candidates for co-ordinates in defining
an adequate semantics for any human language. Exactly what ones should be included
is a matter for debate and empirical research, but at a minimum we would need
worlds, times and places. In this book, however, we will utilise indices that consist
only of worlds and times. The addition of a topographical co-ordinate will only
complicate matters unduly without altering the basic properties of the semantic
system. In principle, any (finite) number of co-ordinates may be added to the model
theory presented below, allowing more and more contextual information to be
included in the semantics, but without altering the fundamental structure of the theory.
The semantics of the modal operators can now be redefined in terms of the
intensions of the formulae they have scope over, i.e. as functions from functions from
co-ordinates to truth values to truth values. Since functions that map some set of
entities into the truth values are characteristic functions of subsets of those entities, the
intensions of formulae may be thought of as sets of indices. Hence, we can define the
semantics of necessity and possibility in terms of functions from (the characteristic
functions of) sets of world/time pairs to truth values. In the definitions in the previous
chapter, Q|> is true (with respect to a model and a reference world) if, and only if, <> |
is true in all accessible worlds in the model. o<|), on the other hand, is true if, and only
if, (() is true in some accessible world. The necessity operator can, therefore, be
interpreted as that function that maps the set of all co-ordinates onto 1 and all other
subsets of W x I onto 0 and the possibility operator is that function that maps the null
set of co-ordinates onto 0 and all other subsets onto 1. In order to differentiate this
approach to necessity and possibility from the earlier approach, we will not use the
modal operators, • and o, but translate the words necessarily and possibly directly
into necessarily' and possibly'. These expressions denote the functions over the
intensions of formulae (over 3[(|)]M-8) which are spelled out in (5) where I is the set of
all temporal intervals defined on a set of moments T, [cc]M.g.wiu is the extension of a
with respect to M, g and index <wn,i> (see Section 10.1.2) and 3[(t)]M<<wn,i>) is the
value of the intension of <|) at the index <wn,i>.
(5) a. [necessarily']M*8wni is that function which maps the intension of a formula

(j) onto 1 iff s[(|)]M8(<wm,i'>) is 1 for all wm e W such that wnRwm and i e
I (i.e. where 3[<f)]M-8 is the set of all possible world/time pairs),
b. [possibly'l 1 ^^ is that function which maps the intension of a formula (j)
onto 1 iff 3 [(|>]M8(<wm,i'>) is 1 for some wm e W such that WnRw™ and some
V e I (i.e. where 3[<()]M'8 is not null).
A question that arises with the interpretation of necessity within models which
have times as well as possible worlds is whether to interpret it as necessity for all
times and worlds, as in (5), or for particular times in all worlds. Should we analyse a
sentence like Necessarily, the Morning Star is the Morning Star as expressing a true
proposition if The Morning Star is the Morning Star is true always and in all situations
or whether it is true now in all situations (and therefore perhaps false at some other
time)? Although it would seem that with sentences like the above it is clear that the
first interpretation is intended, with sentences containing modal verbs interpreted
non-alethically, e.g. Fiona must be singing, it would seem preferable to have an
interpretation in which the proposition expressed by the sentence comes out as true if
Fiona is singing is true in every deontically or epistemically accessible world only at
286
the time of utterance but not at all past and future times in those worlds. Montague
adopts, without argument, the interpretation of necessity in (5.a). However, while this
seems to work as an analysis of the adverb, necessarily, the modal verb must appears
to require the less strict interpretation given in (6.a) which specifies that a formula
must'(§) is true as long as <|) is true in all worlds at the appropriate reference time.
Thus, according to (6.a) [must'(Impf(sing '(fwna ')))]u*.™i is true provided that [(Impf
(sing'(fiona')))]M^mi is true in all worlds wm at i. A parallel interpretation for the
modal verb of possibility may is given in (6.b). This differs from that given for
possibly in (5.b) in that a formula like may'(Impf (sing'(fwna'))) (translating Fiona
may be singing) is true as long as there is at least one world accessible to the
reference world in which Fiona is singing at the time of utterance. Worlds in which
she is singing at other times are irrelevant. Hence, while (6.a) gives a less restrictive
form of necessity than (5.a), the definition of possibility in (6.b) is more restrictive
than that in (5.b).
(6) a. [must']M8'wna is that function which maps the intension of a formula ty onto
1 iff 3[<t)]M-K<wm,i>) is 1 for all wm e W such that wnRwm.
b. [may']M8wni is that function which maps the intension of a formula <> | onto
1 iff s[<|)]M*(<wn,i>) is 1 for some wm e W such that WnRw,,,.
Treating modal expressions as functors over intensional expressions in this way
has the extremely desirable consequence that Leibniz' Law fails to be valid. The
reason for this, of course, is that the intensions of formulae may differ even if their
extensions happen to be the same at some index. Thus, the functions denoted by
necessarily\ possibly \ must' and may' may map the intension of one formula, <(), onto
true (or false) but that of another, \j/, onto the opposite truth value, where s[<t>]Ms is not
identical to 3[\j/]M8. As a concrete example, consider the interpretations of Fiona must
be laughing and Jo must be laughing with respect to M10 and the index <W!,[t2]>.
Suppose that the accessibility relation of the model is one where w3 and w4 are both
accessible to Wi which is accessible to itself, but w 2 is not accessible to w b i.e. where
R is {<Wi,w1>,<Wi,w3>,<Wi,w4>}. The interpretation of must' in (6.a) entails that
[must']M8wl tt2i is that function that maps a function from indices to truth values onto
1 if, and only if, that function maps the indices <Wi,[t2]>, <W3,[t2]>, and <w4,[t2]>
onto 1, i.e. it picks out all those worlds accessible to w, at time [t2]. Thus,
[must']M*vl-iaKz[laugh'(fiona')]M*) is true if, and only if, s[laugh'(fiona')]M* applied
to each of these indices yields the value 1. Looking at the diagram in (4) which
represents the intension of laugh '(fwna'), we can see that this is the case. It is true that
Fiona is laughing in all worlds accessible to Wi at the appropriate time [tj. Hence,
Fiona must be laughing is true. However, although it is true that Jo is laughing is true
in world wt at time [t2], it is not the case that Jo must be laughing is true at that index.
This is because the function z[laugh'(jo')]u* maps the index <w3,[t2]> onto 0, as can
be checked by referring to the diagram in (3) showing the denotation of laugh' in M10,
and so Jo is laughing is not true in every accessible world at time [t 2 ]. Hence,
s[laugh'(jo')]M& differs from z[laugh'(fwna')]M& and so must'(laugh'(jo')) is false
while must '(laugh '(fiona')) is true, despite the fact that Jo and Fiona are both laughing
at the same reference index.
The fact that two expressions are extensionally equivalent at some index has no
bearing at all on their intensional equivalence and so while substitution of the
287
expressions preserves truth in extensional contexts, there is no guarantee that it does

in intensional ones. As we saw at the end of Chapter 9, the inference pattern in (3) of
that chapter fails because the extensional equivalence of the Morning Star with itself
is necessary while that between the Morning Star and the Evening Star is contingent.
The sentences The Morning Star is the Morning Star and The Morning Star is the
Evening Star differ in intension and so are not intersubstitutable in intensional
contexts. Leibniz' Law is thus valid only in extensional contexts and fails for all
expressions that create intensional contexts. Thus, we must revise the definition of the
Law of Substitution in (2) of Chapter 9 to exclude explicitly substitution into
intensional contexts. These contexts also include the scopes of the past and future
tense operators, because the extensions of expressions may vary over time and so
those that are extensionally equivalent at one interval of time may not be at others.
Thus, the truth of (8.c) does not follow from the truth of (8.a) and (8.b) because while
Ethel may now be the Ladies' Scottish Golf Champion, someone else could have been
when Jo met the then Ladies' Scottish Golf Champion. The revised version of the Law
of Substitution is stated in (8), where the term intensional functor covers expressions
like necessarily\ may' and the other opaque predicates discussed in later sections.
(7) a. Jo met the Ladies' Scottish Golf Champion.
b. Ethel is the Ladies' Scottish Golf Champion.
c. Jo met Ethel.
(8) (a = b) —»[())<-» (I)1"*] where a is not in the scope of an intensional functor
or the tense operators Past and Fut.
As we will see in Section 10.2, the analysis of modal expressions given above
can be generalised to all opacity creating predicates. In this way, we can ensure that
the Law of Substitution fails for the reasons shown above and we will see that this
also has the effect of making the other false inference patterns noted in Section 9.1
truly invalid. Before this can be tackled properly, however, we need to construct a
metalanguage that can represent intensional expressions and provide a proper and
fully explicit semantics for this language. This will involve extending our definitions
of possible types and revising the model theory of previous chapters to take account
of intensions.
10.2 The intensional language LIL

In this section, we look at Montague's definition of a co-ordinate semantics for a fully
typed language, called LiL, which is like LQ but with tense and modal operators as
well. First, we must revise the theory of types to take account of intensional
expressions and look at how such expressions should be interpreted. No systematic
attempt, however, is made to revise the grammar fragment so far constructed to take
account of intensional contexts. From now on it is assumed that readers have the skill
to see the relation between expressions in the intensional language and the English
examples they are assumed to translate. While the construction of an adequate
translation algorithm from syntax to LIL is not without its problems or interest, to
attempt the task here would unduly lengthen an already long book and would
not particularly help the reader understand the most important matter, which is the
288
The intensional language LIL
semantic interpretation. Here more than ever the choice of the syntactic framework
determines the precise nature of the translation algorithm and so setting out a
particular algorithm would not necessarily be helpful or enlightening.
10.2.1 Intensional expressions...

The intensional language L,L has the language of Chapter 8, Lr, as its core, but also
contains intensional expressions and intensional operators. To show that an expression
is intensional, an intension operator, symbolised as A , is used and, like all operators,
it is written to the left of an extensional expression. For example, the intensional
counterpart of laugh '(fiona') is Alaugh '(fiona'), that of laugh' is Alaugh \ that of fiona'
is Afiona'. As might be expected, an intensional expression like Alaugh'(fiona')
denotes z[laugh'(fiona')]M*y the intension of its extensional counterpart,
laugh'(fiona'). This gives the general rule for intensional expressions in (9).
(9) If a is any extensional expression, Aoc is an intensional expression which
denotes 3[ot]M-8.
Montague did not restrict the intensional operator to apply only to extensional
expressions but allowed it to be recursive, thus allowing expressions like
AAA
^laugh'(fiona'), laugh'(fiona') and so on. These expressions are not very
interesting semantically, however, as they denote constant functions which pick out
the same denotation for Alaugh'(fiona') or âugh'(fiona') at every index.
Furthermore, higher order intensions appear to be of little or no use in defining the
semantics of any natural language and, as we are interested only in accounting for the
interpretation of natural languages, ignoring the existence of higher order intensions
is justified, at least as a temporary expedient.
Parallel to the intension operator is the extension operator written as v which
applies to an intensional expression a to yield an expression vot which denotes the
extension of a at some index. Thus, at any index the result of applying the extension
operator to an intensional expression, Aoc, is to cancel out the intensionality, making
the expression so formed, VAoc semantically equivalent to the original extensional
expression a. For example, from Alaugh'(fiona') the extensional expression
"Haugh '(fiona') can be formed which denotes at a particular index, say <w3,[t2]>, the
extension of Alaugh '(fiona') at that index, i.e. z[laugh '(fiona ')]M'8(<w3,[t2]>) which is,
of course, equivalent to [laugh'(fiona')]M^wXlt2].The rule for using the extension
operator is given more formally in (10) and the equivalence between ~Aoc and a stated
in (11) holds for all models, indices and variable assignments.
(10) If a is an intensional expression with denotation 3[a]M«, then "a is an
extensional expression whose denotation at any index <wn,i>, fa] M '8 wni , is
the value of 3[a]M«at <w,,,i> (i.e. 3[oc]MK<wn,i>)).
(11) f Aa]M.g,wn,i = [a]M.g,wn,i
While the identity in (11) is valid, a parallel equivalence between A~oc and a is
not valid. As no expressions of this form are considered in this book, the explanation
for this will not be given and readers are invited to try to work it out for themselves;
or they may consult Dowty, Wall and Peters (1981: 175-176).
289
Because Ln. is a typed language, all intensional expressions must be assigned an

appropriate intensional type. These are derived according to the rule in (12) which
introduces a type that has the symbol s as its first element. Mnemonic for sense (after
the Fregean Sinn), s is not itself a primitive type, like e and t, but is just used to
indicate that a type is intensional. The fact that s is not treated as primitive means that
no expressions directly denote indices, a reflection of the fact that natural languages
have no way of referring directly to such things.
(12) Intensional types: If a is a type, then <s,a> is a type.
(12) is added to the rules defining types given in Chapter 4 and givesriseto types like
those shown in (13). Recursive application of the intensional type rule will not be used
in this book, as (10) already rules out the creation of 'higher order' intensional
expressions so that types like that in (13.f) do not appear in the discussions below.
(13) a.
b.
c.
d.
e.
f.
g-
As in the language Ltype, and the languages based on this, LQ and Lr, complex
expressions are derived by the rule of functional application, repeated in (14).
(14) If a is an expression of type <a,b> and p is an expression of type b, then
oc(p) is an expression of type a.
This rule must be strictly adhered to and caution must be taken when utilising it with
expressions in intensional categories. Since there is no type s, there are no expressions
of type s and so intensional types cannot directly combine with any expressions in LIL.
In order to combine an intensional functor with an argument, it is necessary first to
'de-intensionalise' the functor by applying the extension operator,v, to it. Thus, one
cannot directly apply an intensional functor Aoc, of type <s,<a,b», to an argument p,
of type a, but one can apply the extension of the functor v/va to the same argument to
get the well-formed expression CAoc)(P) which, by the equivalence in (11), is
semantically equivalent to cc(p). By the general definition of types given in Chapter
4, intensional types may also appear as the argument or result types of other types.
Examples of such types appear in (13.d,e & f) and care must be exercised in
combining such functors with their arguments. Functors which have the general type
«s,a>,b> can be applied only to argument expressions that are themselves
intensional, i.e. of type <s,a>. Thus, if a is of type «s,a>,b> one cannot directly
apply it to an expression p of type a. First, the latter must be made intensional using
the intension operator to form the expression AP of type <s,a> which can then be the
argument of a to give oc(AP). No problems arise in the application of functors with
intensional result types of the form <a,<s,b», except that it must be borne in mind
that the resulting expression is itself intensional and so may not directly combine with
an argument and may only be an argument of a functor that takes an argument thathas
an intensional type.
290
Exercise 10.2:
Assuming the type assignment that follows, which of the expressions below are
well-formed expressions of W? If they are well-formed say what types they have and
if not, say why not.
Type: Expression:
<s,e> m
e bertie', fiona'
«s,e>,t> change'
<e,t> walk', student'
«s,t>,t> necessarily', possibly
<t,t> ~
«s,<e,t»,<e,t» good'
«e,t>,<e,t» happy'
i. change' (bertie').
ii. good'(Astudent').
iii. ~CAwalk'(bertie')).
iv. necessarily' (A((good' (Astudent' ))(fiona'))).
V. possibly'(change'(Am)).
vi. (happy' (student' ))(vm).
vii. good' (happy' (Astudent')).
The language LIL also contains a full set of variables for each type, including the
intensional types. Formally, the set of variables of any type a is formally symbolised
as Wara and particular variables of type a are formally indicated as vn,a where n is some
natural number. This notation is somewhat cumbersome, however, and certain
notational conventions are usually adopted. The conventions used in this chapter are
set out in (15) and subscript numerals may be used with the various symbols where
there are a number of variables of a particular type. The symbols used to indicate
variables over extensional types (e.g. JC, P, P and R) are the same as they have been
in previous chapters, except for/? and q which are here given an intensional type and
so symbolise variables of type <s,t> and not of type t.
(15) Type: Vanables:
e x, y, z.
<s,e> r.
<e,t> P,Q.
<s,<e,t» A.
<s,t> p>q.
«e,t>,t> p.
<s,«e,t>,t» p-
<e,<e,t» R.
291
Because LiL contains both intensional and extensional variables, the lambda
operator, X, may bind intensional and extensional argument places. The definition of
lambda abstraction remains as it was given in the final form in Chapter 5 (and as it is
repeated in (16.6) below), but one must again be careful to use the correct type of
variable. An extensional variable cannot be used where the position being bound is
intensional, and vice versa. For example, given the type assignments in (15) and
Exercise 10.2, Xr [change '(r)] is well-formed, because r is of type <s,e> and change'
is of type «s,e>,t>, but neither Xx [change '(x)] nor Xx [change '(r)] is well-formed.
The former is ill-formed because x is of type e and so not of a type appropriate to be
the argument of change' while the latter is wrong because there is no instance of the
variable bound by X in the expression in its scope.
I^L also contains the logical quantifiers V and 3 which like X may bind variables
of any type. So, for example, V/? [necessarily '(p)] is a well-formed formula which
asserts that all expressions of type <s,t> are necessarily true while 3A
[(good '(A))(bertie')] asserts that Bertie is a good something or, more accurately, there
is some property with respect to which Bertie is good. The identity operator = in LIL
may also conjoin expressions of any type to yield a formula, as defined in (16.5). In
the rules that construct quantified formulae in (16.13) and (16.14), the universal and
existential operators are operators over extensional formulae (i.e. expressions of type
t) and not over intensional ones (of type <s,t>) and their outputs are again extensional
formulae. The same is true for the tense and aspect operators, Past, Fut, Pres, Perf
and Impf and the propositional connectives, &, v, —>, <-> and ~.
The full definition of Lu. is given in (16) in the form of a recursive definition of
what constitutes the set of meaningful expressions of a certain type. Anything that
does not conform to one of these rules is not a meaningful expression of the logic and
so does not have an interpretation in the model. (16) contains no rules for constructing
modal expressions using the operators • and o, as these these have now been
supplanted by their counterparts necessarily' and possibly \ as discussed in Section
10.1.
(16) a. If a is an expression of type <b,a> and p is an expression of type b, then
oc(p) is an expression of type a.
b. If a is an extensional expression of type a, then A a is an expression of type
<s,a>.
c. If a is an expression of type <s,a>, then "a is an expression of type a.
d. If a is an expression of type a containing a free instance of a variable u of
type b, then Xu [a] is an expression of type <b,a>.
e. If a, P are expressions of type a, then a = P is an expression of type t.
f. If (|) is an expression of type t, then ~(|) is an expression of type t.
g. If <|), \j/ are expressions of type t, then (((> & y) is an expression of type t.
h. If (j>, \|/ are expressions of type t, then (cj) v \\f) is an expression of type t.
i. If ((), \\f are expressions of type t, then ((() —> \\f) is an expression of type t.
j. If (|), \\f are expressions of type t, then (((> <-> \|/) is an expression of type t.
k. If (j) is an expression of type t containing an instance of a free variable u of
any type, then Vu [$] is an expression of type t.
1. If <|) is an expression of type t containing an instance of a free variable u of
any type, then 3u [(()] is an expression of type t.
292
The intensional language L1L
m. If <>t is an expression of type t, then Past (<(>) is an expression of type t.

n. If 4> is an expression of type t, then Fut (^)) is an expression of type t.
o. If 0 is an expression of type t, then Pres (())) is an expression of type t.
p. If <>| is an expression of type t, then Perf ((()) is an expression of type t.
q. If (j) is an expression of type t, then Impf ((()) is an expression of type t.
10.2.2 ... And their interpretation

Having defined the language W , into which expressions from English will now be
translated, we must specify its interpretation, thus, as usual, providing an indirect
interpretation of translated English expressions. First of all, we need a definition of
the sorts of denotation that expressions of different intensional types may have and
then we will move on to a consideration of intensional models and the formal
definition of the model theory that will interpret all expressions in LiL with respect to
appropriate models.
Given the definition of an intension as a function from indices to extensions,
from Section 10.1, the possible denotations of intensional types are easy to define. In
all cases, an expression with an intensional type <s,a> denotes a function from indices
(i.e. world/time pairs) to denotations of type a. The set of all possible world/time pairs
can be symbolised by the Cartesian product of the set W of possible worlds and the
set I of temporal intervals defined by the set of times T. The Cartesian product of two
sets A and B is written as A x B and is equivalent to the set of all ordered pairs where
the first element is a member of set A and the second is a member of set B, as defined
in (17).
(17) Cartesian product: A x B = {<a,b> l a e A & b e B}.
The Cartesian product of W and I is thus, by (17), the set of all pairs of possible
worlds and temporal intervals, i.e. all indices defined by the model. The denotation of
an expression of any intensional type is, therefore, taken from the set of functions
from the Cartesian product of W and I to denotations associated with its associated
extensional type as formally defined in (18.a) and exemplified in (18.b).
(18) a. D<s>a> = Dawxi
b.
A {1,0}
While all intensional expressions have the same sort of formal denotation as in
(18.a), expressions of certain types have a special significance in translating
expressions from a natural language like English and may be associated with
particular philosophical concepts. For example, the intensions of formulae, which are
293
expressions of type <s,t>, have as their denotations functions from world/time pairs
to truth values, i.e. members of {0,l}WxI (cf. e.g. (4) above which gives the denotation
of the expression Alaugh'(fwna') in M10). Because such functions specify the truth
value of a formula in any situation given by the model, they can be thought of as
corresponding to propositions. This captures the intuition that the proposition
expressed by a sentence is that part of its meaning that provides information about the
situations that can be truly described by the sentence.
The intensions of individual constants of type <s,e> are called individual
concepts and denote functions from indices to the set of entities in the model, i.e.
AWx *, a function that picks out an individual entity at any index. In M10, there are five
such functions, one for each of the individual constants ethel \ jo \ bertie \ fiona' and
Chester'. The denotation of the individual concepts of these names is, in each case, a
constant function, i.e. a function that picks out the same entity at every index, i.e.
WOMANj, MANi, MAN2, WOMAN2 and DOG, respectively. More interesting
individual expressions are those that pick out different entities at different indices. The
most obvious examples of such names are titles like Queen of Scotland, President of
the USSR, Mr Universe, Mayor of New York, and so on. The extensions of these
names differ from time to time and from world to world. So, for example, while the
President of the Russian Republic in 1991 was Boris Yeltsin, in 1975 no-one bore this
title and in 1995 it could refer to someone else. Furthermore, we can think of different
situations in which other people bear these titles. So, it is possible to envisage a world
in which the current Queen of Scotland is not also the current Queen of England, i.e.
Elizabeth Windsor, but a descendant of Bonnie Prince Charlie called Morag Stuart;
or one in which Britain had become a republic in the 1860s, and thus where there was
no current Queen of England.
The intensions of one-place predicates are of type <s,<e,t» and so denote
functions from indices to sets of entities, i.e. functions in ({0,l}A)WxI. Such functions
pick out the set of all entities that are in the extension of a predicate at each index and
so may be referred to as properties. Thus, formally, the extension of the predicate red
is a set of entities that are red while the property of redness is a function that specifies
the extension of the predicate in every situation, i.e. Ared'. An example of such a
property is specified for the model M10 in the diagram in (3) which represents the
function that picks out the set of entities that are laughing at any index. We can follow
Montague and generalise the term property to the denotations of expressions of any
type with the structure <s,<a,t» where a is any type. Such expressions denote
functions in ({0,l}Da)WxI and, because such functions pick out sets of entities of the
sort denoted by expressions of type a at every index, they may be said to be properties
of such entities. Thus, expressions of type <s,<e,t» denote properties of individual
entities, those of type <s,«e,t>,t» denote properties of sets, those of type
<s,«s,t>,t» denote properties of propositions, and so on.
Two-place predicates extensionally denote relations between entities, and the
intensions of such expressions, of type <s,<e,<e,t>», are called relations in
intension between entities, denoting functions in (({0,l}A)A)w x! . Like the term
property, this term can also be generalised to cover all expressions that denote the
intensions of two-place functions, i.e. those with the type structure <s,<a,<b,t>»
where a and b are any type. Thus, expressions of type <s,«s,t>,<e,t>» denote
relations in intension between entities and propositions, expressions of type
294
The intensional language L!L
<s,«s,e>,«s,e>,t»> denote relations in intension between individual concepts, and

so on. The different denotations of expressions are summarised in (19) according to
whether they are intensional or extensional. There are, of course, many other
intensional types and we will come across more in the sections that follow, but they
all have the same sort of denotations, i.e. functions from indices (W x I) to extensional
denotations.
(19) Category Extensional Extensional Intensional Intensional
Name Type Denotation Type Denotation
Individual e Entity <s,e> Individual
constant A concept
A WxI
Formula t Truth value <s,t> Proposition

{1,0} {l,0} WxI
One-place <e,t> Set of <s,<e,t» Property

predicate entities of entities
({1,0}A)W*I
Two-place <e,<e,t» Relations <s,<e,<e,t»:> Relation in

predicate between intension
entities between
entities
({0,l}A)A (({0,l}A)A)W*I
Exercise 10.3:
What sorts of denotation do expressions of the following types have? Give the formal
definition of the denotation, describe it in words and try and think of an English
expression that might have such a denotation.
ii.
iii.
iv.
v.
vi.
The models used to interpret expression of LIL consist of the ordered sextuple in
(20).
(20) Intensional model: M = «A,F,W,R,T,o>.
As before, A is a set of entities defining the ontology of the model, containing all the
entities that exist, have existed, will exist and might exist. W is a set of logically
possible worlds and R is some relation of accessibility between them (or indeed a set
295
of such relations; see Section 9.3.2). As in Chapter 8, T is a set of times ordered by

the precedence relation, <. F is again a denotation assignment function, but now
complicated to account for intensions. It assigns an intension to each lexeme in the
object language, thus providing an extension for each basic expression at each
world-time co-ordinate in the model. F can thus be defined as a function from lexemes
to a function in DaWxT, where a is an extensional type. As with temporal models,
there must be no gaps in the intension assigned to a basic expression, either in worlds
or in temporal intervals in those worlds. In other words, there must be no index for
which the extension of any constant is not defined. Thus, the union of all worlds in the
domain of F for some constant a must be equal to W, the set of all possible worlds,
and the union of all time intervals mentioned by F must equal the set of times, T, for
all worlds in W. Thus, F must conform to the two conditions in (21) for all basic
expressions, a.
(21) a. UwnF(a)(<wn,i>) = W.
b. U F(a)(<wn,i>) = T, for all wn e W.
The diagram in (4) that represents the intension of laugh' in M10 can thus more
formally be specified as a set of extension plus co-ordinate pairs (or, equivalently, a
function from indices to characteristic functions of entities), as in (22). The extension
of ¥l0(laugh') need not be specified for every index, provided that there are meaning
postulates that specify how to get to the extension of the constant at any index.
Because laugh is an activity verb, there is a meaning postulate that specifies that the
extensions of the predicate at any interval holds of all subintervals. Thus, in (22) the
extension of laugh' at the index <Wi,[t2]> is the same as ¥Maugh') at the index
(<W!,[ti,t2]>), i.e. {MAN^WOMANJ, since [tj is a subinterval of [tl9tj.
Furthermore, the rule for computing the extensions for intervals not specified by the
denotation assignment function also holds (cf. Chapter 8). Strictly speaking, the
meaning postulates defining the properties of predicates and the general rule
determining extensions at other intervals hold of the denotations assigned by F, but
do not determine them. Formally, therefore, the denotation assignment function should
directly specify the extensions of a predicate at all indices. However, since such a
specification is time consuming and tedious, in the example models in this chapter
the denotation assignment functions give the denotations for only the minimum
indices necessary to determine extensions at other indices, according to the rules given
in Chapter 8.
(22) FI0(laugh') = {«w1,[t,,t2]>,{ MAN1,WOMAN2}>,
«w1,[t3,t4]>,{MAN1}>,
«w2,[t1,t3]>,{DOG}>,
«w 2 ,[t 4 ]>,0>,
«w3,[t2,t3]>,{ WOMAN2}>,
«w3,[t4]>,{WOMAN1, WOMAN, }>,
«w 4 ,[t! ,t4]>, {MANXMAN,, WOMAN,, WOMAN! ,DOG} >}
The extension of a basic expression a at a particular index, <wwi>, is thus the
value of F at that index. Hence, the base of the model theory used to interpret LIL is
given by the statement in (23), where F(oc)(<Wn,i>) is the function F(oc) applied to the
296
index <WnJ> which yields the extension of a at that index. This rule is generalised
to obtain the extension of any intensional expression. Thus, where a has an intensional
type, the denotation of vcc at an index <w,,,i> is determined by the denotation of a (an
intension) applied to the index <w,»i>.
(23) a. If a is a basic expression, then [a]M'«wn'i = F(a)(<wn,i>).
b. fa]M-8wniis [a]M*wni(<wI1,i>).
The intension of a basic expression, a, is F applied to a, i.e. [Aa]M«wn J = F(a).
In general, the intension of an expression Aot in Ln. is a function that picks out the
denotation of a at each index, as formally defined in (24).
(24) If a is an extensional expression of type a, then [Aa]M'«wni is that function,
h, with domain W x I such that for all <wm,P> in W x I, h(<wm,P>) =
[a]M*wmi'.
The intension of an expression is the same at every index, unlike the extension which
may vary from index to index, because intensions are defined over all indices and
make no special mention of the reference index. We may, therefore, equate [Acc]M*wn-i
with s[a]M<8, the intension of an expression with respect to a model and a variable
assignment, irrespective of index.
To ascertain the denotations of complex expressions, whether intensional or
extensional, we simply adopt a variant of the basic rule for interpreting expressions
formed via functional application, relativised, not just to a temporal interval as in
Chapter 8, but also to an index consisting of a world and a time. Thus, the value of
[oc(P)]M'8wni is ascertained by applying the value of the functor a at the reference index,
<wn,i>, to the value of p at the same index. Thus, for example, to ascertain the
extension of the expression laugh'(Chester') with respect to M10 and the index
<w3,[t2,t3]>, we apply the function denoted by laugh' at <w3,[t2,t3]> to the entity
denoted by Chester' at that index, i.e. [laugh'(Chester')]M^V/3^2^ is
[laugh^^^Klchestery*****2*]). The value of [laugh']M*<w3wwis the characteristic
function of the set {WOMANj} and the value of [chester '] M -gw3 -w-w (as at every index)
is DOG. Since the latter is not in the set denoted by the functor at the reference index
<w3,[t2,t3]>, laugh'(chester') is false at that index. Nothing has, therefore, changed in
the interpretation of functor-argument expressions, except that care must now be taken
to ensure that the denotations of the component parts of a complex expression are
ascertained at the same index. Furthermore, caution must be exercised to ensure that
the correct sort of denotation, intension or extension, is selected for each expression.
This will be important in Section 10.3 and discussion of the details is postponed until
then.
As for the rest of the model theory used for interpreting expressions in LIL, very
little differs from the rules we have already come across in previous chapters except
again for the relativisation of denotations to indices. Note that the interpretation of a
variable, w, is, as in Chapter 4 onwards, just the value assigned to u by the value
assignment g, i.e. [u]M'«wni = g(u). In other words, assignments of values to variables
are constant from index to index, as one might expect. Everything else remains much
the same as before and the complete formal specification of the model theory is given
in (25) which includes the interpretations of the expressions necessarily', possibly',
must' and may \ since these are logical expressions whose interpretation does not vary
297
from model to model. Students are advised to take some time to go through this before
going on to the discussion of the important implications of this theory for an account
of opaque contexts in the next section.
(25) a. If a is a basic expression, then [a] M * wn4 = F(a)(<w n ,i>).
b. If a is a variable, then [a]^-*" 4 = g(oc).
c. If a is an expression of type <b,a> and (i is an expression of type b, then
[a(p)]M-8-wni is [a]M'8'wni([p]M'8wni)-
d. If a is an extensional expression of type a, then [Aa]M-8wn4 is that function
h with domain W x I such that, for all <wm,F> in W x I, h(<w m ,i'>) =
[tt]M*wm4' ( = 3[(X]M>8).
e. If a is an expression of type <s,a>, then [/oc]M,g,wn,i j s [a] M '8 wni (<w n ,i>).
f. If a is an expression of type a containing a free instance of a variable u of
type b, then [Xu [a]]M'«wn *is that function h with domain D b such that for any
object a in D b , h(a) = [oc]M'8'wni where g' is exactly the same as g except
perhaps for the fact that g'(ii) = a.
g. If a and |3 are expressions of type a, then [ a = (3]M«wni is 1 iff [a] M « wni is
identical to [p]M,g,wn,i Otherwise, [a = (3]M'8wni is 0.
h. If <|) is an expression of type t, then [~<j)]M« wn îs 1 iff [c))]M,g,wn,i [s Q Otherwise,
[H[)]M'8'wni is 0.
i. If 4> and \\f are expressions of type t, then [((() & \j/)]M,g,wn,i j s x iff [(j)]M,g,wn,i j s
1 and [\|/]M.g,wn,i i s a i s o x. Otherwise, [(<|) & \|/)]M,g,wn.i i s Q
j. If cf> and \|/ are expressions of type t, then [((() v \|/)]M,g,wn,i i s x iff [(|>]M,g,wn,i [s
1 or [\|/]M.g.wn,i i s x. Otherwise, [((() v \j/)]M,g,wn.i j s Q
k. If <() and \|/ are expressions of type t, then [((() -»\|/)]M,g,wn,i i s x iff [<|)]M,g,wn,i i s
0 or [\|/]M,g,wn,i is x. Otherwise, [(()) —»\j/)]M.g.wni is 0.
1. If <|) and \|/ are expressions of type t, then [((() <-> \(/)]M.g,wn,i i s \ iff [(|)]M,g,wn,i a n ( j
[\j/]M,g,wn,i both have the same truth value. Otherwise, [(()) <-> \|f)]M,g,wn,i ls Q
m. If 0 is an expression of type t containing an instance of a free variable u,
then [Vu [a]] M '« wni is 1 iff [<t>]M-8'wni is 1 for all value assignments g' exactly
like g except perhaps for the value assigned to u.
n. If (() is an expression of type t containing an instance of a free variable u,
then [3u [a]] M '8 wni is 1 iff [c|)]M,g\wn,i i s \ for S O me value assignment g' exactly
like g except perhaps for the value assigned to u.
o. If 0 is an expression of type t, then [Pres ((j))]M.g.wn,i i s x iff [(|>]M,g,wn,i i s \
p. If <>| is an expression of type t, then [Past ((())]M-8wni is 1 iff [c()]M<8'wno is 1 for
some j in I where j < i.
q. If <>| is an expression of type t, then [Fut (c|))]M8.wn.i j s x iff [(|>]M.g,wn,j i s x for
s o m e j in I where i < j .
r. If (() is an expression of type t, then [Perf ((())]M-8-wni is 1 iff [<()]M-8.wn,i i s x and
1 contains initial and final points if <|) inherently contains these.
s. If $ is an expression of type t, then [Impf ((|))]M8.wn.i is 1 iff there is an
interval j such that i is a proper subinterval of j and [<J>]M-»o is 1.
t. If <|) is an expression of type t, then [necessarily '(A(j))]M.g,wn,i i s \ iff [(|)]M,g,wtno
is 1 for all j G I and all w m e W such that w n Rw m .
u. If <>| is an expression of type t, then [possibly '(^t))]1*'8'™4 is 1 iff [<|)]M.g,wmo ls
1 for s o m e j € I and some w m e W such that
298
Interpreting opaque contexts
v. If (() is an expression of type t, then [mwsf *(A<|>)]M*wn'i is 1 iff [cj)]M,g.wm,i j s \ for

a l l w m e W such that w n Rw m .
w. If <>| is an expression of type t, then [may '(A<t))]M>8 wn>i is 1 iff [c|)]M,g,wm,i [s \ for
some w m e W such that WnRwm.
10.3 Interpreting opaque contexts

We are now in a position to show to what extent the problems posed by the opaque
constructions discussed in Section 9.1 are solved by the adoption of the intensional
semantics defined above. We begin by considering the interpretation of verbs with
opaque noun phrase or verb phrase object positions like want, try and seem and then
move on to a consideration of verbs of propositional attitude like believe, know, etc.
10.3.1 Oblique transitive verbs

As we saw in Section 9.1, the fact that someone wants or is looking for something
like a unicorn, a drink or a secretary does not entail that there is a specific unicorn,
drink or secretary that they want or are looking for nor that any entity with the
property specified actually exists. This differentiates such verbs from the ordinary
transitive verbs like like and stroke with respect to which quantified object noun
phrases always take wide scope, because the lexical decomposition of the latter
guarantees that they denote relations between entities (cf. Chapter 6). For example, the
sentence in (26.a) translates into LQ as the complex expression in (26.b). After lambda
conversion this reduces to (26.c) from which the entailments in (26.d) and (26.e) can
be straightforwardly derived through conjunction elimination. These verbs also allow
the substitution of referential equivalents in their object position, as shown by the
valid inference pattern in (27).
(26) a. Jo liked a student.
b. XQ [QQo')]QJ> Pu [P(Xy [(like'(y))(x)])]](XP @z [student'(z) & P(z)]])).
c. 3z [student'(z) & (like'(z))(jo')].
d. 3z [student' (z)].
e.
(27) a. Jo liked a student.
b. The student Jo liked is a wealthy Glaswegian.
c. Therefore, Jo liked a wealthy Glaswegian.
As we have seen, opaque transitive verbs do not have the same properties as
transparent ones, as shown in (28): (28.b) cannot be inferred from (28.a) nor can
(28.d) be inferred from (28.c) despite the fact a unicorn and a gryphon are referential
equivalents in the actual world (i.e. they both extensionally denote the null set).
(28) a. Jo sought a student.
b. There is a student that Jo sought.
c. Bertie wanted a unicorn.
d. Bertie wanted a gryphon.
299
Because of the failure of Leibniz' Law in the object position of verbs like want, seek,
imagine and look for, these are analysed as functors with an intensional argument
position. It could be argued that, since ordinary transitive verbs ultimately denote
relations between entities of type <e,<e,t», opaque verbs denote relations between
individual concepts and entities, and so translate into expressions of type
« s , e > , < e , t » . This would go some way towards a solution, since it allows one to
distinguish between the particular individuals being sought or imagined. For example,
the sentence in (29.a) translates into UL as the formula in (29.b), assuming the type of
imagine' to be « s , e > , < e , t » . This translation ignores tense and treats the definite
noun phrase The Ladies' Scottish Golf Champion as translating into the individual
(sublimation of the) constant, sgc\ rather than as a proper Russellian definite
description.
(29) a. Jo imagined the Ladies' Scottish Golf Champion,
b. (imagine'(Asgc))(jo').
According to the model theory specified in Section 10.2, the formula in (29.b) is true
at an index <wn,i>, if the entity denoted by jo' at <wn,i> is 'in an imagining relation'
not with the entity extensionally denoted by the Ladies' Scottish Golf Champion at
that index, but with its associated individual concept, i.e. with [Asgc']M<* rather than
[sgc']M^wn\ If Ethel and the Ladies' Scottish Golf Champion happen to be one and the
same at some index, it does not therefore follow that because Jo is imagining the
Ladies' Scottish Golf Champion, he is imagining Ethel. This is because the individual
concept of ethel', the function that picks out Ethel at every index, may differ from that
of sgc', the function that identifies the Ladies' Scottish Golf Champion at every
index. In model M10, the name Ethel refers to the same entity in every world at every
time. Hence, 3[ethel']M10'8 is the constant function in (30.a). The Ladies' Scottish Golf
Champion, however, is not the same person at every index but in world Wi sgc'
extensionally denotes Ethel (WOMAN0 at all times, but in w 2 it denotes the dog at
the interval [tiytj and Bertie at [t^tj. In world w3, Fiona is the Ladies' Scottish Golf
Champion for the first three moments in the model, but Ethel in the last moment. In
the final world w4, Fiona is the Ladies' Scottish Golf Champion at the interval [ti,t4].
The denotation of Asgc' in M w is, therefore, the function represented in (30.b).
(30) a. 3[ethel']M10-8
MAN 2
<w 3 ,[t 1 ,t 4 ]> WOMANj
<w 4 ,[t 1 ,t 4 ]> WOMAN 2
DOG
300
3[SgC]M10'*
<w2,[t1,t2]> MAN2
<w2,[t3,t4]> WOMANj
<w3,[t1,t3]> WOMAN2
<w3,[t4]> DOG
<w4,[t1,t4]>
Because the intensions oietheV andsgc' differ, it is possible for the denotation
of imagine' at some index to map one of them onto the characteristic function of a set
containing the entity denoted by jo \ but not the other. In other words, it is possible for
Jo to imagine the concept of the Ladies' Scottish Golf Champion without imagining
the concept of Ethel. Hence, Leibniz' Law fails in these contexts and so the inference
pattern in (31) is correctly predicted to be invalid under the assumption that imagine'
is of type «s,e>,<e,t».
(31) a. Jo imagines the Ladies' Scottish Golf Champion.
c. Therefore, Jo imagines Ethel.
Unfortunately, however, treating opaque verbs as relations between individual
concepts and entities is not sufficient, because they may take quantified noun phrases
as direct objects. If the argument type of predicates like imagine' is <s,e>, any
quantifiers in the translation of an object noun phrase still have scope over the verb.
The translation of a sentence like Jo imagined a unicorn is thus parallel to that of Jo
liked a student in (26.c) except that the variable in the object position of imagine' is
in an intensional position, i.e. 3x [unicorn '(x) & (imagine '(Ax))(jo')]. This allows the
application of standard entailment rules like conjunction elimination to get inferences
like those shown in (26.d & e). The type of opaque verbs like these cannot, therefore,
be as simple as «s,e>,<e,t», but must be a functor over full noun phrase
expressions. Indeed, this possibility is implicit in the type-raising of transitive verbs
required by the higher type accorded to full noun phrases in Chapter 6. As we saw
there, in order to maintain transitive verbs as functors over noun phrase denotations,
it was necessary to raise their type from <e,<e,t» to <«e,t>,t>,<e,t». While
transparent verbs like like' are lexically decomposed into a complex expression that
maintains the intuition that they denote simple relations between entities, this is not
the case for opaque verbs like imagine which must be interpreted as functors over the
intensions of noun phrases.
To define the type and possible denotations of intensional transitive verbs, we
need first to ascertain the types of noun phrases in L1L. Montague analysed noun
phrases as denoting, not sets of sets of entities, as we did in Chapter 6, but sets of
properties of individual concepts. In other words, he translated noun phrases into
expressions of type «s,«s,e>,t»,t>, rather than as expressions of type «e,t>,t>.
His analysis, however, seems much more complex than is actually required for the
301
10 lntensional semantics
interpretation of noun phrases. In the first place, there seems to be no reason to take
VP denotations (the arguments of subject NPs) to be sets of individual concepts, i.e.
to assign them the type «s,e>,t>. The subject positions of most verbs are extensional
and so require extensional arguments (of type e), not intensional ones (of type <s,e>).
For those verbs that require intensional subjects (e.g. appear; see below), the
argument position of the VP must be a full intensional NP denotation in order to
include quantifiers, and not just an individual concept. Hence, we can maintain the
translation of (most) VPs into one-place predicates. Furthermore, it seems unnecessary
to require noun phrases to apply to properties (denoted by expressions of type
<s,<e,t») rather than sets. If the translations of noun phrases were intensional in their
argument position we would expect there to be noun phrases that do not allow
substitution of extensional equivalents in this position. It seems likely, however, that
there are no noun phrases of this sort and that wherever a noun phrase applies to a
one-place predicate, truth is preserved if an extensionally equivalent predicate is
substituted. For example, it seems intuitively correct to say that if Every student sings
well is true and that all those things that sing well are also those things that are happy
then Every student is happy is also true. Hence, we can maintain the assumption made
in Chapter 6 that NPs denote extensional generalised quantifiers and so keep the same
type in L^ as they have in LQ.
Keeping the type of noun phrases as «e,t>,t> means that the type of intensional
transitive verbs must be « s , « e , t > , t » , < e , t » , denoting relations between the
intensions of sets of sets of entities and sets of entities. A verb like seek thus
translates into the constant seek\ denoting a function from noun phrase intensions to
sets of entities. The translation of the syntactic rule generating VPs containing
transitive verbs and their objects must thus combine the verb translation with the
intension of the object NP, not its extension. A VP containing a transitive verb and its
object is translated as V'(ANP') and the sentence in (32.a) is translated into the LIL
expression in (32.b) which is equivalent to (32.c) after lambda conversion.
(32) a. Bertie sought a unicorn.
b. Past (kQ [Q(bertie')](seek'(A^P px [unicorn'(x) & P(x)]]))).
c. Past ((seek'(A^P [3x [unicorn'(x) & P(x)]]))(bertie')).
Unlike ordinary transitive verbs, verbs like seek are not further semantically
decomposed and so no further reduction of (32.c) can take place. Informally, (32.c)
asserts of Bertie that he is seeking, not a specific unicorn, but the concept or idea of
a unicorn. Hence, he need be seeking no specific unicorn and no unicorns need exist
for him to be seeking them. Formally, (32.c) is true just in case at some time
preceding the time of utterance in the actual world, the entity denoted by bertie' stood
in a relation of seeking to the intension of a unicorn. The latter is a function that picks
out the set of sets at each index that include some unicorn, i.e. one that picks out the
existential sublimations of unicorn' at each index. Hence, [seek '(AXP [3x [unicorn '(x)
& PMJlXbertie')]"1*'™*. is true if, and only if, the function [seeky]M^ni maps the
intension of the existential sublimation of unicorn\ i.e. z[XP [3x [unicorn'(x) &
P(x)]]]M>* onto a set that contains [bertie']M*wni at <w,,,i>. Because s[XP [3x
[unicorn'(x) & P(x)]]]M& is that function that picks out the existential sublimations
of [unicorn '] M * wn j at each index and not its particular existential sublimation at <w,,,i>,
it follows that no specific entity need be being sought by Bertie at <wmi>.
302
Furthermore, since the intension of an existential sublimation may include indices like
the actual world at the present time where there are no unicorns, i.e. where [XP [3x
[unicorn'(x) & P(x)]]]M^wni is null, no unicorns need exist at the reference index for
the expression in (32.c) to be true. Hence, this analysis properly accounts for the
invalidity of the entailment between (28.a) and (28.b) in its non-specific reading.
Because the object position of seekis an intensional one, Leibniz' Law is not valid and
so we cannot infer from the truth of Bertie sought a unicorn that Bertie sought a
gryphon is also true.
This analysis of verbs which have an opaque direct object position thus captures
the semantic properties of non-specific readings. As we have seen, however, sentences
like Bertie sought a unicorn and Jo imagined the Ladies' Scottish Golf Champion are
ambiguous between the non-specific reading analysed above, and a reading where
Bertie is seeking a specific unicorn and Jo is imagining a specific person who is the
Ladies' Scottish Golf Champion. In these latter cases, one can infer the existence of
an entity with the property specified by the object noun phrase and also that
substitution of extensional equivalents is valid. Since these are semantic properties
associated with transparent transitive constructions (ensured by giving any quantifier
in the object noun phrase wide scope over the verb), the simplest way to analyse
specific readings of intensional verbs is to analyse them in the same way as
transparent verbs. Scope ambiguities between quantifiers in subject and object noun
phrases and between the logical quantifiers and the tense operators were discussed in
Chapters 6 and 8. Such ambiguities seem to be a general property of the semantics of
natural language expressions: wherever sentences and the formulae into which they
translate contain more than one operator the sentence has different readings according
to the different scopes that the operators can have. If this is a general property of
natural languages then we would expect ambiguity between the scopes of the intension
operator A and quantifiers contained in noun phrases to parallel scope variations of the
sort already seen. Since we have already identified an ambiguity in sentences
containing opaque transitive verbs, it seems natural to treat them as deriving from
differences in the scope of operators.
As discussed in Chapters 6 and 8, wide scope quantification can be derived in the
semantics through the rule of quantifying in. In order to derive the specific reading
of the sentence Bertie sought a unicorn using this rule, wefirstcombine the predicate
seek* with an expression containing a free variable of type e of the right type to
combine with an intensional transitive verb. This expression is the intension of the
individual sublimation of a free variable of type e, i.e. AXQ [Q(z)]. The resulting
expression, seekf(AXQ [Q(z)])> is then combined with the subject translation XP
[P(bertie')] to yield XP [P(bertie ')](seek f(AXQ [Q(z)]))- Finally, the lambda operator
is used to abstract on the free individual variable z and the resulting expression is
combined with the translation of a unicorn to give a formula with no free variables,
i.e. ^P, [3y [unicorn'(y) & Pi(y)]](Xz [X? [P(bertie')](seek'(A^Q [Q(z)]))]). This
derivation is shown in (33) in the form of a semantic analysis tree, each line of which
shows a functor to the left combining with an argument expression to its right. The
types of the expressions being combined are written underneath to show that the
combinations are valid according to the rule of functional application and the rule of
quantifying in. (Tense is ignored for the sake of simplicity.)
303
(33)
XP [By [unicorn'(y) & P (y) ]](Xz [XP [P(bertie')](seek'(^Q [Q(z)]))])
XP [3y [unicorn'(y) & P (y) ]] XP [P(bertie')](seek'(A^Q
XP [P(bertie')]
seek'
The expression derived in (33) is equivalent to the formula in (34.a), as can be

shown by applying lambda conversion to it. From the truth of this formula the truth
of those in (34.b) and (34.c) can also be inferred by conjunction elimination.
(34) a. 3y [unicorn'(y) & Xz [(seek'(AXQ [Q(z)]))(bertie')](y)].
b. 3y [unicorn'(x)].
c. 3y [Xz [(seek'(AXQ [Q(z)]))(bertie')](y)]
Because the existential quantifier is outside the scope of A in (34.a), Leibniz' Law
applies and so any extensionally equivalent expression may be substituted for unicorn'
and maintain the truth value of the original formula. If this formula is true and all
unicorns are one-horned horses, then we can infer from the specific reading of Bertie
sought a unicorn that Bertie sought a one-horned horse is also true. The reason for
this is that in the interpretation of (34.a) the extension of unicorn' is assessed at the
reference index, i.e. at the world and time of utterance. Hence, any expression
extensionally equivalent to unicorn' at the relevant index may be substituted salva
veritate.
The interpretation of (34.a) may, however, seem a bit strange: what is it to be
seeking the concept of the individual sublimation of a variable? Because there is one
and only one individual sublimation for any individual at an index, being in a relation
with an individual sublimation is tantamount to being in a relation with the individual
that it is a sublimation of, as we saw in Chapter 6. Furthermore, the concepts of
individual sublimations are also unique to that individual. The modal formula in (35)
is thus valid in Ln, and it follows that predicating something of the concept of an
individual sublimation is equivalent to predicating something of the individual itself.
Thus, to be seeking the concept of an individual sublimation where that individual is
a unicorn is to be seeking an actual individual with that property. Indeed, Montague
provides a general formula to relate relations between entities and concepts of
individual sublimations and the corresponding relations between entities. This is
304
defined in (36.a) which relates an expression 8 of type « s , « e , t > , t » , < e , t » to a

corresponding expression & of type <e,<e,t». How this applies to the example in (33)
is shown in (36.b) to (36.f). This takes the formula in (33) after the translation of
Bertie has combined with the predicate seek '(AXQ [Q(z)]) but before the object noun
phrase translation is quantified in, i.e. the expression in (36.b). Lambda conversion is
applied to get (36.c) at which point the reduction in (36.a) can be applied to get the
expression in (36.d). Quantifying in the translation of a unicorn yields the expression
in (36.e) which is equivalent after lambda conversion to the formula in (36.f).
(35) VxVy D[(x = y) <-> (AA
(36) a. 5* =def Xy [Ax [d(AXJ> [P(y)])(x)]].
b. XP [P(bertie')](seek'(AA,Q [Q(z)])).
c. (seek'(AAQ [Q(z)]))(bertie').
d. (seek\(z))(bertie').
e. XP py [unicorn'(y) & P(y)]](Xz [(seek',(z))(bertie')]).
f. By [unicorn'(y) & (seek\(y))(bertie')].
For reasons that will be discussed in Section 10.3.3, the A-expression Xz
[(seek'(AXQ [Q(z)]))(bertie')](y) in (34.c) cannot be reduced to (seek'(AXQ
[Q(y)]))(bertie'). However, the A-conversion can take place if seek'(AXQ [Q(z)]) is
reduced to seek*(z) using the definition in (36.a). The expression (seek*(y))(bertie') can
be obtained from Xz [(seek*(z))(bertie ')](y)by A-conversion and so the full expression
in (34.c) is equivalent to 3y [(seek*(y))(bertie')]. Hence, from the specific reading of
Bertie sought a unicorn, it can be inferred that there was something that Bertie sought,
as required.
Exercise 10.4:
In order to maintain a regular correspondence between syntactic rules and translation,
it is necessary now to assume that the rule combining any verb with its direct object
translates as the application of the translation of the verb to the intension of that of the
object, i.e. VP' = V'(ANP'). This entails that the translations of transparent transitive
verbs must be of type « s , « e , t > , t » , < e , t » , the same as opaque transitive verbs. In
order to maintain the fact that such verbs basically define a relation between two
entities, they can continue to be semantically decomposed, but must contain a variable
over the intensions of noun phrases. The translation for the verb like is the expression
in i. below, an expression that takes the intension of a noun phrase, makes it
extensional and applies this to the expression denoting the set of all things that are
liked. The output is an expression denoting the set of all entities that like things with
the properties given by the object NP. The two analyses described above for sentences
containing intensional transitive verbs also applies to those containing transparent
ones. Give both derivations for the sentence in ii. Is the ambiguity implied by the two
derivations semantically significant?
i. Xp [Xx Vp(Xy [like'(y)(x)])]].
ii. Fiona liked a student.
305
10.3.2 Control verbs

The interpretation of sentences containing opaque transitive verbs given in Section
10.3.1 is one of the major achievements of Montague (1973). It captures in a very
elegant way the primary semantic properties of the non-specific readings of such
sentences and also provides a means of accounting for their specific counterparts. The
general approach can be extended to account for verbs like want and try which take
verb phrase objects which have an opaque reading (cf. the invalid inference patterns
in Chapter 9). Such verbs, called control verbs in the linguistic literature, may thus
be analysed as denoting relations between properties and entities. Hence, they
translate into expressions of type «s,<e,t»,<e,t», as in the translation of the
sentence in (37.a) given in (37.b).
(37) a. Jo wanted to find a secretary.
b. Past (want'(A>-y [3x [secretary'(x) & (find'(x))(y)]])(jo')).
In sentences containing verbs like want and try, the matrix subject is interpreted
also as the subject of the complement verb phrase. Thus, in (37.a), Jo is both the
person doing the wanting and that doing thefinding,and this needs to be brought out
by the interpretation of the formula in (37.b). Since the identification of the subject of
the matrix clause with the omitted subject of the subordinate infinitive is a lexical
property of these verbs, we may use a meaning postulate to capture this relationship.
This appears in (38.a) and ensures that a relation between properties and entities, like
want\ of type « s , < e , t » , < e , t » is always associated with a relation between
propositions and entities, like want*, of type «s,t>,<e,t», where the subject
argument functions also as the subject of the embedded proposition to which it is
related. According to this meaning postulate (which can also be represented as in
(38.b)), the formula in (37.b) is truth-conditionally equivalent to that in (38.c), which
is interpreted as asserting of Jo that he wants it to be the case that hefindsa secretary.
(38) a. If 8 is an expression of type «s,<e,t»,<e,t», P is an expression of type
<e,t>, and x an expression of type e, then 5(AP)(x) is truth-conditionally
equivalent to 8*(AP(x))(x), where 5* is an expression of type «s,t>,<e,t».
b. VP [Vx [8(AP)(x) <-> 8*(AP(x))(x)]].
c. Past (want*(A3x [secretary'(x) & (find'(x))(jo')])(jo')).
Because the existential quantifier in (37.b) and (38.c) is in the scope of the intension
operator, we cannot infer from the latter that there is a specific secretary that Jo wants
to find, nor, for the same reason, can we substitute any predicate in (37.b) which is
extensionally equivalent to Xy [3x [secretary '(x) & (find'(x))(y)]] or any formula in
(38.c) which has the same truth value as 3x [secretary '(x) & (find'(x)Xjo')]. Hence,
we can account for the fact that if Jo wants tofinda unicorn it does not follow that he
wants to find a gryphon.
The specific reading of (37.a) where there is a particular secretary that Jo wants
to find can be achieved by quantifying in the translation of the object to yield the
expression in (39.a) which, by (38.a), is equivalent to (39.b). (This equivalence does
not result from X-conversion, which is invalid in intensional contexts; see Section
10.3.3.)
306
(39) a. 3x [secretary'(x) & want'(AA,y [(find'(x))(y)])(jo')].

b. 3x [secretary'(x) & wanf(A(find'(x))(jo'))(jo')].
We can infer from the truth of the formulae in (39) that there is a (specific) secretary
that Jo wants to find and that expressions extensionally equivalent to secretary* may
be substituted for it. However, we are not in a position to substitute any other
expression for the formula (find'(x))(jof) in (39.b) or the predicate Xy [(find'(x))(y)]
in (39.a) because these are in the scope of the intension operator. Thus, we cannot
infer from either reading of (37.a) that Jo wants to kiss a secretary, even if the set of
things that Jo finds at some index is the same as the set of things that he kisses, i.e. if
[Xy [(find(y))(jo')]]M*™ is the same as [Xy [(kiss'(y))(jo')]]M*™\
The analysis above provides a way of characterising the meaning of verbs like
want in terms of relations between properties, propositions and entities. It is also
possible to provide an analysis of verbs like seem or appear within the intensional
semantics developed so far. The sentence in (4O.a) has a non-specific reading in which
it cannot be inferred from the truth of (4O.a) that there exists a unicorn that is
approaching nor indeed that anything at all is approaching. Thus, it is the subject
position of such verbs that is intensional when they occur with a VP complement and
it is necessary to make sure that all quantifiers in their subject NPs are in the scope of
the intension operator. Hence, such verbs must denote relations between properties
and the intensions of noun phrases, i.e. translate into expressions of the very complex
type « s , < e , t » , « s , « e , t > , t » , t » . (4O.b) thus provides the translation of (4O.a).
(40) a. A unicorn appears to be approaching.
b. (appear'(Aapproach'))(A^P [3x [unicorn'(x) & P(x)]]).
The sentence in (4O.a) is a paraphrase of that in (41.a) where the verb appear
takes a full complement clause and has as its subject the dummy element it. If we
assume that the latter contributes nothing to the meaning of the sentence, we can
interpret the verb appear as denoting a function from propositions to truth values,
picking out just those propositions which appear to be true. Translating this category
of the verb as the expression appear* of type «s,t>,t>, we thus get (41.b) as the
translation of (41.a).
(41) a. It appears that a unicorn is approaching.
b. appear*(A3x [unicorn'(x) & approach'(x)]).
The paraphrase relation between (4O.a) and (41.a), which depends on the lexical
interpretations of the verb appear, can be captured by the meaning postulate in (42.a)
(with the formal representation in (42 .b)) which relates an expression, 5, of type
« s , < e , t » , < « s , < e , t » , t > , t » to another one, 8% of type «s,t>,t>. The appropriate
relation is stated as one of equivalence between formulae containing 8, predicated of
a property, AP, and the intension of a noun phrase, @, and those containing 8* which
functionally applies to the intension of the formula derived from applying the
translation of the subject to that of the complement, i.e. AC#?(P)). In this way, we can
guarantee that (4O.a) and (41.a) are truth-conditionally equivalent. (43) shows the
steps that are needed to derive (41 .b) from the expression in (43.a) which is
truth-conditionally equivalent to (4O.b) according to (42). The formula in (43 .b) is
derived by 'down-up (~A) cancellation' and that in (43.c) by X-conversion.
307
(42) a. If 8 is an expression of type « s , < e , t » , < « s , < e , t » , t > , t » , P is an

expression of type <e,t>, and p an expression of type «s,<e,t»,t>, then
8( A P)(p) is truth-conditionally equivalent to 8*(ACp(P))), where 8* is an
expression of type «s,t>,t>.
b. VP [Vp [8(AP)(p) <-> 8X A (>(P)))]].
(43) a. appear*(ACAAP [3x [unicorn'(x) & P(x)]](approach'))).
b. appear*(A(^*P \Bx [unicorn*(x) & P(x)]](approach*))).
c. appear*(A(3x [unicorn'(x) & approach'(x)])).
There is more to the semantics of these verbs than is touched on here and the
implications of the intensionality of subjects of verbs like appear deserve more
consideration, since this requires verb phrases to act as functors over their subjects,
rather than vice versa. However, it should be clear that the analysis presented here
does provide a reasonable basis for a satisfactory account of the opaque properties of
these expressions and the way the properties denoted by complement VPs can be
related to propositions involving subject noun phrases.
*Exercise 10.5:
The sentences in (4O.a) and (4O.a) appear to be three ways ambiguous. The sentence
in i. below best corresponds to the reading in (4O.b) and (4O.b). Show how the reading
in ii. may be derived using the quantifying in rule in Chapter 6. What problems arise
in trying to derive the reading that best approximates to the reading in iii. and what
implications does such a reading have for noun phrase interpretation in the analysis
developed in this book?
i. It appears that there is a unicorn and that it is approaching.
ii. There is a unicorn and it appears to be approaching.
iii. There is something approaching and it appears to be a unicorn.
10.3.3 Propositional attitudes

Verbs of propositional attitude like believe, know, doubt and consider also create
intensional contexts in their complement position, as we can see from the fact that the
Law of Substitution fails and ambiguities arise between de re and de dicto readings
of the same sentence (cf. Section 9.1). Because such verbs are opaque with respect to
their complement clauses, they translate into LIL expressions that denote functions
over propositions, in fact relations between propositions and entities, and so translate
into expressions of type «s,t>,<e,t». To translate English sentences like those in
(44) into the formulae in (45), the predicates translating the main verbs, believe or
know, combine first with the intension of the formula translating their complement
clause and then combine with the translation of the subject noun phrase to yield a
formula.
308
(44) a. Fiona believes that the Ladies' Scottish Golf Champion is wealthy.
b. A Glaswegian knows that Ethel is not wealthy.
c. Ethel knows that a Glaswegian is wealthy.
d. Every student believes that every lecturer is wealthy.
(45) a. Pres (believe' (APres (wealthy' (sgc')))(fiona')).
b. Pres (3x [glaswegian'(x) & know'(A(Pres (-wealthy'(ethel'))))(x)]).
c. Pres (know'(A(Pres (3x [glaswegian'(x) & wealthy'(x)])))(ethel')).
d. Pres (Vx [student'(x) ->
believe'(A(Pres (Vy [lecturer'(y) -> wealthy'(y)])))(x)]).
Because the complements of the predicates believe' and know' in (45) are in the
scope of the intension operator, Leibniz' Law does not hold in these positions, and so
one cannot substitute extensionally equivalent expressions within the complement
clause and preserve truth. As an demonstration that this is so, consider the sentences
in (46).
(46) a. Jo believes that the Ladies' Scottish Golf Champion is wealthy.
c. Jo believes that Ethel is wealthy.
The sentence in (46.c) is constructed from the other sentences in (46) by substitution
into the complement ofbelieve. As we have already seen, however, Jo's belief about
the Ladies' Scottish Golf Champion in (46.a) is not the same as a belief about Ethel,
despite the identity of these two entities at a particular index. This can be proved by
showing that the propositions denoted by The Ladies' Scottish Golf Champion is
wealthy and Ethel is wealthy formally define different functions from indices to truth
values. If these functions differ, then the function denoted by believe' will not
necessarily map them both onto the characteristic function of a set containing the
entity denoted by the subject, Jo. In other words, the sentence in (46.c) may fail to be
true even though both (46.a) and (46.b) are true.
The non-equivalence of (46.a) and (46.c) may be illustrated by working through
the interpretations of the complement clauses in (46) with respect to the model Mio.
(47) sets out the intension of the predicate wealthy' in this model.
(47) F1(<wealthy') = {«w 1 ,[t 1 ,t 4 ]>,{MAN 2 ,WOMAN 2 }>,
«w 2 ,[t 1 ,t 2 ]>,{ DOG,WOMAN2}>,
«w 2 ,[t 3 ,t 4 ]>,{ WOMAN 2 }>,
«w 3 ,[t 1 ,t 3 ]>,0>,
«w 3 ,[t 4 ]>,{WOMAN 1 }>,
«w 4 ,[t 1 ]>,{MAN 2 }>,
«w 4 ,[t 2 ,t 3 ]>,{WOMAN 1 ,WOMAN 2 ,MAN 1 }>,
«w 4 ,[t 4 ]>,{MAN 2 }>}
The denotation of the proposition ^(wealthy'(etheV)) in M10 is that function that
maps an index onto 1 if, and only if, the denotation of ethel' at that index is in the set
of entities denoted by wealthy' at that index and maps the index onto 0, otherwise.
Since Aethel' denotes a constant function in M,0(cf. (30.a)), A(wealthy '(ethel')) picks
out the set of indices at which WOMANi is in the extension ofwealthy'. According
to (47), Ethel is wealthy only in world w3 at moment [t4] and world w4 at interval [t2,t3]
309
and so the proposition denoted by ^(wealthy'(ethel')) in M J0 is the characteristic

function represented in (48).
(48) 3[wealthy' (ethel')
<w 2 ,[t 3 ]>

<w 2 ,[t 4 ]>
<w 3 ,[t 2 ]>

<w 3 ,[t 3 ]>
<w 3 ,[t 4 ]>
<w 4 ,[t 2 ]
<w 4 ,[t 3 ]>
<w 4 ,[t 4 ]>
The proposition denoted by A(wealthy'(sgc'))y on the other hand, is the characteristic

function defining the set of indices <wn,i> at which the denotation of sgc' at <Wn,i>
is in the set of entities denoted by wealthy at <Wn,i>. Since the extension of sgc'
changes from index to index, one needs first to ascertain its denotation at a particular
index and then check to see if this is in the set of wealthy entities at that index.
Comparing the intension of sgc' in (3O.b) with (47), we can construct the
characteristic function in (49) defining the proposition denoted by A(wealthy'(sgc')).
The function in (49) is the characteristic function of the set {<w2,[t!]>, <w2,[t2]>,
<w3,[t4]>, <w4,[t2]>, <w4,[t3]>} whilst that in (48) is the characteristic function of
{<w3,[t4]>,<w4,[t2]>,<w4,[t3]>}. Since these two sets are different, the propositions
denoted by Ethel is wealthy and The Ladies' Scottish Golf Champion is wealthy are
different formal objects and so it is possible to believe (or know or doubt) one of them
without believing (or knowing or doubting) the other. Although the function denoted
by the predicate believey may map the function in (49) onto the characteristic function
of a set containing MAN b the extension of Jo, it does not follow that it maps the
function denoted by ^(wealthy'(ethel')) onto a characteristic function of a set
containing MAN t . The proposition expressed by the sentence Jo believes that the
310
Ladies' Scottish Golf Champion is wealthy may thus be true while that expressed by
Jo believes that Ethel is wealthy may be false, despite Ethel's being the Ladies'
Scottish Golf Champion at the reference index.
(49) s[ wealthy' (sgc')] M1°«
<w x ,[t 2 ]>
<w 2> [t 2 ]>

<w 2 ,[t 3 ]>
<w 2 ,[t 4 ]>
<w 3 ,[t 2 ]>

<w 3 ,[t 3 ]>
<w 3 ,[t 4 ]>
<w 4 ,[t 2 ]>

<w 4 ,[t 3 ]>
<w 4 ,[t 4 ]>
The fact that substitution is not valid in the complements of verbs like believe
also accounts for why we cannot attribute a contradictory belief to Jo even if he
happens to believe that Ethel is not wealthy, while also believing that the Ladies'
Scottish Golf Champion is wealthy. If substitution were valid, (5O.b) would be
guaranteed to be true by the truth of (5O.a) together with (46.a) and (46.b) by
substitution and conjunction introduction. The full inference pattern is set out in LIL
formulae (ignoring the present tense operator for simplicity) in (51). (5 La) is the
translation of (46.a), (51.b) that of (5O.a), (51.c) is that of (46.b). Substituting sgc' for
ethel' into (51.b) yields the formula in (51.d) which together with (5La) leads to the
formula in (5Le), the intended translation of (5O.b), via conjunction introduction.
Conjunction introduction, unlike substitution, is valid in intensional contexts provided
that the contexts are kept constant. Thus, if someone believes that <() and also believes
that \j/ then they believe that <|) and \|/.
(50) a. Jo believes that Ethel is not wealthy.
b. Jo believes that the Ladies' Scottish Golf Champion is wealthy and that the
Ladies' Scottish Golf Champion is not wealthy.
311
(51) a. believe'(A[wealthy'(sgc')])(jo').
b. believe'(A[~wealthy'(ethel')])(jo').
c. ethel' =sgc\
d. believe' (A[~wealthy' (sgc')] )(jo').
e. believe'(A[ wealthy'(sgc') & -wealthy'(sgc')])(jo').
As we have seen, the inference pattern in (51) is invalid, because the formula in (51 .d)
cannot be inferred from (51.b) and (51.c) via substitution because of the intension
operator. The truth of (51.e), which ascribes the contradictory belief to Jo that the
Ladies' Scottish Golf Champion is both wealthy and unwealthy, thus does not follow
from the truth of (51 .a), (51 .b) and (51 .c).
There is, however, a reading of (5O.b), that can be derived from the truth of the
premises in (46.a), (46.b) and (5O.a), which does not attribute a contradictory belief
to Jo. In this reading, the first conjunct in (5O.b) is interpreted de dido, as in (51.a),
but the second is interpreted de re. In other words, Jo's belief about the Ladies'
Scottish Golf Champion being wealthy is taken as ascribing to Jo a belief about the
wealth of whoever is named by the expression the Ladies' Scottish Golf Champion,
whose identity he does not know but who exists within his 'belief worlds'. His second
belief, on the other hand, is interpreted as a belief about the particular person referred
to by this expression at a particular index, i.e. Ethel. Thus, while Jo believes de die to
that the Ladies' Scottish Golf Champion is wealthy, he also believes de re, i.e. from
the speaker's perspective, that she is not wealthy.
This second reading of (5O.b) can be derived in much the same way as the
specific readings of opaque transitive verbs were derived in Section 10.2.2 through the
use of the rule of quantifying in. The de re reading of the sentence in (46.a), Jo
believes that the Ladies' Scottish Golf Champion is wealthy, can be constructed by
combining the predicate wealthy' translating the verb phrase is wealthy with the
individual sublimation of a variable, i.e. XQ [Q(x)]. Once this is combined with the
predicate believe' and the individual sublimation ofjo \ we get the formula in (52.a),
which ^-converts into the expression in (52.b). The rule of quantifying in allows us
to abstract on the free variable x and combine the resulting predicate with the
translation of the Ladies' Scottish Golf Champion to get (52.c), equivalent to the
formula in (52.d).
(52) a. XP [P<jo')](believe'(A[~(XQ [Q(x)](wealthy'))])).
b. (believe' (A[~(wealthy'(x))]))(jo').
c. XQ [Q(sgc')]ax [(believe'(A[~(wealthy'(x))]))(jo')]).
d. Xx [believe'(A[~(wealthy'(x))])(jo')](sgc').
It might seem that lambda conversion can be applied one more time to (52.c) to
derive the formula in (51.d) which represents the de dicto reading of Jo believes that
the Ladies' Scottish Golf Champion is not wealthy, from which we could derive the
contradictory formula in (51 .e) via conjunction introduction with (51 .a). Like the Law
of Substitution, however, the rule of Lambda Conversion is invalid in intensional
contexts and must be revised to take account of this as in (53).
(53) Xu [<|)](a) <-> (û provided that u is not in the scope of A
or the tense
operators Past and Fut.
312
To illustrate the fact that lambda conversion is not valid in intensional contexts and
to show that (52.b) does indeed give the de re reading of (46.a), let us consider the
interpretation of this formula with respect to Mi0 and the index <Wi,[t2]>. The crucial
point about this interpretation is that the denotation of sgc' is fixed by the reference
index. This follows because the rule interpreting function/argument structures in LIL
given in (25.3) derives the interpretation of (54.a) from the application of the
extension of the functor at <Wj,[t2]> to that of the argument at the same index, as
shown in (54.b). According to the function in (3O.b), the denotation in M10 of sgc' at
<W!,[t2]> is WOMANi. Hence, by the definition of lambda conversion in Chapter 5,
the result of applying the extension of the ^-expression in (54.b) to that of its
argument sgc' is equivalent to the interpretation of the former with respect to a
variable assignment function where the variable x has the value WOMAN!. Thus,
(54.a) and (54.b) are equivalent to (54.c). In other words, the formula in (52.c) is true
at some co-ordinate if, and only if, WOMANi is in the set of things that Jo believes
are not wealthy. Since WOMAN! is Ethel and Jo believes that Ethel is not a wealthy
woman, we conclude the de re reading of Jo believes the Ladies' Scottish Golf
Champion is not wealthy is true. Since the de dicto reading is false (given the truth of
(46.a)), the de re and de dicto readings of this sentence represented by the formulae
in (52.c) and (51.d) are thus truth-conditionally distinct.
(54) a. [A,x [believe>(A[~wealthy'(x)])(jo')](sgc>)]M-8wl-ra.
b. [Xx [believe>(AhwealthyXx)])Go')]]M'8wl[t2i([(sgc')]M-8'wlit20.
c. [believe'(A[~wealthy'(x)])Go')]M<8WOMAN1/xwl[t21.
The difference in interpretation between the de re and de dicto readings of (46.a)
can be best appreciated by considering what propositions Jo is asserted to believe in
each case. In the de re interpretation just discussed, the relevant proposition that Jo is
said to believe is A[wealthy'(x)] wherex is assigned WOMANi as a value. As we saw
above, this proposition is given by the characteristic function of the set
{<w3,[t4]>,<W4,[t2]>,<W4,[t3]>}. In the de dicto reading in (51.b), however, the
proposition that Jo is said to believe is that denoted by ^[wealthy '(sgc')], the function
in (49) that defines the set {<w 2 ,[tj>, <w2,[t2]>, <w3,[t4]>, <w4,[t2]>, <w4,[t3]>}. The
functions denoted by A[wealthy '(x)] and A[wealthy '(sgc')] are thus different and so Jo
may believe one proposition and not the other without inconsistency. We can thus see
from this example that lambda conversion is indeed not valid in intensional contexts,
because the non-converted formula has a different interpretation (and truth value in
Mio) from the formula that would result after lambda conversion. Furthermore,
because the referent of the embedded subject, the Ladies' Scottish Golf Champion, is
determined by the reference index, the analysis adequately captures the fact that de re
reference occurs outside the belief context and is therefore not dependent on the other
beliefs of the referent of the subject of the main clause.
We are now in a position to provide the reading for the sentence in (5O.b) which
does not ascribe a contradictory belief to Jo. Recall that this sentence was constructed
from a sentence ascribing to Jo a de dicto belief that the Ladies' Scottish Golf
Champion is wealthy and another ascribing to him a belief that Ethel is not wealthy.
Although direct substitution into an intensional context is not permitted, nevertheless
if it is possible to ascribe a de re belief to someone, one can then substitute an
313
extensionally equivalent expression for the directly referring expression. This

obviously follows from the fact that the latter appears outside the intensional context
created by believe' and so is subject to the Law of Substitution. In deriving the
non-contradictory reading of (5O.b), therefore, we construct from the de dido belief
about Ethel in (51.b), repeated in (55.b) below, the de re belief about Ethel in (55.c).
(55) a. believe'(A[wealthy'(sgc')])(jo').
b. believe'(A[~wealthy'(ethel')])(jo').
c. Xx |>elieve'(A[~wealthy'(x)])Oo')](ethel').
It is not in general the case that the truth of the de re reading of a sentence is
guaranteed by the truth of its de dicto counterpart. For example, the de re counterpart
to (55.a), Xx [believe '(^[wealthy '(x)])(jo ')](sgc'), is not true in M10 and so not entailed
by the former. However, where an expression, like etheV and other proper names,
denotes a constant function the de re and de dicto readings of the same belief sentence
turn out to be equivalent. For example, A[-wealthy'(ethel')] denotes that proposition
which picks out the set of indices at which it is true that the extension of ethel' is not
in the set of wealthy entities in M10, i.e. {<Wi,[ti,tJ>, <W2,[ti,t4]>, <w3,[ti,t3]>,
<Wi,[ti]>, <w4,[t4]>}. Since ethel' picks out the same entity at each index, i.e.
WOMANj, this proposition is equivalent to that denoted by A[-wealthy'(x)] where
g(x) is WOMAN!. Hence, where someone has a belief about a particular named
individual, no semantic difference emerges between the de re and de dicto readings
of the sentence. In fact, it is a general property of intensional expressions that denote
constant functions from indices to extensions that lambda conversion continues to
apply, even in intensional contexts. Hence, the rule of lambda conversion in (49)
should be revised to allow the equivalence between Xu [<|)](a) and [ty]**" where the
argument, a, has a constant intension.
Returning to the derivation of the non-contradictory reading of (5O.b), we can
validly infer the de re reading of Jo believes that Ethel is not wealthy in (55.c) from
(55.b) via backwards X-conversion. In (55.c), because ethel' is outside the scope of
the intension operator, any extensionally equivalent expression may be substituted for
it salva veritate. From (55.c) and the premiss in (56.a), therefore, we can construct
(56.b), the de re interpretation of Jo believes the Ladies' Scottish Golf Champion is
not wealthy. From this and the de dicto reading of Jo believes that the Ladies' Scottish
Golf Champion is wealthy in (55.a), we get (56.c), the desired reading of Jo believes
that the Ladies' Scottish Golf Champion is wealthy and that the Ladies' Scottish Golf
Champion is not wealthy, which does not ascribe to Jo a contradictory belief about the
Ladies' Scottish Golf Champion.
(56) a. ether = sgc\
b. Xx [believe'(A[~wealthy'(x)])(jo')](sgc').
c. Xx [believe'(A[wealthy'(sgc') & -wealthy'(x)])(jo')](sgc').
Individual constants with the semantic properties of sgc' are rare (and debatable)
and phrases like the Ladies' Scottish Golf Champion and the President of the USSR
ought more properly to be treated as definite noun phrases as in Chapter 6. However,
the properties of belief contexts ensured by the analysis discussed above carry over
to sentences containing verbs ofpropositional attitude and quantified noun phrases:
i.e. that substitution fails in de dicto readings and that belief sentences are
314
Two problems
ambiguous between de re andde dicto readings. To illustrate this, let us work through
both readings of the sentence in (57). Since no concepts are introduced here that have
not already been met, it will suffice to go over the main points of the analysis and
leave readers to work out the details for themselves.
(57) Fiona believes that a golf champion will be wealthy.
The derivation of the de dicto reading of (57) is straightforwardly derived without the
application of the rule of quantifying in to get the translation in (58.a). The
interpretation of this formula ascribes to Fiona a belief that a golf champion, whoever
they may be, will in the future be wealthy. Because the existential quantifier is within
the scope of the intension operator, no specific golf champion need be believed by
Fiona to be going to be wealthy, nor need there be at the time of utterance any golf
champions at all for the formula to be true. Furthermore, because the translation of the
subordinate subject is within the scope of A no substitution of extensional equivalents
may take place. The de re reading of (57) shown in (58.b) is derived from the
application of the rule of quantifying in. Its interpretation states that there is some golf
champion at the present time (and the actual world) of whom Fiona believes she will
be wealthy. Hence, there must be a specific person who is a golf champion and of
whom Fiona holds her belief. Furthermore, if all golf champions are currently
secretaries then it follows from the truth of (58.b) that Fiona believes of a secretary
that she will be wealthy. Hence, again, the analysis presented above enables one to
distinguish de re from de dicto beliefs and to account for the failure of substitution
within such contexts.
(58) a. {de dicto)
Pres(believe'(A[Fut(3x [golf-champion' (x) & Pres(wealthy'(x))])])(fiona')).
b. (de re)
3x [golf-champion'(x) & Pres (believe'(A[Fut(wealthy'(x))])(fiona'))].
Exercise 10.6:
Work out the semantic derivation of the formula in (58.b) using the rule of quantifying
in. Interpreting the denotation of the predicate golf-champion' as the function in (30.b)
with the entities replaced by unit sets of those entities (so that e.g.
[golf-champion']M1° s wiruJ is {WOMANj}), show that the complement propositions in
(58.a) and (58.b) do denote different propositions with respect to M10 and the index
<Wi,[t2]>. Be careful to take full account of the tense operators in these formulae.
10.4 Two problems

The adoption of a co-ordinate semantics can thus solve the puzzles posed by the
failure of expected entailments in certain contexts. By treating intensionality as an
operator over formulae, we can also, as discussed above, account for ambiguities in
intensional contexts in terms of scope differences between different sorts of operator
(e.g. quantifiers and tense operators). There are, however, two major problems that
315
arise from the analysis of intensional contexts given in Section 10.3. One has to do
with the substitution of intensional equivalents and the other with referring to
individuals at different indices.
10.4.1 Intensional equivalence

We have observed above that the substitution of extensional equivalents fails in
intensional contexts. However, it was also observed in Section 9.2 that for Frege the
intension (Sinn) of an expression is determined by the intensions of its component
parts. Since the intensional semantics defined above preserves the version of the
Principle of Compositionality in Chapter 9, the version of Leibniz' Law in (59) does
hold in LIL. This states that two expressions are intersubstitutable in intensional
contexts if (and only if) their intensions are identical.
(59) (Aa = Ab)-»[<|> <-><)>»*].
Exercise 10.7:
Explain informally why (59) is valid in LIL. Illustrate your answer by showing that the
two sentences below are truth-conditionally equivalent in M10 under the assumption
that 3[robert']M1°8 is the constant function that picks out MAN2 at every index.
i. Jo believes that Bertie is wealthy,
ii. Jo believes that Robert is wealthy.
The validity of (59) in LIL presents a serious problem for the analysis of verbs of
propositional attitude with respect to propositions that are necessarily true. In the
technical sense of the term proposition given in this chapter, there is only one
necessarily true proposition: the characteristic function that maps all possible indices
(i.e. W x I) onto 1. Assuming that all logical and arithmetical truths are true at all
indices, all the sentences in (60), therefore, denote the same proposition, i.e. the set
Wxl.
(60) a. It is snowing or it is not snowing.
b. Four times four is sixteen.
c. All students are students.
Since there is only one necessarily true proposition, it follows that if someone believes
or knows any necessarily true statement, then that person believes or knows all
necessarily true statements. Thus, once someone has grasped the fact that four times
four is sixteen, then they know every arithmetically and logically necessary truth. Not
only do they then know the other sentences in (60) but also that the circumference of
a circle is given by twice the radius of the circle times n and the law of modus ponens,
and so on. There are no new necessarily true statements to be believed or known. This
is patently absurd, of course. People may believe or know some necessary truths but
not others but no-one would argue that since they know that it is either raining or not
raining, then they know all necessarily true statements about the Universe!
The primary problem with logically necessary statements in belief contexts is
316
Two problems
that there are not enough worlds to differentiate the different statements from one
another. As we have seen, contingent propositions are differentiated by the indices at
which they are true which may differ from proposition to proposition. However, if the
set of possible worlds is the set of all logically possible worlds, then obviously there
is no possible world in which a logically true formula is false. Hence, the intensions
of such formulae cannot be differentiated. To remedy this, it has been suggested that
the set of worlds in a model should include not only the logically possible worlds but
also logically impossible ones, i.e. worlds where the laws of classical logic fail to
hold. These would include, for example, worlds where a sentence like It is raining or
it is not raining is false or where it turns out not to be the case that four times four is
sixteen. This certainly allows for differentiation in the intensions of the logically true
(or logically false) statements, but only at the cost of introducing into the ontology
worlds where the familiar rules of inference fail and which are thus so alien to our
understanding that it is unlikely that we could ever comprehend them.
Other ways have been suggested to try and overcome the problem of intensional
equivalence without undermining the whole basis of the possible worlds approach
sketched in previous sections. There is no space here to deal with them all in any
detail and the reader is referred to the suggested reading for fuller information. Most
of them, however, involve the postulation of something more than an intension of the
sort defined in this chapter. For example, Cresswell (1985) develops some ideas of
Rudolf Carnap and David Lewis by defining a structural concept of meaning which
interprets the objects of belief and knowledge as combinations of the intension of a
formula plus all the intensions of its component parts. In other words, the meaning of
a formula <> | in this sense is something like a semantic derivation of the sort we have
been using in deriving scope differences using the rule of quantifying in (cf. e.g. (33)
above). Each node of the analysis tree, however, is labelled with the intension of some
logical expression, rather than the expression itself. The meaning of the sentence Jo
is madin this system is thus something like (61 .a). The difference in meaning between
two logically true statements like It is raining or it is not raining and It is snowing or
it is not snowing results from the differences in intension between the constants they
contain, i.e. rain' and snow'. Hence, one may believe or know (61.b) without
believing or knowing (61.c), since these are different formal objects.
(61) a. <s[Pres (mad>(jo'))]M'8,<j[XPP(jo')]M-8,3[mad']M-8».
b. <s[rain' v ~rain']M^<3[rain']M<8,3[~rain']M'8>>.
c. <s[snow' v-snow']M-8,<3[snow']M-8,3[-snow']M-8».
An alternative solution that does not involve complex intensional structures and
which is somewhat closer to a linguist's intuitions about meaning was suggested in
Section 9.2. This notion of the meaning of an expression defines it as the combination
of the denotation of the expression, both its extension and its intension, and its sense,
as defined in Chapter 7. As briefly discussed in Section 9.3.2, the set of meaning
postulates that determine the sense of an expression can be interpreted as an
accessibility relation R* on possible worlds. A world, wm, is Rz-accessible to another
world, wn, if, and only if, all the meaning postulates associated with a hold in wm and
wn. Since different basic expressions are associated with different meaning postulates,
they define different accessibility relations on the set of possible worlds. For example,
given the three meaning postulates in (62), Rtree> selects all those worlds in which trees
317
are plants and not flowers (but where flowers may not be plants or plants may be
animals) while Rpianf selects those in which no plants are animals (but where trees
may be flowers and flowers may not be plants) and so on.
(62) a. Vx [tree' (x) -> plant' (x)].
b. Vx [tree' (x) -» -flower' (x)].
c. Vx [plant'(x) -> -animal' (x)].
d. Vx [flower'(x) -> plant'(x)].
The sense of a constant a may, therefore, be defined as the set of all indices <wm,i>
such that wRaw' where wn is the reference world. Generalising this to complex
expressions, we can say that the sense of that expression is given by the set of indices
where each world in each index is accessible to the reference world according to the
sense relations of all the constants in the expression. Symbolising the sense of an
expression a as i[a]M8-wni, we can formally define sense as in (63). Under this
interpretation, the sense of a sentence like Every tree is a plant is that set of indices
where trees are plants, are not flowers, and where plants are not animals, i.e. where
each world is Rtree> and Rpianr accessible to this one.
(63) The sense of an expression a, r[a]M* wn<*, is the set of indices, <wm,i>, such
that wRciW*, wRc2w' ... and wRcnw' where Ci, c 2 ,..., Cn are the constants in
a.
A sentence like Every tree is a plant, that conforms to the structure of a meaning
postulate associated with the subject, is thus true at every index in its sense. Hence,
we can account for the apparent tautological (or analytic) properties of sentences like
Every bachelor is unmarried, since it expresses a true proposition at every time in
every sensible world, where the latter is defined as a world in which all meaning
postulates are true for every basic expression in the language. Such sentences are not
logical tautologies, however, because they are not true at every possible index and so
can be distinguished from necessarily true sentences like Every tree is a plant or not
a plant. On the other hand, the sense of the latter sentence is, of course, the same as
that of Every tree is a plant because both sentences contain exactly the same content
expressions, i.e. tree' and plant'. In order to distinguish the meanings of pairs of such
sentences, we therefore need also to take into account their intensions as well as their
senses. We may then call the combination of the sense and the intension of an
expression the meaning of that expression and formally define the concept as in (64).
(64) The meaning of an expression oc, M[a]Mswni is the ordered pair
<z[a]M-8'wni,3[a]M«>.
This gives us sufficient structure to distinguish the meanings of logically true
statements, since sentences containing different lexemes have different senses and
those containing the same lexemes may have different intensions. For example, the
meaning of Every tree is a plant or not a plant differs from that of It is raining or it
is not raining, because, while their intensions are identical, their senses differ because
there are worlds which conform to the patterns on extensions specified by the meaning
postulates associated with tree and plant that do not conform to those of rain and vice
versa. Furthermore, we can distinguish the meanings of two sentences which differ
only in that they contain synonyms in the same position. Thus, while Bertie has a red
318
Two problems
pullover and Bertie has a red jumper both have the same sense, since jumper and
pullover are synonyms (in British English), their intensions differ because there are
indices that do not form part of the senses of these two words where Bertie's red
jumper is not a pullover. Taking the objects of belief and knowledge to be meanings
in the sense of (64), therefore, solves the problem of the substitution of intensional
equivalents into such contexts. Whatever approach is adopted to solve this problem,
however, it is clear that intensions in the technical sense are not sufficient to
distinguish the meanings of all expressions. On the other hand, as illustrated in the
approach to meaning that I have just sketched, this failure is not in itself sufficient to
reject the possible worlds approach to intensionality (and meaning) entirely.
10.4.2 Cross-world reference

A second problem for the theory of intension defined in this chapter has to do with
the interpretation of proper names. Indeed, this whole topic remains a thorny problem
for linguists and philosophers which we can do no more than touch upon here.
Essentially the problem revolves around whether named individuals in one world exist
in others and, if they do, how they can be identified from one world to the next.
In the model M10, the names Bertie, Fiona, Ethel, Jo and Chester are identified
with (the individual sublimations of) individual constants that are assigned a constant
intension by F10. For example, bertie' denotes MAN2 and fiona' denotes WOMAN2
at every index in M10. It has been cogently argued in the literature on this subject that
this is the correct way to treat proper names, particularly as it allows one to refer to
individuals with whom one has no direct acquaintance, like Queen Elizabeth II, or
who may no longer be alive, like Socrates, or who may be fictional, like Jane Eyre.
If proper names always pick out the same entity at whatever world or time, they
always successfully refer and guarantee that when people talk about Jane Eyre,
Socrates or Elizabeth Windsor they are talking about the same entity, even though that
entity may not have corporeal existence in the actual world at the present time. Under
this interpretation, names are rigid designators and the formula in (65) holds true of
any model.
(65) 3x [necessarily'(oc = x)], where a translates a proper name.
Because the existential operator is outside the scope of the modal adverb, (65) asserts
that there is some entity in A that is associated with the constant a at every index and
that this entity is the same one at every index.
Unfortunately, once names are taken to berigiddesignators, a familiar problem
rears its ugly head. If two names denote the same entity at one index, then by (65) they
must denote the same entity at every index. Thus, the Latin version of the Greek name
for the Morning Star was Hesperus and that for the Evening Star was Phosphorus. We
happen to know now that Hesperus and Phosphorus denote the same entity, the planet
Venus. If names are rigid designators, then both Hesperus and Phosphorus must
denote the planet Venus at every index and so have the same intension. Thus, by the
revised version of Leibniz' Law in (59), the term Hesperus can be substituted for the
term Phosphorus in all contexts and so, if the objects of belief are intensions, then we
can infer from The ancients believed that Hesperus is Hesperus that The ancients
believed that Hesperus is Phosphorus (in the de dicto reading), a conclusion that is not
319
intuitively justified, as we have seen.

A further problem arises with sentences that assert the counterfactual identity of
two elements. It is, for example, possible to imagine a person as being someone other
than they are and making perfectly reasonable statements based on this assumption.
All the sentences in (66) are well formed and easily interpretable, yet the interpretation
of names asrigiddesignators means that we cannot take the assumption of the identity
of two entities that are not in fact identical as being true in any possible world at any
time.
(66) a. Clark Kent might not have been Superman.
b. I could have been Shakespeare, if I had lived in the sixteenth century.
Yet another problem concerns the properties that individuals may have at
different worlds or times. For example, it is logically possible that at some index the
entity denoted by Bertie, i.e. MAN2, could be a male human, at another a female frog,
at another a dwarf tree, at another a Mozart sonata, and so on. Yet in what sense could
a piece of music in one world be identified as the same entity as a man in another?
It is not clear that we can use the notion of meaning suggested in Section 10.4.1
to help out here, because it is not clear that proper names have sense in the same way
that other expressions do. Indeed, it seems to be almost axiomatic in philosophy that
names do not have sense. This is not a topic that I wish to get embroiled in here, but
it may be the case that one should follow Aristotle in distinguishing the essential
properties of an entity from its accidental properties. The essential properties of an
individual may then be used to define an accessibility relation between worlds which
excludes indices at which the individual in question does not have these properties.
In that case, we could take the meaning of a proper name to be the set of indices in
which the essential properties of the individual hold coupled with its (constant)
intension. This could help in the cross-world identification of individuals, and, if
identity is defined between essential properties, rather than referents, it might be
possible to solve the problem of counterfactually equating distinct entities.
Other solutions to the problem of interpreting proper names can and have been
put forward, but no one solution solves all the problems without posing more. It is
clear that treating names as rigid designators intensionally is not sufficient, but it is
possible that future research could produce a definition of meaning for proper names
within a semantics based on possible worlds that is robust enough to account for the
persistence of reference across worlds and times, but does not have the unfortunate
consequences mentioned above.
10.5 Postscript
With this discussion of Montague's theory of intensionality, we come to the end of
our excursion into formal semantics. It would seem appropriate, therefore, to end this
chapter with a short review of the theory presented in the preceding chapters with
respect to the criteria of adequacy discussed in Chapter 1.
With regard to the criteria for assessing the adequacy of semantic theories, the
theory of interpretation defined in Section 10.2 comes out well. First and foremost,
the Principle of Compositionality is adhered to throughout, since the denotations of
all expressions are constructed from the intensional or extensional denotations of
320
Postscript
their component parts, depending on how these are put together. The information
contained in an expression is not lost during its interpretation, and information not
connected with the subparts of the expression does not arbitrarily contribute to the
meaning of that expression. Apparent counterexamples to the Principle are treated in
terms of intensions which allows a fully regular way of building up denotations in
terms of functors and their arguments. The theory, therefore, provides an account of
the way the meanings of expressions are built up from the meanings of their parts. The
theory also provides an account of ambiguity. It predicts where this will occur: i.e. in
expressions with homonyms and in those that contain more than one expression that
translates into a logical operator. The interaction of the propositional connectives is
determined by syntactic structure, while scope ambiguities involving the other
operators are handled by the Rule of Quantifying In. The semantics proposed for the
operators ensures that genuinely ambiguous expressions in the object language receive
more than one set of truth-conditions. On the other hand, many genuinely
non-ambiguous sentences (even those that are structurally ambiguous) are assigned
only one set of truth-conditions, as required. The interaction of tense, aspect and
modal operators, however, with each other and with the quantifiers may, as we have
seen, provide too many readings for a single sentence, which is clearly a problem that
needs to be resolved.
In Chapter 7 and subsequently, it has also been shown how truth-conditional
semantics can account for relations of entailment, paraphrase and contradiction
between sentences in English. The use of meaning postulates, interpreted as
constraints on the extensions of related lexemes and phrases, provides the basis for a
theory of sense, as the notion is generally used by linguists, rather than logicians. As
discussed in this chapter and Chapter 9, it is possible to treat sense relations defined
by meaning postulates as structuring the domain of possible worlds in terms of their
accessibility to one another and, in so doing, to provide the basis of an account of
anomaly and the likelihood of a sentence expressing a true proposition. This is
achieved within an explicit and fully formulated theory of denotation, with both
extensions and intensions determined formally in terms of the entities of a model and
worlds (that may be thought of as possible alternative states-of-affairs) and times.
Each basic expression in the object language is associated with a denotation in a
model and the model theory determines how the denotations of all composite
expressions are built up compositionally from these. Hence, the way linguistic
expressions tie up with the world and convey information about possible or actual
situations is made fully explicit within the theory.
Furthermore, in the course of this book, a theory has been provided of the core
truth-conditional meanings of a number of constructions in English. These include:
basic declarative sentences; negative sentences and sentences co-ordinated by and and
or, simple passives; intersective adjectives; quantifiers, both classical and
non-classical; grammatical tense and grammatical and lexical aspect; the modal
adverbs necessarily and possibly and the modal verbs must and may; sentences of
propositional attitude; control verbs and transitive verbs with a referentially opaque
direct object position. There are, of course, many constructions that were omitted from
the discussion (most notable anaphora) and there remain many problems with the
theory of semantics presented in the text. However, formal semantics is now a thriving
field of linguistics and is constantly being developed to analyse more constructions
321
and to solve the problems that emerge. Indeed, there are now many variants of
Montague's basic theory: some are primarily relatively compatible extensions to the
theory and some are fairly radical revisions of it. Yet the theory presented in this
introductory book remains the basis for much of the research that is carried out within
formal semantics and there is no doubt that Montague's conviction that the syntax of
natural languages is as amenable to formal interpretation as that of logical languages
has introduced a valuable and robust strand of research into linguistic semantics.
10.6 Further reading

Lewis (1972) provides a general discussion of the programme adopted by Montague
and discussed in this chapter. Montague discusses his intensional logic and its
interpretation in Montague (1970a; 1970b; 1973), with the latter presenting the
information in the form given in this chapter. Montague's general theory is discussed
in Dowty (1979: ch. 1) and Dowty, Wall and Peters (1981: ch. 8) and the logic and its
interpretation are sketched in McCawley (1981: 401-406), Allwood, Andersen and
Dahl (1977: ch. 8) and Chierchia and McConnell-Ginet (1990: 208-219) and given a
fuller discussion in Dowty, Wall and Peters (1981: ch. 6). The status of intensions as
mathematical or psychological entities is discussed in Partee (1979) and Dowty (1979:
ch. 8). Further discussion of the applications of the logic to intensional constructions
in English presented in this chapter can be found in Chierchia and McConnell-Ginet
(1990:234-255) (modality, complement sentences and infinitives); McCawley (1981:
406-424) (intensional argument positions in e.g. seek and change)', Dowty, Wall and
Peters (1981: ch. 7) (which discusses the grammar of Montague (1973)); Bach(1979),
Klein and Sag (1985) and Dowty (1985) (on control verbs). See also van Benthem
(1988), the articles in Heny (1981), and, in a slightly different framework, Keenan and
Faltz (1985), for discussions of different aspects of intensionality. The problem of
intensional equivalence has generated a good deal of debate. Cresswell (1985b)
provides the approach to the problems of prepositional attitude verbs using structured
meanings discussed in the text and this is taken up in Chierchia (1989). A radically
different approach is taken in Barwise and Perry (1983) where the use of possible
worlds is rejected entirely and subsequent developments in situation semantics have
substantially revised the semantic programme that developed out of Montague's work
(see, for example, Barwise (1986), the articles in Cooper, Mukai and Perry (1990) and,
for some general discussion of the early form of the theory, the papers in Linguistics
and Philosophy 8, no. 1). The analysis of proper names also remains controversial; see
Kripke (1980). For representative examples of current developments in formal
semantics, see the papers in Chierchia, Partee and Turner (1989a; 1989b), Cooper,
Mukai and Perry (1990) and current issues of the journal Linguistics and Philosophy.
322
A N S W E R S TO S E L E C T E D E X E R C I S E S
Chapter 2
Exercise 2.1:
i. poison' (ether ,the-cat')
ii. give'(the-student\ the-cake\ the-lecturer').
iii. crazy' (the-dog').
Exercise 2.3:
i. (ethel' = the-golfer'). True (both ethel' and the-golfer' denote WOMAN,).
ii. (the-student' = the-singer'). False (the-student' denotes MAN! but
the-singer' denotes WOMAN2).
Exercise 2.4:
1. False. 2. True. 3. False.*
4. True. 5. False. 6. False.
7. True. 8. False. 9. False.
10. True. 11. False. 12. True.
13. False. 14. True. 15. True.
16. False.
* D represents the set consisting of the first four positive odd numbers.
Exercise 2.5:
i. laugh'(the-caf). False (CAT e {MAN!, WOMAN,}).
ii. happy'Go'). True (MAN, e {MAN,, WOMAN,, WOMAN 2 }).
iii.r un'(fiona'). False (WOMAN 2 £ {CAT, DOG}).
Exercise 2.6:
i. like'(the-lecturer\ the-caf). True ( < W O M A N 2 , C A T > e
{<MAN 1 ,WOMAN 1 >, <MAN,,MAN 2 >, <MAN,,WOMAN 2 >,
<MAN,,MAN,>, <WOMAN,,WOMAN 2 >, <WOMAN,,MAN,>,
<WOMAN 2 ,CAT>, <WOMAN 2 ,WOMAN,>}).
ii. kick'(ethel',the-student'). False. (<WOMAN,,MANi> € 0 . )
iii. poison'(the-cake',the-caf).False.(<CAKE,CAT>£ {<CAKE,DOG>}.)
Exercise 2.7:
i. give'(the-golfer',the-book' ,the-golfer'). False,
ii. give'(the-student' ,the-book' ,the-lecturer'). False,
iii. give'(bertie', the-book', ethel'). True.
323
Answers to selected exercises
Exercise 2.8:
Truth conditions for n-place predicates: A formula Predn(a!,a2,...,an) is
true if and only if the ordered n-tuple <E! ,E2,...,En> is in the set of ordered
n-tuples denoted by Predn, where E, is the entity denoted by ai, E2 is the
entity denoted by a2 and ... and En is the entity denoted by an. The formula
is false otherwise.
Chapter 3
Exercise 3.2:
i. Three: ~(~(p v q)), ~(~(p) v q), (~(~(p)) v q).
ii. Three: (p v ~(p & q)), ((p v ~(p)) & q), (p v (~(p) v q))
Exercise 3.3:
p q ~p ~q (~P & ~q)

t t f f f
t f f t f
f t t f f
f f t t t
Exercise 3.5:
p q (P •q) (q ~> P) (P q) ((P <-> q) «P-> q) & (q -^ p)))
t t t t t t
t f f t f t
f t t f f t
f f t t t t
Exercise 3.6:
i. ~(give'(jo\the-cat',the-lecturer')). False. (<MAN1,CAT,WOMAN2> e
[give']M1).
ii. (eat'(the-dog',the-cat') v eat'(the-dog',the-cake')).True. (<DOG,CAKE>
e [eat']M1).
iii. (messy'(Chester') -> -(loathe'(the-cat',the-dog'))). True. (DOG £
[messy']M1).
Chapter 4
Exercise 4.1:
i. f(a): well-formed, type t. ii. g(f)' well-formed, type <e,t>
iii. g(a) : ill-formed. iv. h(f): ill-formed.
v. j(f): well-formed, type t. vi. (g(f))(a): well-formed, type t
324
Exercise 4.3:
iii.
[scream']M1
MAN2
WOMAN
WOMAN2
CAT
DOG
CAKE
BOOK
Exercise 4.4:
i. DtD»
iii. D^, iv.
Exercise 4.5:
1.
Chapter 5
Exercise 5.1:
i. The set of all entities y such that Jo liked y and Ethel didn't like y.
ii. The set of all entities z such that z gave the cake to themself.
iii. The set of all entities x that are identical to Jo.
iv. The set of all entities z such that z is a student and z liked Jo.
Exercise 5.2:
1. (12.a) should be (kick'(jo')Xjo').
(12.b) is ill-formed because Xx [howl'(x)](the-cat') does not contain a free
variable (x is bound by Xx to the-cat').
(12.c) is ill-formed because ^y does not bind the x variable in (like'(x))(jo')
and so bertie' cannot be substituted for x.
(12.d) should be crazy'(ether) v drunk'(z).
325
2. i. (like'(the-dog'))(jo') & ~((like'(the-dog'))(ethel')).

ii. (give'(bertie'Xthe-cat')Xbertie').
iii. ~(bertie' = jo').
iv. (like'(the-cat')Xbertie') v ~((like'(the-dog'))(bertie')).
v. (crazy'(bertie') & (like'(bertie'))(jo')) v -(crazy'(bertie')).
Exercise 5.3:
i. [Xy [happy'(y) & laugh'(y)]]M8
MANj
MAN 2
WOMAN 1
WOMAN 2
CAT
DOG
CAKE
BOOK
iv. [Xy [~((y = the-lecturer') v (y = the-student')]M«
MAN 1 .
MAN * \
".Jfc 1
WOMAN! ^ \
WOMAN2 Â
CAT
DOG
/yC
' / *
A
CAKE ' / ^ 0
BOOK '
Exercise 5.4:
i. Xx [(eat'(x))(chester')](the-cake').
ii. Xy [(give'(fiona')(y))(ethel')](the-book').
Exercise 5.5:
i. Ill-formed. (The argument70' is bound by Az not Xx. The correct conversion
is (like'(bertie'))(jo').)
ii. Well-formed,
iii. Well-formed.
326
iv. Ill-formed. (The argument jo* is not bound by the first A,x. The correct
conversion is Xx [scream'(x)] v -(laugh*(jo*)).)
v. Well-formed.
Exercise 5.8:
i. «e,<e,t»,t> ii. <e,<e,<e,t>»
iii. «e,t>,<«e,t>,<e,t»,<e,t>» iv. <t,t>
v. «e,t>,t>vi.<t,<t,t»
Exercise 5.10:
i. [Xp [p <-> rain']]1"2* is that function hi from {0,1} to {0,1} such that for all
elements a in {0,1} hi(oc) is [p <-> rain']* 12 ^.
iii. [X? [slowly*(P)]]M28 is that function h2 from {0,1}* to {0,1^ such that for
all elements a in {0,l}A h3(a) is [slowly'(
Exercise 5.11:
i. (Xp [Xq [p & q]]((poison'(the-dog'))(the-cake')))
((eat'(the-cake'))(the-dog'))
» (poison'(the-dog'))(the-cake') & (eat'(the-cake'))(the-dog').
Chapter 6
Exercise 6.1:
i. Someone liked herself:
3x [(like'(x))(x)].
ii. Someone liked Jo and hated Fiona:
3x [(like'(jo'))(x) & (hate'(fiona'))(x)].
iii. Someone was liked by everyone:
3x [Vy [(like'(x))(y)]].
iv. Everyone gave something to someone:
Vx [3y [3z
Exercise 6.2:
i. False. (Not everyone likes themselves.)
ii. True. (gwoMANi* satisfies [(like'(fiona'))(x) & (like'(x))(fiona')].)
iii. True. (Every value assigned to x satisfies exist'(x).)
iv. True. (Every value assigned to x satisfies (touch'(x))(ethel').)
v. False. (gMMi^ (and other value assignments) fails to satisfy (like' (x))(y) for
any value assigned to y.)
Exercise 6.3:
someone =» XP [3x [P(x)]].
no-one => XP [~(3x [P(x)])]. (or XP [Vx [~(P(x))]]).
Exercise 6.4:
everyone: ( X c AIA =X}
no-one: { X c A I X = 0}
327
Exercise 6.5:
no dog => XP [~(3x [dog'(x) & P(x)])]
(orXP[Vx[dog'(x)-MP(x))]]).
no =* XQ piP [~(3x [Q(x) & P(x)])]]
(or X Q [XP [Vx [Q(x) -> ~(P(x))]]]).
Exercise 6.6:
ii. Some student didn't sing
=> (XQ [XP [3x [Q(x) & P(x)]]](student'))(XP1 [Xy [~(Pi(y))]](sing'))
» 3x [student'(x) & ~(sing'(x))].
iii. If no student sang, then no lecturer screamed
=> (Xp [Aq [p -> q]]((AQ [XP [~(3x [Q(x) & P(x)])]](student'))(sing'))
(XQ, [XP, [~(3y [Q^y) &P1(y)])]](lecturer'))(scream'))
» ~(3x [-(student'(x) & sing'(x))]) -^ ~(3y [lecturer'(y) & scream'(y)]).
Exercise 6.7:
{XcAIXn[dog'] M 'g = 0 }
Negative sublimation of [dog]M
328
Exercise 6 9*
i. XV [P(jo')](Ax [Oq ftp [p & q]](cry'(x)))(laugh'(x))])
» [laugh'(jo') & cry'(jo')],
ii. (XQ [XP [3x [Vy [Q(y) <-> x = y] & P(x)]]](linguist'))(Xx [-(happy'(z))])
» 3x [Vy [linguist'(y) <-> x = y] & -(happy'(x))].
iii. (Xp [Xq [p ->q]](*P [P(ethel')](sing')))
((^Q [AP, [3x [Vy [Q(y) <-> x = y] & P1(x)]]](dog'))(howl'))
» (sing'(ethel') -> 3x [Vy [dog'(y) <-> x = y] & howl'(x)])
Exercise 6.10:
i. Ethel shot a philosopher
=> X? [P(ethel')](XP [Ax [P(Xy [(shoot'(y))(x)])]]
(XQ [XPr [3z [Q(z) & P1(z)]]](philosopher')))
» 3z [philosopher'(z) & (shoot'(z))(ethel')].
Exercise 6.13:
i. False. ii. True. iii. True. iv. False.
Chapter 7
Exercise 7.1:
p q (P ->q) ~q ((p ->q) & ~q) ~P ((P ->q) &~q) ->~p)
t t t f f f t
t f f t f f t
f t t f f t t
f f t t t t t
Exercise 7.2:
i. Let p = Fiona poisoned the cake. Let q = Bertie poisoned the cake. Let r
= Fiona has a bottle of poison in her bedroom. Let s = Fiona has a book
about unsolved murders on her bedside table
a. pvq premiss
b. p -> (r & s) premiss
c. ~q assumption
d. p from a and c by DE2
e. r & s from b and d by MP
f. s from e by CI
g- ~q -» s from c and f by Cdl
329
Chapter 8
Exercise 8.1:
ii. a. [(like'(fiona'))Go')] M8>8lt5] is 1 iff [like'(fiona')]M8^5KLJo']M8*it5]) is 1.
b. [like'(fiona')]M8-8'i<5K[jo']M8-8'[t5])is 1 iff
([like']M8'8ft5K[fiona']M8'8'ft50)(Do']M8-8ft50 is 1.
c. Since [t5] C [tiA], [ j o 1 M 8 ^ is R(jo')([t,Jt 9 ]) = M A N , .
d. Since [t5] c [t,,fe], [fiona']MW5J is F 8 (fiona')([ti,tJ) = W O M A N 2 .
e. Since [t5] c [t 7 ,t 9 ], [like'] M8 ^ t5 i is F 8 (like')([t 7 ,t 9 ]) which is the function
defining {<MAN,,WOMAN,>, <MAN,,WOMAN2>,
<WOMAN1,WOMAN2>, <WOMAN2, WOMAN, >, <WOMAN1,MAN2>,
<MAN 2 ,WOMAN,>}, call this K.
f. K ( W O M A N 2 ) ( M A N , ) = 1.
g. So [(like'(fiona'))Go')] M8 ^ is 1.
Exercise 8.6:
i. [Past (Perf (Bx [(singt'(x))0o')]))]M8^6] is 1. (Jo (MAN,) begins and ends
singing SONG, at interval [t3,t5] which precedes [t6].)
ii. [Fut (Perf (3x [(singt'(x))(ethel')]))]M8.g.[t6] i s 0 . (Although Ethel (WOMAN,)
completes singing SONG2 at [t9], the interval in which she begins the
singing includes [t6]. Therefore, there is no future interval where she begins
and completes the singing of a song.)
iii. [Fut (Perf (3x [sing'(x)]))]M8<grt6] i s 1. (There are intervals of time that follow
[t6] in which someone sings (e.g. WOMAN 2 at [t8].)
iv. [Past (Perf (~(3x [sing'(x)])))]M88ft6J i s 0. (There is no interval of time
preceding [t6] during which no-one is singing.)
v. [Pres(Impf(sing'(]o')))]M'88[t61 is 1. (There is an interval of time properly
containing [t<J (i.e. [t3,t9]) at which MAN, is singing.)
vi. [Fut(Impf(3y [3x [(singt'(x))(y)]]))]M«t6 is 1. (There is an interval of time
following [t6] (e.g. [t8]) which is properly contained in an interval of time
[UM during which MAN, is singing SONG2.)
Chapter 9
Exercise 9.1:
i. [D-laugh'(Chester')]M98wl is 0. (Chester (DOG) laughs in w2.)
ii. [o(~(laugh'(Chester')) -> laugh'(jo'))]M9«wl is 1. (In world w2 in M9,
-(laugh'(Chester')) is 0, so (-(laugh'(Chester')) -> laugh'(jo')) is 1.)
iii. [-Qaugh'(Chester')]M9 «*wl is 1. (Chester does not laugh in w,, so
[•laugh'(chester')]M9* wl is 0.)
iv. [~(o(laugh'(jo') & -laugh'(jo')))]M9'8wl is 1. (There is no world, w, in which
[laugh'(jo') & ~laugh'(jo')]M98w is 1, so [o(laugh'(jo') &
-laugh'(jo'))] M9 « wl is0.)
v. [o~laugh'(fiona')]M98wl is 1. (Fiona does not laugh in w2, so
[-laugh'(fiona')]M9-8-w2 is 1.)
vi. [-olaugh'(fiona')]M9«wl is 0. (Fiona laughs in w,, so [olaugh'(fiona')]M98wl
isl.)
330
Chapter 10
Exercise 10.1:
3[laugh'(jo')]M1°8 ii. s[laugh'(Chester') v laugh'(jo')]M10-g
<w 2 U 2 ]>
<w2,[t3]>
<w2,[t4]>
<w3,[tx]>
<w3,[t3]>
<w3,[t4]>
<w4,[t2]>
<w4,[t3]>
<w4,[t4]>
331
iii. 3[laugh'(fiona') v -laugh'(fiona')]M1°8 iv. s[laugh'(fiona')->laugh'(jo')]M10s

Exercise 10.2:
i. Ill-formed. ii. Well-formed (<e,t>).
iii. Well-formed (t). iv. Well-formed (t).
v. Ill-formed. vi. Well-formed (t).
vii. Ill-formed.
Exercise 10.3:
i. Functions from indices to functions from characteristic functions of entities
to truth values: ({0,1 J^ 1 **) 1 ^ 1 .
ii. Functions from functions from indices to characteristic functions of entities
to truth values: {0,1}«Î}A)WXD
v. Functions from functions from indices to truth values to characteristic
functions of entities: ({O,!
332
REFERENCES
Ajdukiewicz, K. (1935) 'Die syntactische Konnexitat'. Stadia Philosophica 1: 1-27.

(English translation in S. McCall ed. (1967) Polish Logic. Oxford, Oxford
University Press.
Allwood, J., L. Andersen and O. Dahl (1977) Logic in Linguistics. Cambridge,
Cambridge University Press.
Aquist, L (1976) 'A system of chronological tense logic'. In Gunthner and S. Schmidt
eds.: 223-254.
Bach, E. (1974) Syntactic Theory. New York, Holt, Rinehart and Wilson.
(1979) 'Control in Montague Grammar'. Linguistic Inquiry 10: 515-532.
(1986) 'The algebra of events'. Linguistics and Philosophy 9: 5-16.
Ballmer, T. and M. Pinkal eds. (1983) Approaching Vagueness. Amsterdam,
North-Holland.
Bar-Hillel, Y. (1953) 'A quasi-arithmetical notation for syntactic description'.
Language 29: 47-58.
Barwise, J. (1986) 'Conditionals and conditional information'. In Traugott, ter
Meulen, Reilly and Ferguson eds.: 21-54.
Barwise, J. and R. Cooper (1984) 'Generalized quantifiers and natural language'.
Linguistics and Philosophy 4: 159-219.
Barwise, J. and J. Etchemendy (1989) TarskVs World. Stanford, CSLI.
Barwise, J. and J. Perry (1983) Situations and Attitudes. Cambridge, Mass., MIT
Press.
Bauerle, R. (1979) 'Tense logics and natural language'. Synthese 40: 225-230.
Bennett, M. (1977) 'A guide to the logic of tense and aspect in English'. Logique et
Analyse 20: 491-517.
(1978) 'Demonstratives and indexicals in Montague Grammar'. Synthese 39:1-80.
Bennett, M. and B. H. Partee (1972) 'Toward the logic of tense and aspect in
English'. Mimeo, IULC.
van Benthem, J. (1986) Essays in Logical Semantics. Dordrecht, Reidel.
(1988) A Manual of Intensional Logic. Stanford, CSLI.
Blakemore, D. (1987) Semantic Constraints on Relevance. Oxford, Basil Blackwell.
(1989) 'Denial and contrast: a relevance theoretic analysis of BUT. Linguistics and
Philosophy 12: 15-38.
Borsley, R. (1990) Syntactic Theory: a Unified Approach. London, Edward Arnold.
Bradley, R. and N. Swartz (1979) Possible Worlds: an Introduction to Logic and its
Philosophy. Oxford, Basil Blackwell.
Burton-Roberts, N. (1989) The Limits to Debate: a Revised Theory of Semantic
Presupposition. Cambridge, Cambridge University Press.
Carlson, G. N. (1989) 'English generic sentences'. In Chierchia, Partee and Turner
(1989b): 167-192.
Carnap, R. (1956) Meaning and Necessity. Chicago, Chicago University Press
(Revised edn).
333
References
Carston, R. (1988) 'Implicating, explicature, and truth-theoretic semantics'. In R.

Kempsoned.: 155-181.
Chierchia, G. (1989) 'Structured meanings'. In Chierchia, Partee and Turner eds.
(1986b): 131-166.
Chierchia, G. and S. McConnel-Ginet (1990) Meaning and Grammar: an Introduction
to Semantics. Cambridge, Mass., MIT Press.
Chierchia, G., B. H. Partee and R. Turner eds. (1989a) Properties, Types and
Meaning: Volume 1 Foundational Issues. Dordrecht, Reidel.
(1989b) Properties, Types and Meaning: Volume 2 Semantic Issues. Dordrecht,
Reidel.
Chierchia, G. and R. Turner (1988) 'Semantics and Property Theory'. In Linguistics
and Philosophy 11: 261-302.
Church, A. (1956) Introduction to Mathematical Logic. Princeton, Princeton
University Press.
Comrie, B. (1976) Aspect. Cambridge, Cambridge University Press.
(1985) Tense. Cambridge, Cambridge University Press.
Cooper, R. (1983) Quantification and Syntactic Theory. Dordrecht, Reidel.
Cooper, R., K. Mukai and J. Perry (1990) Situation Theory and its Applications.
Stanford, CSLI.
Cresswell, M. J. (1973) Logics and Languages. London, Methuen.
(1985a) Adverbial Modification. Dordrecht, Reidel.
(1985b) Structured Meanings. Cambridge, Mass., MIT Press.
Cruse, D. (1986) Lexical Semantics. Cambridge, Cambridge University Press.
Dahl, O. (1985) Tense and Aspect Systems. Oxford, Basil Blackwell.
Davidson, D. (1967) 'The logical form of action sentences'. In N. Rescher ed. The
Logic of Decision and Action. Pittsburgh, University of Pittsburgh Press: 81-95.
(Reprinted in Davidson (1980).)
(1980) Essays on Actions and Events. Oxford, Clarendon.
(1984) Inquiries into Truth and Interpretation. Oxford, Oxford University Press.
Dowty, D. (1979) Word Meaning and Montague Grammar. Dordrecht, Reidel.
(1982) 'Grammatical relations and Montague Grammar'. In P. Jacobson and G. K.
Pullum eds. The Nature of Syntactic Representation. Dordrecht, Reidel: 79-130.
(1985) 'On recent analyses of the semantics of control'. In Linguistics and
(1989) 'On the semantic content of the notion of "thematic role'". In Chierchia,
Partee and Turner eds. (1989b): 69-129.
Dowty, D. and W. A. Ladusaw (1988) 'Toward a non-grammatical account of
thematic roles'. In W. Wilkins ed. Syntax and Semantics 21: Thematic Relations.
New York, Academic Press: 61-73.
Dowty, D., R. Wall and S. Peters (1981) Introduction to Montague Semantics.
Dordrecht, Reidel.
Dummett, M. (1981) Frege: Philosophy of Language. London, Duckworth (2nd edn).
Eikmeyer, H-J. ed. (1981) Words, Worlds and Contexts: New approaches to Word
Semantics. Berlin, de Gruyter.
334
References
Frege, G. (1892) 'Uber Sinn und Bedeutung'. Zeitschrift fur Philosophie und
philosophisches Kritik: 22-50. (An English translation appears in P. Geach and
M. Black eds. (1980) Translations from the Philosophical Writings ofGottlob
Frege. Oxford, Basil Blackwell.)
Gazdar, G (1979) Pragmatics: Implicature, Presupposition and Logical Form. New
York, Academic Press.
Gazdar, G., E. Klein, G. K. Pullum and I. A. Sag (1985) Generalized Phrase Structure
Grammar. Oxford, Basil Blackwell.
Grice, P. (1975) 'Logic and conversation'. In P. Cole and J. Morgan eds. Syntax and
Semantics, vol. 3. New York, Academic Press.
(1978) 'More on logic and conversation'. In P. Cole ed. Syntax and Semantics, vol.
9. New York, Academic Press.
Giinthner, F. (1976) Time schemes, tense logic and the analysis of English tenses'.
In Gunthner, F. and S. Schmidt eds.: 201- 222.
Giinthner, F. and S. Schmidt eds. (1976) Formal Semantics and Pragmatics for
Natural Languages. Dordrecht, Reidel.
Guttenplan, S. (1986) The Languages of Logic. Oxford, Basil Blackwell.
Hawkins, J. A. (1978) Definiteness and Indeflniteness. London, Croom Helm.
Heim, I. (1989) The Semantics of Definite and Indefinite NPs. New York, Garland
Press.
Heny, F. ed. (1981) Ambiguities in Intensional Contexts. Dordrecht, Reidel.
Hoeksema, J. (1983) 'Plurality and conjunction'. In ter Meulen ed.: 63-83.
Huddleston, R. (1984) Introduction to the Grammar of English. Cambridge,
Hughes, G. E. and M. J. Cresswell (1968) An Introduction to Modal Logic. London,
Methuen.
Hurford, J. R. and B. Heasley (1983) Semantics: a Coursebook. Cambridge,
Kamp, J. A. W. (1975) 'Two theories about adjectives'. In E. L. Keenan ed. Formal
Semantics of Natural Language. Cambridge, Cambridge University Press:
123-155.
Kamp, H. (1981) 'A theory of truth and discourse representation'. In J. Groenendijk,
T. Janssen and M. Stokhof eds. Formal Methods in the Study of Language.
Amsterdam: Mathematical Centre Tracts 135.
Keenan, E. and L. M. Faltz (1985) Boolean Semantics for Natural Language.
Dordrecht, Reidel.
Keenan, E. and J. Stavi (1986) 'A semantic characterization of natural language
determiners'. In Linguistics and Philosophy 9: 253-326.
Kempson, R. (1977) Semantic Theory. Cambridge, Cambridge University Press,
ed. (1988) Mental Representations. Cambridge, Cambridge University Press.
Klein, E. (1980) 'A semantics for positive and comparative adjectives'. In Linguistics
and Philosophy 4: 1-45.
Klein, E. and I. A. Sag (1986) 'Type-driven translation'. Linguistics and Philosophy
8: 163-202.
Kripke, S. (1980) Naming and Necessity. Oxford, OUP.
Leech, G. N. (1974) Semantics. Harmondsworth, Penguin.
Levinson, S. C. (1983) Pragmatics. Cambridge, Cambridge University Press.
335
References
Lewis, D. (1972) 'General semantics*. In D. Davidson and G. Harman eds. Semantics

for Natural Language. Dordrecht, Reidel.
(1973) Counterfactuals. Oxford, Basil Blackwell.
(1986) The Plurality of Worlds. Oxford, Basil Blackwell.
Link, G. (1983) 'The logical analysis of plural and mass terms: a lattice-theoretic
approach'. In R. Bauerle, C. Schwarze and A. von Stechow eds. Meaning, Use
and Interpretation of Language. Berlin, de Gruyter.
L0nning, J. T. (1987) 'Mass terms and quantification'. Linguistics and Philosophy 10:
1-52.
Ludlow, P. and S. Neale (1991) 'Indefinite descriptions: in defense of Russell'.
Linguistics and Philosophy 14: 171-202.
Lycan, W. G. (1984) Logical Form in Natural Language. Cambridge, Mass., MIT
Press.
Lyons, J. (1977) Semantics. Cambridge, Cambridge University Press.
(1981) Language, Meaning and Context. London, Fontana. (New edition to be
published by Cambridge University Press.)
Martin, J. N. (1987) Elements of Formal Semantics. Orlando, Academic Press.
May, R. (1985) Logical Form: its structure and derivation. Cambridge, Mass., MIT
Press.
McCawley, J. D. (1981) Everything that Linguists have Always Wanted to Know
about Logic*. Oxford, Basil Blackwell.
ter Meulen, A. G. B. (1983) 'The representation of time in natural language'. In ter
Meulened.: 177-191.
ed. (1983) Studies in Modeltheoretic Semantics. Dordrecht, Foris.
(1984) 'Events, quantities and individuals'. In F. Landman and F. Veltman eds.
Varieties of Formal Semantics. Dordrecht, Foris: 259-279.
Montague, R. (1970a) 'English as a formal language'. In B. Visentini et al. eds.
Linguaggi nella Societd e nella Tecnica. Milan, Editzione di Comunita: 189-224.
(Reprinted in Montague (1974).)
(1970b) 'Universal grammar'. Theoria 36: 373-398. (Reprinted in Montague
(1974).)
(1973) 'The proper treatment of quantification in ordinary English'. In J. Hintikka,
J. Moravcsik and P. Suppes eds. Approaches to Natural Language. Dordrecht,
Reidel. (Reprinted in Montague (1974).)
(1974) Formal Philosophy: Selected Papers of Richard Montague. Ed. by R. H.
Thomason. New Haven, Yale University Press.
Mourelatos, A. P. D. (1978) 'Events, processes and states'. Linguistics and Philosophy
2: 425-434.
Oehrle, R. T., E. Bach and D. Wheeler eds. (1988) Categorial Grammars and Natural
Language Structures. Dordrecht, Reidel.
Ojeda, A. E. (1991) 'Definite descriptions and definite generics'. Linguistics and
Palmer, F. (1981) Semantics: A New Outline. Cambridge, Cambridge University
Press.
Parsons, T. (1979) Type theory and ordinary language'. In S. Davis and M. Mithun
eds. Linguistics, Philosophy and Montague Grammar. Austin, University of
Texas Press: 127-152.
336
References
(1980) 'Modifiers and quantifiers in natural language'. Canadian Journal of

Philosophy Supplementary Volume VI: 29-60.
(1989) The progressive in English: events, states and processes'. Linguistics and
Partee, B. H. (1979) 'Semantics: mathematics or psychology?' In R. Baiierle, U. Egli
and A. von Stechow eds. Semantics From Different Points of View. Berlin,
Springer Verlag.
(1984) 'Compositionality'. In F. Landman and F. Veltman eds. Varieties of Formal
Semantics. Dordrecht, Foris: 281-311.
(1986) 'Noun Phrase Interpretation and type-shifting principles'. In J. Groenendijk,
D. de Jongh and M. Stokhof eds. Studies in Discourse Representation Theory and
the Theory of Generalized Quantifiers. Dordrecht, Foris.
Partee, B. H. and M. Rooth (1983) 'Generalized conjunction and type ambiguity'. In
R. Bauerle, C. Schwarze and A. von Stechow eds. Meaning, Use and
Interpretation of Language. Berlin, de Gruyter.
Partee, B. H., A. ter Meulen and R. Wall (1990) Mathematical Methods in Linguistics.
Dordrecht, Reidel.
Prior, A. N. (1967) Past, Present and Future. Oxford, Oxford University Press.
Pulman, S. G. (1983) Word Meaning and Belief. London, Croom Helm.
Putnam, H. (1975) 'The meaning of meaning'. In H. Putnam Mind, Language and
Reality: Philosophical Papers vol. 2. Cambridge, Cambridge University Press.
Quine, W. V. O. (1960) Word and Object. Cambridge, Mass., MIT Press.
Reichenbach, H. (1947) Elements of Symbolic Logic. Berkeley, University of
California Press.
Rescher, N. and A. Urquhart (1971) Temporal Logic. Berlin, Springer Verlag.
Reuland, E. J. and A. ter Meulen eds. (1987) The Representation of (In)definiteness.
Cambridge, Mass., MIT Press.
Rohrer, C. ed. (1980) Time, Tense and Quantifiers. Tubingen, Niemeyer Verlag.
Russell, B. (1905) 'On denoting'. Mind 14: 479-493. (Reprinted in I. Copi and J.
Gould eds. (1967) Contemporary Readings in Logical Theory. New York,
MacMillan.)
Sag, I. A., T. Wasow, G. Gazdar and S. Weisler (1985) 'Co-ordination and how to
distinguish categories'. In Natural Language and Linguistic Theory 3: 117-171.
Siegel, M. (1979) 'Measure adjectives in Montague Grammar'. In S. Davis and M.
Mithun eds. Linguistics, Philosophy and Montague Grammar. Austin, University
of Texas Press: 223-262.
Sperber, D. and D. Wilson (1986)Relevance: Communication and Cognition. Oxford,
Basil Blackwell.
Strawson, P. F. (1950) 'On referring'. Mind 59: 320-344.
Tarski, A. (1944) 'The semantic conception of truth'. In Philosophy and
Phenomenological Research 4: 341-375. (Reprinted in A. Martinich ed. (1985)
The Philosophy of Language. Oxford, Oxford University Press.)
Tedeschi, P. J. and A. Zaenen eds. (1981) Syntax and Semantics 14: Tense and Aspect.
New York, Academic Press.
Thomason, R. (1989) 'Motivating ramified type theory'. In Chierchia, Partee and
Turner eds.: 47-62.
337
References
Thomason, R. and R. Stalnaker (1973) 'A semantic theory of adverbs'. Linguistic

Inquiry 4: 195-220.
Tichy, P. (1985) 'Do we need interval semantics?' Linguistics and Philosophy 8:
263-282.
Traugott, E., A. ter Meulen, J. S. Reilly and C. A. Ferguson eds. (1986) On
Conditionals. Cambridge, Cambridge University Press.
Turner, R. (1989) 'Foundations of semantic theory'. In Chierchia, Partee and Turner
eds.: 63-84.
Vendler, Z. (1967) Linguistics in Philosophy. New York, Cornell University Press.
Verkuyl, H. J. (1972) On the Compositional Nature of the Aspects. Dordrecht, Reidel.
(1989) 'Aspectual classes and aspectual composition'. Linguistics and Philosophy
12: 39-94.
Wilson, D. and D. Sperber (1988) 'Representation and relevance'. In R. Kempson ed.:
133-153.
Zwarts, F. (1983) 'Determiners: a relational perspective'. In ter Meulen ed.: 37-62.
338
INDEX
Pages where terms are discussed are indicated in bold
accessibility relation 277-281, 287, 175,314

317, 320 both 187,189
accessible world 276-281, 286, 287
accomplishment 257-258 cartesian product 293
achievement 257-258 categorial grammar 111
actions 12,256-259 characteristic function 96-99, 103,
activities 234,251,254, 257,258, 105-110,113,115-117,123,125,
284, 296 285,286, 310
actual world 268, 270, 273, 277, 278, co-operative principle 20-21
280, 285, 299, 302, 303, 315, 319 co-ordination constraint 195
adjective 9, 11, 31, 32, 34,169-172 commutativity 70,71,138, 224,229
adverb 83, 85,107-110 complement (set) 96-97
alternativeness relation 277 (see also complementaries 217,220-221
accessibility relation) complex formula 55-60complex type
ambiguity 180-186 84, 85, 99, 105, 177, 307
ambiguity 2, 8-9, 26 compositionality 2-6, 24, 25, 30, 33,
ambiguity 36,41, 57-59,65,66, 40,56,78, 82,89,99,107,261,
226, 227, 259- 261, 265, 266, 303, 268,316,320
305-307,321 compound sentence 54-55
analysis tree 38, 57, 91, 138, 165, conditional 70-73, 74, 79-81, 138-
187,303,317 140, 200-202, 204-207, 209, 218,
anomaly 7, 278, 321 229-231
antecedent (of conditional) 70, 72, conjunction 56, 58, 59, 63-66, 78,
73,136,200, 218-220 136, 137, 143, 145, 165,200-203,
antonymy 217, 221 206-209,265,301,311,312
argument 27, 28, 33, 48-52, 75, 76, conjunction elimination 200, 202,
86-87,91,95,99,100,105,106, 207,209,211,213,219
115, 120, 122, 123, 127-131, 160, conjunction introduction 208, 209,
186 212,215,265,311,312
argument (deductive) 17 conjunction operator 56, 58, 70
argument (logical) 198 connective 55-74, 83-85,201,223,
argument structure 28, 83, 89, 106 224
associativity 66, 71 connective (and quantifiers) 163, 166,
asymmetric relation 234 167
at least two 187,189 consequent (of conditional) 70, 72,
attributive modifier 151, 169 73, 136
axiom of extension 46, 47 conservativity 192,193,210
constant function 294, 300, 309, 314,
backwards lambda-conversion 316
144-148, 161, 164, 166, 170-171, context of utterance 18, 22-24, 191
339
Index
contingent formulae 233, 264, 271, disjunction elimination 204, 209

272,274,288,317 disjunction operator 56, 58, 67
contradiction 5-7, 20, 199-202, ditransitive verb 29, 32, 51, 52, 90,
207-213,223,224,268 103, 106, 177
contraries 217 domain (of a function) 90, 94-96,
control verbs 306-308 98-100, 131, 134, 161,175, 176,
conversational implicature 6,21 284, 296-298, 321
conversational maxim 20-21, 228-
230 element (of a set) 45-47, 93-96,
converses 217,221 empty set 45-46,48,49, 101,162
core meaning 6,13-16 entailment 5, 6, 20, 25, 146, 193-
correspondence theory 15,17 195, 197, 199, 200,202, 203, 210,
counterfactual conditional 230, 232, 212, 228, 248, 250, 255, 257, 263,
269 266,274,280,301,303,321
epistemic modality 279, 280
de dicto 265, 308, 312-315, 319 equative construction 31, 34, 39, 43,
de re 265, 308,312-315 78,79
deductive inference 198-200, 203, equivalence 16, 43,44,56,73,74,
222 118, 120, 123, 125, 201, 202, 205,
definite article 6,29,42,174-177, 212,230,306,307,316
195 equivalence operator 56, 73-74
definite description 30, 42, 151, 155, exclusive disjunction 68, 226-228
174-177,300 existential formula 155,181,182
defmiteness 174-177, 195, 199 existential generalisation 211, 212,
deictic expression 22-24, 241, 251, 214,275
285 existential operator 152
denotation 10-14, 16, 26, 32, 39, 40- existential sublimation 168, 169
43, 46-53, 75, 76, 78, 95, 98-103, extension 12, 26, 40, 49-52, 75, 76,
119-121,123,125, 131,132-134, 78-80,93, 98,101,105-110,151,
161, 162,216,218,222,236, 161-163,165-172,216,233,
238-240, 267, 268, 277, 287, 289, 236-240, 263, 264, 267, 269,
293-295, 302, 313, 315, 317, 321 282-284, 289, 290,296, 297
denotation assignment function 39, extension operator 289, 290
40, 42,47, 48, 50, 52, 53, 76-78,
95,103, 105, 236, 238-240, 272, formal language 2, 17, 38, 74, 246
296, 296 function 3, 4, 19, 75, 76, 90, 93-103,
deontic modality 279,280 105-111,113,115-126,130-132,
derivation 30,31,36-38,57,91, 139-143, 167, 168, 173, 188, 193,
127, 140, 146, 148, 179 267-269, 277, 279, 282, 286, 287,
determiner 162-169,171, 174, 187- 293, 296-298, 309, 310,
191 functional application 86, 88, 108,
direct object 30, 32, 50, 82, 88, 89, 171,238,290,297,303
106,126,127,303,305,321 functor 86-89, 99, 105, 106, 161,
disambiguated language 24 164, 178
discontinuous determiner 190 future perfect 248-249
disjunction 56, 58, 67-70, 78, 195, future tense 241, 242, 244- 245, 247-
204, 209, 226-228, 230, 249 250, 252, 259, 288
340
Index
generalised quantifier 187-195, 232, intensional semantics 282-320

302 intensional subject 302
generate 30, 53 intensional type 290, 291, 293, 295,
grammar fragment 25, 28, 29, 43, 53, 297
58, 70, 73, 89, 107, 110, 159, 170, intersective adjective 171, 172, 321
259, 288 temporal interval 234-245, 248, 249,
grammatical aspect 233, 246, 251-257, 260, 262, 283, 284, 288,
251-258, 259, 262, 293, 298 296-298, 300, 309
grammatical function 30, 88-89, 112, interval semantics 234, 235, 262
129 intransitive verb 11, 29, 32, 47, 49,
grammatical subject 30, 32,33, 82, 93, 234
88-89, 93, 106,112, 126, 144, 303,
306-308,315 lambda (k) abstraction 116-120,
127-130, 132, 136, 137, 140, 292
hyponymy 6,217-219 lambda (X) conversion 117-120,
122-125, 126,128, 132, 137, 140,
identity 41-43, 46, 276, 289, 292, 144, 145, 147, 161, 164, 170, 175,
312,316,319,320 178, 299, 302, 304, 305, 312, 313
identity function 139 lambda (X) expression 116-118
identity operator 20, 34, 43, 149,292 lambda (X) operator 112,115-119,
imperfective 251-255, 257, 258, 262 120-149, 152, 153, 164, 178, 188,
imperfective paradox 255, 262 201,292,298,303
impersonal verb 29, 32, 52, 90 Law of substitution 263,266,288,
implicature 6, 26, 68, 199, 200, 228, 308,312,314
229,231,249,250 Leibniz's Law 263 (see also Law of
inclusive disjunction 67, 68,226-228 substitution)
index 283, 285-289, 297, 300-304, lexeme 3, 8, 10-12, 16, 34-36,
307, 309-320 216-219,221,222
indirect object 30, 88, 89, 91, 103, lexical decomposition 195, 217, 221,
126, 129 299
individual constant 28, 32, 42, 43, lexical meaning 3, 5-8, 198-200,
47, 83-85, 90, 151, 173, 174, 294, 207, 215, 216, 218, 222, 224, 231
314,319 LIL 288-293,295-297,300,301,302,
individual sublimation 173, 174, 304,308,311,313,316
186,303,304,312,319 Lp 27-29,31-44,47-49,51-53,
individual variable 115,117,125, 54-56, 58-60, 62, 63, 65, 70, 73-76,
144, 145,186,291,303 78-80,83, 87-89,112, 137,138
inference 6, 17, 20-22, 25, 68, 81, LQ 152-172, 177, 178, 180, 185, 188,
197-231, 263-267, 276, 280, 288, 218, 238, 242, 266, 288, 290, 299,
299,301,306,311,312 302
inference hierarchy 199 Lr 242, 244, 246, 252, 255, 259, 289,
intension 11, 12, 26, 40, 267-269, 290
278, 282-321 ^ 83,86-91,98-110, 112,
intension operator 289, 290, 303, 115-119,126,127,130,131,137,
306,307,309,312,314,315 141,152, 153, 155, 170, 173,290
intensional equivalence 316
intensional model 295
341
Index
material implication 56, 70-73, 74, negation operator 56, 58, 60, 62, 63,
79, 163, 165, 166, 200, 202-205, 76, 84, 85, 87, 90, 107, 212
209, 210, 229-232 negative sublimation 169, 328
maxim of manner 20,21 no 158-159, 164, 169, 193-195
maxim of quantity 21,228-230, 249 nominal modifier 169
maxim of relation 20,21 non-contingent 264,271
meaning postulate 218-223, 257, non-specific reading 266, 303, 306,
258, 278, 279, 296, 306, 307, 318 307
meaningful expression 107, 292 noun phrase 4,7, 9, 29, 31, 36, 42,
metalanguage 16, 17, 19, 74, 75, 150, 159, 164, 167,169,171, 174,
218,288 177, 186, 190-193
modal adverbs 269,270-276, 282, NP co-ordination 143,147-148
321 null set 45, 46, 268, 286, 299
modal verb 269, 272, 287
modality 24, 39, 41, 52,76-78, 80, object language 16-18, 24, 27, 29,
103, 105, 222, 244, 247, 250, 269, 35, 36, 38-40, 55, 56, 75, 78, 95,
270-281, 282, 283, 322 151, 155, 188, 236, 259, 296, 321
model-theoretic interpretation 41, object language
216 oblique transitive verbs 299-305
model theory 18, 19, 39, 40, 43, 75, one-place predicate 32, 48-50, 83-85,
78-80, 98, 103-107, 110, 120, 125, 88,90,98,99, 103,116-118
131,140,141,188, 203, 216,224, ontology 39-41,75-77,236,295, 317
233, 237, 238, 240, 241, 255, 256, opaque transitive verb 299, 303, 305,
286, 288, 293, 296, 297, 298-299, 306, 312 (see also oblique
300, 321 transitive verb)
modus ponens 204-209,212,219, open formula 127, 164, 178, 208,
316 211,212, 246
modus tollens 205, 206, 209, 214, oppositeness 6
215,220,221,229,230 opposites 217,218,220,221
moment 233-237,239,241,243, ordered n-tuple 75
252, 253, 256, 258 ordered pair 50, 51, 76, 82, 84, 101,
monotone decreasing 193-195 102,
monotone increasing 193-195 ordered triple 51, 52
monotonicity 210 overlap (of temporal intervals) 234,
moral necessity 280 235, 240
more than half 188,190, 191
more...than 190 paraphrase 6, 20, 25, 91, 129, 146,
most 188,190-195 148,199-203, 206, 212, 213, 227,
307,321
n-place 32, 44, 52, 78, 79, 83-85 passive 30, 51, 112-114, 126-128,
naming function 39 201
necessarily 264, 269-277, 280, 282, past tense 25,233,241-243,
286-288,291,292,297, 245-247, 249, 250, 252, 259
necessary truth 316 perfective 251-253, 255, 257
negation 4-7, 55-59, 60-64, 79, 84, perlocutionary effect 20
87,89,141, 143,158-159, phrase structure grammar 25, 29, 53
204-207, 217,220 pluperfect 246-248
342
Index
polysemy 8, 26, 36 quantifier 153-155, 157, 159,

possible world 270,272-274, 276, 161-163, 166,167, 175, 177, 181,
277, 281, 282-284, 286, 293, 317, 184,185,188,197,210,211,220,
320 246, 259, 260, 266, 277, 278, 303,
possibly 269, 270-272, 274-276, 282, 304,306,315
283,286,287,291,292,297 quantifier pronoun 159,161,162,
precedence relation 234, 235 167, 197
predicate 27, 28, 32-34, 39, 47-53,
75, 78-81, 82-85,105-107, 128, range (of function) 90, 94-95,
129, 145 98-100, 108
predicate logic 53, 81,153, 189 recursive definition 18,78,98-101,
predicate monotone 193, 194 292
present tense 25,241- 242, 250, 252, reductio ad absurdum 207-209,
259,311 reference 10-12
presupposition 6-7, 26, 199, 226 reference index 287, 297, 303, 304,
primitive type 84-86,99, 290 311,313
principle of compositionality 3-5, 24, reference time 241-243, 247, 248,
30, 33, 40, 78, 82, 99, 261, 268, 255, 287
316,320 reference world 273, 274, 277, 278,
principle of intensional 282,286,287,318
compositionality 268 referentially opaque 267,268,281,
principle of intensional reference 282, 321
268, 269 referentially transparent 267, 268
proper name 172-174, 319, 320 reflexive 277, 280
proper noun 29, 36 relation 34, 39, 40, 44, 49-51, 82,
proper subset 45 93,94, 100-103,112-114,126,179,
proper superset 45 181, 184, 186, 189
properties 2-4, 8, 11, 12, 42, 44, relevance theory 21-22, 26, 231
116, 188-190, 193, 172-174, 193, rigid designator 319
217, 218, 264, 266-269, 276, 294, rule of functional application (rfa)
295, 301, 302, 306, 307, 308, 314, 86-88,160,238,290,303
318, 320 rule of quantifying in 186, 260, 261,
proposition 13-14,200,202,218, 303, 305, 306, 308, 312, 315, 317,
228, 264, 268, 270, 271, 286, 294, 321
295, 306, 309, 310, 313, 314, 316, rule-to-rule hypothesis 5, 25, 33,
317,318,321 89,56, 186,259,261
propositional attitude 265,308-315,
316,321,322 satisfaction (of propositional
propositional content 19, 20, 23 function) 117, 156-159, 161, 165,
propositional function 116-118, 120, 174,176,182-184,208,211-214
122, 127, 130, 131, 137, 152, 153, scope ambiguities 9,180-187,
155, 157-159, 161, 171, 173, 183 259-261, 303, 321
propositional logic 17,53,81,231 selectional restrictions 217, 221
semantic scope 8, 9
quantified noun phrase 12,150,151, sense 3, 10,11, 25, 26,216-219,
159,161,162,167,172,177,301, 222-224, 267, 278, 279, 290, 317-
314 319, 320
343
Index
sense relations 6, 11, 217-222, 224, 42,54,55, 187,188

278,279,318,321 transparent context 267
sensible world 318 truth-conditional semantics 15-17, 74,
sentential co-ordination 57,136-143 95,200,218,224
set intersection 46, 165, 166, 168, truth-conditions 15,16,18-21,23,
171, 173,188-190, 192,193 26, 40, 43, 49, 51-53, 60, 62-70, 73,
set membership 19 74,79, 120,155,157, 163,189,
set theory 18, 19, 44-47, 53, 83, 93- 190-192, 224-229, 241, 243, 245,
98, 188 252, 253, 273, 274,321
set union 46, 236 truth table 61-63, 65, 67, 68, 70,
specific reading 266, 267, 299,302- 72-74, 201-205, 225, 226, 274
305,306,307,312,315 two-place predicate 32, 50, 83-85, 88,
speech act 20,26 100-102, 103,
state-of-affairs 15, 18, 39, 264 type assignment function 90
states 257 type raising 177-180
stative 239, 240, 244, 252, 256- 257 type theory 82, 83-111
subinterval 235, 239, 240, 253,255,
296, 298 universal falsehood 233
subject monotone 193, 194 universal formula 157
subset 45, 46, 96, 108, 109,130 universal instantiation 211, 212, 274
superset 45, 46 universal NP 151
synonymy 6,217,220 universal operator 152
universal sublimation 167, 168
tautology 184, 200-203, 209-211, universal truths 233
228, 230, 248, 249
temporal model 236, 237, 273 validity 17, 197, 198, 200, 201,
temporal reference point 241 203-206, 209-211, 213-215, 223,
tense 9, 23-25, 233, 241-252, 254, 255, 261, 264, 265, 274-276, 280,
257, 259, 260, 262, 263, 272, 283, 287-289, 304
285, 288, 292, 298 valid inference 299
tense operator 242, 243, 260, 311 variable assignment function 120-
three-place predicate 32, 103-104 122, 132
time 23,25,233-241,249,251-255, VP co-ordination 143-146
258-260, 262, 264, 272, 280,
283-288,293, 296-298,312,315, well-formed formula 28, 33, 83, 153,
319,320 163
time of utterance 23,240-242,
244-249, 251-254, 259, 260, 287,
302, 304, 315
transformational grammar 186
transitive relation 234, 277
transitive verb 33, 49-51, 84, 88, 90,
126, 127, 147, 177-180, 186, 299-
305, 306
translation language 24, 27, 35, 54,
55,93,95,110, 141,172,198
translation procedure 28, 32, 36-38,
344

Cann

Uploaded by

Copyright:

Available Formats

Cann

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cann

Uploaded by

Copyright:

Available Formats

This book provides a clear and accessible introduction to formal semantics within a

linguistic framework. It presupposes no previous background in logic, but takes

P. H. MATTHEWS Morphology Second edition

© Cambridge University Press 1993

First published 1993

Library of Congress cataloguing in publication data applied for

ISBN 0 521 37463 4 hardback

Transferred to digital printing 2003

List of figures page xn

5 THE LAMBDA OPERATOR 112

8 TIME, TENSE AND ASPECT 233

Answers to selected exercises 323

The book is on the table page 11

Although billed as an introduction to formal semantics in general, this textbook is

1.1 Semantics and semantic theory

controversial matter, but whatever meanings are taken to be within a theory, it is

1.1.2 Meaning relations

negation of an entailment always derives a contradiction and a sentence Si may be

The book is on the table

1.2 Interpretation and representation

The book is not on the table

1.2.2 Formalisation, models and sets

(26) a. The book is on the table.

1.3.1 Knowledge, behaviour and use

(30) Theory of relevance:

1.4 A note on method

framework that is adopted in this book is that of simple context-free phrase

1.5 Further reading

2.1 Translating English into a logical language

2.7.7 The syntax of LP

(1) t -> PrednCe1^2,...,^), 0 < n.

2.1.2 A grammar fragment for English

The type of syntactic grammar adopted in this book is a simple context-free

Ethel poisoned cat

2.1.3 The translation procedure

(16) a. TIG,: S => W(NP').

poison' (ethel' ,the_cat')

The cat was poisoned by Ethel

project is to interpret the formulae of Lp in terms of their truth-conditions, as

2.2.1 Individuals and identity

(28) a. jo' DENOTES MANL

(29) a. jo' DENOTES MAN,.

2.2.2 A little light set theoiy

In order to show that some element, a, is a member of a particular set, the

2.2.3 Interpreting predicates

(36) a. Jo was happy and laughed.

by the predicate, run\ i.e. {CAT,DOG}. By the same reasoning, run'(bertie') is

Loosely speaking, intransitive verbs and adjectives (i.e. one-place predicates in

If transitive verbs describe relations between two entities, ditransitive verbs

by the third argument forms the third element.

(42) give' DENOTES {<WOMAN2,CAKE,MAN1>,

2.3 Further reading

3.1 Compound sentences

3.2 Complex formulae

(~(rain') —> snow')

If it didn't rain then it snowed

b. it is not the case that S \5GX

(14) a. ((eat'(the-cat',the-cake') v eat'(the-dog\the-cake')) & laugh'(fiona')).

( (loathe'(ethel\ Chester') (happy'(Chester'))) )