Understanding Semantics
Understanding Semantics, Second Edition, provides an engaging and accessible introduction
to linguistic semantics. The first part takes the reader through a step-by-step guide to the
main phenomena and notions of semantics, covering levels and dimensions of meaning,
ambiguity, meaning and context, logical relations and meaning relations, the basics of noun
semantics, verb semantics and sentence semantics. The second part provides a critical
introduction to the basic notions of the three major theoretical approaches to meaning:
structuralism, cognitive semantics and formal semantics.
Key features include:
∑ A consistent mentalist perspective on meaning
∑ Broad coverage of lexical and sentence semantics, including three new chapters
discussing deixis, NP semantics, presuppositions, verb semantics and frames
∑ Examples from a wider range of languages that include German, Japanese, Spanish and
Russian
∑ Practical exercises on linguistic data
∑ Companion website including all figures and tables from the book, an online dictionary,
answers to the exercises and useful links at routledge.com/cw/loebner.
This book is an essential resource for all undergraduate students studying linguistics.
Sebastian Löbner is Professor for general linguistics at the University of Düsseldorf,
Germany. His main research interests are linguistic semantics and cognitive linguistics.
Understanding Language series
Series Editors:
Bernard Comrie, Max Planck Institute for Evolutionary Anthropology, Leipzig, Germany
Greville Corbett, Surrey Morphology Group, University of Surrey, UK
The Understanding Language series provides approachable, yet authoritative, introductions
to major topics in linguistics. Ideal for students with little or no prior knowledge of linguistics,
each book carefully explains the basics, emphasising understanding of the essential notions
rather than arguing for a particular theoretical position.
Other titles in the series:
Understanding Child Language Acquisition (forthcoming)
Caroline Rowland
Understanding Language Testing
Dan Douglas
Understanding Morphology, Second Edition
Martin Haspelmath and Andrea D. Sims
Understanding Phonetics
Patricia Ashby
Understanding Phonology, Third Edition
Carlos Gussenhoven and Haike Jacobs
Understanding Pragmatics
Jef Verschueren
Understanding Second Language Acquisition
Lourdes Ortega
Understanding Sociolinguistics (forthcoming)
Enam Al-Wer
Understanding Syntax, Third Edition
Maggie Tallerman
For more information on any of these titles, or to order, go to www.routledge.com/linguistics
Understanding
Semantics
Second edition
Sebastian Löbner
First published in Great Britain 2002 by Hodder Arnold, an imprint of Hodder Education
Second edition published 2013
by Routledge
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN
Simultaneously published in the USA and Canada
by Routledge
711 Third Avenue, New York, NY 10017
Routledge is an imprint of the Taylor & Francis Group, an informa business
© 2002, 2013 Sebastian Löbner
The right of Sebastian Löbner to be identified as author of this work has been asserted by him in accordance
with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any
electronic, mechanical, or other means, now known or hereafter invented, including photocopying and
recording, or in any information storage or retrieval system, without permission in writing from the publishers.
Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only
for identification and explanation without intent to infringe.
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloging in Publication Data
Löbner, Sebastian.
Understanding semantics / Sebastian Löbner. -- Second Edition.
pages cm. -- (Understanding language)
Previous edition published under Hodder Education, the second edition is now published under Routledge
after Hodder Education Linguistic titles were acquired in 2012.
Includes bibliographical references and index.
1. Semantics. I. Title.
P325.L567 2013
401’.43--dc23
2012048797
ISBN: 978-0-415-82673-0 (hbk)
ISBN: 978-1-4441-2243-5 (pbk)
ISBN: 978-0-203-52833-4 (ebk)
Typeset in 11 on 12pt Minion
by Phoenix Photosetting, Chatham, Kent
Contents
Preface
Acknowledgements
ix
xi
1 Meaning and semantics
1
1.1 Levels of meaning
1.2 Sentence meaning and compositionality
1.3 Semantics: its scope and limits
Exercises
Further reading
2 Dimensions of meaning
2.1 Meanings are concepts
2.2 Descriptive meaning
2.3 Meaning and social interaction: the dimension of social meaning
2.4 Meaning and subjectivity: the dimension of expressive meaning
2.5 Connotations
2.6 Dimensions of meaning
Exercises
Further reading
3 Ambiguity
3.1 Lexemes
3.2 Lexical ambiguity
3.3 Compositional ambiguity
3.4 Contextual ambiguity
3.5 Meaning shifts and polysemy
Exercises
Further reading
4 Meaning and context
Part 1: Deixis
4.1 Person deixis
4.2 Demonstratives and place deixis
4.3 Time deixis
Part 2: Determination
4.4 Definiteness and indefiniteness
4.5 Quantification
1
7
14
16
17
18
18
21
28
33
36
37
39
40
41
41
44
48
49
57
59
60
62
62
63
70
72
74
74
83
Contents
vi
4.6 Generic NPs
Part 3: Presuppositions
4.7 Presuppositions
4.8 Summary
Exercises
Further reading
5 Predication
5.1 Predications contained in a sentence
5.2 Predicate terms and argument terms, predicates and arguments
5.3 Verbs
5.4 Nouns and adjectives
5.5 Predicate logic notation
5.6 Thematic roles
5.7 Selectional restrictions
5.8 Summary
Exercises
Further reading
6 Verbs
6.1 Argument structure, diatheses and alternations
6.2 Situation structure
6.3 Aspect
6.4 Tense
6.5 Selected tense and aspect systems
6.6 Concluding remark
Exercises
Further reading
7 Meaning and logic
7.1 Logical basics
7.2 Logical properties of sentences
7.3 Logical relations between sentences
7.4 Sentential logic
7.5 Logical relations between words
7.6 Logic and meaning
7.7 Classical logic and presuppositions
Exercises
Further reading
8 Meaning relations
8.1
8.2
8.3
8.4
Synonymy
Hyponymy
Oppositions
Lexical fields
90
94
94
101
102
104
106
106
108
111
115
120
122
125
130
131
133
134
135
140
150
157
163
164
165
166
167
167
172
175
184
187
191
197
201
202
203
203
205
208
215
Contents
Exercises
Further reading
9 Meaning components
9.1 The structuralist approach
9.2 Applying the structuralist approach to meaning
9.3 Semantic features
9.4 Semantic formulae
9.5 Semantic primes: Wierzbicka’s Natural Semantic Metalanguage
9.6 Summary and evaluation of the approaches to decomposition
Exercises
Further reading
10 Meaning and language comparison
10.1 Translation problems
10.2 Headache, international
10.3 Relativism and universalism
10.4 Berlin and Kay’s investigation of colour terms
10.5 Consequences
Exercises
Further reading
11 Meaning and cognition
11.1 Categories and concepts
11.2 Prototype theory
11.3 The hierarchical organization of categories
11.4 Challenges to prototype theory
11.5 Semantics and prototype theory
11.6 Semantic knowledge
11.7 Summary
Exercises
Further reading
12 Frames
12.1 Barsalou frames
12.2 Verbs and frames
12.3 Nouns and frames
12.4 Frames and composition
12.5 Frames and cognition
12.6 Conclusion
Exercises
Further reading
13 Formal semantics
13.1 Japanese numerals: a simple example of a compositional analysis
vii
219
220
221
223
226
228
237
242
245
246
246
248
248
251
256
258
262
263
264
265
266
267
276
279
284
292
298
300
300
301
301
311
313
319
321
322
323
324
325
325
Contents
viii
13.2 A small fragment of English
13.3 Model-theoretic semantics
13.4 Possible-world semantics
13.5 The scope and limits of possible-world semantics
Exercises
Further reading
330
342
352
358
362
363
References
364
Index
369
Preface
As for many others, my entrance into semantics was formal semantics, about
forty years ago. At present, this is still the standard approach in many linguistics
departments around the world. Working in the field of semantics my entire academic
life, my conviction has grown that formal semantics is not the ideal framework for
working one’s way into the rich and fascinating reality of natural language meaning.
The perspectives allowed by the formal apparatus developed in formal semantics are
far too restrictive. And the aspects of meaning and the semantic phenomena that are
neglected, or are simply problematic to deal with, are far too numerous. Above all,
formal semantics has little to say about lexical meaning – which, after all, provides
the ultimate basis of all linguistic meaning – and, not by chance, it fails to connect
semantic theory to cognition.
In Understanding Semantics, I have taken a different approach. It is driven by
the idea that students of semantics should first grasp the level of meaning which
linguistic semantics aims to describe and how this level is related to higher levels
of interpretation; they should learn that there are different dimensions of meaning,
in addition to descriptive meaning; they should know about ambiguity and about
the existence of meaning shifts that interfere with lexical meaning; they should get
a notion of the rich inventory of indexical means of expression including deixis,
determination and presupposition carriers; they should learn the basics of lexical
semantics of nouns and verbs; they should know that there are different theoretical
approaches to meaning; and they should get a notion of the fact that linguistic
meaning is ultimately a matter of conceptualizing the things we talk about: when we
put things into words, we are not just giving a one-to-one mapping of what the world
is like – we make a choice by putting things in the particular way we do. Meaning is
not just a matter of logical relations and truth conditions. As to sentential meaning,
the students need to know about the basic semantic functions of NP determination
and the verbal categories of aspect and tense, and they should know the basics of
predication. All this should be discussed from a perspective that also takes a look at
other languages. On this complex background, the more advanced students may start
to work their arduous way into the theory and technicalities of formal semantics.
In order to give an idea of the basic notions of this approach, the book offers a
substantial basic introduction in the last chapter, and a critique.
The second edition of Understanding Semantics is not only a more up-to-date
version of the first edition, but is supplied with new sections that considerably
broaden the coverage of the field. These include:
x
Preface
∑ basic notions of the semantics of word formation (chapters 1 and 12)
∑ deixis and demonstratives (chapter 4)
∑ presuppositions (chapters 4 and 7)
∑ NP semantics (chapter 4)
∑ verb semantics including voice, aspect and tense (chapter 6)
∑ Barsalou frames (chapter 12).
The book is accompanied by a website that provides additional support for
students and instructors (http://www.routledge.com/cw/loebner). Along with a
number of additional minor features, the webpage provides
∑ a checklist of key notions for each chapter, interlinked with
∑ a glossary of all technical terms
∑ pdf versions of all figures and tables for your use in teaching, presentations, term
papers, etc.
∑ solutions to the exercises (instructors only).
Acknowledgements
Among the many people that directly or indirectly were of help in writing this book
and preparing the second edition, I want to express my gratitude to Daniel Hole,
Berlin, who took the trouble of commenting in detail on substantial new parts of
the second edition. I would also like to thank Rachel Daw and Nadia Seemungal at
Routledge for their kind and competent support and guidance for the second edition.
This page intentionally left blank
1
Meaning and semantics
Semantics is the part of linguistics that is concerned with meaning. This is, of course,
a very superficial definition. The crucial term ‘meaning’ on which the definition
rests has several different readings – a first semantic observation which you will
find at the beginning of almost every textbook on semantics. Among the many
uses of the notion ‘meaning’, only some belong to the field of linguistic semantics.
Meaning is always the meaning of something. Words have meanings, as do phrases
and sentences. But deeds may have meaning too. If a government pursues a certain
policy, we may ask what the meaning is of doing so. The ‘meaning’ of an action or a
policy is what sense it makes or what purpose it serves or what it is good for. More
generally, we apply the notion of meaning to all sorts of phenomena that we try to
make sense of.
The first thing to be stated is that linguistic semantics is exclusively concerned with
the meanings of linguistic expressions such as words, phrases, grammatical forms
and sentences, but not with the meanings of actions or phenomena. We will approach
the problem of linguistic meaning step by step, to arrive at a more precise definition
of semantics at the end of this chapter. A more concrete idea of what semantics is
about will result when you learn about the many facets of this fascinating discipline
in the course of this book.
1.1 LEVELS OF MEANING
Even if we restrict the study of meaning to words and sentences, the notion of
meaning has to be further broken down into different levels at which we interpret
words and sentences.
1.1.1 Expression meaning
Let us get started by looking at a simple example that will illustrate what semantics
is about.
(1)
I don’t need your bicycle.
This is an ordinary English sentence. Without even noticing, you have already
recognized it as such, you have interpreted it and you are probably imagining
a situation where you would say it or someone would say it to you. Since you
understand the sentence, you know what it means. But knowing what the sentence
2
Understanding semantics
means is one thing, describing its meaning is another. The situation is similar with
almost all our knowledge. We may exactly know how to get from one place to another,
yet be unable to tell the way to someone else. We may be able to sing a song by heart,
but unable to describe its melody. We are able to recognize tens of thousands of words
when we hear them. But the knowledge that enables us to do so is unconscious.
Uncovering the knowledge of the meanings of words and sentences and revealing its
nature is the central objective of semantics.
Let us now try to determine the meaning of the sentence in (1). We start from the
meanings of the words it contains. The main verb in a sentence occupies a key role. So,
1
what is the meaning of the verb need ? Actually, there are two verbs need: an auxiliary
verb (as in I need not go) and a full verb. In (1) we have the full verb. It is used with
2
a direct object (your bicycle) and roughly means ›require‹. We ‘need’ something if
it is necessary or very important for us. In (1), what is needed is described by an
expression composed of the possessive pronoun your and the noun bicycle. The noun
means some sort of vehicle, usually with two wheels and without a motor.
The words need and bicycle are the main carriers of information in the sentence,
so-called content words. The meanings of most content words are very differentiated
because there are thousands of the same kind. All the other elements in our sentence
are different in that they represent items from a very limited choice of expressions of
the same kind. Such words are called function words and include articles, pronouns,
prepositions, conjunctions and other ‘small’ words. We will examine these elements
one by one.
The subject expression I is one of seven personal pronouns in English (I, you, he, she,
it, we and they). What is the meaning of I ? If Mary says the sentence in (1), it is Mary
who is said not to need the bicycle. If John says (1), it is John. In other words, I is used
for the one who says it; more technically: for the one who produces an occurrence of
this pronoun. The technical term for using an expression for something is reference.
When people use I, they refer to themselves. The entity referred to by an expression
is called its referent. The meaning of the pronoun can thus be described as follows:
I indicates reference to the speaker. Similarly, the pronoun you indicates reference to
the addressee or the addressees.
For each personal pronoun there is a corresponding possessive pronoun: I–my,
you–your, etc. Your in (1) indicates that the bicycle referred to is linked to the
addressee(s). For such a link, there is a broad variety of relations possible. Possession
in the sense of ownership is only one option: the expression your bicycle may also
refer to the bicycle the addressee is just riding or cleaning or repairing, or even the
3
bicycle they have been talking about for the last ten minutes. The meaning of your
can roughly be described as ›linked to the addressee(s)‹.
1 It is common practice in linguistic texts to mark words which are referred to in a sentence, rather
than just used, by using italics. In addition, I use italics for emphasis. Whether a word is referred
to or used emphatically is always clear from context.
2 › … ‹ quotes are used for meanings and concepts.
3 I use they as a gender-neutral 3rd person singular pronoun.
Meaning and semantics
3
The form don’t is a contraction of the auxiliary verb do and the negation particle
not. Don’t contributes two things to the meaning of the sentence. It negates the verb
need and thereby turns its meaning into the contrary. In addition, the form don’t
contributes present tense. Tense is the indication that the sentence refers to a certain
time, e.g. present, past or future time. The actual time referred to depends on when
the sentence is uttered. Due to the present tense in (1), we will by default relate the
situation described to the ‘present’ time, i.e. the time when the sentence is being
uttered. Combining these two components of don’t, we may say: the meaning of don’t
is an indication of reference to the time when the sentence is uttered and it turns the
situation expressed by the main verb into the contrary.
So far this has been an attempt to determine the meaning of each word in the
sentence I don’t need your bicycle. This is typical of the work of a semanticist.
As you will have noticed, it is not trivial. For a content word, the description of
its meaning must be specific enough to distinguish it from all words with other
meanings. It would not suffice to describe the meaning of bicycle merely as ›vehicle
with two wheels‹ because there are other kinds of vehicles with two wheels, such as
motorcycles, kick scooters or kids’ balance bicycles. At the same time, the description
must be general enough to cover all cases in which this word could be used. Since
one usually imagines a particular context when one tries to think of a word and its
meaning, one tends to take the meaning too specifically, disregarding other cases in
which the word can also be used.
As for function words like pronouns and auxiliaries and for grammatical forms
such as present tense, their meanings may at first view seem elusive. But it is possible
to account for them too, as our little discussion may have illustrated.
If we put all the pieces together, we can describe the meaning of the sentence as a
whole. It can be roughly formulated as: ›for the speaker, the two-wheeled vehicle of
the addressee(s) is not required at the time when this is being uttered‹.
It is very important to realize that the sentence leaves open who the speaker and
the addressee(s) are, what particular time is referred to and which bicycle. This is not
part of its meaning. Such questions can only be settled if the sentence is actually used
on a concrete occasion. What is, however, determined by the meaning of the sentence
is how the answers to these questions depend on the occasion when the sentence is
used. First, if it is actually used, it is necessarily used by someone who produces the
sentence. With I in subject position, the sentence ‘tells’ us that it is the speaker who
does not need the bicycle. The use of I functions like an instruction: find out who
produced this sentence, this is the referent of I. Second, the use of your presupposes
that there are one or more addressees. The sentence describes the bicycle as related
to them. Third, if a sentence is uttered, it is necessarily used at a certain time. The
time of utterance serves as the reference time for determining what is present, past or
future. The present tense part of the meaning of the sentence conveys the instruction:
attribute the situation described to the time when the sentence is said. Thus the
meaning of the sentence specifies the way in which its reference is determined if and
when it is used at some occasion.
The meanings of words, phrases and sentences, taken out of any particular context
constitute the level of meaning which will henceforth be called expression meaning.
4
Understanding semantics
Expression is just a cover term for words, phrases and sentences. The term expression
meaning covers in particular word meaning and sentence meaning. As you have
noticed, the determination of expression meaning requires an abstraction from the
use of the expressions in concrete contexts. In this sense, the notion of expression
meaning itself is an abstraction and a theoretical construct. But it is justified in the
way language is conceptualized not only in linguistics but also in common thinking:
we do talk about the meanings of words and complex expressions as such, i.e. we do
address this level of meaning.
1.1.2 Utterance meaning
Let us now examine what happens when the sentence in (1) is actually used. We will
consider two scenarios.
SCENARIO 1
1 August 2012, morning. Mary has been planning a trip to town that
afternoon. Two days before, she talked with her neighbour John about the trip
and asked him to lend her his bicycle. She had lent her car to her daughter and
did not know if she would get it back in time. Meanwhile her daughter is back
and has returned Mary’s car. Mary is talking with John on her mobile, telling
him:
I don’t need your bicycle.
Used in this context, the sentence receives a concrete interpretation. References are
fixed: the personal pronoun I refers to Mary, the possessive pronoun your links the
bicycle to her neighbour John and the time reference is fixed, too: in the given context,
the present tense verb will be taken to refer to the afternoon of 1 August 2012. This is
clear from the fact that Mary could have said: I don’t need your bicycle this afternoon,
without changing the meaning of her utterance. Furthermore, the reference of the
grammatical object your bicycle is fixed: it is the bicycle Mary asked John to lend her
two days before.
This is a different level of meaning, called utterance meaning. It comes about
when a sentence with its expression meaning is actually used in a concrete context
and all references get fixed. When this happens, another central notion comes into
play, the notion of truth. If Mary says (1) in scenario 1, the sentence is true. But
in a slightly different scenario it might be false. As long as the sentence (1) is not
actually used with concrete reference, it fails to be true or false. The question of
truth primarily concerns ‘declarative’ sentences such as the one under review. Only
such sentences, when uttered, are true or false. But it matters also for interrogative
and other types of sentences. For example, if John asked Mary Do you need my
bicycle?, the use of the question form would convey that he wants to know from his
addressee whether it is true or false.
Let us now imagine a different scenario:
Meaning and semantics
5
SCENARIO 2
Same time and place. John’s five-year-old daughter Maggie is playing at her
place with her friend Titus. They are playing with a game of cards that display
all kinds of vehicles. Titus holds a card that shows a snowmobile. Maggie
is eager to exchange this card for one of hers and offers Titus a card with a
bicycle. Titus rejects the exchange:
I don’t need your bicycle.
In this scenario, references of I, your and the present tense are fixed accordingly. What
is interesting is that in such a context the word bicycle can be naturally interpreted
as referring not to a real bicycle but to a card carrying the picture of a bicycle. Are
we to conclude that the lexical meaning of bicycle must be taken as covering not only
real bicycles but also pictures of this kind of vehicle and things that display such a
picture? The answer is ‘No’. The word bicycle literally means real bicycles, but when
used in special contexts it can also mean ›picture of a bicycle‹, ›card with a picture
of a bicycle‹, ›toy bicycle‹, ›replica of a bicycle‹, etc. or also ›someone riding on a
bicycle‹ in utterances like ‘Stop, there’s a bicycle coming!’ This, however, is a matter of
utterance meaning. What happens in such cases is that the lexical meaning is shifted
for obtaining an utterance meaning that fits into the given context. Such shifts are
quite common; there are many shifting-patterns at our disposal.
For a general definition of utterance meaning, we need a notion for what was
called ‘occasion’, ‘context’ or ‘scenario’ above. The technical term for this is context of
utterance. The context of utterance, CoU for short, is the sum of circumstances that
bear on reference and truth.
DEFINITION 1 Context of utterance
The context of utterance (CoU) comprises the following aspects of the
situation in which an utterance is produced:
∑ the speaker (or producer) of the utterance
∑ the addressee(s) (or recipient(s)) of the utterance
∑ the time at which the utterance is produced and/or received
∑ the place where the utterance is produced and/or received
∑ the facts given when the utterance is produced and/or received
We have seen in connection with (1), how utterance meaning may depend on who
the speaker and addressees of an utterance are and at which time it is produced. The
place where an utterance is made matters for the reference of expressions such as
here, there, upstairs, downtown, etc. as well as for the truth of sentences like It’s raining.
Facts matter principally for truth as well as for reference. For example, Mary can only
refer to John’s bicycle in such CoUs where a certain bicycle is related to John. CoUs
may be real or fictitious. If we read a work of fiction or watch a movie, the relevant
facts and figures are those of the story.
6
Understanding semantics
Against this background, utterance meaning can be defined as the meaning that
results from an expression being used and interpreted in a given CoU. Utterance
meaning derives from expression meaning on the basis of the particulars provided
by the CoU.
The notion of utterance meaning does not include all that an addressee may
make of an utterance in a particular CoU. Addressees usually make all kinds of
inferences. For example, in scenario 1, John may infer that Mary is still planning to
make the trip since otherwise she would have told him; that she would have asked
him to lend her his bicycle if she could not have used her car; that, however, her
daughter is back with the car and that Mary is not going to lend her the car again
on that afternoon; that Mary will take the car for her trip; that she considers herself
able to drive, etc. All this is not explicitly said with that sentence, and it need not be
true under different circumstances. In the given scenario, these inferences can be
considered communicated because Mary can rely upon John’s understanding them.
Although these inferences are triggered in the addressee’s mind by the utterance of
the sentence, it is important to separate what is actually being said from what is only
inferred. The investigation of such inferences, their role in communication and how
they are related to the utterance meaning of what is actually said, is an important
part of pragmatics, the scientific study of the rules that govern the use of language.
Within pragmatics, Grice’s theory of ‘conversational implicatures’ and Relevance
Theory by Sperber and Wilson deal with inferences of this kind.
1.1.3 Communicative meaning
Neither the level of expression meaning nor that of utterance meaning is the final and
crucial level of interpretation. In an actual exchange, our main concern inevitably is
this: what does the speaker intend – in particular, what does the speaker want from
me? Conversely, when we say something, we choose our words in pursuit of a certain
communicational intention. Verbal exchanges are a very important form of social
interaction. They will always be interpreted as part of the whole social exchange and
relationship entertained with the speaker.
One and the same sentence can be uttered with quite different communicative
results. The utterance of (1) in scenario 1 will be taken as a withdrawal of a former
request. In scenario 2, the utterance of the same sentence constitutes the rejection of
an offer. In other CoUs, uttering the sentence could serve still other communicative
ends. A theory that addresses this level of interpretation is speech act theory,
introduced in the 1950s by the philosopher John L. Austin (1911–60) and developed
further by others, in particular John R. Searle. The central idea of speech act theory
is that whenever we make an utterance in a verbal exchange we act on several levels.
One level is what Austin calls the ‘locutionary act’, defined as the act of saying an
expression with a certain utterance meaning in the given CoU. In doing so, we
also perform an ‘illocutionary act’, i.e. a certain type of ‘speech act’: a statement,
a question, a request, a promise, an offer, a refusal, a confirmation, a warning, etc.
When Titus in scenario 2 says I don’t need your bicycle, he performs the locutionary
Meaning and semantics
7
act of saying that he doesn’t need Maggie’s card with the bicycle and the illocutionary
act of rejecting her offer. The speech act level of interpretation will be referred to as
communicative meaning.
The three levels of interpretation are connected as follows. Expression meaning is the
level of interpretation which results if the only information we use is the mere linguistic
material. Expression meaning forms the basis for utterance meaning, but does not
determine it. For, as we could see, a sentence with its fixed expression meaning will take
on different utterance meanings if it is used in a particular context. Utterance meaning,
in turn, forms the basis of communicative meaning, without, again, determining it. For
utterances with the same utterance meaning can serve the performance of different
types of speech acts, depending on the ongoing social interaction. Table 1.1 gives a
survey of the three levels of meaning and how they are defined.
Table 1.1
Three levels of meaning
Level of meaning
Definition
expression meaning
the meaning of a simple or complex expression taken in isolation
utterance meaning
the meaning of an expression when used in a given context of utterance
resulting from fixing reference
communicative meaning
the meaning of an utterance as a communicative act in a given social
setting
1.2 SENTENCE MEANING AND COMPOSITIONALITY
1.2.1 Lexical vs compositional meaning
We will now take a closer look at sentence meaning. It is a trivial fact that the
meanings of words and sentences differ in one important point. Meanings of words
must simply be known and therefore learned. In our minds, we host a huge lexicon
where all the words we know and their meanings are stored and at our disposition.
Stored meanings are therefore called lexical meanings.
Words can be combined into sentences. We are usually able to understand the
expression meaning of a sentence without any conscious effort. Nevertheless, this
ability is based on complex cognitive processes which take place in our minds
automatically and unconsciously. The process by which we calculate the meaning
of a sentence is called composition, and the resulting meaning is known as
compositional meaning. In some cases, sentences may have lexical meaning, for
example proverbs such as The early bird catches the worm. This does not mean that
their meanings are merely non-compositional. Rather, such sentences have a regular
compositional non-proverbial meaning plus a special meaning which we have to
learn and store in our lexicon.
Understanding semantics
8
Words can also be used to create new words. This is called word formation,
and its products complex words. For example, you could coin the word joggable,
deriving it from the verb jog by adding the suffix -able, which can be added to a very
large number of verbs. Or you could form the compound carp food. The products of
word formation need not all be stored in the lexicon and yet they can be as easily
and straightforwardly understood as sentences. Obviously, the semantics part of the
brain also knows how to interpret new words formed out of stored lexical material.
Crucially, this holds only for products of word formation that follow general patterns.
There are, on the other hand, very many complex words that carry irregular or
special meanings. Such cases have to be stored in the lexicon in order to receive their
proper interpretation. We’ll return to the semantics of word formation in 1.2.5. For
now, I just want to fix in your mind a terminological convention: the notion of ‘lexical
meaning’ is also to be taken to subsume the meanings of complex words, whether
semantically regular or irregular. This is in line with the ‘dynamic’ view of the lexicon
which sees it not just as a store of entries, but also as comprising components for the
formation of new words and their meanings.
Let us now take a closer look at the composition of sentence meaning in order to
see what is involved.
1.2.2 Grammatical meaning
For a change, we will consider a new example.
(2)
The dog ate the yellow socks.
Let us assume that we have assessed the lexical meanings of the words in (2): the, dog,
eat, yellow and socks. There are no larger units in the sentence with lexical meaning;
the rest of the interpretation is composition. The words in (2) occur in particular
grammatical forms. The verb form ate is past tense – more precisely: simple past
tense rather than progressive (was eating); it is in the indicative mood rather than
in the conditional (would eat), it is active rather than passive (was eaten), it is not
negated (did not eat). The noun socks is plural; dog is singular, by absence of plural -s.
The adjective yellow is neither comparative (yellower) nor superlative (yellowest), but
is given in its basic form called ‘positive’. The forms of the words matter directly for
their meaning, and consequently for the meaning of the whole sentence. The singular
noun dog has a different meaning from the plural noun dogs: dog refers to a single
creature of this kind, and dogs to more than one. Likewise, the meaning of present
tense eat(s) is not the same as that of past tense ate, and the meaning of the simple
form ate is different from the meaning of the progressive form was/were eating. The
meaning of the simple past form with verbs such as eat is the combination of past
tense and perfective aspect. (Tense and aspect will be discussed in chapter 6 on
verbs.)
The meanings of different word forms are not all stored in our lexicon. Rather the
lexicon contains the meaning (or meanings, cf. chapter 3) of only one form of the
word. For a verb, this is a tenseless and aspectless active meaning, for a noun it is
Meaning and semantics
9
4
its singular meaning and for an adjective its positive meaning. The meanings of
particular word forms are derived from the basic lexical meanings by general rules.
These are part of the apparatus we use in composition. We will call the meaning of
grammatical forms grammatical meaning. The meaning of a word in a particular
form is the combination of its basic lexical meaning and the grammatical meaning
of its form.
Word forms are often built by adding an affix to the invariable part of the word, its
stem. Affixes appended at the end of the word are called suffixes, those added at the
beginning are prefixes. It must be noted that not all differences in the grammatical
forms of words matter for their meaning. A certain form may be necessary just for
grammatical reasons. For example, a certain verb form may be obligatory depending
on the subject of the sentence (cf. I am vs she is, she sees vs you see). By contrast, the
grammatical number (singular or plural) of the nouns dog and sock in (2) can be
chosen independently of the construction in which they occur. Therefore, differences
in form only matter for meaning if – in a given construction – different forms can be
chosen freely. In addition, the different forms available must have different meanings.
If, for example, a verb has two different admissible past tense forms like spoiled and
spoilt, the choice of the form would not matter for its meaning. Also, it may be that
one form has a neutral meaning which subsumes the meanings of the alternative
forms. For example, in Korean nouns can take a plural suffix; it is, however, not
necessary to make use of this option in order to refer to a multitude of instances.
Therefore, the basic form of the noun without a plural suffix has a neutral meaning
and does not carry grammatical meaning.
We can sum up these considerations in the following definition of grammatical
meaning:
DEFINITION 2 Grammatical meaning
The form of a word carries grammatical meaning, if
(i) in the given construction alternative forms are admissible
(ii) different forms yield different meanings
(iii) the form chosen does not have a neutral meaning.
Languages differ considerably in what meanings they encode in grammatical forms.
Many languages do not obligatorily mark number on all nouns; others do not have
5
non-positive forms of adjectives or lack tense.
4 In some cases, certain forms of words may have a special lexical meaning, such as glasses as
opposed to glass, or used to as opposed to to use. Some nouns are only used in the plural (trousers);
for certain others the distinction does not matter (logic, logics).
5 See Corbett (2000) for a survey on systems of grammatical number including missing number;
Dryer (2011) offers an online map of forms of number marking for more than 1000 languages
(http://wals.info/chapter/33). For tenseless languages see Comrie (1985: 2.5).
10
Understanding semantics
1.2.3 Syntactic structure
As the last step of composition, the meanings of the words in their given forms are
combined into a whole, the meaning of the sentence. This process is guided by the
syntactic structure of the sentence (this is, for the most part, what grammar is good
for: to guide the interpretation of complex expressions). Let us first determine which
words in (2) belong together. The words the dog form a syntactic unit, a constituent
in syntactic terminology. This kind of constituent is called a noun phrase, NP
for short. The words the yellow socks form another NP; it contains an adjective
in addition to the article and the noun. Actually, the adjective and the noun form
another constituent within the NP. The combination of words into larger syntactic
constituents is governed by the rules of grammar. There is a rule for combining
adjectives with nouns, and another rule for combining a noun, or an adjective-noun
combination, with an article (the article comes first).
Given such rules for forming larger syntactic units we need corresponding
composition rules, in this case:
∑ a rule for deriving the meaning of an adjective-noun combination (yellow socks)
from the meaning of the adjective and the meaning of the noun
∑ a rule for deriving the meaning of an article with a noun (the + dog) or a noun
with adjective(s) (the + yellow socks) from the meaning of the article and the
meaning of the noun, or the noun with an adjective
We will not try to specify such rules now; they will be dealt with in chapters 4, 5
and 13.
Having assessed the dog and the yellow socks as complex constituents, we turn
to the total structure of the sentence. It consists of these two NPs and the verb ate.
These three parts are grammatically related as follows: the verb is the predicate of the
sentence, the NP the dog is its subject and the yellow socks its direct object. The verb
and the direct object form a constituent, known as verb phrase, or VP, which is then
combined with the subject NP to form the complete sentence. We therefore need two
more composition rules:
∑ a rule for deriving the meaning of a VP (ate + the yellow socks) from the meaning
of the verb and the meaning of the direct object NP
∑ a rule for deriving the meaning of a sentence (the dog + ate the yellow socks) from
the meaning of the subject NP and the meaning of the VP
Again, these rules are not trivial. Roughly speaking, the composition works as follows:
the verb eat in its given ‘active’ form means an event which necessarily involves two
elements, an eater and something that is eaten; the direct object NP contributes a
description of the object that is eaten, and the subject NP a description of the eater.
1.2.4 The Principle of Compositionality
Let us sum up the general results we can draw from this example. The syntactic
rules of a language allow the formation of complex expressions from what will be
Meaning and semantics
11
Figure 1.1
The process of composition
GRAMMAR
SEMANTICS
complex
expression
compositional meaning
of the complex
combine expressions
compose meanings
basic expressions
in a particular form
compositional meanings
of the basic expressions
in their given forms
choose grammatical
forms
add grammatical
meaning
basic expressions
lexical meanings
of basic expressions
called basic expressions. (Basic expressions are expressions with a lexical meaning.)
The meaning of complex expressions is determined by semantic composition. This
mechanism draws on three sources:
1 The lexical meanings of the basic expressions
2 The grammatical forms of the basic expressions
3 The syntactic structure of the complex expression
The general scheme in Fig. 1.1 shows that semantic composition is thought of as a
so-called bottom-up process: it proceeds from the basic units to the complex ones.
The lexical meanings of the smallest units serve as input for the rules of grammatical
meaning, whose output is the input for the composition rules. The converse of a
bottom-up process is a top-down process. If semantic interpretation were a topdown process, this would mean that the meanings of words are derived from the
meanings of sentences.
That complex expressions receive their meaning by the process of composition is
the central idea underlying semantics. This is called the Principle of Compositionality.
The principle is attributed to the German philosopher, logician and mathematician
Gottlob Frege (1845–1925), and sometime called Frege’s Principle. Although he
obviously applied the principle, there is no passage in his publications that could
serve as a quotation.
12
Understanding semantics
DEFINITION 3 Principle of Compositionality
The meaning of a complex expression is determined by the lexical meanings
of its components, their grammatical meanings and the syntactic structure of
the whole.
The principle implies that the meanings of complex expressions are fully determined
by the three sources mentioned, i.e. by the linguistic input alone. The principle,
therefore, does not apply to utterance meaning and communicative meaning because
at these levels non-linguistic context knowledge comes into play.
1.2.5 The semantics of word formation
Just like sentence formation, word formation has two sides: a form side and a meaning
side. Along with the formation of grammatical word forms, the form side of word
formation belongs to the domain of morphology, the area of linguistics concerned with
the forms of words. The two main mechanisms of word formation are derivation and
compounding. In the case of derivation, a given word is modified, usually by adding
an affix. Examples are the adjective drink-able, the noun drink-er or the verb en-slave.
There is also conversion – the use of a word in a different word class without changing
its form, for example the noun drink converted from the verb drink or the verb bottle
converted from the noun bottle. In English, conversion is very frequent. The other major
mechanism, compounding, is the combination of two or more words into one. Some
examples in this chapter are word meaning, content word, speech act and composition
rule. The products of word formation are called complex words, including the products
of conversion, even though nothing is added to the word form.
The semantics of word formation is a field that is much less clear than sentence
semantics. There certainly are patterns, however: certain conversions, derivations
and compounds are regularly interpreted in a particular way. Let me just give four
examples:
∑ suffixing a verb V with -able produces an adjective meaning ›one can V it‹: drinkable, read-able, compar-able, deriv-able
∑ suffixing a verb V with -er produces a noun meaning ›someone who Vs‹ : driv-er,
los-er, speak-er, bak-er
∑ prefixing a verb V with un- produces a verb which denotes the reversal of what is
done by V-ing: un-do, un-zip, un-plug, un-tie
∑ prefixing a verb V with out produces a verb which means ›doing V better, or
more‹: out-run, out-play, out-number, out-weigh
Such patterns can be defined as pairings of formation rules and corresponding
interpretation rules. If they can be applied to an open input class, they are called
productive. Productive patterns crucially rely on the fact that there is a general
interpretation rule for their productions; otherwise they could not be freely applied.
Meaning and semantics
13
However, the semantics of word formation is complicated by the fact that there
are often several interpretation rules for the same formation rule. This is a situation
quite different from the composition of sentence meaning where, as a rule, there
is only one composition rule for each syntactic rule. Compounds in particular are
notoriously manifold in how the meanings of the parts are related: a ‘book title’ is
a title that names a book, a ‘book shelf ’ is a shelf for books, a ‘book shop’ is a shop
where books are sold, and so on.
Another reason for the complexity of word formation semantics is the existence
of so many cases with unpredictable, lexicalized meanings. These blur the picture of
the semantic rules of word formation. If you take a look into a standard dictionary
you’ll find that very many entries are complex words. They are listed in the dictionary
because their meanings are irregular – otherwise there would be no need to include
them for explanation. Extreme cases are words like butterfly; the word looks like a
compound of butter and fly, but its meaning has nothing to do with ›butter‹ and
little to do with ›fly‹. Very frequent are complex words with special meanings that
are narrower than their regular meaning, e.g. drinker, which does not just mean
›someone who drinks‹, but rather ›someone who regularly drinks too much alcohol‹.
When two nouns occur in a sentence, like book and shelf or book and shop in (3),
often the sentence tells us how they are to be linked semantically:
(3)
a.
Ann put the book onto the shelf.
b.
She had bought it in the shop around the corner.
In (3a) we see that the shelf is something the book is put onto, in (3b) that the
shop is something where the book was bought. The respective verb specifies a
semantic relation between the two nouns. With compounds, there is nothing like
this to help us to know what the two parts have to do with each other. They are
just jammed together. There are authors who suggest that compounding is a ‘relic
of protolanguage’ (Jackendoff 2009:111). According to the theory of protolanguage
(Bickerton 1992), there was a stage in the evolution of language where there were
words with meanings but not yet grammar. Words were just somehow combined; no
functional elements, no word classes, no rules for sentences. According to Bickerton,
protolanguage still exists, for example as an early stage in language acquisition or
in the speech of aphasics who have lost grammar. Jackendoff adds compounding as
another case.
If the way in which two elements are to be semantically combined is not explicitly
expressed, there is more room for interpretation. Nevertheless, the interpretation of
compounds is not just mere guessing. It cannot be, for we obviously agree on how to
interpret regular compounds. The reason for this is that the way semantically to link
a noun to another noun is often built into their meanings. It is part of the meaning
of shop that a shop is a place where things are sold, and it is part of the meaning of
book that it is some kind of artefact which can be bought. Putting this information
together leads us to the interpretation of book shop. We will turn to the semantics of
word formation in various places in this book (8.2.2, 11.6.2, 12.2.2 and 12.3.3).
14
Understanding semantics
1.3 SEMANTICS: ITS SCOPE AND LIMITS
Obviously, we are able to understand sentences in isolation, for example if we find
them written on a piece of paper or appearing on a screen of our computer. We
know word meanings; we know the meanings of grammatical forms; we know how
to combine the meanings of the words in a complex expression. We need not have
a concrete CoU in order to interpret a sentence. We can just stop interpreting at the
meaning level of expression meaning.
One way of explaining what semantics is about is to say that it describes this part of
our linguistic knowledge: the cognitive equipment that enables us to interpret regular
linguistic expressions out of context, simply as linguistic forms. We acquire this
cognitive equipment along with grammar when we learn our native language. And
we have practical evidence for assuming that this equipment is, by and large, uniform
among our speech community. We normally understand what others say and we are
normally understood – at least at the level of expression meaning.
Of course, there are considerable differences in lexical knowledge among speakers
of the same language. Other people may know certain words and their meanings
which we don’t, and vice versa, and sometimes people disagree on word meanings.
But even so, we know that these differences can be removed by explaining or
discussing word meanings. As to grammatical meanings and composition rules, we
very rarely seem to disagree with others. The cognitive equipment for interpreting
linguistic expressions is essentially shared within a speech community. Agreement
within the community is the result of a permanent synchronization of language
usage, and thereby linguistic knowledge.
This mentalist approach to semantics is adopted here. It takes the so-called
‘I-language’ as its essential object. ‘I-language’ means the mental language apparatus
of the individual speakers of a language. Making I-language the point of departure
allows one to pose the relevant questions: What are the meanings which we have
stored in our minds or compute compositionally? What kind of semantic operations
do we apply to meanings, such as the combination of meanings, or meaning shifts?
Which kind of information is involved at the three levels of interpretation? Such
questions help to say more precisely what belongs to the scope of semantics and what
is beyond its scope.
1.3.1 The scope of semantics
1.3.1.1 Explaining sentence meaning
According to the Principle of Compositionality, the following sub-disciplines of
semantics are required in order to account for our ability to interpret sentences in
isolation:
∑ lexical semantics: the investigation of expression meanings stored in the mental
lexicon (dog, sock)
Meaning and semantics
15
∑ semantics of regular word formation: the investigation of the regular meanings
of words that are formed by the productive rules of word formation (joggable,
carp food)
∑ semantics of grammatical forms: the investigation of the meaning contribution
of grammatical forms
∑ compositional sentence semantics: the investigation of the rules that determine
how the meanings of the components of a complex expression interact and
combine
Often semantics is subdivided into two sub-disciplines only: lexical semantics is
then understood as also comprising regular word semantics, and the semantics of
grammatical forms is subsumed under sentence semantics.
1.3.1.2 Interfaces of expression meaning with other levels of interpretation
The domain of semantics is not confined to sentence meaning. It also concerns the
links of sentence meaning to utterance meaning and communicative meaning. The
example in (1) illustrated two things that we do with the meaning of a sentence when
we try to make sense of an utterance of it.
First, we will try to fix references for those elements of the sentence that require
this. Elements such as the pronoun I call for fixing their reference and guide us in
doing so. Remember that the meaning of I is an instruction for determining the
referent of the pronoun as the one who has produced the utterance of this particular
‘I’. One can think of such an element as an ‘anchor’ in the sentence meaning which
tries to find a purchase in the context for anchoring the sentence meaning in the CoU.
Other such anchors are provided by definite articles, by tense or by specifications of
time or location like here and then. It is a task for semantics to investigate the nature
and function of such anchors. These anchors are technically called indexicals. They
will be dealt with in more detail in chapter 4. Indexicals are part of what is called
the interface between expression meaning and utterance meaning. This interface
consists of those components which link the level of sentence meaning to the level of
utterance meaning. It, too, belongs to the domain of semantics.
Second, there is the aspect of what expressions can mean when used in context,
in addition to what they literally mean. It may happen that we are led to shift lexical
meanings when we try to fix the reference of a sentence in a satisfactory way. This is
what we saw happen to the word bicycle in the second scenario, where it was taken in
the sense ›card with a picture of a bicycle‹. Such shifts take place before the reference
is fixed. They, too, belong to our cognitive semantic equipment. They constitute
another component of the interface between expression meaning and utterance
meaning. We will deal with them in the second part of chapter 3.
There is also an interface of expression meaning with communicative meaning. It
is constituted by elements that guide the interpretation of an utterance as a speech
act. These elements include grammatical means of indicating different types of
sentences such as declarative, interrogative or imperative. Also to be included are
16
Understanding semantics
elements such as the indication of formality or politeness or elements that directly
serve certain speech acts, like please or goodbye (see 2.3 on social meaning).
We can now define the discipline of semantics much more precisely:
DEFINITION 4 Semantics
Semantics is the study of the meanings of linguistic expressions, either
simple or complex, taken in isolation. It further accounts for the interfaces of
expression meaning with utterance meaning and communicative meaning.
1.3.2 Beyond semantics
The approach to semantics taken here draws a clear boundary. Semantics ends
where contextual knowledge comes in. Thus, utterance meaning and communicative
meaning are beyond semantics; they fall into the domain of pragmatics.
Another important point of demarcation concerns the lexical meaning of content
words. Obviously, people vary greatly in how much knowledge they have about certain
things. Is all we can know about avocados, for instance, to be considered the lexical
meaning of the word avocado? This cannot be. Otherwise the lexical knowledge of
most content words would vary too much between experts and plain language users.
Rather, the meaning of a word is something like the common denominator of the
content shared in a speech community. The actual conventional meaning is a lean
bundle of information, just enough to secure efficient communication. Therefore,
word meaning needs to be distinguished from general world knowledge. This issue
will be taken up when we discuss connotations in the next chapter (2.5), and later in
11.6, when we will distinguish between semantic knowledge and cultural knowledge,
in the context of cognitive semantics.
EXERCISES
1. Explain the distinction between the three levels of meaning in your own words.
Part of the exercises will always be to define central terms introduced in the chapter
in your own words. In order to acquire new words, it is very important that you
really make active use of them. Do not copy the definitions from the text, but try to
explain them out of your own understanding. It may be very useful to work on the
answers together with others.
2. Sue tells Martin: ‘I don’t care if you use my microphone.’
Which one of the three levels of meaning is addressed in the following questions?
Give your reasons for the answer.
a. What is the meaning of microphone?
b. Which microphone is meant?
c. Is this an offer?
d. Does the sentence relate to one or several microphones?
Meaning and semantics
17
3. The square brackets in the following sentence indicate constituent structure.
(The words within a pair of brackets form a constituent.)
[I [[don’t [know]] [where [[the kids] are]]]]
a. How many composition rules are needed for this sentence? (You need not
formulate the composition rules.)
b. Determine word by word if the word form carries grammatical meaning. (You
need not specify the grammatical meanings.)
4. Describe the meanings of the following compounds. Try to explain how the
meanings come about:
a. schoolboy
b. schoolmate
c. schoolhouse
5. Read chapter 1 of Isac & Reiss (2008) on I-language.
FURTHER READING
1.1 Lyons (1995, ch. 1) on levels of meaning. Verschueren (1999) ch. 1 on inferences
and speech acts, ch. 3 on the role of context in interpretation. Levinson (1983) on
Austin’s speech act theory and Grice’s theory of implicatures; Huang (2007) on
speech act theory and relevance theory.
1.2 Van Valin (2001) and Tallerman (2011) on elementary syntax. On word formation
see Plag (2003) and ch. 19 by Laurie Bauer and Rodney Huddleston, in Huddleston &
Pullum (2002), Olsen (2012) on the meaning of compounds.
1.3 Isac & Reiss (2008, ch. 1) for an introductory definition of I-language; Huang
(2007, ch. 7) on the distinction between semantics and pragmatics.
2
?
Dimensions of meaning
This chapter will try to convey a more precise idea about expression meaning. In
the first part about descriptive meaning, we will consider the relationship between
meaning, reference and truth. The second part is concerned with non-descriptive
meaning, i.e. dimensions of lexical meaning that are relevant on the level of social
interaction or for the expression of subjective attitudes and evaluations.
2.1 MEANINGS ARE CONCEPTS
In order to understand what kind of entities word and sentence meanings are, the
best thing we can do is consider the role that meanings play in actual communication.
We will discuss another concrete example and assume a CoU that takes up scenario 1
from 1.1.2: Mary, just back from her trip, finds her daughter Sheila quite upset. Sheila
has spent the time with Mary’s dog Ken, and the two do not like each other. When
asked what happened, Sheila answers:
(1)
The dog has ruined my blue skirt!
Let us suppose that what Sheila says is true and that Mary believes what Sheila
says. Mary will then know something she did not know before: that Ken has ruined
Sheila’s blue skirt. She knows this because Sheila said (1) and because this sentence
has the meaning it has. Let us take a closer look at how the transfer of information by
such a sentence works, first for a single word and then for the whole sentence.
2.1.1 The meaning of a word
We assume that Sheila is referring to Ken. What enables Mary to recognize that?
Sheila used the words the dog: the definite article the and the noun dog. Both play
an important role. The main information is conveyed by the noun. It specifies the
referent as an entity of a certain kind, namely a dog. What entitles us to say so? It
is the fact that the word dog means what it means. When you were asked to explain
what the word dog means, you will probably say that dogs are a certain kind of
medium-sized animal with four legs and a tail, that they are often kept as pets, that
they bark, that they may bite, etc. In other words, you will most likely give a general
description of dogs. This is an adequate reaction: giving a general description of dogs
may well count as an explanation of the meaning of dog. At least roughly, the meaning
Dimensions of meaning
19
of such words may safely be regarded as a description of the kind of thing the word
can be used for.
Now, a very important point to realize is this: the word does not carry this
description with it. This can be seen from the trivial fact that words which we do not
know do not have any meaning to us. What a word in fact carries with it when it is
spoken and heard is its sound form (or its spelling, if it is written). When Sheila says
the word dog, she produces a certain sound pattern. The respective sound pattern
is stored in Mary’s mind as part of her linguistic knowledge and enables her to
recognize the word when she hears it.
The meaning of the word dog, i.e. the description of dogs, must also be something
residing in Mary’s mind. It must be knowledge directly linked to the sound pattern of
the word. The meaning is therefore a mental description. For mental descriptions in
1
general, the term concept will be used. A concept for a kind, or category, of entities
is knowledge that allows us to discriminate entities of that kind from entities of other
kinds. A concept should not be equated with a visual image. Many categories we have
words for, like mistake, thought, noise, structure, mood, are not categories of visible
things. But even for categories of visible things such as dogs, the mental description
is by no means exhausted by a specification of their visual appearance. The dog
concept, for example, also specifies the behaviour of dogs and how dogs may matter
for us (as pets, watch dogs, guide-dogs, dangerous animals that may attack us, etc.).
We can now give a partial answer to the question of how Mary is able to recognize
that Sheila is referring to Ken: Sheila acoustically produces the word dog; Mary
recognizes the sound pattern; in her mind the pattern is linked to the meaning of the
2
word dog, the concept ›dog‹; the concept is a mental description of a potential referent.
So due to the use of the word dog, Mary knows what kind of entity Sheila is referring to.
That Mary has the concept ›dog‹ linked to the sound pattern of dog in her mind
is, of course, only part of the story. Sheila must have the same concept in her mind
linked to the same sound pattern. More generally, a word can only be considered
established if its form and meaning are linked in the minds of a great number of
language users.
Still, we have not explained how Mary is led to assume that Sheila refers to this
particular dog. The crucial clue to an explanation is the definite article the. Had Sheila
used the indefinite article a instead, Mary would not have concluded that Sheila was
referring to Ken. What is the meaning of the definite article? It does not provide a
direct cue to Ken, but it signals that the description supplied by the following noun
applies to an entity in the given CoU which the addressees are supposed to be able to
sort out. Therefore the article will cause Mary to ask herself which entity in the given
3
CoU fulfils these conditions.
1 In Chapter 11 on cognitive semantics the notions ‘concept’ and ‘category’ will be treated in more
depth and detail. For the present purposes you may take a category as a set of entities of the same
kind.
2 This kind of quote will be used for concepts, and meanings in general: ›x-y-z‹ is the concept that
constitutes the meaning of the expression x-y-z.
3 We will have a much closer look at the meaning of the definite article in 4.4.2.
20
Understanding semantics
This is how far the meanings of the words the dog take us in reconstructing the
communication between Sheila and Mary with respect to the reference to this
dog Ken. For the conclusion that it is Ken which Sheila is referring to, Mary needs
extra-linguistic context information. The fact that Sheila is using the definite article
restricts the choice of candidate dogs to those Mary and Sheila both know. The
family’s dog Ken is a privileged candidate, since he is firmly established as ‘the dog’
in the given context.
2.1.2 The meaning of a sentence
In her mind, Mary has the forms and meanings of all words in (1) at her disposal.
She also knows the grammatical meanings of the singular form, of the positive form
of the adjective and of the indicative present perfect form of the verb (recall 1.2.1).
Applying all this and her knowledge of grammar to the linguistic input, she will be
able to compose the expression meaning of the whole sentence (1.2). The result is one
complex concept which combines the meanings of all elements of the sentence. Let
us call this a concept for a situation. The main component of the situation concept
is the concept ›ruin‹ contributed by the verb. It is of central importance because
4
it connects all other elements. As a concept for an event of the kind ‘x ruins y’ it
involves three elements: the event itself, the ruiner x and the ruined object y. In the
complete situation concept, the event is described as one of ruining, the ruiner is
described as a dog that is identifiable in the given CoU, the ruined object is described
as a skirt, a blue one, linked to the speaker (recall 1.1.1 for the meaning of possessive
pronouns like my and your); the present perfect tense contributes the specification
that the situation at the time of utterance results from a previous event of the kind
indicated. Thus the meaning of the sentence as a whole is a concept for a specific kind
of situation. It can roughly be described as in (2). The description does not contain
an explanation of the content-word meanings, but it makes explicit the contribution
of the functional elements (1.1.1).
(2)
›the situation at the time of utterance results from a previous event in which a
dog that is uniquely determined in the CoU ruined a blue skirt which is uniquely
determined by its being linked to the speaker‹.
What was said about the meanings of words and sentences can be summed up as
follows:
DEFINITION 1 Meanings of content words and sentences
The meaning of a content word (noun, verb, adjective), is a concept that
provides a mental description of a certain kind of entity.
The meaning of a sentence is a concept that provides a mental description of
a certain kind of situation.
4 See chapter 5 on predication.
Dimensions of meaning
21
2.2 DESCRIPTIVE MEANING
In the previous section it was established that expression meanings are concepts.
Actually the discussion here and in chapter 1 was confined only to a certain
dimension of meaning, namely the dimension that bears on reference and truth. It is
called descriptive meaning, alternatively ‘propositional meaning’. We will elaborate
on descriptive meaning now, making more explicit how it is related to reference and
truth. We will turn to non-descriptive meaning in the second half of the chapter.
2.2.1 Descriptive meaning and reference
2.2.1.1 Reference and the descriptive meaning of words
When dealing with reference, the first thing to be observed is that, strictly speaking, it
is usually not simply words that have referents. If the sentence in (1) is true, it involves
reference to five things: the dog (an object, in the widest sense), the speaker’s blue skirt
(another object), the speaker herself, the ruining of the skirt (an event) and the time of
utterance (a time). Table 2.1 shows which elements of the sentence refer to these five
things. The subject NP and the object NP each have a referent; the possessive pronoun
within the object NP has a referent of its own. The finite verb contributes reference to
a certain kind of event and, due to its tense, to a certain time. The adjective blue has no
referent of its own, but it contributes to the description of the referent of the NP my blue
skirt. The example shows that the referring elements of the sentence can be phrases (e.g.
NPs), words (the verb) or grammatical forms (tense).
Table 2.1
Five referents of sentence (1)
Type
Referent
Referring element
object
the dog
NP
the dog
object
the speaker’s blue skirt
NP
my blue skirt
object
the speaker
poss. pronoun my
event
ruining
verb
time
utterance time
tense has ___ed
ruin
All this notwithstanding, it makes sense to talk of the potential referents of
content words. Since the referent of an NP is essentially described by the noun, we
may loosely speak of it as the ‘referent of the noun’. Analogously, we can talk of the
‘referent of a verb’. Adjectives never have a referent of their own, but they always
describe the referent of some NP (see 5.4 for details). Thus, still more loosely
speaking, we may extend the notion of referent to adjectives, keeping in mind that
their ‘referents’ are borrowed, as it were. In simple words, a potential referent of
a content word is whatever can be called so. If the notion of potential referent is
22
Understanding semantics
extended to all content words, descriptive meaning can be defined in a way that
relates it directly to reference:
DEFINITION 2 Descriptive meaning of content words
The descriptive meaning of a content word is a concept for its potential
referents.
When a sentence is used in a particular CoU, the addressees will try to fix concrete
referents that match the descriptions. However, and this is a very important point, it
may be impossible to fix referents, if the sentence is not true. Consider the sentence
in (3):
(3)
There is a letter for you.
Let us assume that Sheila says so to her mother, but that she is not telling the truth:
there is no letter for Mary. There may be a letter, but not for her, or no letter at all. In
any event, if the sentence is not true, the NP a letter for you lacks a referent. Usually,
the finite verb of the sentence has a concrete event referent only if the sentence is true.
For example, if (1) is false in some CoU, then the dog has not ruined the speaker’s
blue skirt and hence the verb ruin, in that CoU, fails to have a referent.
2.2.1.2 The descriptive meaning of sentences: propositions
There is no generally accepted notion for what a sentence as a whole refers to in a
given CoU. For lack of a better term it will be called the situation referred to. The
referents of the referring elements of the sentence are components of the situation
referred to; together, in their particular constellation, they form a complex state of
affairs. For the sentence in (1), the situation referred to in the given CoU comprises
the five referents listed in Table 2.1: (i) the dog Ken, (ii) the speaker Sheila, (iii) her
blue skirt, (iv) its being ruined by the dog, (v) the time referred to. The particular
constellation of these five referents is as the sentence has it: at the time referred to, the
blue skirt is Sheila’s and its condition is the result of it being ruined by the dog Ken.
The notion of the situation referred to only makes sense if the sentence is true: as
we have seen, some elements of the sentence may lack a referent if it is not true. Thus,
only if a sentence is true in a particular CoU does it properly refer to a situation of the
kind it describes. Therefore, whenever the term situation referred to is used, it will be
assumed that the sentence is true.
In analogy to the notion of potential referents we can talk of the situations
potentially referred to. These are all those situations that fit the mental description
provided by the meaning of the sentence, i.e. all the situations for which the sentence
is true. The descriptive meaning of a sentence can now be defined as in Definition 3.
In accordance with common terminology, the descriptive meaning of a sentence is
called its ‘proposition’. Alternatively, the proposition of a sentence will be referred to
as the ‘situation expressed’, or the ‘situation described’.
Dimensions of meaning
23
DEFINITION 3 Proposition
The descriptive meaning of a sentence, its proposition, is a concept that
provides a mental description of the kind of situations it potentially refers to.
As we have seen, it is not only content words that shape the descriptive meaning
of the sentence. Functional elements such as pronouns and articles or tense, a
grammatical form, contribute to the proposition as well (recall the description of the
meaning of (1) given in (2)). Making use of Definition 3, we can give the following
general definition:
DEFINITION 4 Descriptive meaning of a word or a grammatical form
The descriptive meaning of a word or a grammatical form is its contribution
to the descriptive meanings of the sentences in which the word or grammatical
form may occur.
To sum up, the descriptive meaning of a sentence is a concept for a certain kind of
situation. If the sentence is true in a CoU, such a situation actually exists and can
be considered the referent of the sentence. The situation referred to contains the
referents of all referring elements of the sentence. Table 2.2 gives a survey of different
types of potentially referring expressions, their respective descriptive meanings and
types of referents.
Table 2.2
Descriptive meanings of the elements of sentence (1)
Expression
(type)
Descriptive meaning
(* definitions adopted from The New Oxford Dictionary
of English)
Referent
(type)
skirt
(noun)
›a woman’s outer garment fastened around the waist
and hanging down around the legs‹ *
object
eat
(verb)
›put (food) into the mouth and chew and swallow it‹*
event
blue
(adjective)
›of a colour intermediate between green and violet, as
of the sky or sea on a sunny day‹*
object [borrowed]
the [noun]
(article)
the referent of the noun is uniquely determined in the
given CoU
–
I
(pronoun)
the referent is the speaker
object
The dog has ruined my
blue skirt.
(sentence)
see (2)
situation
24
Understanding semantics
2.2.2 Denotations and truth conditions
2.2.2.1 Denotations
The descriptive meaning of a content word is a concept for its potential referents. As
such it determines, or mentally describes, a category of entities. The meaning of dog
5
is a concept that determines the category DOG of all dogs, the verb concept ›ruin‹
determines the category RUIN of all events of ruining. The category determined by
the meaning of a content word is called its ‘denotation’; a word is said to ‘denote’ this
category.
DEFINITION 5 Denotation
The denotation of a content word is the category, or set, of all its potential
referents.
The denotation of a word is more than the set of all existing entities of that kind. It
includes fictitious referents as well as real ones, usual exemplars and unusual ones,
6
maybe even exemplars we cannot imagine because they are yet to be invented.
The relationship between a word, its meaning and its denotation is often depicted
in the semiotic triangle, a convenient schema which will be used in this volume in
a variety of forms. Figure 2.1 gives the semiotic triangle for the descriptive meaning
of content words as such. The arrow that connects the word with its denotation is
drawn with a broken line. This is to indicate that a word is not directly linked to its
denotation, but only indirectly via its descriptive meaning.
Figure 2.1
The semiotic triangle for a content word in general
content word
descriptive meaning:
a concept
denotation:
a category
5 SMALL CAPITALS are used for categories.
6 Yet the totality of existing exemplars of a category certainly is representative of the category, and
primarily shapes our concept for this kind of thing. We will occasionally use the term actual
denotation for the subset of the denotation that is formed by its real existing members. In 11.6
the notion of cultural category will be introduced for actual denotations. There we will see more
clearly what the relation is between the total denotation of a word and the corresponding cultural
category – and why the distinction is important.
Dimensions of meaning
25
When a content word is actually used in a concrete CoU, we deal with a token of the
word, i.e. a particular spoken or written realization. The semiotic triangle then yields
a relationship between the word token, its meaning and its referent: the meaning
describes the referent to which the word, in the given CoU, refers (Fig. 2.2).
Figure 2.2
The semiotic triangle for a content word token
content word
descriptive meaning:
a concept
referent in the
world
2.2.2.2 Truth conditions
There is no established term for what would be the denotation of a sentence.
In analogy to the denotation of a content word it would be the set, or category, of
all situations to which the sentence can potentially refer, i.e. the category of
all situations in which the sentence is true. There is, however, another notion that
is quite common and directly related to the would-be denotation of a sentence: its
7
so-called truth conditions.
DEFINITION 6 Truth conditions
The truth conditions of a sentence are the conditions under which it is true.
We will say that a sentence has the truth value ‘true’ if it is true, and the truth value
‘false’ if it is false. In order to see what is meant by ‘truth conditions’, let us consider
sentence (1) once more. Obviously, this sentence taken as such is not merely true
or false; rather, its truth value depends on the CoU in which it is used: the question
8
of its truth or falsity arises only in relation to a given CoU. Sentence (1) is true in
a given CoU if there is a uniquely determined dog and a uniquely determined blue
7 The definition given here is sufficient for more general purposes. It will be made more precise in
4.7.3 in order to integrate presuppositions.
8 There are exceptional sentences that have the same truth value in all possible CoUs: these are called
‘logically true’ or ‘logically false’ (see 7.2); examples are sentences like ducks are birds, which is
logically true, or two times three is seven, a logically false sentence. Only for this kind of sentences,
can the truth value be determined independently of a given CoU.
26
Understanding semantics
skirt belonging to the speaker. Additionally, the dog must have done something to the
skirt such that at the time of utterance the skirt is ‘ruined’. If all these conditions are
fulfilled in a given CoU, the sentence is true in this CoU, and vice versa: if the sentence
is true in a CoU, then all these conditions are fulfilled.
This is the reason why it is possible to communicate things by saying a sentence. We
utter a sentence like (1) or any other declarative sentence, and communicate, by the
way in which we speak, that what we say is true in the given CoU. The addressee(s)
will take the truth of the sentence for granted – assuming that they have no reason
to doubt what is being said – and conclude that its particular truth conditions are
fulfilled in the given CoU.
If sentence (1) were uttered by someone else and/or at some other time and/or
under different circumstances, it might well be false. Of course, this does not mean
that there is only one CoU where (1) is true. Many blue skirts belonging to somebody
have been, or will have been, ruined by some dog and in all these cases the owner
may truly utter sentence (1).
For an explicit formulation of the truth conditions of (1) we can resort to the
description of the meaning of (1) given in (2) above.
(4)
Truth conditions of sentence (1):
The sentence The dog has ruined my blue skirt is true in a given CoU if and only
if the following holds:
‘The situation at the time of utterance results from a previous event in
which a dog that is uniquely determined in the CoU ruined a blue skirt
which is uniquely determined by its being linked to the speaker.’
A proper definition of the truth conditions of a sentence S always takes this form:
‘S is true in a given CoU if and only if …’
‘If and only if ’ – abbreviated ‘iff ’ – means ‘if ’ in both directions: ‘S is true if the truth
conditions hold, and if the truth conditions hold, then S is true.’
We can now see that the notion of truth conditions is equivalent to the notion of
the denotation of a sentence. The denotation of a sentence would be the category, or
set, of all the situations the sentence can refer to. If one knows the truth conditions of
a sentence, one knows which situations the sentence can refer to: to those which are
part of a CoU where the sentence is true. Conversely, if one knows which situations a
sentence can refer to, one is able to define its truth conditions: the sentence is true in
a CoU iff this CoU comprises a situation that the sentence can refer to.
In analogy to Fig. 2.1, the connection between a sentence, its proposition and its
truth conditions can be put as follows: the descriptive meaning of the sentence is its
proposition, and the proposition determines the truth conditions of the sentence.
The resulting picture is given in Fig. 2.3, another variant of the semiotic triangle.
Dimensions of meaning
27
Figure 2.3
The semiotic triangle for a sentence
sentence
proposition:
a complex concept
truth conditions
2.2.3 Proposition and sentence type
So far in this discussion of sentence meaning one aspect has been neglected: the
grammatical type of the sentence too contributes to its meaning, and this contribution
is non-descriptive. Compare, for example, sentence (1) to its interrogative counterpart
(5):
(1)
The dog has ruined my blue skirt.
(5)
Has the dog ruined my blue skirt?
The question describes exactly the same sort of situation. Hence it is considered to
have the same proposition as (1). Yet the total meaning of (5) is, of course, different
from the meaning of (1): (5) renders a question while (1) renders an assertion. The
difference in meaning is due to the grammatical forms of the sentences or, technically
speaking, to differences in grammatical sentence type. (1) is a so-called declarative
sentence. Declarative sentences in English have a certain word order: the finite verb is
in the second position of the sentence, usually after the subject. (5) is an interrogative
sentence of the yes-no-question type: in English, the finite verb is in the initial
position and has to be an auxiliary verb.
The semantic contribution of the grammatical sentence type is not part of
the proposition. For declarative sentences, it consists in presenting the situation
expressed as actually pertaining. This sentence type is therefore used for making
assertions, communicating information, etc. The interrogative sentence type, by
contrast, leaves open whether or not the situation pertains. It is therefore the
standard option to be chosen for asking questions.
Imperative sentences represent a third grammatical sentence type:
(6)
Don’t ruin my blue skirt!
In English imperative sentences, the finite verb fills the first position of the sentence
and normally there is no explicit subject. Imperative sentences are used for
commands, advices and similar speech acts. The proposition of (6) would be that the
addressee(s) do not ruin the speaker’s blue skirt.
28
Understanding semantics
The meaning contribution of grammatical sentence type is a first example of
non-descriptive meaning. We will now consider two more types: social meaning and
expressive meaning. The meaning of sentence type belongs to neither of them.
2.3 MEANING AND SOCIAL INTERACTION:
THE DIMENSION OF SOCIAL MEANING
Talking to others is social interaction, i.e. an activity exerted in coordination with
others. Any verbal utterance will receive an interpretation as a communicative act
(1.1.3) in the current social network, and in this sense it always has a social function.
Language as such can be said to serve first and foremost social functions. (This
does not contradict the view that language is primarily a means of communication:
communication, in particular the communication of information, is of course a very
important type of social interaction.)
2.3.1 Expressions with social meaning
The term social meaning does not refer to this general aspect of verbal interaction,
and is thereby not to be confused with the communicational meaning of a verbal act.
Rather, social meaning is on a par with descriptive meaning: it is a dimension of the
lexical meaning of certain words, phrases or grammatical forms. If an expression
has social meaning, it has so independently of the particular CoU. Most expressions
and grammatical forms do not have social meaning, but some do. If expressions with
social meaning are built into sentences, these inherit a dimension of social meaning
due to the principle of compositionality. Let us consider an example. Sheila is on the
train in Germany and is addressed by the ticket inspector:
(7)
a.
Ihre Fahrkarte, bitte! – Danke.
(German)
b.
Deine Fahrkarte, bitte! – Danke. (German)
c.
‘Your ticket, please! – Thank you.’
(7a) would be appropriate if Sheila is an adult and no acquaintance of the inspector.
9
The 3rd person plural form of the possessive pronoun, Ihre, literally ‘Their’, is
required for the formal, or ‘polite’, style of speech used for addressing adults. By
contrast, (7b) contains the simple 2nd person singular possessive pronoun dein and
would be the proper, informal, form of address if Sheila were a child, a relative or a
close acquaintance of the ticket inspector. Otherwise, using (7b) would be considered
rude. If the inspector addressed Sheila in English, (7c) would be adequate in all cases.
But when speaking German, the inspector is forced to choose between the formal and
the informal way of address (or to avoid the use of pronouns altogether).
The formal German pronoun Sie, when used as a term of address, has the same
descriptive meaning as English you: it indicates reference to one or more addressees.
9 In their ‘polite’ use the 3rd person plural pronoun and the corresponding possessive pronoun are
written with a capital letter: Sie for you, Ihr for your.
Dimensions of meaning
29
But, in addition, it has a non-descriptive dimension of meaning which English you
lacks: the social meaning of indicating a certain kind of social relationship between
the speaker and the addressee. Correspondingly, the informal variants du (singular)
and ihr (plural) have the same descriptive meaning as Sie (if one disregards the
differentiation in number) but differ in social meaning. The distinction between the
two kinds of relationship relevant for choosing either Sie or du in German is also
relevant in other respects: it coincides with the use of surnames with title (along
with Sie) vs first names as vocative terms of address. For example, the unmarked
way of address and reference to the addressee would be (8a) or (8b), while the mixed
variants in (8c) and (8d) are marked under normal circumstances:
(8)
a.
Ist das Ihr Fahrrad, Herr Schmidt?
formal
b.
Ist das dein Fahrrad, Helmut?
informal
c.
?? Ist das Ihr Fahrrad, Helmut?
mixed
d.
?? Ist das dein Fahrrad, Herr Schmidt? mixed
‘Is that your bicycle, Helmut/Mr Schmidt?’
(7a) and (7b) above contain two further expressions with social meaning: bitte
›please‹ and danke ›thank you‹. Unlike the German terms of address, these two
expressions have exclusively social meaning; they are the first expressions we
encounter which lack descriptive meaning altogether. (English thank you, containing
you, might be considered as referring to the addressee(s). To this extent it also has
descriptive meaning.) The sentences in (7) are elliptical: they lack anything but
the direct object and could be spelt out as Geben Sie mir Ihre Fahrkarte, bitte (lit.
‘give you me your ticket, please’). At least they will be interpreted in this sense
and can thus be considered elliptical imperative sentences with the proposition
›addressee gives speaker addressee’s ticket‹. The addition of bitte does not change
the proposition. Rather it marks the request as modestly polite. Bitte, in this use, is a
mere formality marker, indicating, similar to the forms of address, a certain kind of
social relationship between speaker and addressee(s).
Danke ›thank [you]‹ has no descriptive meaning either. It just serves as a slightly
formal response that counts as a recognition of a favour or service. Given different
social rules, the ticket inspector might have bowed instead of saying anything. Since
the utterance lacks a proposition, the question of truth does not arise. One might
think that, instead, the question arises whether or not the speaker is in fact grateful.
But their emotions are not relevant. If in a relevant situation people say ‘Danke!’ or
‘Thank you!’ they have automatically performed the act of thanking, regardless of
what they actually feel. Verbal acts of thanking obey specific social rules, and these
are observed if one merely utters one of the conventional phrases of thanking.
All languages have set phrases with a clear-cut social meaning and no other
dimension of meaning: phrases of greeting (Hi) or saying goodbye, phrases of
apologizing (sorry), recognizing (thank you), or answering the phone. For each such
phrase in each language, there is a social rule that defines the circumstances under
30
Understanding semantics
which it is properly used and what it then means. For example, the use of goodbye
phrases is only adequate if both speaker and addressee(s) are in some form of social
contact; saying goodbye then means a certain step in ending the current contact.
Generally, the defining criterion for an expression carrying social meaning is this:
DEFINITION 7 Social meaning
An expression or a grammatical form has social meaning if and only if it
conventionally serves the indication of social relations or the performance of
conventionalized social interaction; there must be rules for social interaction
that govern its use.
By contrast, the use of expressions with descriptive meaning is governed by rules
of factual adequacy; for example, a declarative sentence is used correctly if its truth
conditions are fulfilled in the given CoU.
In 1.1, three levels of meaning were introduced. Expressions like please or goodbye,
with exclusively social meaning, belong to the interface between expression meaning
and communicative meaning (1.1.3, 1.3.1). They are elements in the utterance that
provide the crucial clues for the kind of speech act which is performed. The social
rules that govern their use are treated in Austin’s speech act theory under the label
‘felicity conditions’ (i.e. conditions necessary for the speech act actually to come
about; see Further Reading in chapter 1 for references). Table 2.3 displays examples
of different kinds of expressions with social meaning.
Table 2.3
Social meaning
Expression (type)
Social meaning
Descriptive meaning
German du
(pronoun of address)
informal relationship
the person addressed
German Sie
(pronoun of address)
formal relationship
the person or persons addressed
English you
(pronoun of address)
–
the person or persons addressed
Sheila
(proper name as term of address)
informal relationship
the person called Sheila
Mr Murple
(proper name as term of address)
formal relationship
the person called Mr Murple
please
(adverb)
formal demand
–
Hi!
(complete utterance)
informal greeting
–
Dimensions of meaning
31
In the linguistic literature, phenomena of social meaning are mostly treated under
the label of honorifics. These include forms of address and self-reference, special
grammatical forms such as the Japanese masu-form (see next section) and other
linguistically conventionalized forms of paying respect.
2.3.2 Social meaning in Japanese
In European and European-influenced societies, social differentiation of speech
is only moderately reflected in the language system. Apart from set phrases,
social meaning is essentially restricted to terms for reference to the addressee.
Furthermore, the system of differentiation is usually restricted to two levels of
formality, an informal and a more formal way of speaking. In other cultures, however,
the social differentiation of expression pervades language to a much greater extent.
Such languages include Japanese, Korean, Vietnamese, Thai, Burmese and former
imperial Chinese. We will take a closer look at Japanese.
In English or German, you do not find any formality markers in sentences unless
the speaker refers explicitly to the addressee. There would be no simple way of
reformulating an utterance like The dog ate the yellow socks to express your relationship
with the addressee along the lines relevant for the use of surname vs first name or
the use of Sie vs du in German. (The only possibility would be to add a vocative to
the sentence: The dog ate the yellow socks, John / Mr Murple.) In Japanese, however,
one would have to choose among two or more levels of formality in any sentence
whatsoever, because formality is obligatorily marked on the finite verb. In Japanese, the
sentence the dog ate the yellow socks would correspond to either (9a) or (9b):
(9)
a.
inu
wa
kiiroi
sokkusu
o
tabe-
ta.
dog
TOP
yellow
socks
ACC
eat
PT
10
‘the dog ate the yellow socks’
b.
inu
wa
kiiroi
sokkusu
o
tabe-
mashi- ta.
dog
TOP
yellow
socks
ACC
eat
FORM
PT
‘the dog ate the yellow socks’
The plain past tense form of the verb in (9a) represents the informal way of talking
which is only appropriate in rather intimate relationships like those within a family,
between lovers or good friends. Normal formal talking, which is the standard between
people of the same status, is expressed by inserting a formality marker, here in the form
10
= topic marker (‘as for … ’), ACC = accusative case marker, FORM = formality marker, PT =
past tense, PS present tense. In Japanese, particles follow the NP; there are no articles. We use
the Hepburn transcription for Japanese which is oriented at the pronunciation (consonants are
roughly pronounced as in English, vowels as in Spanish or German).
TOP
32
Understanding semantics
-mashi-, between the stem of the verb and its tense ending (9b). The resulting form is
called the masu-form. The marker -mashi- has no descriptive meaning.
In addition to marking the level of formality on the finite verb, Japanese has rich
differentiation among pronouns, or pronoun-like expressions, for the addressee as
well as for the speaker. Instead of the single 1st person singular pronoun I in English,
a Japanese speaker has to choose among plain omission (usual and neutral), boku or
ore (informal, men only), watashi (formal, standard), watakushi (very formal) and
other, more special expressions. Furthermore, there are different nouns and verbs
for things belonging to, or actions done by, oneself vs others. Let us consider another
example in order to get an idea of how the differentiation of formality works:
(10) a.
Uchi
ni
i-
ru.
b.
Uchi
ni
i-
mas-
u.
c.
Uchi
ni
ori-
mas-
u.
home
in
be
FORM
PS
‘… am/is/are at home’
The sentences in (10) are typical colloquial Japanese sentences, lacking a subject
term. The word uchi for someone’s home is an informal term. The verb iru ›to be
[temporarily]‹ is used in its plain, informal form in (10a). (10b) is more formal,
replacing plain iru with its masu-form imasu. In addition to the grammatical form,
the choice of the verb itself matters for social meaning. The verb iru as such has no
social meaning attached to it. In (10c), however, it is replaced by the verb oru in its
masu-form orimasu. Oru also means ›to come/go‹, but is ‘humble’: by using it, its
subject is socially lowered. There is also a variant of ›to be/come/go‹ that raises its
subject referent in social status, the verb irassharu with its masu-form irasshaimasu:
(10) d.
Otaku ni
irasshai- mas-
u.
home
be
PS
in
FORM
‘… is/are at home’
If, for the sake of simplicity, the meanings ›to go/to come‹ of oru and irassharu
are disregarded, the four sentences in (10) all have the same descriptive meaning:
›[someone not specified] is at home‹. The term uchi for someone’s home is informal,
while otaku is a formal term that marks the house as belonging to someone superior.
It is now interesting to see how the differences in social meaning restrict the use and
interpretations of the sentences. The employment of terms with social meaning in
Japanese is governed by two principles: (i) the addressee is never inferior, and (ii)
the speaker is never superior. Given these principles, (10d), although it contains no
subject, cannot be used in the sense ›I am at home‹. (10c) amounts to ›I am at home‹
or that somebody is at home who, in face of the addressee(s), ranks equal with the
speaker, e.g. members of the speaker’s family. By contrast, (10a) and (10b) can mean
Dimensions of meaning
33
›I/you/he/she/it/we/they am/is/are at home‹ since the verb is neutral with respect to
the social status of the subject.
2.4 MEANING AND SUBJECTIVITY:
THE DIMENSION OF EXPRESSIVE MEANING
Anything we say will also be taken as the expression of a personal emotion, opinion
or attitude. When you imagine Sheila saying ‘The dog has ruined my blue skirt’, you
will probably have assumed a manner of speaking that displays anger, indignation
or frustration. But the sentence can also be spoken in a neutral manner or in a
way that exhibits different emotions, for example desperation or concern, but also
delight, amusement or relief. It could be pronounced in a mean or a kind manner. Or
hastily, or particularly slowly. All these ways of shaping an utterance would serve the
expression of Sheila’s feelings and attitudes towards the dog, or Mary, or the entire
situation. But even if she chose a neutral way of speaking, Sheila would inevitably
display certain traits of her personality, by choosing the term ruin for what the
dog did to her skirt, by not calling the dog by its name, by reporting damage of her
property to the dog owner, etc. In this sense, every utterance serves, consciously or
not, the expression of personal feelings, evaluations and attitudes.
2.4.1 Expressive meaning
Again, this general aspect of language use, its expressive function, is not what is
meant with the notion of expressive meaning. On a par with descriptive and social
11
meaning, expressive meaning is part of the lexical meaning of certain expressions,
a semantic quality of words and phrases independent of the CoU and of the way
they are being spoken. Again, there are two kinds of expressions with this special
meaning: some have only meaning and others have both descriptive and expressive
meaning. There do not seem to be expressions with both social and expressive
meaning. We will first turn to the former.
DEFINITION 8 Expressive meaning
An expression or a grammatical form has expressive meaning if and only if
it conventionally serves the immediate expression of subjective sensations,
emotions, affections, evaluations or attitudes.
Let us call expressions with expressive meaning simply expressives. The most
typical instances are words and phrases such as ouch, wow, oh. Such interjections
11 Be careful not to confuse the terms expression meaning and expressive meaning. Expression
meaning is a cover term for the semantic meaning of words, phrases and sentences, since all
these are expressions. Expressive meaning is a dimension of expression meaning, alongside with
descriptive meaning and social meaning.
34
Understanding semantics
are language-specific. Languages may differ in how many such expressions they
have, and an interjection may have different meanings in different languages. Here
are some interjections from Hungarian: fuj [fuj] (disgust), au [u] (sudden pain),
jaj [jj] (sudden pain or fright), jajaj [jjj] (sadness or concern), hüha [hyh]
(admiration, warning, fright), hú [hu:] (admiration), ejha [ejh] (astonishment).
Other examples of expressives are exclamations of various sorts, such as Gosh!,
Goddammit!, Jesus!, Oh my goodness!, and so on.
At least some feelings, sensations, attitudes and evaluations can thus be expressed in
two ways: subjectively and immediately by means of expressives, and propositionally
by forming sentences with the respective descriptive meaning. For example, the
difference between saying ‘Ouch!’ and ‘That hurts!’ is this: That hurts! is a sentence
with a regular descriptive meaning, consisting in the proposition that ‘that hurts’.
‘Ouch!’ expresses sudden pain, as would a certain grimace or moaning. You can react
to someone saying ‘That hurts!’ with ‘Really?’, ‘That’s too bad’ or ‘I don’t believe it!’,
which are replies relating to the proposition of the sentence and its being true. You
cannot reply to ‘Ouch!’ in the same way because there is no proposition.
The rules governing the use of expressives are simple. Since all expressives serve
to express personal feelings, attitudes or sensations, which are perceptible only to the
holder, their correct use is just a matter of personal judgement. For instance, saying
ugh! is generally taken as an expression of disgust. In order to decide whether the
use of this expression is correct in a given CoU, the speaker only has to make up their
mind whether or not they want the addressee(s) to believe that they find something
disgusting.
Interjections and exclamations can be used as complete utterances. Other
expressives such as hopefully, (un)fortunately or thank God can be inserted into
a sentence in order to add a personal attitude to the situation expressed. These
additions do not contribute to the proposition, as the following examples may
illustrate:
(11) a.
b.
Fortunately, Bob will arrive tonight. – Really?
I’m glad Bob will arrive tonight. – Really?
The questioning reply ‘Really?’ is always directed at a proposition asserted before.
In (11a), the remark can only be related to the proposition that Bob will arrive that
night (‘Will Bob really arrive tonight?’), not to the speaker’s attitude expressed by
fortunately. In (11b), however, the same attitude is described by the descriptive
formulation I’m glad ... and is hence part of the proposition of the sentence.
Therefore, the reply ‘Really?’ can be used to question the speaker’s claim that they are
glad that Bob will arrive (‘Are you really glad that Bob will arrive tonight?’), i.e. their
attitude described.
More drastic cases of expressives are swear words. Consider the following
12
passages from the film Pulp Fiction:
12 Quentin Tarantino, Pulp Fiction (London: Faber and Faber, 1994), pp. 140 and 13, respectively.
Dimensions of meaning
(12) J:
35
But me, my eyes are wide fuckin’ open.
V:
What the fuck does that mean?
J:
That’s it for me. From here on in, you can consider my ass retired.
(13) Any of you fuckin’ pricks move and I’ll execute every motherfuckin’ last one of
you.
The adverbial insert the fuck in (12) serves the expression of an emotion that is
probably neither necessarily positive (like joy) nor negative (like indignation) but
just strong: it adds an emotional charge to the whole utterance. Other expressives,
like bloody or the highly productive fucking add the same kind of emphasis to
single components of the proposition. None of these expressives make additions
to the propositional content. However, in the examples two other expressions with
expressive meaning do: my ass in (12) is an emotionally charged term for reference to
the speaker. It must have descriptive meaning because it is an NP in object position.
Its descriptive meaning is just ›I/me‹. Similarly, the expression you fuckin’ pricks in
(13) has the same descriptive meaning as the plain pronoun you, plus the specific
expressive content of the rest of the phrase. This expression is not just emotionally
charged but clearly offensive when used with reference to others.
In English and presumably most other languages, there exist many emotionally
charged terms for persons, such as idiot, bastard, motherfucker or ass-hole for the
offensive part, or darling, baby, honey as terms of affection. They all have the same
unspecific descriptive meaning ›person‹. The phenomenon extends to expressions for
men, women, children, body parts or other things of central importance in everyday
life, such as certain animals, vehicles, housing, food, clothes, as well as to the most
common activities: walking, talking, working, eating, drinking, sleeping or sexual
activities. Most expressive terms add a negative attitude to the meaning, others are
just emotional. Positive expressive meanings are rare. Many of these expressions have
a neutral, purely descriptive, meaning variant besides the expressive one. For example,
paw in its neutral meaning denotes the feet of certain animals. The expressive variant
of the word means ›hand‹ for its descriptive meaning part while its expressive
meaning part is an unspecific emotional emphasis.
Table 2.4
Expressive meaning
Type
Expression
Expressive meaning
Descriptive meaning
interjection
ouch
sudden pain
–
adjective
stupid
dislike, contempt
–
adverb
fortunately
preference
–
noun
bastard
dislike
›person‹
noun
paw
emotional emphasis
›hand‹ (of a person)
verb
cram
emotional emphasis
›eat‹
36
Understanding semantics
2.4.2 Social vs expressive meaning
Many semanticists consider expressive meaning and social meaning as not clearly
separated. The distinction is, however, not as difficult as it might appear. The use of
terms and forms with social meaning is governed, for this part of their meanings,
by rules of social interaction. They define what kind of social circumstances make
suitable occasions for using the expression and they define what its use is taken
for: a greeting, an apology, a polite or intimate way of referring to other persons,
etc. By contrast, the use of terms with expressive meaning is governed (again, for
this part of their meanings) by criteria which concern only subjective adequacy
with respect to expressing one’s personal feelings, attitudes etc. Sure enough, there
are rules of conduct constraining the expression of feelings or attitudes under
certain social conditions and the use of expressive terms, in particular swear
words as terms of address, may have severe social consequences. But while there
are clear social rules for the use of, say, the first name vs the surname plus title
as terms of address, there is no such rule for addressing someone with idiot or
motherfucker.
Some expressions with predominantly social meaning can be considered as means
of the socially ritualized expression of feelings and attitudes, e.g. terms of thanking,
wishing well, congratulation, condolence or apologizing. Phrases such as I’m sorry
or nice to meet you, which literally represent descriptions of attitudes, point in this
direction. Still, these phrases are primarily social and not expressive. One’s real
subjective feelings are not, and cannot be, socially relevant. What counts is the actual
expression of feelings, and behaviour consistent with having them.
Table 2.5
Dimensions of meaning
Dimension
Function
Criteria for correct use
descriptive meaning
description of referents and situations
agreement with facts
social meaning
indication of social relations and
performance of social acts
social rules of conduct
expressive meaning
immediate expression of personal
sensations, feelings, attitudes or
evaluations
subjective choice
2.5 CONNOTATIONS
If an expression has descriptive meaning, any mention of it will activate not only the
concept for its potential referents but together with it a host of further associations.
Among the associations, some are conventional. They are called connotations
and often considered to be something like a secondary meaning in addition to the
primary lexical meaning. However, connotations such as ‘dirty’ for pigs are neither
Dimensions of meaning
37
part of the descriptive meaning of pig (clean pigs can perfectly be referred to as
‘pigs’) nor do they constitute expressive meaning (the word pig can be used in
an expressively neutral way). Often the connotations of a word change, while its
meaning remains the same. For example, the connotations of the word computer
have changed considerably since the 1960s (when computers had the connotations of
dangerous super-intelligent machines threatening to escape human control and take
over), but the word still means the same. What has changed dramatically in the last
fifty years is the actual denotation (recall fn. 6) of the word, and it is these changes
which have given rise to the change of connotations. It is therefore more appropriate
to consider connotations to be connected not to the word itself (like meaning) but
rather to the actual denotation.
While a distinction is drawn in this volume between a word’s meaning and the
connotations associated with its actual denotation, it should be mentioned that
expressive meaning, e.g. the negative attitude expressed by derogative terms, is
called connotation by other authors. Indeed, connotations play a role for the semantic
motivation of swear words. For example, the conventional attribute of dirtiness is
the basis of the use of pig or the German equivalent Schwein as an offensive term for
people. But the attitude expressed by the swear word is not part of the meaning of pig
as a mere animal term, nor is the conventional attitude towards pigs identical with
the expressive meaning conveyed by the swear word.
Negative connotations, together with social taboos, are responsible for what
is called euphemisms: roughly, good or indirect terms for bad or taboo things.
Political language is full of euphemisms (just take the vocabulary of warfare),
as are the semantic fields of death (pass away for die) or sexuality (cf. indirect
terms like intercourse, sleep with s.o., make love, etc.) Negative connotations are
also at issue in matters of political correctness. Certain labels like homosexual
are considered discriminatory due to their connotations and replaced by
connotatively neutral terms or terms used by the members of the social group
themselves. But as long as the discrimination itself persists in society, the new
expressions will soon take on the old connotations. Due to this inflationary
process, we are observing a rapid succession of ‘politically correct’ expressions
for certain social groups (such as handicapped being replaced by disabled being
replaced by challenged).
2.6 DIMENSIONS OF MEANING
We have had a detailed look at three dimensions of meaning: descriptive, social and
expressive meaning. In addition, we saw in 2.2.3 that the grammatical meaning of
sentence type is another non-descriptive dimension of meaning: it is neither social
nor expressive meaning. These four dimensions of meaning, if given, are equally
parts of expression meaning. The distinction of dimensions of meaning is orthogonal
to the distinction of levels of meaning which we made in 1.1 (see Fig. 2.4).
38
Understanding semantics
Figure 2.4
Dimensions and levels of meaning
Dimensions
Levels
expression
descriptive social expressive (others)
utterance
speech act
The levels of utterance meaning and communicative sense build on the level of
expression meaning. Therefore, these levels inherit descriptive, social and expressive
meaning (and further dimensions of meaning) from the basic level of expression
meaning (if it is present there). Consider an utterance like this:
(14) ‘Jack, take your damn coat off my seat!’
The sentence carries the descriptive meaning ›addressee take their coat off the
speaker’s seat‹, the grammatical meaning of an imperative sentence, the social
meaning of informal address by the given name and the expressive meaning
conveyed by damn. At the level of utterance meaning, this amounts to an imperative
with informal reference to the addressee and pejorative reference to their coat. The
communicative sense amounts to a rather rude request.
Semantics has always been mainly concerned with descriptive meaning. The
exploration of social and, even more so, expressive meaning is not yet very advanced.
For this reason, these dimensions will not be further elaborated in this book.
Mainstream semantics, in particular in the field of sentence semantics, investigates
linguistic meaning by means of logical analysis (see chapters 7 and 13). Applied to
sentences, it aims at describing their truth conditions and the rules of composition by
which they derive; for content words and complex expressions, it is the denotations
that matter. Widely disregarding other dimensions of meaning, expression meaning
is identified with descriptive meaning, and this in turn is reduced to truth conditions
and denotations.
In this book we take a mentalist approach to meaning: we deal with meanings as
concepts. These concepts determine truth conditions and denotations (as depicted
in the semiotic triangles), but they cannot be identified with them. For as we shall
see in 7.6 and 13.5, it is possible that words with different meanings have identical
denotations, and sentences with different propositions may have the same truth
conditions. There may be conceptually different ways of denoting the same set of
cases or expressing the same kind of situation.
If we want to understand the connection between language and cognition, it is
important to investigate the conceptual level of meaning as this is the interface
between language and thinking. Of course, the dimension of descriptive meaning
Dimensions of meaning
39
is the one that is by far the most differentiated and interesting one; it constitutes
the way in which we represent the world. But if we also want to understand the way
in which the use of language is embedded in social interaction and in which it is
connected to our feelings and emotions, we should spend more effort understanding
social and expressive meaning.
EXERCISES
1. Try to define the descriptive meaning of the following words and compare your
definition to the definition given in a monolingual English dictionary, e.g. Oxford
English Dictionary: fish, milk, red, pregnant, follow.
2. What are the appropriate terms in English for (i) asking somebody to repeat
something you did not understand, (ii) for accepting an apology, (iii) for
answering the phone, (iv) for New Year’s greetings? What kind of meaning do
these expressions have?
3. Try to find three interjections other than ouch and ugh and determine what they
express.
4. Try to find expressives with the descriptive meaning ›mouth‹, ›car‹, ›child‹, ›walk‹,
›work‹.
5. Try to find five examples of modern euphemisms.
6. Try to determine the presently politically correct terms for people who are
politically incorrectly called ‘blind’, ‘black’ or ‘fat’.
7. Discuss the roles of words uttered in communication: how is it possible that
words ‘convey’ meaning?
8. Discuss the connection between descriptive meaning, reference and truth.
9. Most sentences of a natural language like English can be used in different CoUs
for communicating different information, although their meaning remains the
same. How is this possible?
10. Try to determine word by word the descriptive, social and/or expressive meaning
of the expressions in the following dialogue:
A:
Hi, Velma. Are you going to that stupid lecture?
B:
Well, yes. I’m sorry, Sweety.
11. Discuss the relationship between expression meaning and connotations for
words such as pig.
40
Understanding semantics
FURTHER READING
Lyons (1977, ch. 7) on reference, sense (his term for meaning) and denotation.
Lyons (1995, chs 6.6–7) on sentence type and its meaning. Palmer (2001) has a
comprehensive account of non-descriptive sentence meaning. Levinson (1983,
ch. 2.2.5) on social deixis. Brown and Gilman (1960) for a classical study about
pronouns of address. Brown and Levinson (1978) for a general account of social
meaning. Suzuki (1978, ch. 5) for a comparison of forms of address and self-reference
in Japanese and English; Sohn (1999: 9.14) for social meaning in Korean; Cooke
(1968) on the rich choice of terms for speaker and addressee in Thai, Burmese and
Vietnamese; Kummer (2005) on honorifics in Thai. Watts (2003) offers a more recent
general work on politeness from a sociolinguistic perspective. Lyons (1995, ch. 2.3),
Cruse (1986, ch. 12.2) for expressive vs descriptive meaning. Andersson and Trudgill
(1990) and Hughes (1992) for comprehensive studies of swear words in English, also
Bryson (1990, ch. 14) on swearing and taboos in English.
3
Ambiguity
In dealing with meaning in the previous chapters, expressions were treated as though
they had only one meaning (though possibly composed of different dimensions).
This is, of course, not the case. Many, if not most, words have more than one meaning
and even complete sentences may allow for several readings. The technical term for
this phenomenon is ambiguity: an expression or an utterance is ambiguous if it
can be interpreted in more than one way. The notion of ambiguity can be applied to
all levels of meaning: to expression meaning, utterance meaning and communicative
meaning. In dealing with ambiguity, we will first turn to the level of lexical meaning.
This is in accordance with the general bottom-up perspective on interpretation taken
in this book (1.2, 1.3).
3.1 LEXEMES
In 1.2.1 we distinguished lexical meaning from compositional meaning. Lexical
meaning must be learned and stored in our mental lexicon; compositional meaning is
the meaning of composite expressions and is derived ‘on demand’ from the meanings
of the components by applying general semantic rules. Those linguistic units which
carry lexical meanings are called lexemes. Typically, lexemes are single words, but
there are also composite expressions with a special lexicalized meaning. Therefore
‘lexemes’ cannot just be equated with words. Examples of composite lexemes are
so-called idioms like throw in the towel meaning ›give up‹ or fixed adjective-noun
combinations such as white lie, broad bean or little finger. Composite lexemes need
not be idioms like the ones mentioned. Less spectacular cases are particle verbs such
as give up, fill in, look forward to, put on, figure out, etc.
Lexemes are stored in the lexicon of the language, a huge complex structure in
the minds of the language users. Lexical meaning is not to be confused with the
meaning you may find in a dictionary. Dictionaries describe the meanings of their
entries by means of paraphrases. For example, you may find the meaning of bird
described as ‘feathered animal with two wings and two legs, usually able to fly’ (New
Oxford Dictionary of English). In order to understand the description, you have to
know what these other words, feathered, animal, etc. mean. If you look these up in
turn, you will find yet other words used for their description. Dictionaries are in this
sense circular. No matter how carefully they are compiled, they will always contain an
irreducible set of words the meaning of which is not, in fact cannot be, explained in
this way. By contrast, the lexical meanings we have in our mental lexicons are not just
42
Understanding semantics
paraphrases. They are concepts. Whatever these are (this is a question for cognitive
psychology), they are not words.
To a lexical item, there is more than its form and its meaning. Above all, lexemes
are linguistic units within the language system; they can be built into phrases and
sentences according to the grammatical rules of the language. They are assigned to
1
different grammatical categories which differ in their grammatical behaviour. For
example, a noun can be combined with an adjective, and the whole with an article to
form an NP. The NP in turn can be combined as a direct object with a verb to form a
VP, and so on (cf. the discussion of the example in 1.2.3).
In English, many expressions can be used as members of more than one category:
for instance, light is used as a transitive verb (light the candle), a noun (a bright light)
and an adjective (a light colour); walk may be a noun (take a walk), an intransitive
verb (walk in the park) or a transitive verb (walk the dog); too may be a particle
(›also‹) or an adverb (too much), and so on. Although these expressions may be very
similar in meaning, they are considered different lexemes. There are (at least) three
different lexemes light, three lexemes walk and two lexemes too. In general the same
word in different grammatical categories constitutes as many different lexemes.
The members of certain grammatical categories in a language may exhibit
inherent grammatical properties, such as gender in languages like German,
Russian or Latin (see 4.1.1 on gender). The gender of a noun is not a grammatical
form it may take freely: rather, it is an inherent property of the lexeme that may also
determine the form of the article and preceding adjectives (cf. German masculine
der Computer, feminine die Maus (Engl. mouse), neuter das Motherboard). In Bantu
languages such as Swahili, nouns belong to one of five to ten noun classes, e.g. terms
for persons, for animals, for plants or for artefacts. Russian and the Slavic languages
have inherent aspect for verbs, distinguishing between perfective and imperfective
verbs (see 6.3 on aspect).
The grammatical category determines the range of grammatical forms a lexeme
can take. Some categories of lexemes have just one form, e.g. adverbs (here, then),
particles (already, too, only) or prepositions (on, after, without), while the forms of
expressions belonging to other categories may vary. English nouns have a singular
and a plural form (child, children) as well as genitive forms of the singular and the
plural (child’s, children’s). Adjectives have a positive, comparative and superlative
form (light, lighter, lightest or precise, more precise, most precise) and an adverbial
form (lightly, precisely). Verbs exhibit a fairly wide variety of forms which mark
(among other things) grammatical person (sings vs sing), tense (sings vs sang)
and aspect (sings vs is singing). Composite forms such as have been singing are also
considered forms of the main verb sing. For each grammatical category, there are
morphological rules for building the forms in the regular cases. But certain lexemes
may be exceptional. Irregular verbs have special past tense and past participle forms
(sang, sung instead of singed); some adjectives have special comparative, superlative
and/or adverbial forms (better, best, well instead of gooder, goodest, goodly). A few
1 Another term for grammatical categories is parts of speech. We will occasionally use the more
informal term word class.
Ambiguity
43
nouns have non-standard plural forms (child–children, mouse–mice, foot–feet, leaf–
leaves, sheep–sheep). Exceptional forms are also part of the definition of a lexeme.
If a lexeme has different forms, one of them will be used as its citation form or
dictionary form, i.e. the form in which it will be listed in a dictionary or cited when
it is spoken about. Usually it is the simplest form of the lexeme. For example, the nongenitive singular form of a noun is used as its citation form.
Each grammatical form of a lexeme has a spoken form and an orthographic form
(if there is a written standard for the language). Let us use the terms sound form
and spelling, respectively. The sound form of the three grammatical forms kids, kid’s
and kids’ is the same, while their respective spellings differ.
To sum up, a lexeme is a linguistic item defined by the following specifications,
which constitute what is called the lexical entry for this item:
DEFINITION 1 Lexeme
A lexeme is defined by the following constitutive properties:
∑ its sound form and its spelling (for languages with a written standard)
∑ the grammatical category of the lexeme (noun, intransitive verb, adjective,
etc.)
∑ its inherent grammatical properties (for some languages, e.g. gender)
∑ the set of grammatical forms it may take, in particular irregular forms
∑ its lexical meaning
These specifications apply to both simple and composite lexemes. Composite
lexemes too, such as throw in the towel or red light, have a fixed sound form, spelling
and lexical meaning. They belong to a grammatical category (intransitive verb for
throw in the towel and noun for red light), they have inherent grammatical properties
and the usual range of grammatical forms. For example, the grammatical forms of
throw in the towel are those obtained by inserting the grammatical forms of the verb
throw: throws in the towel, threw in the towel etc.
In principle, each of the specifications of a lexeme is essential: if two lexical items
differ in one of these aspects, they are considered different lexemes. There are,
however, exceptions. Some lexemes have orthographic variants, e.g. rhyme/rime,
others may have different sound forms, e.g. laboratory may be stressed on either the
first or the second syllable. The American and the British varieties of English differ
in pronunciation and spelling for many lexemes. As long as all the other properties
of two orthographic or phonetic variants of a lexeme are identical, they will not be
considered different lexemes but lexemes with a certain limited degree of variation.
As far as the component of lexical meaning is concerned, there may be variation of
different extent. There are cases of the same form being associated with completely
unrelated meanings, e.g. bank in the sense of a financial institution and in the sense
given in river bank. These are considered cases of two lexemes which happen to have
the same form. The phenomenon is called homonymy. Sometimes these differences
in meaning come with minor differences in form; for example, the German word
44
Understanding semantics
Bank, too, has two unrelated meanings, namely the financial institution meaning
and the meaning ›bench‹; the plural of the first is Banken, of the second Bänke.
Homonymy contrasts with cases where the different meanings of a form are clearly
related; for instance, body may denote the whole physical structure of a human being
or an animal, or just the trunk; it may denote a corpse, or a group of people working
or acting as a unit (like a physical body). These are considered cases of one lexeme
with different variants of meaning, i.e. as cases with ‘minor’ variation in the meaning
part of the lexeme. This phenomenon is called polysemy.
To sum up, the definition of a lexeme will tolerate, and incorporate, minor variation
in some of its components, while substantial differences for one or more aspects will
give rise to the distinction of different lexemes.
3.2 LEXICAL AMBIGUITY
3.2.1 Homonymy
The adjective light can be used with two meanings. Let us talk of lightA1 if the
adjective is taken as the opposite of dark, and of lightA2 if it is the opposite of
heavy. LightA1 and lightA2 have not always had the same form. LightA1 derives from
a historical source which in German developed into the present-day adjective licht
(meaning, in one of its meaning variants, approximately the same as lightA1). Words
with the same historical origin are called cognates. LightA2 is a cognate of a different
German word, the adjective leicht (›light, easy‹). Due to their different origins, lightA1
and lightA2 are considered two different words by most linguists. In general, different
meanings are assigned to different lexemes if they have different historical sources.
The idea is that, as long as their meanings remain distinct, different words do not
develop into one, even if their sound forms and/or spellings happen to coincide for
2
independent reasons. In addition to lightA1 and lightA2 there is a noun lightN which
is related to lightA1 and means the kind of visible radiation as well as certain sorts of
objects that emit light. A verb lightV is also related to lightN and lightA1.
The two adjectives lightA1 and lightA2 are an instance of what is called total
homonymy: two lexemes share all distinctive properties (grammatical category and
grammatical properties, the set of grammatical forms, sound form and spelling) yet
have unrelated meanings. One would talk of partial homonymy if two lexemes with
unrelated meanings coincide in some but not all of their grammatical forms, e.g. the
verbs lie1 (lay, lain) and lie2 (lied, lied). Partial homonyms can give rise to ambiguity
in some contexts (don’t lie in bed!) but can be distinguished in others (he lay/lied in
bed).
2 It has often been questioned if this historical criterion is really relevant. Average speakers do not
know the origin of the words they use. All that matters for them is some ‘feeling’ as to whether
or not the two meanings have anything to do with each other. For instance, to many speakers of
English the words ear for the body part and in an ear of corn appear to be the same word, although
historically they are of different origins.
Ambiguity
45
DEFINITION 2 Homonymy
Two lexemes are totally homonymous if they have unrelated meanings, but
share all other constitutive properties.
Two lexemes are partially homonymous if they have unrelated meanings, but
coincide in some of their grammatical forms.
Homonymy can be related either to the sound forms of the lexemes or to their
spellings: homonymy with respect to the written form is homography; if two
lexemes with unrelated meanings have the same sound form, they constitute a
case of homophony. The nouns bow1 [b] (cf. bow and arrow; German cognate
Bogen), bow2 [ba] (›front of a ship‹, German cognate Bug), bow3 [ba] (›bending‹;
German cognate Ver-beug-ung) are all homographs, but only bow2 and bow3 are also
homophones. Examples for words that are total homophones but not homographs
would be the noun pairs tail/tale, story/storey or cue/queue. Partial homophones are
numerous: threw/through, write/right, there/their, whole/hole, to/two/too and so on.
In Japanese, homophones are very frequent, while homographs are rare: most
homophones are disambiguated by using different characters. For example, the sound
form seichō [seio:] has more than ten unrelated meanings, each written differently
when written with Chinese characters; among them there are ᡂ㛗 ›growth‹, ᨻᗇ
›government office‹, Ύ⃈ ›clear, serene‹, ኌㄪ ›tone (of voice), style‹, ᡂ㫽 ›adult
bird‹, ᩚ⭠ ›medicine for internal disorders‹, 㟼⫈ ›listening quietly‹ and ṇᮅ
3
›legitimate dynasty‹. Each word is written with two Chinese characters, the first of
which is pronounced sei, and the second chō. If you look at them, each sei character
and each chō character is different. Since Chinese characters carry meaning, each of
the eight two-character combinations has a different meaning. The high frequency in
Japanese of homophones written with different characters is one reason why writing
in Chinese characters is considered a necessity by most Japanese. One might think
that the high number of homophones would cause problems in oral communication.
But if you look at the different meanings of seichō (and similar cases), you will see
that they are so different that they are very unlikely to occur in the same context.
3.2.2 Polysemy
While homonymy is a rare and accidental phenomenon, polysemy is abundant. It is
rather the rule than the exception.
DEFINITION 3 Polysemy
A lexeme is polysemous if it has two or more interrelated meanings.
3 Writings and translations are taken from http://jisho.org/, 15 June 2012.
46
Understanding semantics
4
We will refer to the interrelated meanings as meaning variants. Each of these
meaning variants has to be learnt separately in order to be understood. The
phenomenon of polysemy results from a natural economic tendency of language.
Rather than inventing new expressions for new objects, activities, experiences, etc.
to be denoted, language communities usually opt for applying existing terms to new
objects, terms hitherto used for similar or related things. Scientific terminology is
one source contributing to polysemy on a greater scale. Some scientific terms are
newly coined, but most of them will be derived from ordinary language use. Among
the terms introduced here, lexeme, homonymy, polysemy are original scientific terms,
while others, such as meaning, reference or extension, are ordinary expressions for
which an additional technical meaning variant was introduced. There are, of course,
lexemes which are not polysemous, e.g. the terms for the months or for the days of
the week; these are called monosemous and the phenomenon, monosemy.
As an example of polysemy let us consider the noun light: it means a certain sort
of visible radiation, but also electric lamps, traffic lights or illuminated areas (cf.
light and shadow). Clearly, these meanings are interrelated. Likewise, the different
meanings of lightA2 which correspond to the opposites heavy and difficult are
somehow interrelated although the relation is harder to define. Note that heavy itself,
and with it its opposite light, is again polysemous (cf. a heavy stone, heavy rain, a
heavy meal).
3.2.3 The relationship between homonymy and polysemy
It is only from one particular perspective that homonymy and polysemy present
an alternative: namely, when we ask whether a particular form that is used with
more than one meaning is to be assigned to one or to more lexemes. Apart from this
question, polysemy and homonymy are independent phenomena; two homonymous
lexemes may well at the same time be themselves polysemous. For example, lightA1
and lightA2 are homonymous, and both homonyms are in themselves polysemous.
Different meaning variants of lightA2 are involved in combinations like a light bike, a
light meal, a light exercise or a light breeze. LightA1 has two meaning variants, too, the
opposite of dark and something like ›pale‹ in expressions such as light blue.
The relationship between polysemy and homonymy is illustrated in Fig. 3.1. The
difference is this: in the case of homonymy, multiple meanings are assigned to as
many lexemes; in the case of polysemy, to the same one. Homonymous lexemes
may themselves be polysemous. The fact that the multiple meanings (i.e. meaning
variants) of polysemous expressions are related is indicated by depicting them as
clusters.
3.2.4 Vagueness
There are no gradual transitions between the meaning variants of a polysemous
expression. The meaning variants are distinct. Polysemy therefore is not to be
4 The different meanings of homonyms or of polysemous lexemes are often referred to as ‘senses’.
The term ‘meaning variant’ chosen here is restricted to cases of polysemy.
Ambiguity
47
Figure 3.1
Homonymy and polysemy
Homonymy
lexeme1
Polysemy
lexeme2
lexeme
is realized as
is realized as
sound form/
spelling
sound form/
spelling
means
meaning1
(variants)
means
meaning2
(variants)
means
meaning
variants
confused with flexibility of use. For very many lexemes, their proper application to a
given case is a matter of degree. For example, whether or not we will refer to a child as
a ‘baby’ depends on criteria such as the age of the child and its developmental stage.
Both criteria are gradual. What one person considers a baby need not be considered
so by another person. As a consequence, the denotation (2.2.2) of the word baby
has flexible boundaries. This does not mean that the word baby has infinitely many
meanings that differ in how the borderline is fixed between babies and ex-babies, as
it were. Rather the concept ›baby‹ is in itself vague: it allows for adaptation to the
given CoU.
DEFINITION 4 Vagueness
A lexical meaning is vague if it allows for flexible adaptation to the given CoU.
Vagueness can be observed with all concepts that depend on properties varying on a
continuous scale. Colour terms like red have a vague meaning, because we conceive
the range of colours as a continuum with fuzzy transitions. Whether something is
‘big’ or not, or ‘good’ or not is a matter of degree. In general, all gradable adjectives
(i.e. adjectives with a comparative and superlative form) are vague.
Widespread vagueness in the lexicon should be considered another economic
trait of language. For example, with the pair tall/short, language provides us with
a rough distinction on the scale of body height. This is much more efficient for
everyday communicative purposes than expressions with a more precise meaning,
48
Understanding semantics
say ›between 6 and 7 feet‹. The issue of vagueness and the important role it plays will
be taken up again in 11.5.
Vagueness may occur in combination with polysemy. For example, the meaning
variants of lightA2 are a matter of different underlying scales (of weight, difficulty,
etc.). These scales can be distinguished quite clearly. But for each scale, the meaning
of light describes just a low degree on this scale, whence each meaning variant in
itself is vague.
3.3 COMPOSITIONAL AMBIGUITY
It is not only lexemes that may be ambiguous at the level of expression meaning, but
also complex expressions, in particular sentences. As was stated in 1.2, the meaning
of a sentence is derived in the process of composition and is thereby determined by
its lexical components, their grammatical meaning and the syntactic structure of
the sentence. Each one of these three ingredients can give rise to ambiguity of the
sentence. If a sentence contains an ambiguous lexeme, the process of composition
will yield as many meanings of the sentence as the ambiguous item has, for example
two readings (at least) in the case of sentence (1), since window may also mean a
window on a computer screen. If the sentence contains more than one ambiguous
lexical item, the meanings will multiply.
(1)
Open another window.
As we will see below, not all these meanings will reach the level of utterance meaning.
But strictly speaking all these are possible expression meanings. Since this type of
ambiguity concerns compositional meaning, I would like to call it ‘compositional
ambiguity’.
Independently of lexical ambiguities, the syntactic structure of a sentence may be
ambiguous. Consider the following examples.
(2)
a.
She met the man with her friend.
b.
Flying planes can be dangerous. (Chomsky)
c.
John and Mary are married.
In (2a) the PP with her friend can be related to the verb meet (meaning roughly the
same as she and her friend met the man), or it can be taken as an attribute of the NP
the man (›the man who was with her friend‹). In (2b), the phrase flying planes can
be read as ›flying in planes‹ and as ›planes that are flying‹. (2c) can mean that John
and Mary are married to each other or that they are both married, each to a different
spouse. Such sentences are syntactically ambiguous. Their structure can be analysed
in different ways, and consequently the composition of sentence meaning can follow
different paths. Syntactic ambiguity usually results in semantic ambiguity, i.e. in
different meanings.
Ambiguity
49
Grammatical forms, too, may be ambiguous. For example, the English past tense
forms of verbs have an indicative past tense reading referring to past events along
with a conditional reading with present or future time reference. The form were has
the past tense meaning in (3a) and the conditional reading in (3b):
(3)
a.
I knew you were here.
b.
I wish you were here.
3.4 CONTEXTUAL AMBIGUITY
3.4.1 Interpretation in context
The process of composition yields one or more compositional meanings of the
sentence. When it comes to interpreting words and sentences in their context, i.e.
when one proceeds from the level of expression meaning to the level of utterance
meaning (1.1.2), the expression meanings of words and sentences may be modified.
A sentence actually uttered in a CoU must fulfil certain requirements in order to
qualify as a reasonable message. First, as a minimal requirement, it must not be selfcontradictory, i.e. false in all possible CoUs, because in this case, it cannot be applied
5
to any concrete situation whatsoever. Second, it must in some way be relevant in the
given CoU. These three conditions will be embodied in the Principle of Consistent
Interpretation to be introduced in 3.4.4. Utterance meanings of a word or a sentence
that pass these conditions are called possible readings.
Due to these additional constraints, the set of compositional expression meanings
of the sentence may undergo considerable changes. Three things can happen to a
particular compositional expression meaning:
(a) The expression meaning may be taken over as it is and enriched with
contextual information, e.g. by assigning it a concrete referent.
(b) The expression meaning may be refuted and eliminated if it is contradictory
or does not fit the CoU.
(c) The expression meaning may be modified by some kind of meaning shift in
order to fit the CoU, and subsequently enriched with contextual information.
As we will see right now, this not only holds for sentence meanings but also for
the expression meanings of all components of the sentence, i.e. lexical meanings,
grammatical meanings and the compositional meanings of complex components of
the sentence.
Option (b) may lead to a disambiguation of the sentence at utterance level, i.e. to
a reduction of the number of possible readings. Polysemous lexemes are often such
that their respective meaning variants occur in different types of contexts. As a result,
mechanism (b) allows us to keep the number of lexical forms lower by packing them
5 In 7.2 the term self-contradictory will be replaced by the technical term logically false.
50
Understanding semantics
with clusters of meaning variants that will normally be reduced to a single one when
used in a specific context.
The meaning shifts involved in option (c) create new expression meanings and,
out of them, utterance meanings. For example, when the sentence I don’t need your
bicycle in 1.1.2 was interpreted in the second scenario, bicycle was taken to refer to a
playing card with a picture of a bicycle. This interpretation rests on a meaning shift
of the word bicycle, by which its lexical meaning is replaced with a closely related
new expression meaning. Thus for lexical items, the application of meaning shifts
is another source of ambiguity. The resulting additional meanings only occur in
contexts that trigger the respective meaning shifts. Therefore, this type of ambiguity
shall be dubbed ‘contextual ambiguity’. Meaning shifts follow general patterns which
belong to the semantic repertory of the language users. Three basic patterns will be
described in the next subsection. Meaning shifts constitute another mechanism that
keeps our lexicon economical. Thanks to them, lexemes can actually be used in many
more senses than need to be stored in the lexicon.
In actual communication, however, there may be instances in which the
compositional meaning of a sentence does not fit the context and does not lend itself
to sense modification in order to make it fit. If we assume that our interlocutors are
not talking nonsense, there is a fourth possibility:
(d) The context is revised or enriched as to make the meaning fit into it.
We will return to this issue in 4.7.5, under the label of ‘accommodation’.
3.4.2 Disambiguation
The following sentences contain the ambiguous lexeme letter (›alphabetic character‹
vs ›written message‹):
(4)
a.
Johnny wrote a letter.
b.
Johnny wrote a letter to Patty.
c.
Gamma is the third letter of the Greek alphabet.
(4a) has two readings because the rest of the sentence, Johnny wrote a ____, allows
for both meanings of letter. (4b), however, has only one reading. The addition to
Patty requires the message meaning of the word. Likewise in (4c), the rest of the
sentence would not make sense unless letter is understood as ›character‹. Thus
the immediate sentential environment of a word may call for particular meaning
variants and eliminate others. Along with the meanings of the parts concerned, the
respective compositional meanings of the sentence will be eliminated. A sentence
may also have no possible sensible reading at all if its parts do not fit together. For
example, (5) below is self-contradictory and therefore disqualified at utterance
level. Due to its lexical meaning, the verb shiver requires a subject referent that is
animate and has a body. But the lexical meaning of age does not allow for this sort
of referent.
Ambiguity
(5)
51
Johnny’s age shivered.
Such sentences will hardly be used (except as examples for self-contradictory
sentences, or maybe in poetry). But they serve to illustrate an important point:
observing the rules of grammar and semantic composition does not guarantee that
the result ‘makes sense’.
3.4.3 Meaning shifts
3.4.3.1 Metonymical shift
The following example is borrowed from Bierwisch (1982):
(6)
James Joyce is difficult to understand.
The sentence has at least four readings. If you relate the sentence to James Joyce the
writer, you may first of all take it as meaning that (i) the writings of James Joyce are
difficult to understand. But if you imagine a context where the author is still alive, the
sentence might also mean that (ii) the way he talks, (iii) the way he expresses himself
or (iv) the way he acts is difficult to understand. In the first reading, the name James
Joyce refers to Joyce’s work. In the other readings, it refers to the writer himself. Yet the
proper name James Joyce is not polysemous: we do not have to learn about the lexical
meaning of this particular name to know that these interpretations are possible. In
principle, all names of persons can be used for referring to their published work
(if there is any). The interpretation in context is due to a semantic shift generally
available for all names of people.
Similar shifts are very common. Consider the readings of university in the
following examples:
(7)
a.
The university lies in the eastern part of the town.
b.
The university has closed down the faculty of agriculture.
c.
The university starts again on April 15.
The subject the university refers to the campus in (7a), to the institutional body in
(7b), and to the courses at the university in (7c). Again, this is not a case of polysemy.
The word university lends itself naturally to the meaning shifts that create these
readings. We do not have to know each of them. Many other words with similar
meanings exhibit the same kind of variation: school, theatre, opera, parliament and
so on. The same kind of variation is also paralleled in other languages. This kind of
variation is not rooted in lexical ambiguity. Its source is more general.
If we take a closer look at the meaning shifts involved, we see that in each case the
term the university refers to something that somehow belongs to a university. Let
us assume that the word lexically denotes a certain kind of educational institution.
Such an institution (unless it is a virtual university on the web) must be located
somewhere, it must have an administration, and it must offer courses. It is in this
52
Understanding semantics
sense that its premises, its committees and administration, and the courses offered
belong to the university. Apparently, a term that denotes objects of a certain kind can
also be used to refer to certain things that usually belong to such objects. The term,
then, is felt to ‘stand for’ those things which belong to its referents proper: in (7) the
university in this sense ‘stands for’ the campus, its administration and the courses. In
(6) James Joyce stands for his work. This use of terms is called metonymy: a term
that primarily refers to objects of a certain kind is used to refer instead to things that
belong to objects of this kind.
DEFINITION 5 Metonymy
An expression is used metonymically if it is used to refer to things that belong
to the kind of objects to which the expression refers in its literal meaning.
The corresponding type of meaning shift will be referred to as metonymical shift.
The crucial condition of ‘belonging to an object of this kind’ can be made more
precise if we use the notion of a concept. The word university is linked to a concept
for universities as its lexical meaning. The concept specifies that a university is an
educational institution with premises, teaching, teachers, students, an administration
and so on. A metonymical shift shifts the reference of the word from a standard
referent, a university, to an essential element of the underlying concept. We will get a
clearer picture of metonymy in the chapter on frames (12.3.1).
3.4.3.2 Metaphorical shift
The four sentences in (8) are the opening lines of an article in an American news
magazine (Newsweek, 19 October, 1998, p. 30):
(8)
a.
They were China’s cowboys.
b.
The swaggering, fast-talking dealmakers threw around grand projects and
big figures as if the money would never stop flowing.
c.
Then the sheriff came to town.
d.
Last week Beijing said it was shutting down one of the flashiest investment
institutions, [name of the institution].
The sentences are about Chinese investment institutions and they are full of
metaphorical language. Although there is a literal reading for the first sentence, it will
not be taken in that sense. Rather the next sentence tells us that (8a) refers to certain
‘dealmakers’. We will therefore take the expression China’s cowboys in a metaphorical
sense: the persons referred are not claimed to be cowboys, but to be in some way
like cowboys. In this case, according to (8b), they resemble cowboys in that they
are swaggering, fast-talking and throwing things around. The metaphor is further
developed in (8c) with the appearance of the sheriff, another typical ingredient of a
Ambiguity
53
Wild West setting. Sentence (8d) explains who the ‘sheriff ’ is: Beijing (the name of the
Chinese capital metonymically stands for the Chinese government). This sentence
takes us back from the metaphor to literal interpretations.
Let us define more explicitly what a metaphor is: concepts, notions, models,
pictures for things from one domain, the source domain, are borrowed for the
description of things in another domain, the target domain.
DEFINITION 6 Metaphor
An expression is used metaphorically if it is used to refer to things that are in
crucial aspects similar to the kind of objects to which the expression refers in
its literal meaning.
In (8) the source domain is the Wild West and the target domain is the international
investment scene of China at the time when the article was published. To the majority
of the magazine’s readers, the source domain is better known than the target
domain. Hence, concepts taken from the Wild West domain may help to describe to
this particular readership what’s going on in China. (A Wild West metaphor would
probably be of less help to Chinese readers.) Every metaphor is the construction
of a parallel: the dealmakers are likened to cowboys in certain respects, mainly
their public behaviour, and the Chinese government takes the role of the sheriff in
exerting its authority. In general, metaphorical language can be characterized as
talking about things in the target domain in terms of corresponding things in the
source domain.
A metaphor yields a new concept in the target domain, a concept that is similar
to the original concept of the source domain in that it contains certain elements,
although not all, of the source concept. Metonymy is quite different from metaphor.
When we talk metonymically, we remain within the same domain. We borrow an
element from the original concept, but the links to the other elements remain.
University in the ›campus‹ meaning remains immediately related to university in its
›institution‹ meaning, James Joyce’s work remains related to the person James Joyce.
The relations between the general objects and the things, or aspects, belonging to it
are only possible within one domain.
3.4.3.3 Differentiation
The James Joyce example (6) is relevant in one more respect. The four readings
mentioned differ in the way the verb understand is interpreted in context: it may
relate to the author’s work, to his articulation, his way of expressing himself and the
way he behaves, respectively. Likewise, in the first reading of the sentence, relating to
his work, the verb will be taken in one particular meaning. It is reasonable to assume
that the verb understand in all these cases just means ›understand‹. If we attributed
the different readings of understand to polysemy, we would end up in countless
distinctions of lexical meaning variants of the majority of words. Note, for example,
that understanding a sentence may relate to its articulation when uttered, its syntactic
54
Understanding semantics
structure, its descriptive meaning or its utterance meaning. The different readings can
be better explained if one assumes that to understand means to understand someone
or something in a certain respect that is determined by the context.
The following example (taken from Bierwisch 1982: 11) can be explained in the
same way:
(9)
a.
John lost his friend in the overcrowded subway station.
b.
John lost his friend in a tragic car accident.
c.
John lost his friend, as he could never suppress bad jokes about him.
The common part John lost his friend has three different readings due to the
respective sentence context. In (9a) lose means a loss of contact, in (9b) John’s friend
stops being his friend because the friend no longer exists, and in (9c) the friend is
supposed to live on but stops entertaining a friendly relationship with John. In each
case, the verb lose can be taken to mean something like ›stop having, due to some
event‹. What the context contributes to this is the meaning in which the ›having‹
component is interpreted and the kind of event that causes the loss.
The examples illustrate a third common kind of meaning shift. Bierwisch calls it
conceptual differentiation. In this book, the simple term differentiation is preferred.
DEFINITION 7 Differentiation
Differentiation adds content to a given concept.
It can be defined in general as a meaning shift which results in a special case of what
the expression denotes in its lexical meaning.
There are several more types of meaning shifts, but we will not go further into the
matter.
Table 3.1
Kinds of meaning shifts
Lexical meaning
Metonymy
the university starts in April
›educational
institution‹
Metaphor
→
›courses at the
university‹
they were China’s cowboys
›man who herds
cattle‹
Differentiation
Shifted meaning
→
›person behaving
like a cowboy‹
James Joyce is hard to understand
›perceive the
meaning‹
→
›interpret the text
meaning‹
building a new concept out of an
element of the original concept
building a new concept in the target
domain by borrowing parts of the
concept in the source domain
adding conditions to the original
concept
Ambiguity
55
3.4.4 The Principle of Consistent Interpretation
The driving force of the meaning modifications due to interpretation in context is a
very basic principle:
DEFINITION 8 Principle of Consistent Interpretation
At the level of utterance meaning, a composite expression is always interpreted
in such a way that its parts fit together and that the whole fits the context.
This principle, if appropriately generalized, probably governs all interpretation
whatsoever, because interpretation usually concerns a complex input and is always
interpretation in some relevant context. As we have seen, its application to sentence
interpretation at utterance level may lead to elimination of compositional meanings as
well as to the creation of new ones. The principle generally rules out self-contradictory
readings: they are always due to parts within a sentence that do not fit together (recall
(5)). It also rules out irrelevant readings: these do not fit the wider context.
The notion of ‘context’ here is not identical with the notion of CoU. Parts of a
composite expression, e.g. a sentence, are embedded into the whole expression, and
the whole is embedded into the CoU as well as into larger, more general contexts.
In order to see what is involved, let us return to one of the examples above. The
immediate context of a lexical item is primarily the syntactic phrase it is a part of.
For example, the whole NP a letter to Patty is the phrase context of the noun letter
in (4b). Since the PP to Patty is attached to the noun, it must denote something that
can be directed to somebody; this is the reason why the reading ›written character‹
of letter is rejected. Every further syntactic combination of a sentence component
containing a word embeds it into an additional context level. In (4b), a letter to Patty
is combined with the verb wrote to form the VP of the sentence; since letters in the
sense of written messages are things that are written, the message interpretation of
the NP is accepted in the context of the VP wrote a letter to Patty, and along with it
the respective meaning of letter. When the VP is combined with the subject of the
sentence, in this case Johnny, the reader checks whether this meaning of the VP fits
the expression meaning of the subject. This is a matter of the logical requirements
of the verb write for its subject referent. These requirements are fulfilled in the given
context if the name Johnny refers to a person (rather than a pet or a teddy bear).
In a wider context, the interpretation of the sentence may be modified further. For
example, Johnny might be a four-year-old boy not yet able really to write. He might
be playing that he is writing a letter to his friend Patty. In such a context, the sentence
might be used to describe the situation ‘in quotes’, as it were.
In general, when an expression is built into a sentence, it is combined with other
expressions in accordance with certain syntactic rules, and the result may be
combined with other parts of the sentence until the sentence is complete. Each step
of syntactic combination embeds the original expression in a new, and wider, context,
and each context may impose new requirements on the meaning of the expression.
56
Understanding semantics
In addition, even grammatical forms require certain meanings of the expressions
that carry them, and consequently they may trigger meaning shifts, too. For example,
the English progressive form ‘be + V-ing’ requires a verb that denotes some process
or activity that is going on; it cannot be used with verbs that denote just a state.
Therefore, if the progressive form is used with state expressions such as be polite
(for example in John is being polite), the lexical meaning of the expression is shifted
to something like ›act in a polite way‹, which does not describe a state but some sort
of active behaviour. The context which a sentence provides for an expression can
be called its sentential context. It is the sentential context of the subject in (5) that
eliminates any available meanings and the sentential context of the word letter in
(4b) and (4c) that leads to the elimination of one or the other meaning of the word.
In other cases, the sentential context triggers meaning shifts in order to make all
parts of the sentence fit together. In (7a, b, c) the metonymical shifts of the subject
NP the university are due to the logical requirements of VPs (discussed as ‘selectional
restrictions’ in 4.7.2.2 and 5.7). The metaphorical shifts in (8) come about in different
ways. In (8a) it is the combination with China’s which leads to a reinterpretation of
cowboys (because, supposedly, there are no real cowboys in China); in (8b) threw
around grand projects requires a metaphorical interpretation of threw around in
order to come up with new selectional restrictions that fit the kind of object (grand
projects) given here; the analogue holds for the combination of money and flowing.
In (9a–c) it is obviously the sentential context which leads to the differentiated
interpretations of the VP lost his friend.
Finally, the context of the whole sentence, the context of utterance and the wider
context in which it is embedded, influence the interpretation of the sentence at
utterance level. For example, the actual CoU of the James Joyce sentences in (6)
will determine whether we choose one interpretation or the other. In (8c) the
whole sentence Then the sheriff came to town is taken metaphorically to fit the
extrasentential context; without doing so the context would not provide a referent for
the definite NPs the sheriff and town.
In 1.2 semantic composition was characterized as a bottom-up process, in which
the expression meaning of the whole is derived step by step from the expression
meanings of its elements (lexical items, grammatical forms and syntactic structure).
In other words, the output of the process, the sentence meaning(s), is determined by
the input. When a sentence is interpreted in context, i.e. when its possible utterance
readings are determined, meaning shifts and meaning eliminations interfere with
the process of composition. This interference constitutes a top-down element of
the interpretation: the input may be re-interpreted in terms of appropriate outputs.
Thus, to a certain extent, the output determines the input – at the level of utterance
meaning. It is important to see that the meaning modifications at the interface
of expression meaning and utterance meaning do not contradict the Principle of
Compositionality. Rather, the modifications of meaning that are triggered top-down
by the context guarantee that the bottom-up process of composition works even if
special output is required from the process. When the required shifts of the input
expression (!) meanings are arranged, the process of composition produces its
output in a perfectly regular way. This shows that the rules of composition are not
Ambiguity
57
simply sacrificed if they produce an unwelcome result, but, on the contrary, they are
preserved even at the cost of sacrificing (to a certain extent) the expression meanings
of their input.
3.5 MEANING SHIFTS AND POLYSEMY
3.5.1 Contextual vs lexical ambiguity
The fact that almost all lexemes can undergo meaning shifts in certain contexts
can be considered a further dimension of their abundant ambiguity. However, this
type of ambiguity is fundamentally different from polysemy. In principle, polysemy
is a matter of single lexemes in single languages. Consider the colour adjectives in
English. Many of them are polysemous, with meaning variants not primarily relating
to colour properties. For instance, green may mean ›unripe‹. This is motivated by the
fact that green colour of many fruits indicates that they are not yet ripe. From this,
in turn, derives the meaning variant ›immature‹ due to a metaphor that establishes a
parallel between the development of personality and the process of ripening of fruits.
This meaning variation is an accidental matter of English green. Due to the same
motivations, it might, but need not, occur in other languages provided they have a
word for the colour green. But there is no parallel for exactly this kind of variation
in the case of the other colour words. Although the colour of very many fruits is red
when they are ripe, red cannot mean ›ripe‹ or ›mature‹. Likewise in German, blau
(›blue‹) also means ›drunk‹, but English blue does not; nor does any other colour
adjective in German or English mean ›sober‹. Sometimes, words given as translation
equivalents in different languages may have parallel meaning variants, but usually
their variation will not match.
All this does not hold for interpretations resulting from meaning shifts triggered
by observing the Principle of Consistent Interpretation:
∑ Interpretations that result from meaning shifts triggered by the context need not
be written into the lexicon.
For example, we will not assume that (8c) provides evidence that the lexical meaning
of sheriff exhibits a variant ›Chinese government‹; likewise uses of bicycle such as
in Scenario 2 in 1.1.2 do not force us to assume that bicycle can mean ›playing card
that carries a picture of a bicycle‹. Consequently, we cannot resort to our stock of
lexical meanings in order to arrive at the interpretations that result from metaphor,
metonymy, differentiation or other types of meaning shifts. These shifts can only be
successfully employed in communication if they constitute general patterns available
to all language users.
∑ Contextual meaning shifts are based on general patterns.
These general patterns can be uniformly applied to very many cases; they are
available for all lexical meanings that fulfil the input conditions for these conceptual
operations. The respective types of shifts belong to the interface between expression
58
Understanding semantics
meaning and utterance meaning (recall 1.3.1). They form part of our general
linguistic competence.
∑ The patterns of meaning shifts can be observed systematically with open classes
of lexemes.
The patterns are not even language-specific; the examples discussed in 3.4 can be
literally translated into other languages without getting stripped of the meaning
shifts they undergo.
∑ The same patterns of meaning shift can be observed to operate cross-linguistically.
Psychologists and cognitive linguists consider conceptual operations like metonymy,
metaphor and differentiation to constitute not just semantic faculties, but more
generally fundamental mechanisms of concept formation. This would explain the
fact that the same types of shifts can be observed cross-linguistically.
The phenomenon of systematic meaning variation is currently dealt with in
various frameworks from different theoretical perspectives. You will find it discussed
under labels such as ‘systematic polysemy’, ‘principled polysemy’, ‘facets of meaning’
and ‘microsenses’, as well as ‘coercion’.
3.5.2 Polysemy and meaning shifts
The types of meaning shifts mentioned correspond to types of meaning relations.
For example, the relationship of the shifted meanings of university to the lexical
(‘institution’) meaning of the word is metonymical and the reading ›lose by death‹ of
lose is a differentiation of ›lose‹ in the general sense. The same kind of relationships
exist between the meaning variants of polysemous lexemes: in very many cases
of polysemy, meaning variants are interrelated by way of metonymy, metaphor or
differentiation.
As to metonymy, recall the case of green with its secondary meaning ›unripe‹. Here
green colour is taken as metonymically standing for a certain stage of biological
development. Other cases of lexicalized metonymy are the following:
(10) a.
The asshole did not even apologize. part for the whole, ‘pars pro toto’
b.
He talked to celebrities.
property for a person with the
property
c.
His last date was in a bad temper.
event for person involved
d.
I wrote a paper.
carrier for content
e.
the Green Berets
clothing for wearer
Parts of the body (10a) belong to a person or animal. A property belongs to the
one who has the property (10b): celebrities are persons who have the property of
celebrity. To a date belongs the person one dates (10c). To a piece of paper with
Ambiguity
59
something written on it belongs what is written on it (10d). Pieces of clothing have
their wearers; they are associated with the people wearing them (10e).
The metaphorical use of the verb flow for money that we encountered in (8b) is so
common that it can be considered a lexicalized meaning variant, in addition to the
literal meaning of the flow of liquids. Other examples of lexemes with metaphorical
meaning variants are terms like mouse for a computer mouse, lightA2 in a light meal,
or the majority of idiomatic expressions like throw in the towel, kick the bucket or
make a mountain out of a molehill. Most proverbs are metaphorical, e.g. Birds of a
feather flock together or A rolling stone gathers no moss.
Differentiation too is a common source of polysemy: a lexeme may have a meaning
variant that applies to a special case of what the basic meaning of the lexeme
applies to. Car denotes some sort of vehicle in general, but nowadays preferably an
automobile; glass may mean a kind of material, but one meaning variant is ›glass‹
in the meaning of a certain kind of drinking container made of that material (as in
beer glass).
Although polysemy is a matter of single lexemes in single languages, we see that
the variation of meaning follows general principles. Apparently, we command an
intuition that allows us to recognize that the meaning variants of a polysemous
expression are interrelated. This is evidence for the existence of general patterns in
variation, and these are the very patterns to be found in contextual meaning shifts.
One difference is that the meaning variation encountered as polysemy often
involves more than one meaning shift. A good example is provided by the word film.
Originally it denoted a thin layer or membrane (cf. a film of sweat). From this derives
the metonymic variant that denotes a thin flexible transparent strip of celluloid or
similar material that carries such a layer of light-sensitive emulsion (i.e. the type
of ‘film’ that can be put into a camera, exposed, developed and fixed). A second
metonymy produces the meaning variant ›film (exposed to light, developed and
fixed) that carries photographic pictures‹. If the pictures form a movie, this can again
be called ‘film’, now due to a third metonymy of the type ›carrier medium for the
visible projection‹; this ‘film’ can be projected on a screen and watched in the cinema.
The projection tells a story and this, again, is called a ‘film’ (metonymy level 4). In a
‘film’ in this sense, characters appear and act their parts and a story takes its course.
A fifth metonymy derives the meaning of ‘film’ in the sense of an art or a medium,
a sixth produces the reading ›film industry‹ (go to the film, etc.). Note the amazing
semantic distance between the original meaning present in a film of sweat and the
last variants mentioned.
EXERCISES
1. Which properties determine a lexeme?
2. Find three composite lexemes (idioms) of each of the categories noun, intransitive
verb and transitive verb.
3. What is the difference between homonymy and polysemy?
60
Understanding semantics
4. Discuss the ambiguity of the following words with the meanings indicated: do
they constitute a case of polysemy or homonymy? Try to determine the historical
sources.
a. fraud ›act of deceiving‹ vs ›person who deceives‹
b. calf ›young of cattle‹ vs ›fleshy back part of the leg below the knee‹
c. sole ›bottom surface of the foot‹ vs ›flat fish‹
d. point ›sharp end of something (e.g. a knife)‹ vs ›dot used in writing‹
e. character ›mental or moral qualities of a person‹ vs ›letter‹ (e.g. Chinese
character)
f. palm ›inner surface of the hand‹ vs ›palm tree‹
g. ring ›circular band of metal‹ vs ›telephone call‹
5. What is the relation between the meaning of a word in its lexical meaning and
the word in a metaphorical meaning? What is the relation between a word in its
lexical meaning and in a metonymical meaning?
6. Find out the instances of metaphor and metonymy in the following passage
(from the Newsweek article cited above):
Sound like Asian contagion? So far, China has escaped economic disaster. But
even in China, the mighty can fall. … Can China reform its financial system,
but avoid the social unrest that has crippled the rest of Asia?
7. Find examples where two meanings of a polysemous lexeme are related by
metaphor, metonymy or differentiation (three of each kind).
8. Discuss the meaning shifts underlying the use of bean for the head, paw for the
hand, snotnose for a child.
9. Does the polysemy of so many words constitute an advantage or a disadvantage
for communication?
10. Discuss the difference between polysemy and the variation of meaning due to
metaphorical shift, metonymical shift or differentiation.
11. Discuss the ways in which the Principle of Consistent Interpretation affects the
interpretation of a sentence in context.
12. Think of examples where university is used metonymically for more things
connected to a university than its premises, courses and administration.
FURTHER READING
Tallerman (2011, ch. 2) on lexemes, grammatical categories and their connection with
syntax. Cruse (1986, ch. 2 and ch. 3) on lexemes and ‘lexical units’. Lyons (1995, ch. 2)
and Lyons (1977, ch. 1) on ambiguity. Lakoff (1987) and Ungerer & Schmid (2006)
and Kövecses (2006, chs 7–10) on textbook level, on metaphor and metonymy in
Ambiguity
61
the context of cognitive semantics; Tylor and Takahashi (2011) is a recent handbook
article. For a different account of homonymy, polysemy and contextual ambiguity, see
Croft & Cruse (2004, ch. 5).
4
Meaning and context
This chapter is relatively large. It deals with three phenomena which are normally
1
not treated together in other textbooks on semantics or pragmatics: deixis, the
determination of NPs and presuppositions. These three phenomena are, however,
more closely connected than is commonly recognized. They all contribute to the
indexicality of language and belong to the interface between expression meaning and
utterance meaning. In addition, they are partially embedded in each other: deictic
reference is a special case of definite NP determination; definite NPs are, in turn, a
paradigm case of expressions carrying presuppositions.
PART 1: DEIXIS
We have already seen in 1.3.1 that a sentence taken as such does not tell you what
it refers to. It is only when a sentence is actually used in a concrete CoU that we
establish its reference. You say, for example, ‘This sushi doesn’t taste good!’, referring
at that very moment to a particular dish of sushi in front of you. It is only by concrete
reference that the sentence expresses anything at all: one fixes the reference of the
subject NP, and thereby the predicate of the sentence is applied to an object in the
world and tells the recipient something about it. Fixing the reference is the basis for
any verbal communication.
For efficient communication it is most important that speaker and addressee agree
on what is being referred to. Therefore, the fixation of reference has to be based on
givens that are equally apparent to all interlocutors. The most obvious at hand is
the immediate context of utterance. It consists of things directly and automatically
given for speaker and addressee: who it is who produces the utterance, to whom
it is addressed, at which time and at which place it is produced. At least this is the
case in the standard situation of face-to-face communication. All languages have
means of directly referring to immediate elements of the CoU. Among these means
are pronouns such as I and you or expressions like here and now. Linguists call such
expressions deictic and the general phenomenon deixis.
The use of deictic expressions not only anchors an utterance in the world, it also
imposes the perspective of the speaker on the utterance. The speaker forms the
so-called deictic centre; the ‘I’ is the one who decides who is being addressed; where
1 Pronunciation [dakss], also [dekss].
Meaning and context
63
the speaker is is ‘here’; when they speak is ‘now’. In the first part of this chapter,
we will consider the three most important deictic relations: relation to the persons
involved in the utterance (4.1), to the spatial situation (4.2) and to the time of
utterance (4.3).
4.1 PERSON DEIXIS
Person deixis is deixis based on the linguistic category of person. The category of
person relates to the roles that the participants take in an individual utterance: in
the singular, the ‘1st person’ is the speaker, or producer, of the utterance, S for short;
2
the ‘2nd person’ is the addressee, or A; the term ‘3rd person’ subsumes everybody
3
who is neither first nor 2nd person. These roles are called discourse roles. In most
languages the discourse roles play a role in grammar; these are languages that have
the category of ‘grammatical person’. English is one of these languages: grammatical
person plays a role in the choice of personal pronouns and the forms of finite verbs
(she says; I am; you are; etc.).
The most salient linguistic means of person deixis are systems of personal
pronouns. They may differ in the grammatical categories of person, gender, number
and formality. Table 4.1 displays the personal pronouns in German. They form a
paradigm, i.e. a closed system in which each slot is filled with a particular form. It
often happens that there is the same form for more than one slot (this is known as
syncretism) as is the case for the form sie/Sie in the German paradigm: sie is both 3rd
person feminine singular (‘she’) and 3rd person plural (‘they’); Sie (with capital S)
is the 2nd person pronoun of formal address (recall the discussion in 2.3.1), in both
singular and plural function. It is grammatically 3rd person plural, but semantically
2nd person, since it is a pronoun of address.
Table 4.1
German paradigm of personal pronouns (nominative case)
Singular
Person
Plural
Masc.
1
Fem.
Neut.
ich
wir
du
ihr
Social
meaning
informal
2
formal
Sie
3
er
sie
es
sie
2 You will often find the notions ‘hearer’ or ‘recipient’, but these are imprecise because what matters
is who is addressed. Someone may hear or receive an utterance without being addressed.
3 For plural reference, e.g. with we and you, the correspondence between grammatical person and
S and A is less direct (see the discussion of number at the end of 4.1.1 The meaning of personal
pronouns).
64
Understanding semantics
4.1.1 The meaning of personal pronouns
The descriptive meaning of personal pronouns can be described in terms of
definiteness, person, number and gender; in addition, personal pronouns may carry
social meaning, like German du, ihr and Sie.
Definiteness. All personal pronouns are definite. Their referents are always uniquely
4
determined in the given CoU. I refers to the speaker, YOUsg to the unique addressee,
not just to ›someone addressed‹. HE, SHE or IT is always used for someone or
something certain. This is also the case with plural pronouns: WE, THEY and YOUpl
always jointly refer to a certain group of persons or things. In such cases, too, one
talks of one referent.
Person. For 1st or 2nd person singular pronouns, the descriptive meaning is clear:
they refer to S and A, respectively. For 3rd person pronouns, the only restriction is
that their reference excludes S and A. Since 3rd person pronouns are definite, the
uniqueness of reference relies on the CoU. They are mainly used anaphorically: they
refer to something which has been mentioned before. As they do not provide much
descriptive content (except as regards number and gender), the previous mention
of their referent needs to be immediately before the utterance, if not in the same
sentence. An anaphoric expression is called an anaphor, the kind of reference,
anaphora. The so-called antecedent is the expression in the preceding discourse by
which the referent was previously mentioned. Let us have a look at a sentence and
two possible continuations. Both contain the pronoun it in anaphoric use. In (1a), the
antecedent is the video, in (1b), it is the big tsunami.
(1)
a.
The video shows the big tsunami.
A friend of mine took it himself.
b.
The video shows the big tsunami.
A friend of mine survived it when he was in Malaysia.
In both cases it would be impossible to determine the utterance meaning of the
second sentence without the first. In this regard, 3rd person pronouns differ
fundamentally from 1st and 2nd person pronouns: I and YOUsg can be interpreted
without any preceding text. One only needs to know the immediate context of the
utterance that contains the pronouns. In order to construe an anaphoric 3rd person
pronoun (or any other anaphor) one also needs to know the CoU of the preceding
utterance that provides the antecedent. The only exceptions are sentences that
contain the anaphor along with its antecedent as in (2):
(2)
If you watch this video [antecedent], you’ll never forget it [anaphor].
4 I use italic capitals when I refer to words independent of a particular language. For example, SHE
stands for English she, French elle, Japanese kanojo, and so on. The context will help to decide if I
means 1st person singular pronoun in general, or the English one. For the distinction of singular
and plural YOU we use YOUsg and YOUpl.
Meaning and context
65
Gender. English does not have the grammatical category of gender. The distinction
between he and she relates to biological sex. In (3a) she refers to a female person, in
(3b) he to a male one. In the interpretations that first come to mind, it is a female
5
teacher, or male teacher, respectively, speaking of themselves.
(3)
a.
My teacher said she would be off for a week.
b.
My teacher said he would be off for a week.
The situation is different in gender languages such as Spanish, French, Russian or
German. In German and Russian, every noun has one of three genders, masculine,
feminine or neuter. For most nouns, there is no apparent semantic motivation for
the gender. The only exception is person terms (and sex-specific terms for animals,
like Kater ›tomcat‹ or Stute ›mare‹): nouns for females are feminine and nouns for
6
males, masculine. The converse does not hold: masculine or feminine nouns for
persons do not necessarily refer to male and female persons, respectively. If a noun
that denotes persons is unspecific as to the sex of the person, it may carry any
grammatical gender: Gast ›guest‹, Nachkomme ›descendant‹ are masculine, Person
›person‹, Flasche ›twerp‹ (epithet, lit. ›bottle‹) are feminine, Kind ›child‹ and Mitglied
›member‹ are neuter. These data show that grammatical gender as such does not
have descriptive meaning in a gender language like German. If a person or animal
term specifies the biological sex of a potential referent, this is part of the meaning
of the noun. In such cases, grammatical gender will match with biological sex. This
produces the illusive impression that grammatical gender carries the meaning of
biological sex.
This also holds for personal pronouns in gender languages: gender does not
carry descriptive meaning. When 3rd person pronouns are used anaphorically, their
gender matches the gender of the antecedent NP. For this reason, the gender of a
pronoun can be an important clue for determining the antecedent.
Number. Plural has a different meaning with 1st person pronouns than with nouns.
When one talks of ‘children’, one means several instances of what is described by the
singular form child. The plural multiplies, as it were, the reference of the singular.
However, when you use WE instead of I, you do not refer to several ‘I’s who would
jointly produce the utterance. Rather, WE means: ›S and those who belong to S‹. Who
it is that belongs to S depends on the given CoU. Those belonging to S may have the
status of 2nd or of 3rd person. Similarly, plural 2nd person pronouns may refer to A
and those belonging to A. Let us assume that Mary asks John:
5 Actually, the situation in English is more complicated because there are remnants of grammatical
gender from earlier stages of English, such as feminine pronouns being used for vessels, motor
bikes, or nations.
6 The only exceptions are the neuter nouns Mädchen (›girl‹) and Weib (›woman‹ with an expressive
meaning component). The neuter gender of Mädchen is due to the fact that the word is a
diminutive. Diminutives formed with the suffixes -chen or -lein are neuter, independent of the
gender of the noun they are derived from.
66
(4)
Understanding semantics
What are you doing tonight? We wanted to ask you out for a drink.
Using we she refers to herself plus 3rd persons belonging to her in the given context,
maybe her husband; using you, she may refer to John alone, or to John plus 3rd
persons belonging to him.
This special plural in the case of WE and YOUpl is called associative plural. Plural
2nd person pronouns in addition have a normal plural reading which refers to a
plurality of simultaneously addressed persons. The plural of 3rd person pronouns
is always the normal plural: THEY refers invariably to a set of persons that does not
include the speaker or any addressees.
Social meaning. In addition to their descriptive meaning, personal pronouns may
carry social meaning, for example the meaning of informal or formal interaction (see
2.3.1 and the next subsection).
4.1.2 Paradigms of personal pronouns compared
The languages of the world differ considerably in the structure of their systems of
personal pronouns. The differences concern the categories of person, number, gender
and social meaning.
Person. Some languages have fewer distinctions, some more. There are languages
with only 1st and 2nd person pronouns. On the other hand, some languages
distinguish between ‘exclusive’ and ‘inclusive’ WE: inclusive WE includes A in
addition to S, while exclusive WE refers to S and 3rd persons, excluding A. In (4),
Mary would have to use an exclusive WE; inclusive WE would have been adequate for
a question like ‘What shall we do tonight?’
Number. Almost all languages distinguish between singular and other grammatical
numbers. English only has singular and plural; other languages also have dual,
for reference to two cases, or further numbers for small sets of referents (trial for
reference to three, paucal for reference to a small number of cases).
Gender. Gender distinction is more frequent in the singular than in the plural
and more frequent in the 3rd person than in the 2nd or even 1st person. There are
languages like Hungarian which have no gender distinction at all in their personal
pronoun system.
Social meaning. Many languages have personal pronouns that carry social meaning.
These are most commonly pronouns that distinguish between informal and formal
address, such as German informal du and ihr vs formal Sie. These pronouns often
historically developed as special uses of other personal pronouns. For example,
German Sie was originally a 3rd person plural pronoun (‘they’), which, over the
course of history, also evolved a 2nd person usage. Therefore, in present usage the
plural pronoun form sie/Sie has two functions: 3rd person plural without social
meaning and 2nd person (both singular and plural) with the social meaning of
formality. In general, there are three strategies for using existing pronouns in a
special way for achieving a social meaning of formality:
Meaning and context
67
Table 4.2
Strategies of pronominal formal address
One addressee
More than one
informal
2nd singular
2nd plural
Strategy 3
3rd singular
3rd plural
Strategy P
2nd plural
2nd plural
Strategy 3+P
3rd plural
3rd plural
Strategy 3
3rd person pronouns for formal address, with 3rd person singular for a
single addressee and 3rd person plural for more than one
Strategy P
2nd person plural pronoun for formal address
Strategy 3+P
3rd person plural pronoun for formal address
Strategy 3 is applied in Hungarian and Italian; French and Russian apply Strategy
P, German and Danish Strategy 3+P. Strategy P cannot distinguish between formal
and informal pronominal address if more than one is addressed, since the formal
pronoun of address coincides with the informal 2nd person plural pronoun. The
3+P strategy of German and Danish does not allow any distinction between formally
addressing one or more people.
For a while, English made use of Strategy P: the language featured the informal
2nd singular thou as well as the formal you. In the course of time, the use of formal
you was extended to more and more situations and finally became the appropriate
pronoun of address for all occasions while thou dropped out of use except for
certain religious contexts. The result is the present situation in which there is no
distinction between formal and informal 2nd person pronouns and, concomitantly,
no distinction between singular and plural 2nd person pronouns. Some varieties
of English have developed new 2nd person plural pronouns, e.g. you all or you guys
(Wales 1996: 73). In these varieties, the meaning of you is shifting to 2nd person
singular.
Sometimes pronouns of formal address developed from expressions other than
pronouns. For example, Spanish usted (formal 2nd person singular) is a contraction
of the original vuestra merced (‘your mercy’). Being derived from a common noun,
usted and its plural ustedes is grammatically 3rd person. Spanish can thus be
considered a variant of strategy 3.
Languages without paradigms of personal pronouns. Some languages have
expressions for indexical reference to discourse roles, but lack a rigidly structured
paradigm with a single entry for each case. Typically, the personal pronouns are
more like ordinary nouns. There are a large number of expressions for speaker and
addressees, and these differ strongly in social meaning. These languages include
Japanese, Korean, Vietnamese, Thai and Burmese; Imperial Chinese, too, used to be
among these. Thai speakers choose from about thirty expressions when referring to
themselves. The choice depends on the gender and status of the speakers themselves
68
Understanding semantics
and those of their addressees. For example, there is a special 1st person expression
khâa´phraphúd´thacâaw to be used when addressing the king; the literal meaning is
›majesty’s servant‹.
4.1.3 Person deixis in verb inflection
Person deixis is not only accomplished with personal pronouns. Often verbs are
inflected for person and number and the verb forms form a similar paradigm to the
personal pronouns. If the form of the verb agrees with the subject in number and
person and if a pronoun subject can be omitted, person deixis is exerted by the verb
form alone. This is the case in Spanish, as in many other languages (cf. Table 4.3).
Table 4.3
Form paradigm of the Spanish verb (indicative present active)
beb-o
I drink
beb-emos
we drink
beb-es
you drink (sg, informal)
beb-éis
you drink (pl, informal)
beb-e
he/she/it drinks
you drink (sg, formal)
beb-en
they drink
you drink (pl, formal)
4.1.4 Possessive pronouns
In most European languages there is a parallel paradigm of possessive pronouns
(MY, YOUR, etc.). Other languages do not have extra possessive pronouns; for
example, Japanese uses the personal pronouns with the genitive particle instead,
literally ‘I’s’, ‘you’s’, ‘he’s’, etc. instead of my, your, his. Hungarian has a paradigm of
noun suffixes instead of possessive pronouns (cf. Table 4.4).
Table 4.4
Possessive suffixes in Hungarian
hajó-m
my ship
hajó-nk
our ship
hajó-d
your (sg, informal) ship
hajó-tok
your (pl, informal) ship
hajó-ja
his/her ship
your (sg, formal) ship
hajó-ju
their ship
your (pl, formal) ship
Possessive pronouns and affixes extend person-deictic reference to other things
than the discourse participants themselves. I can only refer to the speaker, but with
MY plus noun one can refer to all sorts of things that belong to the speaker. For
example, the descriptive meaning of my ship or hajóm combines the concept ›ship‹
with the deictic relation to S. The result is the concept ›ship that belongs to S‹.
7
The meaning of possessive pronouns. Possessive pronouns express that the noun
referent belongs to the person or thing indicated by the pronoun. This person or
thing is called the possessor and the thing or person that belongs to the possessor,
7 The following equally applies to possessive affixes.
Meaning and context
69
the possessum (Latin, ›possessed‹). In the case of my ship, the possessor is S and
the possessum is S’s ship. The meaning of possessive pronouns has two aspects: the
specification of the possessor, and the relation between possessor and possessum.
Specification of the possessor. Possessive pronouns determine the possessor in the
same way as personal pronouns determine their referent. This meaning component
is identical with the meaning of the corresponding personal pronouns. Everything
said in 4.1.2 carries over to this part of the meaning of possessive pronouns. It
may involve social meaning along with descriptive meaning. For example, German
Ihr and French votre have the descriptive meaning component of the possessor
being the addressee(s) and the social meaning component of formality toward the
addressee(s).
Relation between possessor and possessum. Talking of ‘possessor’ and ‘possession’
and of ‘possessive pronouns’ makes one think of ownership. However, the relation
between possessor and possessum is by no means always of this kind. ‘My son’, ‘my
leg’, ‘my age’, ‘my mistake’, ‘my name’ are not things which I possess in the ordinary
sense of the word; rather, they are things which somehow can be connected to me.
The meanings of these five nouns describe the potential referents as standing in a
certain connection to a possessor, but the respective relations are very diverse. The
son’s relation to S is a kinship relation; the leg’s relation to S is a part-of relation; the
age is a temporal measure of its possessor; S’s mistake is something S did; his name
is what S is called by. For each of these nouns, a specific relation between possessor
and referent is directly written into its descriptive meaning. Such nouns are therefore
called relational nouns. They not only specify a relation between the referent and
a possessor, their referent also depends on the possessor: in order to determine the
referent of a relational noun, one needs to know the possessor to which it relates.
Of course, there are many nouns which do not specify a relation of their referent
to some possessor, e.g. terms for arbitrary sorts of things, such as ship, dog, bicycle,
chewing gum or stone. Most nouns are of this type. They are called sortal nouns.
They describe their potential referents in terms of what they are like. If one uses them
with a possessive pronoun, this does not necessarily express possession. As we have
noted in 1.1.1, S can use my bicycle to refer to a bicycle which is connected to S in
one way or other: it may be the bicycle which S owns, or uses all the time (without it
belonging to S), it may be the bicycle which S was assigned in some connection, or it
may be the bicycle which S is notoriously talking of. If the noun denotes an article of
daily use, there may be a privileged relation to a particular user. For example, without
a special context, one will assume that ‘my bicycle’ is the bicycle which S uses. But
this is not necessarily so. Ultimately, the relation to the possessor is a matter of the
particular CoU.
We can summarize the meaning component of the relation to the possessor as
follows: the possessive pronoun indicates that the referent of the NP stands in some
relation to the possessor, but it does not indicate which kind of relation this is. If the
possessive pronoun is combined with a relational noun, the relation is provided by
the meaning of the noun. If it is combined with a sortal noun, the relation is a matter
of the CoU.
70
Understanding semantics
These considerations concerning the relation between possessum and possessor not
only hold for possessive pronouns and affixes, but also for possessive constructions
in general, e.g. Anna’s studio or the meaning of possessive pronouns.
4.2 DEMONSTRATIVES AND PLACE DEIXIS
Place deixis relates to the spatial situation in which an utterance takes place. The
points of reference are the location of S and, in some cases, also the location of A.
The immediate categories of place deixis are HERE and THERE. HERE serves as
the reference to the location of S, THERE refers to where S is not. Place deixis is not
restricted to reference to places. Arbitrary objects of reference can be determined
by place deixis. The linguistic means for this purpose are demonstratives such as
English this and that: this denotes something that is where S is (except S herself) and
that refers to something that is where S is not.
When I use English here or this, I only give the information ›where I am‹ or ›thing
that is where I am‹. This is a very vague localization: ‘where I am’ can extend to
a very different radius around myself (depending on context) and there may be
several possible referents within this radius. That is why demonstratives are often
8
accompanied by a gesture of pointing, meant to help the identification of the referent.
4.2.1 Japanese demonstratives
Let us take a look at the Japanese system of demonstratives. It is richer than the
English system (Table 4.5).
Table 4.5
Japanese demonstratives
Close to
speaker
Close to
addressee
Close to
neither
pronominal
kore
sore
are
‘this’
adnominal
kono
sono
ano
‘this’
place noun
koko
soko
asoko
‘this place’
adjective
konna
sonna
anna
‘such’
adverb
k
s
‘so’
Japanese demonstratives distinguish three deictic categories: close to speaker, close
to addressee and close to neither speaker nor addressee. Thus, the deictic orientation
is person deictic. The forms kore, sore and are are nouns; they can be used to refer
to things, not to persons. For demonstrative reference to persons, one uses the forms
8 The terms demonstrative and indexical are of Latin origin, deixis is from ancient Greek. They all
contain the meaning of ‘pointing’.
Meaning and context
71
kono, sono and ano combined with a noun meaning ›person‹ or some kind of person,
e.g. kono hito ‘this person’. The -no forms are combined with nouns like articles; they
are what is called determiners. Demonstratives like kore which can be used as full
NPs are called pronominal, those combined with nouns, adnominal. Koko, soko
and asoko are demonstrative place terms. They are nouns, unlike English here and
there which are adverbs; Japanese koko means ‘this place’, not ›here/at this place‹.
Konna, sonna and anna are adjectival demonstratives which combine with nouns.
Konna hon means approximately ‘a book such as this one here, a book like this’. Kō,
sō and ā are adverbs like English so; ‘kō yatte’ (yatte ‘do it’) would mean ‘do it this
way’. The case of Japanese illustrates two dimensions of systems of demonstratives:
the semantic dimension of deictic relation (e.g. proximity to speaker) and the
grammatical dimension of their function as noun, determiner, adjective or adverb.
4.2.2 Systems of demonstratives compared
Most languages distinguish two or three deictic categories of demonstratives. In
cases where it is two, the distinction is between proximal (close to the deictic centre)
and distal (not close to it). Since the deictic centre is defined as the location of S,
‘proximal’ and ‘distal’ can also be defined as ‘close/not close to S’. English is of this
type: here and this are proximal, as opposed to distal there and that.
Spanish has a system with three levels of distance: proximal este/esta, medial (i.e.
not in the deictic centre, but also not far from it) ese/esa and distal (far from the
deictic centre) aquel/aquella. Thus, its organization is different from the Japanese
system.
German is a rare exception. There are no deictic distinctions for pronominal or
adnominal demonstratives; there is just one form dies(er). This demonstrative can,
however, be combined with demonstrative adverbials hier (proximal), da (distal) and
9
dort (distal): dieses Buch da (›this book there‹). French is quite similar.
4.2.3 The meaning of demonstratives
Demonstratives have three semantic dimensions: deictic relation, type of reference
and definiteness. The deictic dimension has just been discussed. The type of
reference concerns the function of the demonstratives. If they are used as full NPs,
they refer by themselves. In adnominal use, the deictic information is combined
with the descriptive meaning of the noun. For example, the referent of this dog is
conceptually described as a dog by means of the noun, while the demonstrative adds
the information that it is a certain something, close to S. In this way, it is possible
to refer to a particular dog. The function of adnominal demonstratives is thus
analogous to the function of possessive pronouns, where the descriptive meaning of
the noun is combined with a person-deictic relation.
9 The WALS survey on distance contrasts in demonstratives, which covers 234 languages, has only
seven with no contrast, among them German and French (Diessel, chapter 41 in WALS, http://wals.
info/chapter/41).
72
Understanding semantics
Adjectival demonstratives like such or Japanese konna, etc. add to the nominal
concept a particular quality which is determined deictically. Demonstrative spatial
adverbs or nouns refer to a place that is deictically determined.
Definiteness is a common component of all demonstratives. With this book we do
not just refer to any book close to S, but to the specific book close to S, and likewise
for pronominal and spatial demonstratives. Adjectival demonstratives, too, are
definite in referring to the quality deictically indicated.
4.2.4 Anaphoric use of demonstratives
In most languages, demonstratives also have anaphoric uses, others have separate
anaphoric pronouns. In the second sentence in (5), both that and there are used
anaphorically.
(5)
I saw a hedgehog in Frank’s garden the other day. Do you think that hedgehog is
living there?
In anaphoric use, the deictic distinction between proximal, distal or medial is not
relevant in most cases. If a language has a system with two demonstratives, usually
the distal one is used for anaphora; languages with a three-part system use the
medium one. Anaphoric uses of demonstratives loosen the tie to the immediate CoU.
By using anaphora one can refer to things that are not immediately present, but were
mentioned earlier on. Anaphors are bound to the given discourse; they do, however,
constitute an important first step towards a kind of reference which is purely
definite. In fact, definite articles in most languages developed from demonstratives.
For example, the definite articles in Spanish (el/la), Italian (il/la) and French (le/la)
emerged from reductions of the Latin distal demonstrative ille/illa. English the and
German der/die/das, too, derive from demonstratives.
4.3 TIME DEIXIS
Time deixis relates to the time when an utterance is produced, the ‘time of utterance’.
For a CoU, the time of utterance is the present: the time before it is the past, and
the time after it, the future. There are two phenomena in the domain of time deixis:
grammatical relation to time by means of verb tense, and temporal reference by
lexical means such as temporal adverbs, specifications of time, etc. Many languages,
for example Chinese, do not have a real tense system of verb forms; Chinese verbs
need not carry tense, and temporal reference is mainly a matter of the given context.
Tense. The grammatical category of tense will be dealt with in chapter 6.4. Just for
the sake of illustration, consider three examples from English:
(6)
a.
past tense:
She knew the e-mail address.
b.
present tense:
She knows the e-mail address.
c.
future tense:
She will know the e-mail address.
Meaning and context
73
The past tense form knew of the verb to know expresses a state in the past: the state
expressed obtained at a particular time before the time of utterance. The present
tense relates the same predication to the ‘present’ time, i.e. the time of utterance,
while the future tense relates to a particular time after the time of utterance.
Frequently, the tense forms in a language have more than one use. For example, the
past tense forms in English can also be used for counterfactual reference in present
time, and the form will plus infinitive is alternatively used for expressing probability
(of a present state).
Lexical means of time deixis. The central expression of time deixis is NOW. It refers
to the time of utterance. Just as HERE may refer to a place of rather vague extension,
NOW may refer to time intervals of very different length. English has another timedeictic adverb, then, which is distal and can refer to a certain time either in the past or
in the future. There are a number of adverbs that enable more fine-grained temporal
localization: just, recently, formerly, once, at once, soon, later, etc. Other adverbs refer
to days: today, tomorrow, yesterday; ‘today’ is the day which contains the time of
utterance, ‘tomorrow’ the day after today, etc. There are also adnominal expressions
of time deixis, such as next, last, former or future. When combined with a noun, they
add a temporal localization of the referent.
Stopover
We have taken a look at the central means of deixis. They serve to bring about
reference by relating the utterance to immediate elements of the CoU. These
elements are those things that are automatically given with any utterance in
face-to-face communication: speaker, addressee(s), the time of utterance and
the place where it is produced. Part of the deictic expressions serves the direct
reference to these elements. These expressions are independent parts of the
sentence with their own reference, e.g. pronouns or adverbs. Along with these
independent expressions, there are adnominal deictic means. They allow us
to combine the descriptive meaning of a noun with a deictic indication of the
referent. Table 4.6 gives a short survey of pronominal/adverbial and adnominal
deictic expressions.
Table 4.6
Pronominal and adnominal deictic expressions
Kind of deixis
Pronominal/adverbial
Adnominal
person deixis
I, you, …
my ship
place deixis
here, there
this, that
this ship
time deixis
now, tomorrow
the next ship
74
Understanding semantics
If we were confined to deictic reference, we could only talk about things
which are immediately given in the CoU. With anaphoric use of demonstratives,
we can expand our radius of reference considerably. Still, in order to free
ourselves from the limits of the given context of discourse, we need more
general linguistic means of anchoring utterances. The most important one is
general definite reference, for example marked by definite articles. As we shall
see, definite reference still anchors the utterance in the CoU, but only in a very
general way. A definite article only conveys the message ‘the referent is uniquely
determined in this context.’
We will proceed by further pursuing the adnominal track. In addition to
the deictic adnominals – demonstratives and possessives – more general
determiners will enter the stage: definite and indefinite articles and so-called
quantifiers.
PART 2: DETERMINATION
If one wants to use a noun for reference, one has to form an NP with it. There are
various ways to do this, and to indicate thereby the kind of reference intended. In
many languages one can choose the grammatical number of the noun, e.g. singular or
plural, expressing if one refers to one or more than one instance. In many languages
the noun can be combined with a definite or indefinite article to indicate whether the
referent is something given beforehand or not. The noun can also be combined with
other determiners: with demonstratives, possessive pronouns or elements such as
each, every, all, both or no. It is also possible to use an NP without concretely referring
to something given in the CoU. In all the examples we have considered so far, the NPs
referred concretely (to a particular dog, a skirt, a bicycle, etc.), but NPs can also be
used for relating to general cases without talking about particular things given in
the context. (7) is an example of this. The NP books does not refer to any particular
books. This use of NPs is called generic.
(7)
Books are getting cheaper and cheaper.
These aspects of NPs are subsumed under the notion of determination. The
following sections discuss the core phenomena: definiteness and indefiniteness (4.4),
quantification (4.5) and genericity (4.6).
4.4 DEFINITENESS AND INDEFINITENESS
4.4.1 An example
Let us assume Sheila did not say (8a) to Mary, but (8b) instead:
Meaning and context
(8)
a.
The dog has ruined my blue skirt.
b.
A dog has ruined my blue skirt.
75
When you compare the two sentences, it is easy to see the difference between the
definite and the indefinite article. By using the definite article, Sheila indicates that
she is referring to a particular dog given beforehand in the context of the utterance,
to the dog. For both Sheila and Mary, this is the only dog that comes into question in
the context described, the family dog Ken. The definite article points to this context.
It is indexical.
If Sheila had used the indefinite article, she would have expressed that she was
not talking about a dog which is uniquely determined beforehand. The addition
‘beforehand’ is crucial: objectively, the dog is uniquely determined by having ruined
Sheila’s skirt; but what matters for the choice of the article is the fact that this dog
comes into play between Sheila and Mary only with this very utterance. Before that,
the creature did not belong to the common context of the two of them. After having
said (8b), Sheila is able to talk anaphorically of ‘the dog’, continuing, for example, in
the following way:
(9)
A dog has ruined my blue skirt. It happened when I was walking through the
park. It suddenly came running from the back and snapped at me. I was really
frightened. I had no idea who the dog belonged to.
The example illustrates a frequent interplay of indefinite and definite article: first
a referent is introduced in the discourse with an indefinite article, later the same
referent is taken up with a definite article. Correspondingly, it is often said that the
function of the indefinite article is to introduce new referents into the discourse,
and that the definite article serves to anaphorically refer back to referents already
introduced. These descriptions of the functions of the articles are, however, much
too narrow.
4.4.2 The meaning of the definite article
The combination of a noun with the definite article is called a definite description.
A definite description can contain additional material such as adjectives or relative
clauses attached to the noun: the blue skirt, the skirt on the chair or the skirt that the
dog has ruined. In a definite description, the article is combined with material that
provides semantic information about the referent: the referent is a skirt, it is blue, it is
on the chair, it has been ruined by the dog, etc. What the definite article itself adds to
the description of the referent is the information that the given description is unique
in the given context: there can be only one thing it applies to.
The crucial question, then, is this: when does a definite NP constitute a unique
description in the given context? There are essentially two possibilities: the uniqueness
can be due to the very meaning of the NP, independently of the particular context,
or it can be due to the special circumstances in the given CoU. In the first case, we
have what is called semantic uniqueness, in the second, pragmatic uniqueness
76
Understanding semantics
The definite description the dog in the fifth sentence of (9) is pragmatically unique.
The preceding sentences create a special context in which one particular dog is
introduced and plays a prominent role. Without this context, the referent of the dog
would not be uniquely determined.
DEFINITION 1 Semantic and pragmatic uniqueness
An NP is semantically unique if the description of its referent is unique
independently of the given CoU.
An NP is pragmatically unique if the description of its referent is unique only
in the special circumstances given in the CoU.
4.4.2.1 Semantic uniqueness
In order to illustrate the notion of semantic uniqueness, we will take a look at a
newspaper text. The text is a translation of parts of a news article in the German daily
Tageszeitung from 17 October 2011. The article relates to the debate at that time about
the succession of the king of Saudi Arabia.
(10) […] ‘The king is suffering from chronic back pain,’ said one Western diplomat.
[…] The monarch’s health and age recently gave rise to speculations about his
succession. In Saudi Arabia, the entire ranks of the leadership are about eighty
and are all ill, out of the country or unable to govern.
10
The noun king is relational, the possessor being the respective kingdom. Since for
a country, if it has a king at all, it has only one, the description king is semantically
unique. Uniqueness of reference is built into the meaning of the noun king. One
therefore says: the noun king is inherently unique. Since the news article is about
Saudi Arabia and the country had a king when the article appeared, the possessor
of the king (the kingdom) is uniquely determined and with it the king at that time.
The noun monarch, however, though of a related meaning, is neither relational
nor inherently unique, but plainly sortal. The concept ›monarch‹ comprises all
queens, kings, empresses, tsars, pharaohs, etc. of all times and all countries under a
sortal notion of a particular kind of person, namely one who solely rules a country
(under certain conditions). The noun monarch is not relational because its potential
referents can be determined without fixing a possessor. It is not inherently unique
because in a given CoU there may be many monarchs (or, by chance, just one, or none
at all). For this reason, the description the monarch in the text is not semantically
unique, but pragmatically. In (10) the noun can be used with the definite article here
10 Note that the possessor for a relational noun like king is what appears in the of phrase after the
word: in King of Saudi Arabia, Saudi Arabia is the possessor of the king, in the grammatical sense
(rather than the king being the possessor of the country).
Meaning and context
77
only because a particular monarch, the king of Saudi Arabia, was established in the
context of the article (and no other monarch).
Like king, the nouns health and age (in the meanings given here) are relational
and inherently unique. Health and age are always health and age of some possessor,
and a given possessor, at a given time necessarily has one health and one age only. If
the possessor of health or age is uniquely determined, so is the single referent of the
nouns. The last definite description in the article, the entire ranks of the leadership, is
also semantically unique: for a given country or other institution, there is only one
group of persons which forms the leadership ranks.
Most inherently unique nouns are relational. The most frequent sub-types are role
terms, part terms and terms for all kinds of attributes. Role terms encompass nouns
such as king, president, director, etc. They denote the head of some organizational unit
which is their possessor – a country, a company, a university, etc. Other role terms
include words like mother or boss: the possessor is a person, and for a given possessor
the referent is a uniquely determined other person (usually a person has only one
mother and at most one boss). Examples of inherently unique part terms are bodypart terms for parts of which there is only one: head, nose, mouth, throat, back, etc.,
or terms for unique parts of objects: mouthpiece, lid, bottom, surface and so on. Terms
for attributes denote various aspects, or dimensions, by which their possessor can be
described: size, weight, width, structure, shape, colour, meaning, price, name, content,
character, age, profession, sex, etc. Inherently unique relational nouns are called
functional nouns. They are so called because there is a function in the mathematical
sense which assigns a referent to every possible possessor: her mother, her back, her
address, her height, etc. Relational nouns, in the narrow sense of the word, are the
remaining relational nouns that are not inherently unique. Relational nouns, too,
include role terms (friend, neighbour), part terms (finger, tooth, side) and attribute
terms (attribute, property, feature), the crucial difference being that for one possessor
there may be no, one or more referents.
There are also non-relational nouns that are inherently unique. They denote
‘institutions’ in the world, in the broadest sense: things of which – in a given type
of context – there is only one, playing a particular role in that context. Such nouns
include sun (in the sense of the sun of Earth), moon (in the corresponding sense),
pope (the head of the Catholic Church), weather (in a given CoU, which includes a
particular place and time, there is only one weather), date (day, month and year), etc.
Inherently unique non-relational nouns are called individual nouns.
The four types of nouns mentioned so far are arranged in Table 4.7. ‘Relational
nouns’ are to be taken in the narrow sense of the word.
Table 4.7
Types of nouns
not inherently unique
inherently unique
non-relational
sortal nouns
boy, dog, skirt, rock
individual nouns
pope, sun, date, weather
relational
relational nouns
son, foot, property
functional nouns
father, head, size
78
Understanding semantics
Sortal or relational nouns can be turned into inherently unique concepts by adding
certain expressions, such as superlatives (the fastest notebook), ordinal numerals and
similar expressions (the third/next/last chapter), or appositions that are themselves
inherently unique (cf. the year 1984, my daughter Emma, the word akimbo). The
nouns notebook, chapter, year, daughter and word are not inherently unique, but
when complemented by these adjectives and appositions they represent semantically
unique descriptions.
There are things or persons which are established in certain social contexts and
play a unique role there. For these, often functional or individual terms exist: for ‘the
dean’ of a faculty, ‘the driver’ in a bus, the ‘Mum’ in a family. But very often we use
just sortal nouns in such cases, in particular if the social context is rather restricted:
we talk about ‘the bus stop’, ‘the pizza shop’, ‘the playground’, ‘the pharmacy’ in a
neighbourhood, ‘the kitchen’, ‘the postbox’, ‘the dog’, ‘the car’ in a household, ‘the
waitress’, ‘the menu’, ‘the meal’, ‘the bill’ when we visit a restaurant. In all these cases,
the definite article can be used without establishing unique reference beforehand.
The referent is established in the given type of context and identified by the function
or role it plays.
When one uses an NP with an inherently unique noun or with a sortal noun in
a context where it denotes something with a unique role, its referent is uniquely
determined beforehand. The definite article only confirms this fact. Nevertheless,
the definite article is not superfluous: it is indexical in that it points to the given type
of context. It is the CoU that ultimately determines the respective referent of the
definite description: ‘the pope’ is the pope in office at the time referred to, his ‘health’
is the state of health at that time, ‘the weather’ is the weather at the given time and
place, ‘the dog’ is the dog presently kept in a household, ‘the playground’ is the kids’
playground in the neighbourhood where they live, etc.
4.4.2.2 Pragmatic uniqueness
Pragmatically unique definite NPs only refer uniquely due to the particular context
they are used in. This uniqueness, too, is established in a social context – in the current
discourse – but its lifespan is restricted to this situation. Once the discourse is over, or
has moved on, the ground for unique reference is no longer given. One way to create
pragmatic uniqueness is by deictic use of definite descriptions. Unlike adnominal
demonstratives, the definite article does not carry a deictic differentiation, but it still
indicates that the reference of the NP is determined by the CoU. For example, if you
encounter somebody on a walk and there is a dog around, you can say to her:
(11) Is the dog yours?
You may accompany your utterance with a gesture pointing to the dog you mean. The
dog may also be in the focus of attention of the interlocutors anyway, in which case
a gesture is not necessary.
A further case of pragmatic uniqueness involves definite NPs in anaphoric use. An
example is the dog in Sheila’s story in (9). Its antecedent is a dog in the first sentence.
Meaning and context
79
The three NPs a dog, it in the third sentence, and the dog all have the same referent;
they are coreferent. Sheila sets up the referent as a character in her story and thereby
establishes it in a unique role. In the news article in (10), the monarch is an anaphor
with the antecedent the king. The referent of the anaphor and its antecedent was
introduced previously in the specific context of the article.
4.4.2.3 A test for pragmatic vs semantic uniqueness
There is a simple test for distinguishing between semantic and pragmatic uniqueness.
In the case of pragmatic uniqueness, the definite article can be replaced by an
adnominal demonstrative. This is due to the fact that the article in these uses has
the demonstrative function of pointing to the immediate CoU. Replacement by a
demonstrative is not possible with semantic definites. One would not say, ‘I’ll take
this dog for a walk’ when just referring to the family dog, or, ‘This king of Saudi
Arabia is already old and frail’. At least, the use of the demonstrative would require
a special context, or express a special attitude. By contrast, Sheila in her story in (9)
could equally well say: ‘I had no idea who this dog belonged to’, and the news article
in (10) might have a demonstrative with monarch: ‘The health and the age of this
monarch recently gave rise to speculations about his succession.’
4.4.2.4 The range of application of the definite article
As this test shows, the range of application of definite descriptions overlaps with the
one of NPs with adnominal demonstratives. Both have deictic and anaphoric uses.
In the case of deictic demonstratives, the deictic content and/or a gesture of pointing
helps to identify the referent uniquely. Deictic and anaphoric uses essentially make
up the range of pragmatic uniqueness. In the domain of semantic uniqueness,
however, only definite descriptions can be used (Fig. 4.1). This is the basis for the
test just described. For demonstratives, the anaphoric use is an extension of their
original deictic use. Definite descriptions, in turn, extend unique determination to
semantic uniqueness.
Figure 4.1
Uses of demonstratives and the definite article
Demonstratives
deictic
anaphoric
pragmatic uniqueness
semantic uniqueness
definite article
We can fix the following as the general function of the definite article:
80
Understanding semantics
DEFINITION 2 Function of the definite article
(i) The definite article indicates that the NP refers to something uniquely
determined in the given context, either in the particular or in a wider
context.
(ii) The definite article is indexical in relating to the relevant context in which
the referent is uniquely determined.
4.4.3 Singular and plural, count and mass nouns
In order to better understand NP reference, we need to have a look at the function
of singular and plural and the distinction between count nouns and mass nouns.
These distinctions are not relevant for understanding definiteness, but they will play
a central role for indefiniteness, which we will deal with next.
Singular and plural. Most English nouns can be used in the singular and in the
plural. For a noun like skirt, the singular form skirt refers to a single object, the plural
form skirts to more than one. These multiple instances form one referent; a statement
like (12) refers to a complex referent consisting of several skirts:
(12) These skirts have become too tight.
Consequently, the denotation of the singular noun skirt is different from the
denotation of plural skirts. The denotation of skirt is the set of all single skirts,
while the denotation of skirts is the set of all cases of more than one skirt. In formal
semantics, such cases are called ‘sums’ or ‘groups’.
Count nouns and mass nouns. Count nouns are nouns which can be used in
the plural, and in most cases, with numerals; apart from a few exceptions (e.g.
pants), they can also be used in the singular. By contrast, mass nouns are used
in the singular to refer to an unspecified quantity of something: juice, flour, air,
garbage, metal, clothing, furniture are mass nouns, as are abstract nouns such as
anger, love, heat or grudge. Several instances of ‘apple’ are referred to as ‘apples’, but
several instances of ‘juice’ may form just one case of ‘juice’. One can therefore say, for
example, that the guests at a party ate ‘many apples’ (plural) and consumed ‘much
juice’ (singular). The denotation of a mass noun encompasses single instances of
juice as well as collections of several such instances.
The difference between mass nouns and count nouns is due to the fact that their
meanings are concepts of different type. The meaning of the count noun skirt is a
concept which clearly delineates its potential referents; it provides a criterion for
what is one skirt: one skirt is a single, whole skirt. Neither does a part of a skirt
constitute an instance of a skirt, nor do several skirts form a case of one skirt.
By contrast, a mass noun juice describes its potential referents in a way that is
appropriate for distinguishing juice from other kinds of stuff, but does not yield a
criterion for what would constitute one potential referent. It is not possible to decide
Meaning and context
81
if some quantity of juice is one case of ‘juice’, multiple cases or just part of one case.
Mass nouns do not ‘individuate’ their potential referents.
A special sub-type of mass nouns, so-called aggregate mass nouns, comprises
nouns like literature, furniture, equipment, staff, etc. Aggregate mass nouns collect
different kinds of items into one concept. For the single items there may be count
noun terms, such as chair, table and bed for items of furniture. Aggregate mass
nouns leave open how many items are referred to. You can add items to a referent of
furniture, or remove items, and it still remains a possible referent of furniture. Often
there are pairs of plural count nouns and aggregate mass nouns with roughly the
same meaning: clothes and clothing, shoes and footwear or employees and staff.
Since count noun concepts define clearly delineated single cases, these cases can
be counted, and count nouns can therefore be combined with numerals and used in
the plural. Mass nouns lack the preconditions for these uses.
Actually, mass nouns are sometimes used in the plural, but then they take on the
meaning of count nouns. ‘Juices’ are different kinds of juice (kinds are countable)
and ‘beers’ are servings of beers, or kinds of beer. Without being reinterpreted as
count nouns, mass nouns can only be used in the singular and not with numerals.
They combine with vague quantity specifications such as much, a little or more. For
a precise specification of quantity they must be combined with a measure unit word
such as litre, kilogram, etc. Unit words form count expressions with mass nouns –
litre juice, kilogram flour – which can be combined with numerals.
4.4.4 Indefinite NPs
English has an indefinite article a(n), but only for singular count indefinites. No
article is used for simple mass or plural indefinites. Sometimes they are said to
contain a ‘zero article’, but where there is nothing there is also no zero article. Thus,
one should prefer to talk of bare mass nouns and bare plurals. Consider the
following example:
(13) a.
I had an apple and __ strawberries and __ orange juice.
In French, indefinite NPs with mass or plural nouns are marked explicitly, with a
combination, or fusion, of the partitive particle de and the definite article forms le/
11
la/les. The French equivalent of (13a) is this:
b.
J’ai mangé une pomme et des fraises et j’ai bu du jus d’orange.
formally: ‘I have eaten an apple and of the strawberries and I have drunk of
the orange juice.’
The French indefinite markings for mass and plural NPs contain the definite article,
but this is not reflected by their meaning. The referent of the NPs need not be part of
11 The respective forms are: du > de le (masculine), de la (feminine) and des > de les (plural).
82
Understanding semantics
a quantity given beforehand; des fraises does not mean ›some of the strawberries‹, as
the formal one-to-one translation suggests but just ›(some) strawberries‹.
In colloquial English, there may be an indefinite article for mass and plural nouns
presently evolving: a reduced form of some, often described as sm in the semantic
literature and pronounced [sm/sm]. It is to be distinguished from full some [s m],
which constitutes a quantity specifier (see 4.4.5 and 4.5.2)
The meaning of indefinite NPs. The meaning of the indefinite article is the same
as the lack of an article, or sm, with mass nouns and plurals: it indicates that the
referent of the NP is not given beforehand. It is not uniquely determined, but only
given a qualitative characterization.
4.4.5 Definite and indefinite NPs in general
Definiteness and indefiniteness are not tied to the presence of definite or indefinite
articles; some of the cases have already been discussed. In English, the following are
definite NPs:
DEFINITION 3 Definite NPs
a. NPs with
- a definite article (definite descriptions) – the cup
- adnominal demonstratives – that cup
- possessive pronouns – her cup
- Saxon genitives – Fred’s cup
b. proper names
c. pronominal demonstratives
d. personal pronouns
NPs with a possessive pronoun or a Saxon genitive are only definite if they are not
used predicatively. In (14a) my brother and Kate’s brother are in predicative use; the
NPs do not refer and are not definite. (14a) means the same as ›Ben is a brother of
me/ a brother of Kate‹. In (14b), the respective NPs are used referentially; here they
mean the same as ›the brother of me/of Kate is coming for a visit‹.
(14) a.
b.
Ben is my brother/Ben is Kate’s brother.
My brother/Kate’s brother is coming for a visit.
Indefinite NPs, too, comprise more than the types just discussed:
Meaning and context
83
DEFINITION 4 Indefinite NPs
a. simple indefinite NPs: singular count nouns with indefinite article; bare
mass nouns, bare plurals; mass nouns and plural nouns with sm.
b. count nouns with quantity specifications:
numerals, no, many, (a) few, some [s m], any, several, …
c. mass nouns with quantity specifications:
measure specifications 3 litres of, no, much, (a) little, some [s m], any, …
d. indefinite pronouns: somebody/-one/-thing, no/-one/-thing, anybody/-one/thing, etc.
4.4.6 Articles across languages
Definite or indefinite articles, or equivalent markings on nouns, do not exist in all
languages. Some languages only mark definiteness, some only indefiniteness, many
do not mark either. Russian and most of the other Slavic languages have no articles,
likewise Japanese. Turkish and Chinese mark only indefiniteness (in certain cases);
Irish only definiteness. In addition, languages differ considerably in when they make
use of the markings available. For example, in Modern Greek and colloquial German,
personal proper names are used with the definite article, but not so in English and
written standard German. If a language does not mark definiteness, there will be
other ways to indicate unique reference. The most important means is sentence
structure. In Chinese, for example, a language with SVO word order, it is essentially
assumed that the subject has definite reference while the object is indefinite. If these
conditions are not given, special constructions are used.
4.5 QUANTIFICATION
4.5.1 Quantifier, domain of quantification and quantified predication
In theoretical semantics, no NP phenomenon has been studied as deeply as so-called
quantification. Some theories consider all types of determination as instances of
quantification, including definite, indefinite and demonstrative determination.
However, I consider it important to apply a more restricted notion of quantification,
in order to properly understand the phenomenon. Let us have a look at an example;
the sentences are to be construed coherently:
(15) a.
Next week there will be a big birthday party at John’s place.
b.
Sue is going to invite nine kids.
c.
Each kid is to get a present.
d.
For some kids, John has not got anything yet.
e.
No kid will be disappointed.
84
Understanding semantics
In the second sentence, certain ‘kids’ are introduced in the discourse, those nine kids
that Sue is going to invite. From this point on, one could refer anaphorically to these
kids as ‘the kids’. In (15c) these kids are referred to: each of them is to get a present.
The reference to ‘the kids’ is implicit, though. But we might as well have the following
in (15c, d, e):
(16) c.
Each of the kids is to get a present.
d.
For some of the kids, John has not got anything yet.
e.
None of the kids will be disappointed.
Such constructions with a pronoun and an of PP following are called partitive
constructions. The fact that in (15c, d, e) the three NPs each kid, some kids and
no kid can be replaced by partitive constructions reveals that they involve definite
reference to ‘the kids’. They relate to a given total of cases which could be referred to
with a definite plural NP.
The three NPs mentioned are quantifying NPs. All quantifying NPs relate to a
total of cases, the so-called domain of quantification (DoQ for short). In (15c, d, e),
the domain of quantification is formed by the nine kids introduced in (15b).
A quantifying NP is combined with a predication (usually the rest of the sentence)
about the single cases in the DoQ. If we use the variable x for the single cases in the
DoQ, the predications in the three sentences are: ‘x is to get a present’, ‘for x, John has
not got anything yet’ and ‘x will be disappointed’. The predication is simply obtained
by replacing the quantifying NP with a variable. Let us call the predication the
quantified predication.
Now, the contribution of the determiners each, some and no is the following: they
quantify the number of cases within the DoQ for which the predication is true. This
is why they are called quantifiers. In the case of each, the predication is true without
exception, in the case of some it applies to an unspecified part of the DoQ and for
no it is not true for any case in the DoQ. Table 4.8 displays the three components –
domain of quantification, quantified predication and the quantification expressed by
the quantifier – for the examples in (15).
Table 4.8
Components of quantification
Domain of
quantification (DoQ)
Quantified predication
(QP)
Quantification:
the QP is true …
(17c) the nine kids
x is to get a present
for each x in the DoQ
(17d) the nine kids
for x, John has not got
anything yet
for some x in the DoQ
(17e) the nine kids
x will be disappointed
for no x in the DoQ
What is at issue when quantification is applied is the fact that the quantified
predication might be true for some cases within the DoQ and might be false for
Meaning and context
85
others. Using each, every or all, we exclude that the predication is false for any cases;
by using no, we exclude that it is true for any cases; some expresses that there are
cases within the DoQ for which the predication is true – leaving open whether there
are also cases where it is false.
In English there are four determiners which are exclusively used as quantifiers:
each, every, all and both. In addition, certain forms of NPs with other prenominal
elements can be used for quantification, but only under certain conditions, as we
will see next.
4.5.2 Quantification and indefinite NPs
Indefinite NPs are sometimes quantificational, sometimes not. There is a simple
criterion to determine whether a given indefinite NP in a particular utterance is
quantifying or not: it is quantifying iff it is explicitly or implicitly partitive. Let us
return to the examples in (15). As we have seen in (16), the three quantifying NPs
in (15c, d, e) can be replaced by partitive constructions; they are therefore implicitly
partitive. The five sentences in (15) also contain two indefinite NPs that do not
quantify: nine kids in (15b) and a present in (15c). In the context assumed here, they
cannot be interpreted as implicitly partitive and therefore they cannot be replaced by
explicit partitive constructions (the sign § marks semantic inadequacy):
(17) b.
§ Sue is going to invite nine of the kids.
c.
§ Each kid is to get one of the presents.
In the case of (15c), a straightforward partitive paraphrase is not even grammatically
possible; the indefinite article would have to be replaced by the numeral one.
It is not a coincidence that the quantifying NPs in (15) have partitive paraphrases,
while the others do not. NPs which quantify over, and thereby refer to, a certain DoQ
can always be paraphrased with a partitive construction. Conversely, a (referential)
NP which can be construed as partitive always produces quantification. The definite
NP within the partitive paraphrase explicitly refers to the DoQ. Selecting some of the
cases of the domain has the effect of quantification: the predication about the NP is
true for as many cases in the DoQ as indicated, and may be false for the rest.
We can therefore tell precisely when an indefinite NP is quantificational. As we have
seen, this is not merely a matter of the form of the NP, but crucially depends on the
context.
DEFINITION 5 Quantificational indefinite NPs
An indefinite NP is quantificational in a given utterance iff it can be construed
as partitive. It can be construed as partitive iff it can be replaced by the
corresponding partitive construction.
86
Understanding semantics
It remains to be stated that not all types of indefinite NPs are eligible for partitive
interpretation – for some types there is no corresponding partitive construction. The
partitive construction corresponding to an indefinite NP has a pronominal element
which corresponds to an adnominal element of the NP. Usually the adnominal
elements have identical pronominal forms, but sometimes they do not; for example,
none is the pronoun corresponding to adnominal no. In order to have a corresponding
partitive construction, an indefinite NP needs to contain an adnominal element with
a pronominal counterpart. Bare mass nouns and bare plurals lack such an element,
and the indefinite article, too, has no pronominal counterpart. The same holds for sm
with mass nouns and plurals. The weak form sm has no pronominal use. Pronominal
some [s m] is the counterpart of adnominal some [s m], not of sm. Therefore, simple
indefinite NPs lack a corresponding partitive construction – and cannot be used for
quantification. Only indefinite NPs with a quantity specification (cf. Definition 4 b
and c) are eligible for quantificational use.
(18) Indefinite NPs with corresponding partitive constructions
Only indefinite NPs with a quantity specification have a corresponding partitive
construction.
When indefinite NPs are not implicitly partitive, they simply refer to something not
yet uniquely determined in the given CoU; if the NP contains a quantity specification,
the latter adds information about the quantity of the referent (cf. nine kids in the
example above).
Indefinite mass NPs are analogous to count NPs. They can be used for quantification
iff they contain a specification of quantity and the context allows for a partitive
interpretation. An example would be:
(19) Mary:
John:
‘Did you have enough cake?’
‘Oh yes, there’s even some cake left.’
The indefinite pronouns somebody, none, etc, are like indefinite NPs with a quantity
specification (in fact their form tells that they originated from such NPs). They can be
used in a partitive construction and therefore for quantification if the context allows
for this interpretation, for example as someone of us, none of the cake, etc. They also
have, of course, non-quantificational uses as in There’s somebody at the door.
4.5.3 Quantification and definite NPs
Unlike indefinite NPs, definite NPs never quantify. This is obvious with singular
count definites:
(20) The kid will be brought.
The definite subject NP refers to a single case and the predication – in this case ‘x will
be brought’ – applies directly to this single case.
Meaning and context
87
Very similar are so-called collective plurals. Actually, collective plurals are not
a special variant of plural, but a variant of predication; I therefore prefer the term
collective predication. A collective predication is applied to a complex object; it can
be applied only to the whole as such, not to the elements that make it up. Examples
are (21a, b) with plural NPs and (21c) with a mass NP:
(21) a.
The kids know each other.
b.
The kids gather in front of the school.
c.
The literature on the topic is very diverse.
A single kid cannot gather, or know each other, a single item of literature cannot
be diverse. These are predications which can apply only to complex things. Thus,
such sentences, too, are predications about just one case, the complex referent of the
definite plural or mass NP.
The case of plural and mass definites is less obvious with distributive predication.
Distributive predication, too, applies to a complex object. However, it is ‘distributed’
over the individual parts that make it up; if it is true, it is true for every single part,
and thereby for the whole in general. (22a,b) are examples:
(22) a.
b.
The kids will be brought.
The literature on the topic is in Russian.
(22a) is true if for each of the kids the predication ‘x will be brought’ is true. Some, or
all of them may be brought together, but this is left open here, and does not matter.
Likewise, (22b) says that every single item of literature on the topic is written in
Russian.
Crucially, distributive predication, though about a multitude of single cases,
is essentially one predication about the complex object. This can be seen if one
considers such sentences together with their respective negation. There is a simple
method for determining the negation of a sentence: transform the sentence into a
yes–no question, answer it with ‘No’ and form the sentence which exactly expresses
what the negative answer says. Applied to (22a, b), this yields the following:
(23) a.
b.
‘Will the kids be brought?’ – ‘No.’
Negation: The kids will not be brought.
‘Is the literature on the topic in Russian?’ – ‘No.’
Negation: The literature on the topic is not in Russian.
Obviously, the negation is achieved by negating the VP, i.e. the predication of the
sentence. Now, if you look at the negations, you realize that they constitute the same
kind of predication: distributing down to the single parts that make up the whole.
The point of distributive predication is therefore whether or not it applies uniformly
to all parts of the whole. Therefore, distributive predication and its negation form an
88
Understanding semantics
all-or-nothing contrast: either the kids will all be equally brought, or all equally not,
analogously for the Russian or non-Russian literature. Treating all elements of the
complex referent uniformly makes them form a unit, a single complex case. If the
parts of the whole are not uniform, distributive predication does not make sense. If,
for example, some of the kids will be brought and some will not, we can say neither
(22a) nor its negation (23a).
To sum up, sentences with predications about the referent of a definite NP always
constitute a predication about just one case. The case is complex if the definite NP is
mass or plural, but it is still just one case. Distributive predication carries down to the
elements of the complex referent, but it carries down uniformly and therefore, too, it
treats the complex referent as one uniform whole. For this reason, predications about
the referent of a definite NP are never quantifications.
If the definite NP is mass or plural, and the predication distributive, the
sentence shares with instances of quantification the reference to a complex domain.
But it differs crucially from quantification in that it excludes the possibility of
positive and negative cases in the domain. This very possibility, however, is what
quantification is all about. Every single case in the domain is to be considered
individually; quantification always involves parallel predications with potentially
different outcomes.
Table 4.9 summarizes the results of the discussion of quantifying, indefinite and
definite NPs.
Table 4.9
Quantificational and non-quantificational NPs
NPs of the form …
quantify …
every, each, all, both N
(genuine quantifying NPs)
… always
indefinite NP with
quantity specification
… if construed as partititve
simple indefinite NP
(a(n) N, bare mass N, bare plural)
… never
definite NP
… never
4.5.4 Quantification and negation
With quantifying NPs, negation yields a different picture than with definite NPs. The
negation of (24a) is (24b):
(24) a.
b.
Each kid will be brought.
Negation: Will each kid be brought? – No. Not each kid will be brought.
Meaning and context
89
The negation is not placed at the verb, but at the quantifying determiner (at least
12
it can be placed there). The result is an all-or-not-all contrast. Unlike the all-ornothing contrast we observed with (22), it covers all possibilities. The kids in the
DoQ are either all brought (24a) or not all (24b); the mixed case of some kids being
brought and some not is also covered by the negation. Recall that it is covered neither
by (22a) – the kids will be brought – nor by its negation (22b) if quantification is
absent.
The quantifiers every, all, and both behave the same way as each. With quantifying
indefinite NPs, we also get a contrast that covers all possibilities. For example, the
negation of a quantifying NP of the form ‘some N’ is formed by replacing some with
no:
(25) a.
b.
Some kids will be brought.
Negation: No kids will be brought.
Here, the mixed case is covered by the positive sentence. Figure 4.2 displays the
different contrasts under negation for sentences with definite NPs and quantifying
NPs. Quantifying sentences and their negations together cover the whole range of
possibilities, while sentences with a definite plural or mass NP leave a gap, due to the
all-or-nothing contrast between global truth and global falsity. These observations
explain the function of quantification: it serves to fill this gap which arises with
simple predication over complex referents.
Figure 4.2
Negation contrasts for definite and quantifying NPs
negative cases only
mixed cases
positive cases only
the N not
the N
not every N
no Ns
every N
some Ns
For sentences with a definite subject NP, negation is formed by negating the VP,
as you saw in (22) and (23) above. For sentences with a quantifying subject NP, VP
negation yields not the negation of the sentence, but what is called internal negation
as opposed to external negation in (24b) and (25b).
12 There is also the syntactic variant EACH kid will NOT be brought, with a special intonation contour
on each and not, which serves to indicate that the negation relates to each rather than to the verb,
where it is syntactically placed.
90
Understanding semantics
(26) a.
b.
Each kid will not be brought. (= No kid will be brought.)
Some kids will not be brought. (= Not all kids will be brought.)
Sentences with a definite subject NP do not offer the possibility of two ways of
negation. VP negation is the only negation available and it yields the negation of the
sentence as a whole.
4.5.5 The nominal onion
The discussion of determination shows that an NP may have several layers of
structure; it forms an onion, metaphorically speaking. The core is formed by the
noun; next to the core are adjectives. The next layer consists of quantity specifications
such as numerals, several, many, few, etc. To these, determination is added next; this
layer comprises definite and indefinite articles, demonstratives and possessive
determiners. The outermost layer is constituted by quantifiers such as all, every, each
and both. Most quantifiers do not combine with an article; all is an exception which
shows that quantification supersedes determination. Figure 4.3 displays the nominal
onion with an example and with the elements belonging to the different layers.
Figure 4.3
The nominal onion
all
(
quantifier
all
every
both
the
(
determiner
the
a
sm
this / that
my / etc.
three
(
quantity
blind
mice
( adjective ( noun ) ) )
) )
one, two, …
several
some
much, many
no
…
4.6 GENERIC NPs
4.6.1 Generic sentences with simple indefinites
There is a special use of NPs, where they do not serve concrete reference but the
expression of general conditions. Examples would be the simple indefinite NPs in
sentences like these:
(27) a.
A millipede has a thousand legs.
b.
An icosahedron consists of twenty triangles.
c.
Catholic priests are not allowed to marry.
Meaning and context
d.
Bare mass nouns are indefinite.
e.
Money does not buy happiness.
91
Using such sentences, one does not refer to things in a concrete CoU, but one talks
in general terms about, for example, millipedes, icosahedrons, Catholic priests,
bare mass nouns or money. Because of their ‘general’ character, such sentences, and
the NPs in them, are called generic. Non-generic sentences are called episodic, or
particular. Let us consider the following dialogue between Mary and John about
their neighbour Gerald, a Catholic priest; (28a) is episodic, (28b) generic.
(28) a.
Mary:
‘Just imagine, Gerald is going to get married in October.’
b.
John:
‘Catholic priests aren’t allowed to marry, are they?’
John replies with this generic sentence, because it states a general condition relevant
for Gerald in the given CoU. But using this sentence he is not explicitly referring to
Gerald, as he would if he said, for example, ‘As a Catholic priest, Gerald isn’t allowed
to marry, is he?’
Talking in generic sentences is best understood as a special mode of speech in
which elements of a sentence are used without concrete reference in the CoU. By
doing so, a certain ‘general’ quality of the sentence is accomplished. For generic
elements in a sentence, utterance meaning does not depend on the CoU – unlike the
utterance meaning of referring NPs. There may be something like concrete cases in
the CoU to which a generic statement applies, but these cases are not directly referred
to. The possible cases of application are all treated alike. This is what leads to the
blanket character of generic sentences.
In the semantic literature, generic indefinite NPs are mostly analysed as universal
quantifiers. According to this view, Millipedes have a thousand legs would essentially
mean ›all millipedes have a thousand legs‹. This analysis is not correct. To see this,
consider again the contrast between such a sentence and its negation. The negation
of sentences with generic indefinites is formed by negating the VP of the sentence.
The sentences (29a, b, d) are the negations of the respective sentences in (27), while
the negative sentences (27c, e) are the negations of (29c, e):
(29) a.
A millipede does not have a thousand legs.
b.
An icosahedron does not consist of twenty triangles.
c.
Catholic priests are allowed to marry.
d.
Bare mass nouns are not indefinite.
e.
Money buys happiness.
Again, the sentences and their respective negations form an all-or-nothing contrast.
Consequently, generic indefinites cannot be considered universal quantifiers since
universal quantifiers would yield an all-or-not-all contrast. Thus, the statement in
92
Understanding semantics
Table 4.9, that simple indefinite NPs do not quantify, carries over to simple indefinite
NPs in generic use.
4.6.2 Generic quantification
Quantification is also possible in generic sentences – if one uses explicit quantifiers.
The quantification in generic sentences does not refer to a domain of concrete cases
given in the CoU, but to this kind of case in general. Examples would be sentences
like:
(30) a.
Some think money does not buy happiness.
b.
No millipede really has a thousand legs.
c.
Every Catholic priest can marry.
The respective negations are:
(31) a.
Nobody thinks money does not buy happiness.
b.
Some millipedes really have a thousand legs.
c.
Not every Catholic priest can marry.
In each case, the contrast between the positive and the negative is not all-or-nothing,
but covers all possibilities. Like with episodic quantification, there is the possibility
of internal and external negation.
Indefinite NPs with a quantity specification usually do not yield quantification in
generic use:
(32) a.
Two heads are better than one.
b.
Two’s a company, three’s a crowd.
c.
Too many cooks spoil the broth.
d.
Many children only have one parent.
The first three sentences are not construed as quantifications (over heads, people or
cooks), but as generic statements about cases of two heads vs one head, of two people,
or three, and of too many cooks working on the same broth. Only generic indefinite
NPs with vague quantity specifications such as many, much, more, most, only a few,
etc. may have a quantifying reading. (Note that many in (32d) is quantificational,
while too many in (32c) is not.) The quantificational readings concern the relative
frequency of the cases described, among the totality of cases of the kind. Therefore,
such statements can equivalently be made using adverbs of frequency:
(33) Adverbial generic quantification:
Children often/rarely/frequently/mostly/always/never have one parent.
Meaning and context
93
In principle, the possibility of expressing quantification by means of adverbs instead
of quantificational NPs is also given for episodic quantification. For example, the
quantificational sentence in (15c) can be equivalently phrased with quantifiers in
adverb position:
(34) Adverbial episodic quantification:
The kids are each to get a present.
(= Each kid is to get a present.)
The kids are all to get a present.
(= All the kids are to get a present.)
The adverbial variants of quantification in (33) and (34) display two points
of difference between episodic and generic quantification. First, only episodic
quantification involves reference to the totality of cases, namely the DoQ. Note that
the DoQ is directly referred to with a definite NP (the kids) in (34), while for generic
quantification the kind of cases related to is denoted by a generic plural NP (children)
in (33). Second, the choice of relevant adverbs shows that only generic quantification
is in terms of the frequency of relevant cases among the kind of cases related to.
Stopover
The determination of an NP determines the way in which the NP refers (if
it refers). The central phenomenon is definiteness. NPs can be definite or
indefinite or quantifying. Quantifying NPs represent a semantically complex
mode of expression: episodic NP quantification combines definite reference to
the DoQ with information about the proportion, or number, of cases within it
for which the predication is true. If you like, episodic NP quantification can be
considered a combination of simultaneous definite reference to the DoQ and
indefinite reference to the positive cases within it.
By no means all languages have quantifying determiners like English every.
The majority of languages use adverbial quantifiers for the expression of
quantification.
Generic NPs cannot be straightforwardly aligned with indefinite or definite
reference. All the English examples we discussed above are indefinite NPs,
but the expression of genericity by indefinite NPs is not representative across
languages. Many, if not most, languages would use definite NPs for this purpose.
Chinese, for example, which does not mark definiteness explicitly, uses the
same constructions for generic NPs as for definite NPs. The heterogeneity of
the grammatical determination of generic NPs can be explained by the fact
that generic NPs exhibit traits of both definite and indefinite reference. They
share with definite reference that what they relate to is given independently of
the CoU; with indefinite reference they have in common that they are not about
a particular single case.
94
Understanding semantics
PART 3: PRESUPPOSITIONS
4.7 PRESUPPOSITIONS
We will now turn to the third part of the phenomenon of indexicality –
presuppositions. It concerns indexical relation to the facts that make up the CoU.
The facts that are potentially relevant in a given CoU are much more manifold and
complex than the other components of a CoU (i.e. speaker, addressee(s), time and
place). Correspondingly, the relevant linguistic means are much richer. There is a
connection to the matter of determination: we will once again have to deal with
definite NPs.
4.7.1 An example
By way of illustration, let us consider the following short text to be read coherently:
(35) a.
Tony rose from the breakfast table.
b.
He went into his room and opened the window.
c.
He was sorry that he had not replied to Lucy.
Not only are we able to construe these sentences coherently, we also construct a
possible CoU that this passage would fit into. We will assume that in this context
there are two persons named Tony and Lucy, respectively. Tony has a room; the room
has a window; the window was closed until Tony entered the room to open it; and so
on. How are we led to these assumptions? This will become clearer if we take a closer
look at the wording of the text.
a. Tony rose from the breakfast table.
The sentence contains the name Tony as subject NP. As definite NPs express unique
reference, there must be a uniquely determined something called ‘Tony’ in the CoU.
There are many things which can bear the name Tony – male and female persons,
pets, cuddly toys, boats or songs; however, in this first sentence, it must be a person,
since only a person is capable of ‘rising from the breakfast table’. Later we are led to
assume that Tony is a male person because we will relate the anaphoric masculine
pronoun he to Tony; but from the sentence given we cannot tell the sex of the
bearer of the name. Turning to the VP rose from the breakfast table, we obtain more
information about the CoU. The definite NP the breakfast table informs us that there
is a uniquely determined table where people had breakfast; this, in turn, provides
information about the time of the day. Also, Tony must have been sitting at the table;
otherwise he could not be said to have risen from it.
All this information is not made explicit in the sentence. It does not state: ‘There
is somebody with the name of “Tony”; it is morning; there is a table for breakfast;
Meaning and context
95
Tony has been sitting at that table.’ The statement proper only states that Tony
rose from the breakfast table. The information about the CoU which we can read
off (or reconstruct from) the sentence is conveyed subliminally, as it were, by
certain elements of the sentence. They presuppose conditions that must be given
beforehand. With these implicit presuppositions, the sentence inconspicuously links
to the conditions in the given CoU. If these conditions were not given, uttering the
sentence would not make sense; it would be pointless. These presuppositions are
also technically termed thus:
DEFINITION 6 Presuppositions
Presuppositions of a sentence are those conditions pertaining to the CoU
which it must meet in order for an utterance of the sentence to make sense in
that CoU.
Presuppositions are part of the meaning of a sentence. When a proper name is used
as the subject or object of a sentence, the sentence presupposes that in the given CoU
there is a uniquely determined something which carries this name; to rise, related to
persons, presupposes that the person was not standing before, etc. The following two
sentences contain more cases of presuppositions.
b. He went into his room and opened the window.
The personal pronoun he is definite and therefore presupposes that it refers to a
male person, uniquely determined by this description. Analogously, the possessive
pronoun his presupposes that it refers to a male possessor uniquely determined by
using his. The NP his room is used referentially here and is therefore construed as
definite (recall 4.4.5); thus it presupposes that the description his room uniquely
singles out a particular room as its referent. It is said that ‘he’ went into his room. This
presupposes that ‘he’ was not in this room before. Opened the window contributes
two more presuppositions: the NP the window must uniquely refer to a particular
window, and this window was not open before.
One will construe the text automatically in the way that ‘he’ is Tony, that ‘his’ room
is Tony’s room, and that the window he opened is in that room. This interpretation
suggests itself if we interpret the second sentence in the context of the first. It is,
however, not semantically cogent. For example, if we knew that Tony is a woman,
we would not interpret he as coreferent with Tony; also the owner of the room could
be different from the one who is said to enter it. Similarly, the sentence does not say
that the window is in the room mentioned. We establish these connections within
the sentence and between the sentences by following general strategies of creating
coherence. Such coherence relations do not derive from the presuppositions of
the sentences. Presuppositions are general conditions that arise from the expression
meaning of the sentence. They cannot comprise conditions that are only given in
special CoUs.
96
Understanding semantics
c. He was sorry that he had not replied to Lucy.
The two occurrences of he make the same semantic contribution. Again, they
need not cogently be referring to the same person. The construction be sorry that
presupposes that what is described in the that clause is a fact: one cannot be sorry
about things that did not really happen. Therefore, the construction presupposes
that ‘he’ had not replied to Lucy. This in turn presupposes that there is something
uniquely determined by the name Lucy; this Lucy must have taken some action
‘he’ could have replied to, e.g. asked a question. For the sake of simplicity, we will
assume that this, together with the requirements of the verb reply to, amounts to the
presupposition that Lucy is a person.
4.7.2 Presupposition carriers
13
The expressions and constructions that carry presuppositions, can be grouped. It
would take us too far afield if we were to discuss all known kinds of presupposition
carriers. We will confine ourselves here to definite NPs and certain classes of verbs.
4.7.2.1 Definite NPs
When definite NPs are used for reference, they generally carry the presupposition that
the description of the referent thereby given is unique. That means first of all that a
referent exists in the given CoU (existential presupposition), but in addition also that
the given description is unique (uniqueness presupposition). This holds for definite
descriptions as well as for all other types of definite NPs: proper names, personal
pronouns, NPs with adnominal demonstratives and referential possessive NPs; Table
4.10 lists the different types of definite NPs and the presuppositions they carry.
Analogous presuppositions can be stated for the pronouns you (plural) and they,
as well as for the other possessive pronouns besides my.
Episodic quantifying NPs carry the same presuppositions as a definite NP referring
to the DoQ (cf.4.5.2); thus, each child has the same presupposition as the children.
4.7.2.2 Verbs
The example in (35) contained several verbs that carry presuppositions of different
types. For one, every verb imposes certain conditions on the things and persons it
predicates about. We saw in the examples that Tony must be a person; otherwise
he could not be said to ‘rise from the breakfast table’. At closer inspection, every
verb comes with such conditions for the referents of its subject, object and other
complements. The verb reply is restricted to persons; go is rather unspecific as to the
13 In the literature you will often find the term presupposition trigger for what is here called
presupposition carrier. The notion of a ‘trigger’ of presuppositions derives from the idea that
presuppositions are some kind of inference drawn when certain utterances are produced and
interpreted. This is not the understanding of presuppositions given here. Presuppositions are just
part of the meaning of the respective expressions or constructions, not something to be derived by
inference.
Meaning and context
97
Table 4.10
Definite NPs and their presuppositions
Definite NP
Presupposition
proper name
There is somebody or something which is uniquely determined by carrying this
name.
the dog
There is a dog and it is uniquely determined which dog is referred to.
this apple
There is an apple which is referred to and it is uniquely determined by deixis which
apple is referred to.
my skirt
There is a skirt which belongs to the speaker and it is uniquely determined which of
the speaker’s skirts is referred to.
I
There is somebody who produces this utterance and it is uniquely determined who
it is.
you (singular reading)
There is a person to whom this utterance is addressed and it is uniquely determined
who it is.
he
There is a male person referred to and it is uniquely determined who it is.
we
There are persons referred to, among them the speaker, and it is uniquely
determined who they are.
kind of referent it requires for the subject, but in the construction ‘go to one’s room’,
it is restricted to persons, too. These kinds of presupposition are called selectional
restrictions. They will be discussed in more depth in 5.7.
The verb open is representative for a large number of verbs used to express a
certain state coming about. ‘To open the window’ means to do something which
leads to the result that the window is open. Thus, on the part of the window the verb
denotes a transition from the window not being open to its being open. All such
verbs of change presuppose that the resulting state did not apply before. You cannot
open something which is already open, and so on. Table 4.11 lists examples.
Table 4.11
Verbs and their presuppositions
Verb
Presupposition
she opened the door
before, the door was not open
she left the kitchen
before, she was in the kitchen
she went into her room
before, she was not in her room
she lay down
before, she was not lying down
she became calmer
before, she was less calm
she stopped musing
before, she was musing
she picked up her guitar
before, she was not holding her guitar
she had a glass of water
before, she had not had this water
98
Understanding semantics
Some verbs with a that clause complement presuppose what is said in the that clause.
An example occurred in (35): He was sorry that he had not replied to Lucy. Such verbs
are called factive verbs. They include among many others regret, admit, report,
recognize and know.
(36) Tony regrets/admits/recognizes/knows that Lucy does not want to see him.
presupposes: Lucy does not want to see him.
By presupposing the content of the that clause, constructions with factive verbs also
presuppose whatever the that clause itself presupposes.
Certainly not all verbs with a that clause complement presuppose its content. For
example, say, think, hope, expect, fear, claim, plan, request, etc. do not:
(37) Tony says/believes/expects/fears/claims that Lucy does not want to see him.
does not presuppose: Lucy does not want to see him.
There is a further condition, which we already encountered in connection with
distributive predication in 4.5.3, namely the condition that the parts of the whole
to which a distributive predication is applied form a uniform whole – i.e. that
the predication is either uniformly true or uniformly false for all parts. Since the
character of a predication in most cases depends on the verb, we may mention this
presupposition here. It is called the presupposition of indivisibility and it requires
that the object of a distributive predication is uniform, or homogeneous, with
respect to what the predication says about it. For collective predications or simple
predications about just one case, this condition holds anyway. It can therefore be
considered a general presupposition of any predication whatsoever. The formulation
uses the technical term argument (see chapter 5) for the object of a predication:
DEFINITION 7 Presupposition of Indivisibility
The argument of a predication is indivisible with respect to the predication; a
predication is true or false of its argument as a whole.
As we have seen in the discussion of (22) and (23) above, it is this condition which
is responsible for the fact that a distributive predication and its negation form an
all-or-nothing contrast.
4.7.3 Presuppositions and truth conditions
Presuppositions immediately bear on the issue if a sentence is true or false. As we have
already stated in 1.1.2, this question arises only when a sentence is actually uttered
in a CoU. When uttered, the presuppositions must be given in the CoU. Otherwise
the question as to whether the sentence is true or false cannot be answered. Let us
consider a simple example to illustrate the point:
Meaning and context
(38) a.
99
The dog stopped barking.
The sentence carries the presupposition of existence and uniqueness of the referent
of the dog. If there is no dog in the given context, or if there are several dogs and
the dog referred to is not uniquely determined, it cannot be decided whether the
sentence is true or false. One just would not know which dog is being talked about.
If we assume that the presuppositions for the dog are fulfilled, there remains the
presupposition contributed by stopped barking: the dog has been barking. If the dog
had not been barking, the question as to whether it had stopped doing so or not
would not arise, i.e. whether the sentence is true or false.
The same holds for the negation of the sentence:
(38) b.
The dog did not stop barking.
The negation of the sentence carries the same presuppositions: the sentence refers
to a uniquely determined dog, and this dog has been barking. For if it had not been
barking, there would be no point in saying that it did not stop barking.
The same presuppositions also hold for the corresponding yes–no question:
(38) c.
Did the dog stop barking?
Only if (38a) is true, can the question be answered ‘Yes’; only if (38b) is true, can it
be answered ‘No’.
This shows that for declarative sentences presuppositions are necessary conditions
for the sentence having a truth value at all in a given CoU. We have, therefore, to
modify the definition of truth conditions we gave in 2.2.2 in order to take account of
presuppositions.
DEFINITION 8 Truth conditions (revised)
The truth conditions of a sentence are the general conditions under which it is
true, provided that all presuppositions of the sentence are fulfilled.
The presuppositions are not themselves part of the truth conditions, but exist prior
to them. This is clear from the fact that they also hold when the sentence is false.
In the literature, presuppositions are often treated as some sort of inference,
14
similar to logical entailment (see 5.3.1) or to so-called conversational implicatures.
However, this is a confusion of cause and effect. Presuppositions are conditions
14 Conversational implicatures are a central notion in Grice’s theory of language use (mentioned
in 1.1.2). According to Grice, people engaged in a conversation will normally assume that their
interlocutors adhere to basic social rules of conversation, such as telling the truth, being relevant,
being economical, etc. Conversational implicatures are inferences drawn on the basis of this
assumption. For instance, we will normally believe what a speaker says because we assume he or
she tells the truth.
100
Understanding semantics
on the CoU. If someone utters a sentence in a concrete CoU, his interlocutors will
assume that the utterance is meant to make sense in the given CoU and that the
presuppositions therefore are to be taken as given. Unless there is evidence to the
contrary, the addressee(s) will assume that this is the case. Thus, the presuppositions
of a sentence will normally cause an inference to the effect that they are fulfilled in
the given CoU.
4.7.4 Presupposition tests
These considerations lead to two tests for deciding whether a condition is a
presupposition or not. The negation test can be used to check whether the condition
also holds for the negated (or un-negated sentence). The question test checks if the
same condition holds for the corresponding yes–no question. If the result is positive,
the condition is a presupposition. The three sentences in (38a, b, c) illustrate the two
tests for the sentence the dog stopped barking. Let us return for the last time to the
examples in (35).
(35) a.
sentence:
Tony rose from the breakfast table.
negation:
Tony did not rise from the breakfast table.
question:
Did Tony rise from the breakfast table?
Obviously the negation and the question carry the same presuppositions as we stated
for the original sentence.
In the case of the second sentence – he went to his room and opened the window –
the tests would be applied separately to the two sub-statements it contains: he went
to his room and he opened the window. Again, it turns out that the negations and
corresponding questions carry the same presuppositions as were stated above for
the positive sentence.
For the third sentence, one needs to observe that forming the negation and the
question leaves the that clause untouched:
(35) c.
sentence:
He was sorry that he had not replied to Lucy.
negation:
He was not sorry that he had not replied to Lucy.
question:
Was he sorry that he had not replied to Lucy?
The negation test and the question test can both only be used for declarative
sentences. But this is sufficient for identifying most presupposition carriers.
4.7.5 Presuppositions and the Principle of Consistent Interpretation
The Principle of Consistent Interpretation (3.4.4) requires that an utterance fit its
context. Part of this requirement is that the presuppositions of an utterance be
fulfilled in the given CoU because otherwise the utterance would not make sense. The
Principle has two consequences concerning presuppositions. First, it may happen
Meaning and context
101
that the context as it has been understood up to the time when a sentence is uttered
did not positively support its presuppositions. In this case, very often so-called
accommodation takes place: the context is adapted to the utterance in order to do
justice to the presuppositions. Second, the requirement that certain presuppositions
be fulfilled may trigger meaning shifts.
Accommodation occurs all the time. It is such a common thing to do that we often
exploit the mechanism in conversation. For example, assume Sheila is just about
to leave home for town and Mary asks her, ‘Would you post my letter?’ Sheila will
accept the presupposition that her mother has a letter to be posted, even if she had
not known about that letter; otherwise she would have to assume that her mother’s
request is pointless. Mary would have relied on this inference by Sheila, which saves
her an explanation.
To see the key role of presuppositions in triggering meaning shifts, have a second
look at the examples for metaphor, metonymy and differentiation that we considered
in 3.4.3. Let me give just three examples:
(39) a.
Metonymy:
Moscow declared the Chechen rebels defeated.
b.
Differentiation:
She had forgotten to charge her card.
c.
Metaphor:
His courage evaporated.
The selectional restrictions of the verb declare for its subject referent ask for a human
being. In addition, the whole VP declared the Chechen rebels defeated requires some
authority. This triggers the metonymic shift from the original meaning ›(city of)
Moscow‹ to ›Russian government (residing in Moscow)‹. In the second sentence, the
verb requires a chargeable type of card as the object referent, whence the general
meaning of card will be shifted to a differentiated meaning, say ›SIM card‹ or ›money
card‹. In (39c), too, there is a mismatch between the type of subject referent and the
selectional restrictions of the verb. In this case however, the mismatch is dissolved
by giving the verb a metaphorical reading, approximately ›vanish‹ with selectional
restrictions that fit the given type of referent.
4.8 SUMMARY
We have come a long way in this chapter, starting from deixis and demonstratives,
visiting definiteness, quantification and genericity, to finally arrive at presuppositions.
What unites all these phenomena is that they belong to the interface between
sentence meaning and utterance meaning. The deictic means relate to the immediate
context of utterance for establishing reference out of it. The most important linguistic
means of reference are NPs. NPs are subject to determination; in most cases, they
are either definite or indefinite. Definite NPs are indexical in that they depend on
the given context for determining their reference. Likewise all episodic quantifying
NPs are indexical in referring to a uniquely determined domain of quantification. In
turn, definite NPs are a special case of presupposition carriers. Presuppositions in
102
Understanding semantics
general are indicative of the context of the utterance. By imposing conditions on the
CoU, they indicate what kind of context is required for the sentence to make sense at
all. Presupposition carriers not only serve to establish reference; more generally they
serve to integrate the content of a sentence into the context in which it is used.
EXERCISES
On deixis
1. Consult a Spanish grammar or textbook for the system of personal pronouns. Set
up a table like Table 4.1.
2. What do personal and possessive pronouns have in common, and what is the
difference between them?
3. Use WALS (http://wals.info/) in order to find three languages on different
continents which distinguish inclusive and exclusive WE. Use the Ethnologue
website (http://www.ethnologue.com/) to classify the three languages. For
example, the classification for English is: Indo-European, Germanic, West.
4. Consult Siewierska (2004) to find two languages which do not have 3rd person
pronouns. Use Ethnologue to classify them.
5. Try to find out if there are formal pronouns of address in Finnish and, if so,
determine which of the strategies mentioned in 4.1.2 is used in Finnish. Try to
find out what the rules are for using the formal type of address (interview a native
speaker, if possible).
6. Use the Oxford online dictionary (http://oxforddictionaries.com/) to determine
if, according to the definitions given there, the following nouns are relational or
not: spouse, bed, frog. If more than one definition is given, answer the question for
each of them separately. The answers may differ.
7. Discuss with fellow students the space deictic quality of the verbs bring and fetch.
Give a tentative written description.
8. a. It is very rare to find different feminine and masculine 1st person singular
pronouns, while a difference in gender for 2nd person singular pronouns is
more frequent. What might be a reason for the difference?
b. In languages with 1st person pronouns that differ in social meaning, there
may be 1st person pronouns exclusively used by women. Think of a possible
reason.
On determination
9. Explain the function of the definite article in your own words.
Meaning and context
103
10. Consult WALS to find two languages, each from different continents, which
a. have definite articles but no indefinite articles,
b. vice versa.
Use Ethnologue to classify them.
11. In the following text
a. determine which NPs are definite
b. for the definite descriptions among them, which are pragmatically unique,
and which semantically unique?
The Mole had been working very hard all the morning, spring-cleaning his
little home. First with brooms, then with dusters; then on ladders and steps
and chairs, with a brush and a pail of whitewash; till he had dust in his throat
and eyes, and splashes of whitewash all over his black fur, and an aching back
and weary arms. Spring was moving in the air above and in the earth below
[…] (from Kenneth Grahame, The Wind in the Willows, Amazon Kindle
edition, originally published by Charles Scribner’s Sons, New York, 1913)
12. a. Determine all the indefinite NPs in the previous text. Are there quantifying
NPs among them?
b. Identify all the nouns in the text and determine whether they are count nouns
or mass nouns.
13. Determine which NPs in the following text (a passage from chapter 10.1) are
indefinite. Decide which ones are quantificational and which ones are not.
English has a count noun wood1 and a mass noun wood2. A wood1 consists
of trees, while wood2 is the substance trees largely consist of. Hence, the two
meanings can be clearly distinguished on linguistic grounds. Likewise, the
Japanese meanings of ki1 ›tree‹ and ki2 ›wood2‹ are clearly different. It follows
that neither of the two English expressions matches with either of the two
Japanese words. They only match in one meaning variant, respectively.
14. In the following text (a passage from chapter 3), determine which indefinite NPs
are generic and which are not.
In dealing with different aspects of meaning in the previous two chapters,
expressions were treated as though they had only one meaning (though possibly
composed of different parts). This is, of course, not the case. Many words have more
than one meaning and even complete sentences may allow for several readings.
On presuppositions
15. Consider the sentence Nicola left her bike in front of the supermarket. Which
among the sentences a. to e. state a presupposition of the sentence?
a. There is a uniquely determined person called ‘Nicola’.
b. There is a uniquely determined bike which belongs to a uniquely determined
female person.
104
Understanding semantics
c. The bike belongs to Nicola.
d. The owner of the bike had been riding on it before it was left at the
supermarket.
e. After Nicola left it there, the bike was in front of the supermarket.
16. Try to identify the presuppositions involved in the following three sentences:
(a) The father got back to the hospital after midnight. (b) The wife was sitting
in the chair by the bed. (c) She looked up at the husband and then she looked
back at the child.
(from Raymond Carver, ‘The Bath’, in What We Talk About When We Talk About
Love, p. 41, London: Vintage, 2003)
17. What does the presupposition of indivisibility mean for the following sentence?
The eggs are raw.
FURTHER READING
Deixis
Anderson and Keenan (1985) and Diessel (2012) provide cross-linguistic surveys.
On deixis and anaphora in English see chapter 17 by Lesley Stirling and Rodney
Huddleston in Huddleston and Pullum (2002).
4.1 Siewierska (2004) is a comprehensive survey on the topic of grammatical person:
ch. 3 on person deictic systems, ch. 6 from the perspective of social meaning (‘social
deixis’). Cooke (1968) on terms of address in Thai, Burmese and Vietnamese. In the
World Atlas of Language Structures online (WALS) you will find surveys on systems
of personal pronouns in several hundred languages: ch. 44 (Siewierska) on gender
distinctions; ch. 35 (Daniel) on plurality; ch. 39 (Cysouw) on the inclusive/exclusive
distinction; ch. 45 (Helmbrecht) on politeness distinctions; ch. 43 (Bhat) on 3rd
person pronouns and demonstratives. For the more general aspects: ch. 36 (Daniel,
Moravcsik) on associative plural; ch. 31, 32 (Corbett) on gender systems; ch. 33
(Dryer) and ch. 34 (Haspelmath) on plural marking; on inalienable possession see
chs 58 (Bickel and Nichols) and 59 (Nichols and Bickel).
4.2 Diessel (1999) is a general typological overview on demonstratives; there are
chapters in WALS by the same author on distance contrasts in demonstratives
(ch. 41) and on pronominal and adnominal demonstratives (ch. 42). On place deixis,
see Levinson (1983: 2.2.3).
4.3 Levinson (1983: 2.2.2) on time deixis.
Determination
For determination and NPs in English, in particular definiteness, quantification, the
mass/count distinction, genericity, number and gender see the elaborative account
Meaning and context
105
provided in ch. 5 by John Payne and Rodney Huddleston in Huddleston and Pullum
(2002).
4.4 Löbner (1985) on definiteness. WALS chs 37 and 38 (both by Dryer) on definite
and indefinite articles.
4.5 There is a lot of literature from the field of formal semantics on quantification
(see ch. 13). Most of the literature adopts the approach of Generalized Quantifier
Theory which subsumes indefinite and definite NPs under quantifiers.
4.6 Carlson (2012) on genericity.
Presuppositions
Seuren (1990) on presuppositions, Levinson (1983: 4.2) on presupposition ‘triggers’
(= carriers).
5
?
Predication
We will now have a look at the way in which different kinds of words interact to
form meaningful phrases and sentences, addressing in more detail the mechanism
of composition (1.2). We will see that a phrase or a sentence is by no means a simple
sequence of words, one added to the other like beads on a string. Rather, sentences
have a sophisticated structure in which each part plays its own role and interacts
with the others in its own way. Here, we will focus on one central semantic property
of verbs, nouns and adjectives: the way they provide predications about one or more
of the potential referents of the sentence.
5.1 PREDICATIONS CONTAINED IN A SENTENCE
Let us take a look at the following example, a simple English sentence, and try to
determine the proposition it expresses.
(1)
Johnny sent money to a dubious company.
The first word, Johnny, is a proper name. It forms a complete NP. The VP contains
a finite verb, i.e. a verb which carries tense and, in English, ‘agrees’ (see 5.6.2)
with the subject of the sentence. Here the subject is the NP Johnny and verb tense
is simple past. The bare mass noun money forms another NP, the so-called direct
object of the verb. The last three words, a dubious company, form a third NP, in this
case also containing an adjective between article and noun. This NP is part of a PP
(prepositional phrase) headed by the preposition to. The PP is the so-called indirect
object of the verb. The total VP consists of the finite verb, the direct object NP and the
indirect object PP. In Fig. 5.1 the sentence is analysed into its syntactic components.
Figure 5.1
Grammatical structure of the sentence in (1)
[ [JohnnyNP] [sentV [moneyN] [toP [a dubiousA companyN]] ] ]
NP
NP
NP
Sentence
VP
PP
Predication
107
The three NPs each provide a description of one referent. The subject NP describes
its referent as an entity with the name Johnny, the direct object NP characterizes
its referent as ‘money’ and the indirect object NP provides the information that its
referent is a dubious company. Let us call the three referents rj, rm and rc, respectively.
The verb contributes the information that they participate in an event of sending,
rj being the sender, rc the recipient and rm the object sent. Thus the meaning of
the sentence constitutes complex information about these three referents. If an
expression provides information about a referent, it is said to make a ‘predication’
or to ‘predicate’ about it. In Table 5.1, it is sorted out which words contribute which
predications about which referents. For the time being, we disregard the event
referent of the verb (but see 5.3.2) and we do not treat the proper name Johnny as
contributing a predication. The preposition to is omitted from the list because it does
not contribute a predication. Similarly, articles would not be included.
Table 5.1
Predications contributed by the words in (1)
Word
Predication
sent
rj sent rm to rc
rm was sent to rc by rj
rc was sent rm by rj
money
rm is money
dubious
rc is dubious
company
rc is a company
The simple analysis shows two interesting things. First, while the nouns and the
adjective relate to only one of the referents, namely the referent of the NP they belong
to, the verb relates to all three referents. Second, different words in the sentence
contribute predications about the same referents. Figure 5.2 depicts these relations.
Figure 5.2
Predication structure of sentence (1)
words
NPs
referents
[ JohnnyNP] sentV [ moneyN] toP [ a dubiousA companyN]
rj
rm
rc
108
Understanding semantics
A broken line connects each NP to its referent; arrows point from the single words to
those referents they provide information about. The figure illustrates the central role
of the verb: by predicating about all three referents, the verb ties the whole sentence
together. The three NPs, in contrast, independently supply information about one
referent each.
5.2 PREDICATE TERMS AND ARGUMENT TERMS,
PREDICATES AND ARGUMENTS
Words that contribute a predication are called predicate terms.1 Predicate terms
can be verbs, nouns, adjectives or adverbs (like rapidly). Whether or not proper
names like Johnny constitute predicate terms is a controversial issue. Many formal
semanticists hold that they are just names, i.e. direct expressions for their referents.
Alternatively, proper names can be considered predicate terms that express the
property of ‘being called [Johnny etc.]’. Both positions have their merits. We will
adopt the first position here.
The meanings of predicate terms are concepts that concern one or more entities:
›money‹, ›dubious‹ and ›company‹ concern one entity each, the event concept ›send‹
concerns three. Such concepts are called predicates; the entities they concern are
their arguments. Predicates are ‘applied’ to their arguments: for example, in (1),
›dubious‹ is applied to rc and ›send‹ is applied to rj, rm and rc. Predicates with one
argument are called one-place predicates, predicates with two arguments two-place
predicates and so on. One-place predicates are also called ‘unary’ predicates, twoplace predicates ‘binary’, etc.; therefore the term ‘arity’ is used for the number of
arguments of a verb.
The arguments of a predication are those entities in the world to which the
predication is to be applied. In (1) the arguments of the predicate ›send‹ are
described by the three NPs of the sentence: the subject, the direct object and the
to-PP. These NPs are argument terms belonging to the verb – three constituents of
the sentence which are grammatically related to the verb. In this chapter, the term
complement will be used for all argument terms that form a separate syntactic
2
constituent and have their own referent. Figure 5.3 displays the verb and its three
complements.
1 In the literature, you will often find the term ‘predicate’ used indiscriminately for predicate terms
as well as for their meanings.
2 In syntactic approaches, a distinction is drawn between obligatory and optional argument terms.
Only the former are called complements, the latter ‘adjuncts’. A general definition of the distinction
is notoriously difficult and depends on the theory adopted. We will not tackle the problem here and
use the term complement in a wide sense that also includes optional argument specifications.
Predication
109
Figure 5.3
A predicate term and its complements
predicate term
Johnny
subject
sent
money
to a dubious company
direct object
to-PP
three argument terms (complements)
While predicate terms and argument terms both belong to the level of expressions,
predicates and arguments belong to different levels: predicates are concepts, and
arguments, being referents, are things in the world. Thus, predication always involves
three levels: (i) the level of linguistic expression: predicate terms and argument
terms; (ii) the conceptual level of the meanings of the predicate terms and argument
terms involved; (iii) the referential level of arguments. Figure 5.4 shows the three
levels involved with the predication expressed by a verb; the scheme basically adds
the conceptual level to the diagram in Fig. 5.2. The double horizontal line separates
the level of concepts from the level of the world. The simple vertical lines connect
the expressions in the first line to their meaning in the second; dashed lines connect
these concepts to the referents of the NPs; the thick arrows mark the relation between
the verb predicate and its arguments.
Figure 5.4
Levels involved in a predication by means of a verb
the company
kept
the money
expression level
argument term
NP
predicate term
verb
argument term
NP
conceptual level
argument description
›company‹
predicate
›keep‹
argument description
›money‹
referential level
argument
the company
argument
the money
110
Understanding semantics
Analysing predication by means of NPs that function as argument terms calls for a
different picture (Fig. 5.5). In this case, the argument of the predication expressed by
the noun is not described in a separate argument expression, but is just the referent
of the NP; it is therefore called a referential argument. A predicate term can have
at most one referential argument. Later, we will also consider referential arguments
of verbs (5.3.2).
Figure 5.5
Levels involved in a predication by means of a noun
money
expression level
NP
conceptual level
predicate
›money‹
referential level
argument
(and referent)
the money
When adjectives are combined with a noun, we have yet another situation. The
argument of ›dubious‹ in (1) is the referent of the NP a dubious company. Since
reference is considered a matter of nouns, not of adjectives, the argument of ›dubious‹
is not a referential argument of the adjective, nor is it specified by a complement of
the adjective. Rather, the adjective ‘borrows’ its argument from the noun. In cases like
these, I propose to talk of a parasitic argument.
The analysis of predication demonstrates an important point: a predication only
comes about at the level of utterance meaning, since it is ultimately about referents of
the sentence. Predication only becomes effective in a concrete CoU, and only then it
delivers a truth value by application to the given arguments. This is the reason why the
question of truth of a sentence only arises at utterance level (1.1.2): sentences are true
or false because of the predications they convey about the referents of the sentence.
Since predication is a matter of the utterance level, it is subject to the Principle
of Consistent Interpretation (3.4.4). Elimination and modification of readings can
interfere with the conceptual level of predication. In fact, all the examples of meaning
shifts discussed in 3.4 involve predications and the problem of making predicates
applicable to their arguments. We will return to this topic in the section about
selectional restrictions (5.7.3).
Predication
111
The considerations on the levels involved in predication give rise to another variant
of the semiotic triangle. The configuration in Fig. 5.6 represents the relationships for
a single predication in a given CoU.
Figure 5.6
The semiotic triangle for predication
predicate term
predicate:
a concept
set of
arguments
5.3 VERBS
We will now turn to the three major word classes and the ways in which they function
as predicate terms. Verbs are combined with a separate argument term for each of
their arguments. They also have a referential argument not mentioned so far: the event
to which the verb refers. We will discuss this later in 5.3.2. The referential argument
is not counted when we classify a verb as one-place, two-place, etc. The discussion
will be confined to the most frequent types of verb. If verbs, nouns and adjectives are
categorized as one-, two- or three-place predicate terms, this is based on the number
of arguments in standard constructions. Obviously, it is almost always possible to
add further optional argument specifications to a given predicational construction. It
would make little sense to try to count all possible arguments that might be specified
for a given argument term. The issue of how many arguments a given lexical item has,
however, is not trivial and will be taken up again at the end of this section.
5.3.1 Major types of verb
Intransitive verbs are one-place predicate terms. The only argument is specified
with an NP which, in English, is always the subject of the sentence.
(2)
a.
The cat is sleeping.
b.
The snow has melted.
c.
The door opened.
d.
The author of the love letter did not succeed.
112
Understanding semantics
Transitive verbs are two-place predicate terms with two argument terms, the subject
and the direct object of the sentence.
(3)
a.
The cat is eating the dog’s food.
b.
He wants your help.
c.
The dog cannot open the door.
d.
Thirty-one students filled in the questionnaire.
Note that in (3d) the words in the questionnaire do not form a PP. In is part of the
particle verb fill in, rather than a preposition. You will have noticed that the verb open
appears as an intransitive verb in (2c) and as a transitive verb in (3c).
Ditransitive verbs have three argument terms. For one group, the third argument
term is an ‘indirect object’. In standard English word order, an indirect object is
either placed before the direct object or it is marked with to and placed after it.
(4)
a.
He’ll give my sister the keys.
He’ll give the keys to my sister.
b.
I showed them the photograph. I showed the photograph to them.
Other syntactic categories of complements. It should be mentioned that verb
complements are not confined to NPs. Many verbs have prepositional complements
with a lexically fixed preposition, cf. depend on, refer to, differ from, etc. Other verbs, e.g.
verbs of motion like go or put, can be combined with a wide range of PPs. Some verbs
take that-clauses as complements (know, believe, assume, say), infinitives (try/manage/
begin to) or gerunds (start/stop/keep -ing), to mention only the most common types.
Alternative grammatical means of specifying arguments. There are other ways
of specifying arguments of verbs than combining the verb with a complement. For
example, English imperative sentences usually have no subject. The corresponding
argument is then construed as the addressee(s). The imperative sentence in (5a)
expresses the same proposition as the corresponding declarative sentence (5b). Put
is a three-place verb with a direct and a prepositional object.
(5)
a.
Put the keys on the window-sill.
b.
You put the keys on the window-sill.
In languages such as Spanish or Italian, grammatical person and number are
reflected in verb inflection. The subject
itself can be omitted in suitable CoUs and
3
the resulting sentences are equivalent to sentences with the corresponding personal
pronouns in subject position (recall 4.1.3).
3 Because the pronoun is usually omitted, its use results in some sort of emphasis on it. In this
sense, the pronoun, if present, is not redundant.
Predication
(6)
a.
b.
habl-o
árabe
speak-1S
Arabic
habl-as
árabe
speak-2S
Arabic
≈
≈
yo
hablo árabe
I
speak-1S Arabic
tú
hablas árabe
113
(Spanish)
you speak-2S Arabic
Mechanisms similar to the one described for Spanish can be observed in languages
like Korean or Japanese, which have an elaborate system of social meaning. For
example, certain honorific forms of verb would only be used if the subject argument
is the speaker themselves or, in other cases, an addressee to be treated formally. In
such cases, the subject argument can be inferred from the form of the verb and
the rules of social meaning. Recall, for example, the Japanese sentences (10a-d)
discussed in 2.3.2.
5.3.2 Referential verb arguments
It is widely assumed that verbs also predicate about a referential argument for the
event described, in addition to the arguments hitherto mentioned. For example,
the referential argument of the predicate ›send‹ is what is referred to as an instance
of ‘sending’ when the verb send is used. As a general notion comprising actions,
processes and other occurrences, the term event is commonly used in semantics for
the referents of verbs.
There are good reasons for assuming a referential argument for verbs. First, the
event argument can be considered to serve as the argument of the tense element of
the meaning of the finite verb, which reasonably constitutes a predication of its own
(see 6.4). For example, in (1) the past tense form sent of the verb expresses that the
event of sending happened before the time of utterance. Second, the referential verb
argument can be considered the argument of certain adverbs, e.g. carefully in John
closed the envelope carefully: it is the closing, i.e. the event referred to, which is said to
be ‘careful’. Third, there are many deverbal nouns which denote an event of the kind
the verb denotes, e.g. the simple -ing derivation in (7).
(7)
Johnny’s sending money to the company was a mistake.
If we assume that the verb sent in (1) has an event argument, this argument is the
event of Johnny’s sending money to a dubious company. If (1) and (7) are both
related to the same CoU, the referential argument of the verb in (1) is identical to the
referent of the NP Johnny’s sending money to the company in (7). Thus, the meaning
of the derived noun sending can partly be captured by stating that it is a predicate
with the same referential argument as the verb concept ›send‹ (see also the analysis
of deverbal nouns in 12.2.2).
5.3.3 Deciding on the number of arguments
The question as to how many arguments a predicate term involves is often difficult
to settle. Let us mention just two aspects of the problem. Very many verbs appear in
114
Understanding semantics
more than one construction with a different number or a different quality of argument
terms. One sort of variation can be dealt with in a relatively straightforward way: the
occurrence of the same form in different grammatical categories, e.g. the abovementioned form to open with an intransitive and a transitive use. These must be
considered two different verbs as they belong to different word classes (intransitive
vs transitive verbs) and have different meanings. Intransitive open predicates of its
subject argument a certain change of state. Transitive open predicates of its subject
argument an action which leads to a corresponding change of state of the direct
object argument. Thus intransitive and transitive open express different predications
about their respective subject arguments. Their meanings are clearly different.
The second type of variation is more difficult to handle. A verb like eat can be used
in a wide range of constructions including the types instantiated by the following
examples:
(8)
a.
Fred is eating spaghetti.
b.
Fred is eating spaghetti with a plastic fork.
c.
Fred is eating spaghetti with a plastic fork from a big bowl.
d.
Fred is eating with a plastic fork from a big bowl.
e.
Fred is eating from a big bowl.
f.
Fred is eating.
In all these constructions, unlike in the case of intransitive and transitive open,
the verb eat predicates the same of its subject argument. Obviously, each of the
constructions contains a subject complement; in fact, English does not allow for its
omission. Other languages do, for example Japanese, but if the eater argument is
omitted, it must be clear from the given context who it is who eats. We may therefore
assume that the predicate ›eat‹ necessarily involves an eater argument. In view of
(8f) one might feel that all other complements are optional, since none of them is
grammatically obligatory. But the direct object argument is different. The concept
›eat‹ necessarily also involves a second argument. There can be no eating without
something being eaten. Therefore that argument is understood to be involved in the
situation described, even if it is not specified. Fred is eating is interpreted as Fred is
eating something. This is why we feel that the direct object is omitted in (8d, e, f).
The other arguments that can be added have a different status: Fred is eating is
not necessarily interpreted as Fred is eating with something or Fred is eating from
something. Neither something eaten from nor something eaten with constitutes
a necessary component of an eating event. Accordingly, specifications of such
arguments are not syntactically missing if they are absent. Thus the basic number of
arguments for ›eat‹ is two, and eat is a transitive verb, although its direct object can
be omitted. It must be added that not all transitive verbs allow the omission of the
direct object. Along with many others one such verb is eat’s close relative devour. We
will return to the question of the number of verb arguments in 6.1.
Predication
115
5.4 NOUNS AND ADJECTIVES
5.4.1 Major types of nouns
5.4.1.1 One-place nouns
The majority of nouns constitute one-place predicate terms. Unlike verbs, one-place
nouns are not combined with a separate argument term. They are primarily used as
the head of referring NPs that function, for example, as a verb complement (but see
5.4.3 for the ‘predicative’ use.)
(9)
The dog managed to open the door.
In this sentence, the argument of the one-place noun dog is the referent of the
subject NP the dog, and analogously for the noun door. Thus, both nouns have a
referential argument. In (1) above, the assumed person Johnny is the referent, and
the only argument of the proper name Johnny. Personal pronouns can be considered
as one-place predicate terms, too: I expresses the predication ›is the speaker‹, you
with singular meaning conveys the predication ›is the addressee‹. The predicational
content of 3rd person pronouns is confined to meaning components such as gender
(recall 4.1.1 on the meaning of personal pronouns). The definiteness component of
personal pronouns is not part of their predicational content.
In general, the determination of an NP is independent of its predicational content.
In the case of common nouns, the same predicational content can be combined with
definite, indefinite, quantificational or generic determination. For proper names and
pronouns – which are usually used as full NPs – a particular determination is part of
their lexical meaning. Different cases are displayed in Table 5.2; the pronoun one here
is the impersonal pronoun as in One cannot be too careful in these matters.
Table 5.2
Predication and determination of proper names and pronouns
Predication
Determination
Sheila
has the name ‘Sheila’
definite
she
female
definite
someone
person
indefinite
everything
non-person
quantificational
one
person
generic
5.4.1.2 Relational nouns
In 4.4.2.1, we introduced the distinction between sortal, individual, relational and
functional nouns. Sortal and individual nouns (e.g. priest and pope, respectively) are
one-place predicate terms, while relational and functional nouns (sister and mother,
116
Understanding semantics
respectively) are two-place. The aspect of inherent uniqueness that distinguishes
sortal from individual, and relational from functional nouns, does not matter in this
case. We will therefore use the term ‘relational noun’ in this chapter in the broader
sense that also includes functional nouns.
One group of relational nouns is kinship terms, e.g. uncle and sister in (10).
(10) My uncle is going to marry Molly’s sister.
The two NPs my uncle and Molly’s sister each have a referent: the speaker’s uncle
and Molly’s sister, respectively. These are the referential arguments of the relational
nouns uncle and sister. In addition to the referential argument, each noun has an
argument for the one the referent is a relative of. In the terminology of kinship
relations, this is called the ‘propositus’. In the case of Molly’s sister, the propositus is
Molly. It is specified by the NP Molly’s, which is linked to the noun sister by means
of the possessive ’s. In the case of my uncle, the propositus argument is the speaker,
specified by the possessive pronoun my. Thus both NPs refer to an additional
argument, the speaker and Molly, respectively. We will call the additional arguments
of relational nouns relational arguments.
In addition to the relational nouns uncle and sister, (10) contains the two-place
verb marry. Thus the proposition expressed comprises three two-place predications:
that someone is an uncle of the speaker, that this someone is going to marry someone
else, and that this other person is the sister of Molly. An analysis in the style of Fig.
5.2 may help to illustrate how the predicates are related to their arguments. In Fig. 5.7
the subscript ‘NR’ is short for ‘relational noun’. The four arguments ai, au, am and as are
short for ‘I’, the speaker’s uncle, Molly and her sister, respectively.
Figure 5.7
Predication structure of sentence (1)
words
NPs
[ my uncleNR] is going to marryV [ [Molly’s] sisterNR ]
arguments
ai
au
am
as
The relational argument of two-place nouns is usually specified with a possessive
construction. In English, there are three common alternatives. In the simplest one,
a possessive pronoun precedes the relational noun (my uncle). Alternatively, a full
NP with a possessive ’s can be placed before the noun (the children’s uncle). Finally, a
‘possessor’ can be specified by an of-PP following the noun (an uncle of my mother);
in this case there is also the alternative of the so-called ‘double genitive’ (a friend of my
mother’s), where the possessor NP is in the genitive.They are counted as arguments
Predication
117
proper because the meaning of the respective nouns cannot be defined other than in
terms of two arguments. (Recall the discussion of relational nouns in 4.1.4.)
There are many more relational nouns: words like friend, neighbour, rival, boss;
abstract nouns like name (of), height (of), occupation (of); or linguistic notions
such as meaning (of), pronunciation (of), subject (of), argument (of). In some cases,
relational arguments are specified by non-possessive PPs, cf. ticket to, attack on,
discontent with, equivalence to. Some relational nouns have two additional arguments,
e.g. difference, relationship or distance or nouns derived from transitive verbs, such as
conquest, discovery or invention.
In English, possessive arguments of nouns are not syntactically obligatory. In
many languages, however, possessive constructions with relational nouns differ in
form from those with non-relational nouns. In the case of relational nouns one talks
of inalienable possession, and in the case of non-relational nouns of alienable
possession. (11) is an example from Lakhota, the language of the Sioux. Lakhota
has possessive prefixes, e.g. ni- for 2nd person singular. Possessive prefixes can only
be attached to relational nouns. This happens with the relational noun nági ›spirit‹
in (11a). Ninági means ›yoursg spirit‹ and instantiates the inalienable possessive
construction, ‘possessive prefix + relational noun’. By contrast, wowapi ›book‹ in
(11b) is a sortal noun. It cannot take a possessive prefix; but it can be turned into
a relational noun and then it can. The prefix thá makes a relational noun out of
sortal nouns. tháwowapi is relational and means ›book of‹. Prefixed with ni it yields
›yoursg book‹. nitháwowapi is an example for the alienable possession construction:
‘possessive prefix + thá + nonrelational noun’.
(11) Lakhota (Sioux, Robert D. Van Valin p.c.)
a.
inalienable
ni-nági ki
2S-spirit DEF
‘yoursingular spirit’
b.
alienable
ni-thá-wowapi ki
2S-REL-book
DEF
‘yoursingular book’
5.4.2 Major types of adjectives
5.4.2.1 One-place adjectives
Adjectives in their basic form, the ‘positive’ (recall 1.2.2), are mostly one-place
predicate terms; as we will see, this does not hold for the comparative form.
Grammatically, three ways of using adjectives can be distinguished: the attributive
use, the predicative use and the adverbial use. We will only deal with the attributive
and the predicative use here. In the attributive use, the adjective, within an NP, is4
attached to the head noun, e.g. a dubious company, a red balloon, the stupid driver.
As we saw in 5.2, the argument of the adjective in attributive use is the referential
argument of the noun.
4 In English, attributive adjectives precede the noun. In other languages, the adjective follows the
noun. For example, in Italian ‘stupid driver’ would be conducenteN stupidoA.
118
Understanding semantics
The second way of using adjectives is the predicative use. In the most common
predicative construction in English, the adjective is combined with the so-called
copula verb be, or a similar verb such as become, to form the VP of the sentence. In
such cases, the predication expressed by the adjective relates to the subject referent.
Thus, the subject NP is a complement of the complex expression consisting of the
copula verb plus the predicative adjective. Many languages do not use a copula with
predicative adjectives. They would combine the subject NP 5directly with an adjective
which serves as the grammatical predicate of the sentence.
(12) a.
b.
John is clever.
John umnyj.
‘John is clever’
(Russian)
‘John is clever’
(Japanese)
John clever
c.
John
wa
kashikoi.
John
TOP
clever
5.4.2.2 Two-place adjectives
Some adjectives, e.g. other than, different from, similar to, fond of, satisfied with,
keen on, have a second argument. It is invariably specified by a PP complement, i.e.
by a separate argument term. With respect to their first argument, these adjectives
behave like one-place adjectives. Combined with the PP complement they can only
be used in predicative constructions such as a copula sentence or the relative clause
construction in (13b) (which constitutes a second variant of predicative use). In
attributive use as in (13c), adjectives with complements are ungrammatical.
(13) a.
My uncle is very fond of Molly’s sister.
b.
She wore a sweater similar to yours.
c.
*She wore a similar to yours sweater.
A special case of two-place adjectives are one-place adjectives in their comparative
form. The comparative adds a further argument to the adjective, an entity the first
argument is compared to. Thus big is one-place, and bigger is two-place. The second
argument is specified by a than-PP, i.e. a complement proper. If the comparative is
formed of a two-place adjective, the result is a three-place predicate term (cf. (14c)).
(14) a.
Her hair was oilier than Julio’s.
b.
I hardly can imagine a book more boring than this one.
c.
Actually, my uncle is more fond of Molly than of her sister.
5 Russian has a copula for the past tense, Japanese uses a copula with one type of adjectives but not
with the type represented by kashikoi.
Predication
119
5.4.2.3 Non-predicational adjectives
You should be aware, however, that some adjectives do not serve as predicate terms:
for example, the ‘modal’ adjectives alleged, potential, actual, would-be. They can only
be used attributively: the alleged terrorist, a potential referent, etc. Their predicative
use is impossible because they do not express a predication of their own: it would
not make sense to say of something that it is ‘alleged’ or ‘potential’. Rather, modal
adjectives serve to modify the predication expressed by the noun by relativizing
whether or not it is true. An ‘alleged terrorist’ is someone for whom the predication
‘terrorist’ is not plainly true in the given CoU, but only ‘allegedly’ so, i.e. in some
different context. Similarly, a ‘potential’ referent is something for which the predicate
›referent‹ is only potentially true. A similar group of adjectives relates the predication
to different times or places, e.g. former, future or local, etc. These are instruments
of temporal or spatial deixis (recall 4.2 and 4.3). Yet another different type of
non-predicative adjectives is represented by combinations such as atomic war,
educational disaster, doctoral dissertation. Here, too, the adjective does not provide
a predication of its own (the war is not ‘atomic’, the disaster not ‘educational’, etc.).
Rather the adjective modifies the meaning of the noun by adding a specification to
some conceptual component of its meaning: ›war with atomic weapons‹, ›disaster in
education‹, ›dissertation written to acquire a doctor title‹, and so on. Consequently,
not all adjectives are predicate terms, and not all combinations of an attributive
adjective and a noun express a joint twofold predication.
5.4.3 Nouns in predicative use
Not only adjectives but also NPs, and therefore nouns, can be used predicatively.
Predicative NPs do not refer.
(15) John is a teacher.
In this sentence, only the NP John has a referent. Syntactically the subject NP is a
complement of the copula. Semantically it is passed on, as it were, to the predicate
term a teacher that forms the VP with the copula, and thus serves as an argument
term for teacher. The same mechanism applies for predicative adjectives. The copula
itself is not a predicate term; it is a linguistic means of forming a predicate term out
of an NP or adjective. The argument of the resulting predicate can then be specified
in the subject position. Thus, for example, the construction in (15) can be used to
apply two NP predications to the same argument.
Relational arguments of predicative nouns are specified in the same way as in their
referential use. The predicative use does not affect the syntactic structure of the NP.
(16) a.
b.
This is my uncle.
This is a ticket to Novgorod.
Table 5.3 displays the main types of predicate terms and the way in which their
arguments are specified in the sentence. Predicative uses are not included. Verbs,
120
Understanding semantics
nouns and adjectives differ in the way their first argument is integrated into the
sentence structure, but they are similar in that further arguments take the form of
complements, usually NPs or PPs. For the second argument of relational nouns there
is a special group of (possessive) constructions.
Table 5.3
Types of predicate terms and how their arguments are specified
Type
First
argument
Further arguments
verb intransitive
the bell rang
complement
–
verb transitive
she opened the door
complement
complement
noun 1-place
the postman
referential
–
noun relational
asking her name
referential
possessor
a letter from Johnny
referential
complement
adjective 1-place
a pink envelope
parasitic
–
adjective compar.
thicker than the last one
parasitic
complement
adjective 2-place
full of promises
parasitic
complement
Before the topic of predication is continued, a brief section about predicate logic
is inserted. Ultimately derived from the predicational structure of natural language
(i.e. Ancient Greek), more than two thousand years ago by Aristotle, it is a formal
system widely used in semantics for the analysis and representation of sentence
meaning.
5.5 PREDICATE LOGIC NOTATION
In so-called predicate logic (PL), a simple notation has been developed for
predication. The basic expressions of predicate logic are one-place, two-place, etc.
predicate constants (i.e. predicate terms that do not have the status of variables) on
the one hand, and so-called individual terms on the other. Individual terms serve
as argument terms and are interpreted as referring to ‘individuals’, where these are
whatever may serve as an argument of one of the predicates. There are two sorts of
individual terms, individual constants and individual variables. Roughly speaking,
individual constants can be thought of as proper names and individual variables
as something like 3rd person personal pronouns, i.e. expressions that refer to some
particular individual determined by the CoU. For current purposes, we define a
predicate logic language with the following basic expressions (again, referential verb
arguments are disregarded).
Predication
121
6
one-place predicate constants
cat, money, dubious, company, sleep
two place predicate constants
marry, sister, uncle
three place predicate constants
send_to
individual constants
j [for Johnny], m [for Molly], i [for the speaker]
individual variables
x, y, z
The predicate terms are combined with argument terms in a uniform way. In the
most common notation, predicate terms are followed by the appropriate number of
individual terms enclosed in parentheses and separated by commas. The following
would be simple PL formulae with the meanings indicated.
(17) a.
money(x)
x is money
b.
uncle(j, m)
Johnny is an uncle of Molly
c.
send_to (j, x, y)
Johnny sends x to y
The notation of predicate logic reflects the view that predicates are incomplete
propositions with empty slots for arguments. When combined with the appropriate
number of argument terms, a predicate term yields a formula (a sentence in PL).
Formulae can be combined to form more complex
formulae by using sentential
7
connectives such as the conjunction Ÿ ‘and’. This allows us to analyse the
predicational part of a natural language sentence by ‘translating’ it into a predicate
logic formula. The single predications are connected by truth-conditional conjunction.
In the following examples, referents not specified by a proper name are represented
by variables. Tense, aspect (progressive form of the verb) and articles are neglected.
(18) a.
The cat is sleeping.
8
cat(x) Ÿ sleep(x)
b.
Johnny sent money to a dubious company.
send_to(j, x, y) Ÿ money(x) Ÿ dubious(y) Ÿ company(y)
c.
My uncle is going to marry Molly’s sister.
uncle(x, i) Ÿ sister(y, m) Ÿ marry(x, y)
The method makes transparent which predications a sentence contains, to which
arguments they are applied and how the different predications interact by sharing
arguments. In chapter 13, predicate logic will be treated in more depth and detail
and the method of semantic analysis by translation into a formal language will be of
central concern.
6 It is common practice to use bold type for predicate and individual constants that correspond to
natural language words.
7 More conjunctions will be introduced in 7.4 on sentential logic
8 A formula such as ‘sleep(cat)’ would violate the rules of PL syntax, because the predicate term cat
is not allowed to be used as an individual term in the argument position of sleep.
122
Understanding semantics
5.6 THEMATIC ROLES
5.6.1 The arguments of verbs as thematic roles
The different arguments of a verb predicate are referred to as its roles or
participants. A transitive verb has two roles, for example the eater and the thing
eaten, the opener and the opened, the helper and the helped. Grammar consistently
distinguishes the different roles of a multi-place verb. When the verb eat is used
in its active voice, the eater is always specified by the subject of the verb eat and
the thing eaten by its direct object. The analogue holds for multi-place nouns and
adjectives.
An important question then is whether there is cross-verb consistency in the way
the roles of verbs are marked. Is there something common to the role of the eater
and the role of the helper that is responsible for their appearing in the subject
of the sentence? Is there something in common to all subjects, or to all direct or
indirect objects? Can the roles that are encoded in tens of thousands of verbs be
consistently categorized into a small number of abstract roles? Are these abstract
roles universally applicable to all roles of all verbs in all languages? Semanticists
and syntacticians have tried to answer these questions positively. There is a good
chance of succeeding, but things are not straight and simple. A first look at the
data clearly shows that the subject does not always denote the same role. Consider,
for example, intransitive and transitive open in the following sentences:
(19) a.
The doorO opens.
b.
This keyI opens the doorO.
c.
The childA opened the doorO.
d.
The childA opened the doorO with her own keyI.
While these sentences represent different concepts of opening, it is intuitively
clear that they all fit into one scheme with three roles: (i) an animate agent A
opening something; (ii) an object O that becomes open, (iii) an instrument I
used to open O. In (19a), the subject specifies O, in (b) I and in (c) and (d) A.
Conversely, O is specified by the subject in (a) but by the direct object in (b–d).
The instrument I appears as the subject in (b) and as a prepositional adjunct
in (d). The patterns do, however, exhibit certain regularities: if A is specified, it
appears as the subject. O is always specified in the object position as long as it is
not the only argument term.
Since the first attempts, back in the 1960s, at establishing a set of universal roles,
many theories have been developed in different theoretical frameworks. It is now
common practice to speak of thematic roles ( -roles, theta-roles, with the Greek
letter ‘theta’ for thematic) or semantic roles. Some draw a distinction between
thematic roles and semantic roles, but the difference need not concern us here. The
Predication
123
Table 5.4
Thematic roles
Role
Description
Examples
agent/actor
performs the action expressed by the verb, controls the
event
Johnny wrote a love letter
the cat has eaten the egg
she gave me the keys
you put the keys on the desk
my uncle marries Molly
theme/
patient
undergoes the action/change/event expressed by the verb Johnny wrote a love letter
the cat has eaten the egg
she gave me the keys
you put the keys on the desk
my uncle marries Molly
the door opened
the snow is melting
experiencer
experiences a perception, feeling or other state
I heard him
the outburst surprised her
instrument
an instrument, or a cause, by which the event comes
about
this key opens the door
he opened the door with a key
she was shaking with fear
locative
a location
the keys are on the desk
goal
goal of a movement
put the keys on the desk
path
path of a movement
she rode through the desert
inventory of thematic roles differs from theory to theory, but the roles in Table 5.4
9
are uncontroversial.
General thematic roles are useful in several respects. For example, they help to
account for the meaning relations holding between the three different verbs open
used in (19a), (19b) and (19c, d), respectively. For predicate terms, a description of
their arguments in terms of roles, their so-called argument structure constitutes
an important part of their distinctive properties. Thematic roles also allow a proper
description of phenomena like passive, which change the argument structure of a
verb in a specific way (see 6.1).
5.6.2 Linking
The mechanism by which a language distinguishes the different arguments of
predicate terms is called linking. We will not go into this complex matter here. As far
as English is concerned, some simple linking rules can be stated that were already
indicated in connection with (19): an AGENT role always appears in subject position,
9 We follow the practice of writing thematic roles with small capitals (THEME etc.).
124
Understanding semantics
a THEME can only be the subject if no AGENT is present. These rules hold for active
sentences. In passive sentences, THEMES and other roles can be made the subject of the
sentence. The AGENT complement is deleted, but it can be specified by an additional
by-PP.
(20) a.
b.
active:
The dog (AGENT, subject, obligatory) opened the door
(THEME, direct object, obligatory).
passive:
The door (THEME, subject, obligatory) was opened [by
the dog (AGENT, PP, optional)].
The effective linking of argument terms depends on a proper grammatical distinction
between the subject, the direct objects and other ‘grammatical functions’. In English
sentences, the subject and the direct object differ in three ways, which illustrate three
general linking strategies to be observed in the languages of the world.
∑
Agreement. The finite verb ‘agrees’ with the subject NP in certain
grammatical properties.
In English, the finite verb agrees with the subject NP in grammatical person (1st, 2nd
and 3rd) and number (singular or plural). Agreement shows up in the 3rd person
singular -s of full verbs in the present tense (he/she/it speaks vs I/you/we/they speak_)
and in the forms of the verbs be and have (I am, you are, she is; I was, we were; have vs
has). In other languages, e.g. Georgian and Hungarian, the verb may in addition agree
with the direct object. Yet other languages such as Chinese, Korean and Japanese do
not exhibit agreement between the verb and its complements.
∑
Case. The complements of the verb are distinguished by case.
In English, the subject is in nominative case, the object in objective case. However,
the difference shows up only with some pronouns (I, he, she, we, they vs me, him, her,
us, them and who vs whom). Prepositions, too, can be subsumed under this linking
strategy. As a rule, there is a small set of cases and/or prepositions, e.g. nominative,
objective and to in English, that are used for those complements which specify the
arguments proper. Additional complements that specify original arguments – such
as an instrument of eating or a vehicle of motion – are marked with prepositions or
cases that are semantically more specific. For example, the role of the instrument in
English is marked with by or with (19d). With is not restricted to marking instrument
terms; it is polysemous. The PPs in (21) are specifications of special roles not
contained in Table 5.4:
(21) a.
b.
Johnny ate the salad with disgust.
Johnny ate the salad with his friend.
Predication
∑
125
Word order. Often different positions in the sentence are provided for
different grammatical functions, in particular for subject and direct object.
In English, the subject NP precedes the finite verb, the direct object follows it. Since
English has only rudimentary case marking and agreement, word order is its main
linking strategy. Some languages, such as Chinese, that lack case and agreement
altogether rely completely on word order as grammatical linking mechanism.
Of course, grammar is not the only resource for determining which roles of the
verb are specified by which complements in the sentence. If the grammatical linking
information is insufficient, the assignment of roles is a matter of semantics or context.
For example, the German sentence (22) contains two singular NPs (Kaffee and John)
which could both be any case except genitive. The verb is 3rd person singular, like
both NPs, too. Unlike English, German allows, under certain circumstances, the
inversion of subject and object. Thus the grammatical information is insufficient
for distinguishing between the AGENT complement and the THEME complement. In
normal contexts, the sentence will of course be interpreted as saying that John drank
coffee rather than the other way round.
(22)
Kaffee
trank
John
coffee
drank
John
If the determination of the roles is left to semantics or context, this is not a case of
linking. The notion of linking is restricted to grammatical means of role distinction.
5.7 SELECTIONAL RESTRICTIONS
A predicate cannot be applied to arbitrary arguments, and consequently, a predicate
term cannot be combined with arbitrary complements. In addition to the requirements
of grammar and conditions such as the Principle of Consistent Interpretation (3.4.4),
complements must fulfil certain semantic, or logical, conditions and this leads to
restrictions on possible argument terms. The discussion here will be restricted to
verbs, but analogous considerations apply to adjectives and nouns in predicative use.
5.7.1 Selectional restrictions of verbs
Two somewhat strange examples may illustrate what kinds of conditions are involved.
(23) a.
b.
The cook has murdered an eggplant.
The potatoes are frying the cook.
If taken literally, the two sentences describe impossible situations. The verb murder
requires a living being as its THEME/PATIENT argument, usually a human being. Only
a minority of speakers of English would use the verb murder also for the killing of
126
Understanding semantics
animals, probably none would use it for plants or even fruits. Likewise, the verb fry
requires an AGENT argument capable of acting. It need not be a person – one could
imagine an animal putting, and keeping, something on a hot stone in order to get
it fried. But potatoes cannot fill this role. There are fewer problems with the THEME
argument of (23b). Although highly unlikely, the THEME of a frying event can be a
person: persons are friable. But the THEME role too underlies logical restrictions. For
example, words, numbers, properties or addresses cannot be fried.
The logical conditions on arguments are called selectional restrictions (also
‘selection restrictions’). These were briefly introduced in 4.7.2.2. The term is
motivated by the idea that a predicate term selects, and thereby restricts, the range
of possible arguments. Let us assume, for example, that the verb vaccinate requires a
human being as its AGENT. Then, in appropriate CoUs, the following sentences comply
with the selectional restrictions:
(24) a.
b.
The doctor himself vaccinates John.
The next one vaccinates John.
The choice of the subject term in (24a) guarantees that the selectional restrictions
are fulfilled: doctors are persons. It is, however, not necessary that the noun in the
complement NP entails human being. The selectional restrictions only require that
the referent of the complement is a person. The potential referents of the subject NP
the next one in (24b) are by no means necessarily persons: the NP can refer to almost
anything, because the pronoun one can replace an arbitrary count noun. But if the
subject refers to a person in the given CoU, (24b) is semantically correct.
To work out the selectional restrictions of a particular predicate term can be a very
difficult task. Take, for example, the THEME argument of the transitive verb open. What
kinds of things can be opened? We can open a door, e.g. by sliding it open, and thereby
create an opening in the wall that can be passed through. We can open a room by
opening a door to the room. We can open our mouth. We can open our eyes. Or our
arms. Or a fist. We can open a bottle or a tube by removing or opening its lid. We can
open an envelope by slitting it open, or a letter by unfolding it, or a book. We can open
a bank account or a business. We can open a ceremony. We can open a computer file.
We can open perspectives. These are not only different kinds of objects, but in addition
open in almost each case means something slightly different. If I open a door or a lid,
the argument that is ‘opened’ is moved, or removed, in order to create an opening in the
enclosure of a spatial region (e.g. a room or the interior of a bottle). If I open a bag, or
an envelope, the THEME argument is the enclosure and the result of the act is an aperture
in the enclosure. If I open a room, a trunk, a garden, a shop, a box, or my mouth, I refer
to an object which has an enclosure in which then an aperture is brought about. So,
actually, there are two or three different roles involved in these kinds of opening events:
(i) a spatial region (e.g. a room) which is made accessible; (ii) its enclosure; and (iii),
possibly, a (re)movable part of the enclosure that provides, or blocks, access to the
region. Each of these roles goes with its own selectional restrictions. In the variants
open a fist, open a letter, open the arms, open the wings (of a bird), the THEME arguments
Predication
127
play yet a different role and the verb is used to express a different kind of process
similar to spreading or unfolding. In the variant represented by open a ceremony, the
selectional restrictions of the THEME argument require an event or a procedure. Opening
it, we start the ceremony and thereby ‘enter’ a certain sequence of events. Yet other
selectional restrictions govern the use of open in I opened the style file, you must open a
bank account or she opened a computer business.
It would be wrong to conclude that the selectional restrictions for the THEME
argument of transitive open are so general that they cover almost anything. Rather
the verb is multiply polysemous. In each of its meaning variants, the verb expresses
a different process with respect to its THEME argument and imposes different
selectional restrictions. If we take these restrictions seriously, we will be able to
explain how the meaning variants are related, e.g. the readings of open in open one’s
eyes and open one’s eyelids.
5.7.2 The process of fusion
The combination of a predicate term with a complement results in two sources
of information about the argument. First, the complement provides an explicit
specification of it. Second, the predicate contributes implicitly the selectional
restrictions for the argument. These two pieces of information are conjoined when
the meaning of the sentence is composed. We can think of the process as logical
conjunction (i.e. combination by and) of the two pieces of information. Let us call the
process fusion, adopting a term from Jackendoff (1990). It is illustrated in Fig. 5.8.
Figure 5.8
The mechanism of argument description
predicate term
contributes
sentential context
contributes
SELECTIONAL RESTRICTION
ARGUMENT SPECIFICATION
fusion
TOTAL DESCRIPTION
OF THE ARGUMENT
Consider the following simple example (25):
(25) The dog opened it.
The THEME of the opening is merely specified as ‘it’ – which might refer to anything
inanimate. In addition, the referent of it must fulfil the selectional restriction of
128
Understanding semantics
being something ‘openable’. This restricts the range of possible referents of the
pronoun it considerably. In cases such as these, the selectional restrictions contribute
substantially to the description of the argument.
The same does not hold for the AGENT argument the dog in (25). The specification of
the AGENT as a dog is so specific that the selectional restrictions for the AGENT argument
are thereby automatically fulfilled: dogs in general are potential AGENTS of openings.
Here, the selectional restrictions do not add to the specification of the argument.
A sentence like (26) provides a third possibility.
(26) She will marry next June.
The predicate ›marry‹ requires a person as an AGENT; the pronoun she contributes
that the subject argument is female (not necessarily a person) or one of those rare
nonhuman items in English (such as ships) for which feminine gender is used.
Fusion of the restriction ‘person’ with feminine gender yields ‘female person’ as
specification of the argument.
Finally, the fusion of the two descriptions of an argument may lead to a conflict –
if they are incompatible as in (23a) the cook has murdered an eggplant or (23b) the
potatoes are frying the cook. In such cases, fusion results in a logical contradiction.
To sum up, fusion may have different results:
(i) If the selectional restrictions are not more specific than the argument
specification, the total description is identical with the argument specification.
(ii) If the selectional restrictions are more specific than the argument specification,
the total description is identical with the selectional restrictions.
(iii) If the selectional restrictions and the argument specification neither entail
(cases i and ii) nor contradict each other (case iv), fusion results in a total
description of the argument which is more specific than both sources.
(iv) If the selectional restrictions and the argument specification are incompatible,
the total description is contradictory, i.e. inapplicable to any concrete
situation.
In actual communication, contradictions will not be accepted (except as examples for
contradictions as in the current context). Application of the Principle of Consistent
Interpretation (3.4.4) will either eliminate the corresponding reading or lead to an
adjustment with a meaning shift. The principle applies because a predicate term and
its complements constitute immediate sentential context for each other. For example,
in the case of (27), we would assume that the verb drink imposes the selectional
restriction ‘liquid’ on its direct object argument:
(27) She drank the coffee.
The object NP the coffee, taken for itself, does not necessarily denote a liquid. ‘Coffee’
can also be coffee powder, coffee beans or coffee shrubs. In (27c), the selectional
Predication
129
restriction ‘liquid’ rules out all but the drink variant of the meaning of coffee. The
result is a disambiguation of the noun coffee.
Conversely, we can observe disambiguation of the verb:
(28) a.
b.
She corrected her uncle.
She corrected the error in the style sheet.
In (28a) the specification of the THEME/INSTRUMENT argument requires a meaning
variant of the verb in which it can be applied to persons, ruling out a reading such as
the one required in (28b).
5.7.3 Selectional restrictions and meaning shifts
If fusion leads to contradiction, it may be possible to obtain an admissible reading
by means of a meaning shift. Let us first consider an instance of a metonymical shift
(recall 3.4.3).
(29) Moscow declares the Chechen rebels defeated.
The verb declare requires a human being or organization as its AGENT argument, but
the given specification Moscow is the name of a geographic entity. We will change the
meaning of Moscow by a metonymical shift (location → institution located there)
in order to meet the selectional restrictions of the verb. Note how the selectional
restrictions serve as a guide that indicates the direction of the meaning shift.
In the case of metaphorical shifts, it is often the predicate term whose meaning
is shifted, as can be observed in (30). The literal meaning of the verb evaporate
requires some kind of physical substance. The subject, however, refers to a mental
state. To remedy the conflict, the verb meaning is shifted to the more general meaning
›vanish completely‹ which carries selectional restrictions that allow for this kind of
argument.
(30) His courage evaporated.
The processes of metaphor and metonymy regularly affect selectional restrictions.
If an argument term undergoes a metonymical shift, the resulting referent usually
is of a different logical sort, cf. ‘university’ as a location vs ‘university’ as an
institution vs ‘university’ as the university personnel. Likewise, metaphorical
interpretation of an argument causes a shift into a different conceptual domain,
usually also of a different logical sort, e.g. when ‘money’ is conceived of as a liquid
that may ‘flow’. If a predicate is interpreted metaphorically, as e.g. evaporate in (30),
the selectional restrictions change too, as to match with the sort of objects that
make up the target domain.
130
Understanding semantics
5.7.4 Semantic irregularity
The massive occurrence of meaning shifts in the interpretation of actual sentences
blurs a question that is central to semantic analysis, the question of semantic
irregularity. The notion of selectional restrictions provides us with one clear type
of cases: if a specification of an argument in the sentence is logically incompatible
with the selectional restrictions, then the construction is semantically irregular.
Simple as this seems to be, we have seen that semantic regularity is a question of
the readings assumed for the predicate term and its argument specifications. For
instance, sentence (23a) above, the cook has murdered an eggplant, is semantically
irregular only if we assume the original lexical meanings of murder and eggplant.
It becomes regular, i.e. interpretable, if we allow an appropriate meaning shift of
either the verb or the direct object. (Possible interpretations are left up to your
imagination.) It therefore makes more sense to avoid the simple notion of semantic
acceptability and to replace it by a description of the conditions under which it is
possible to make sense of a complex expression. We may then distinguish between
degrees of acceptability, such as (i) interpretable on the basis of the lexical meanings
of all components, (ii) interpretable by means of common types of meaning shifts,
(iii) interpretable only by means of uncommon types of meaning shifts. Probably, a
fourth category, ‘interpretable by no means at all’, does not exist.
5.8 SUMMARY
This chapter focused on predication, the semantic function of the main word
classes, verbs, nouns and adjectives. Built into a sentence, each of these ‘content
words’ adds a predication to the total proposition, about one or more referents.
The three major word classes differ in how their arguments are specified. Verb
arguments, except for the event argument, are specified by complements, i.e.
separate syntactic constituents with a referent of their own. One-place nouns are
mainly used as referring expressions that predicate about their referents. One-place
adjectives are parasitic for their argument. One of the most important insights
concerning sentence meaning is the fact that the predications contained in a
sentence are interconnected by argument sharing. If you take a look back at the
examples, in particular the analyses in Fig. 5.2 and Fig. 5.7, you will see that the
network of predications includes all referents of the sentences. It is this network
structure that makes a sentence a coherent semantic unit. The verb has the key
role in this structure. It occupies the centre of the network like a spider in its web
holding all the threads. This role of the verb corresponds to the fact that most verbs
are two- or multi-place predicates.
Since the meaning of a sentence is a network of predications about its referents
(including reference time and event referents), the sentence as a whole can be
considered one complex predicate expression about the situation referred to. For
example, sentence (1) renders a complex predication about a situation: an event e
Predication
131
takes place, at some time t; e is an event in the past (cf. past tense); it is an event of
sending that involves three referents rj, rm and rc, with rj the AGENT, rm the THEME and
rc the GOAL of sending; rj is someone called ‘Johnny’, rm is money and rc is a company
and dubious. This is what the sentence, due to the sophisticated grammar of English,
is able to express in not more than seven words. Slightly accommodating Fig. 5.6 for
the application to sentences, we obtain the picture 10in Fig. 5.9 for the sentence as a
complex predication about the situation referred to.
Figure 5.9
The sentence as a complex predication
sentence
the proposition,
a complex predicate
the situation
referred to
The study of predication also sheds light on the mechanism of composition.
First, composition is for the most part a matter of integrating all the predications
contained in the word meanings into an overall structure. Second, we have seen how
predicate terms and complements interact in the process of fusion in providing the
total description of the argument. We will later return to the underlying mechanisms
of composition that bring about the complex network of predications contained in a
sentence (cf. 12.4 and 13).
EXERCISES
1. What is the difference between a predicate and a predicate term, between an
argument and an argument term? Discuss the four notions and explain how they
are connected.
2. Discuss the ways in which arguments are specified for verbs, nouns and
adjectives. For nouns, distinguish between referential and predicative uses. For
adjectives, relate to attributive and predicative uses. For predicative uses, assume
a simple copula construction.
10 Actually, the sentence carries even more meaning than this: the grammatical meanings expressed
by aspect and mood of the verb and the meaning of the declarative sentence type.
132
Understanding semantics
3. Give an analysis in the style of Fig. 5.7 of the following two sentences.
a. Arthur opened the door of the kitchen.
b. My daddy met your mother in the supermarket.
4. Express the predications contained in the two sentences in a predicate logic
formula like those in (18). Use the individual constants a, i and u for Arthur, the
speaker and the addressee, respectively.
5. Determine the thematic roles of the verb arguments in the two sentences.
6. The following examples illustrate three types of predicative NP constructions.
Discuss the ways in which the NP a born loser gets its argument in each type of
construction.
a. Peter is a born loser.
b. She called him a born loser.
c. Peter, a born loser, managed to get fired within three hours.
7. What are the linking rules for the following relational nouns? In c, refer to the
respective arguments of the verb discover.
a. distance
b. difference
c. discovery
8. Try to describe the correlations between the two uses of the verbs provide, wipe
and load in terms of thematic roles and grammatical relations (which role
appears as subject etc.):
a. She provided Sheila with boots.
vs She provided boots for Sheila.
b. He wiped the crumbs off the table. vs He wiped the table.
c. She lost.
vs She lost the match.
9. Try to formulate the selectional restrictions (consult the online Oxford English
Dictionary http://oxforddictionaries.com/ for meaning variants and their
definitions, and discuss all of them):
a. of the verb conquer for its direct object argument
b. of the adjective yellow for its only argument
10. Explain in terms of selectional restrictions how the metonymical readings of the
following sentences come about:
a. The university lies in the eastern part of the town.
b. The university has closed down the faculty of agriculture.
c. The university starts again on April 15.
Predication
133
FURTHER READING
Givón (1993, ch. 3) for an extensive discussion of verb types and thematic roles.
Tallerman (2011, ch. 6) on the basic facts of linking and ch. 7 on passive and causative
constructions. Radford (1988, ch. 4) on complements and adjuncts, ch. 7 on thematic
roles and selectional restrictions. You will find a broader discussion of semantic
roles in Kearns (2011, ch. 10), Riemer (2010, ch. 10) and Van Valin (2005, 2.4). Palmer
(1994) is a monograph about the topic. For a recent survey see Davis (2011).
6
?
Verbs
Nouns and verbs are the two basic word classes of language. Almost every sentence
contains both. Verbs are primarily used for predication, and nouns for reference,
where the referring NPs provide the arguments for the predication expressed with
the verb. Chapter 4 dealt with the major aspects of reference and NP semantics. Now
we will turn to the semantics of verbs.
As we saw in the previous chapter, sentences may contain several nested
predications. One of them is the central predication around which the whole sentence
is organized. Recall the first example from chapter 5, repeated in (1a) below. The
sentence contains five predications. The central predication is the one expressed by
the verb sent, while the other predications, expressed by Johnny, money, dubious and
company serve to specify the arguments of the central predication. It is the central
predication that is subject to negation if the sentence is negated, as it is in (1b). And it
is the central predication to which the temporal information applies that is expressed
by aspect and tense, which are not accidentally realized by the choice and the form of
the verb, in this case past perfective (see 6.3 and 6.4 below).
(1)
a.
Johnny sent money to a dubious company.
b.
Johnny didn’t send money to a dubious company.
There are languages which, in certain cases, can express the central predication
without using a verb. These would be predications by means of adjectives, nouns or
PPs in languages which do not use a copula. For example, in Russian the equivalents
of the English sentences I’m a linguist, I’m in Moscow, I’m stupid would just be
expressed as ‘I – linguist’, ‘I – in Moscow’, ‘I – stupid’. Most sentences, however,
contain a verb for their central predication. In many languages, including English,
there are no sentences without a verb.
In all languages, verbs constitute a large part of the lexicon; English has tens
of thousands of them. Verbs express actions to walk, processes to melt, events to
explode and states to know. The situations expressed are of a temporary nature, they
may apply for long or short periods; they are located and ordered in time. Also the
situations may have an internal temporal structure such as one state changing into
another, an activity coming to an end or a process beginning. Therefore, temporal
aspects of predications with verbs are one central issue of verb semantics. We will
turn to them in the sections 6.2 to 6.5 on situation structure, aspect and tense. The
other central issue is the fact that verbs almost always express a predication about
not only their own referent but also about one or more additional arguments. In
Verbs
135
this respect they can be compared with relational nouns (5.4.1.2). Thus, while most
nouns are one-place predicate terms, most verbs are multi-place. The situation to
which a verb refers does not exist independently, but it manifests itself in certain
temporary conditions of the participants. The only exceptions are zero-place verbs
such as rain or thunder which apply immediately to the situation referred to. Verbs
differ in the number of arguments and the roles these inhabit (5.3). We will turn to
general aspects of argument structure in a minute.
Investigating the argument structure of verbs on the one hand and the temporal
structure of the situations expressed on the other will enable us to distinguish major
sub-classes within the vast number of verbs. In addition, this chapter will deal with
grammatical meaning in more depth than any other chapter of the book. We will
treat three general issues of verb grammar: diatheses (such as passive), aspect and
tense. What we will not deal with is the complex issue of mood and modality, i.e. of
elements like the modal verbs and auxiliaries written in italics in (2):
(2)
a.
She can/may/must/shall go home.
b.
She could/might/should/would go home.
Dealing with general aspects of verb semantics, you will realize a very important
general trait of language: language is not just a matter of depicting reality one-toone, but rather a matter of describing reality selectively and from one particular
perspective out of many possibilities.
6.1 ARGUMENT STRUCTURE, DIATHESES AND ALTERNATIONS
6.1.1 Argument structure
Verbs not only differ in the number of arguments, but also in the constellations
of thematic roles. Action verbs have an agent argument, others do not. Verbs of
movement have locational arguments such as GOAL, SOURCE or PATH, in addition to
the role of the AGENT or THEME moving. Verbs of sensation or perception have an
EXPERIENCER argument. Among the possible arguments of a verb predication, there
are core arguments (one, two or at most three). These are typically syntactically
obligatory and appear as subject, direct object or indirect object. AGENT and THEME/
PATIENT are always core arguments. The other arguments are of a more peripheral
nature; they appear as oblique complements (or adjuncts), for example in the
form of prepositional phrases. They are typically optional and not included in the
lexical verb concepts; such peripheral optional arguments include, for example,
INSTRUMENTS. Syntax provides for a privileged status of one argument of the verb,
featuring it as the subject of the sentence. There is a hierarchy among the arguments
of a verb. If there is an AGENT or EXPERIENCER, i.e. a participant that is typically
human, it will appear as the subject of the sentence. In the absence of such a role,
other arguments may be assigned the subject status. Recall the variants of the verb
open in (19), chapter 5: if there is an AGENT involved, it will appear as the subject; in
136
Understanding semantics
absence of an AGENT, an INSTRUMENT, if involved, will assume subject status; if there
is neither an AGENT nor an INSTRUMENT specified, lowest ranking THEME appears as
subject.
6.1.2 Voice, diatheses and alternations
The core argument structure of a verb can vary and it can be changed by certain
grammatical operations. We encountered verbs with varying argument structure in
chapter 5 (eat, give and open). These variations are technically called alternations.
Alternations come with different syntactic constructions, e.g. with or without a direct
object, but without a change of the form of the verb; we will turn to them in 6.1.3. The
more general term for changes of argument structure is diathesis (plural: diatheses);
it also covers morphologically marked variation of the argument structure, like
the passive in English. We will briefly have a look at four diatheses: passive and
antipassive, causative and anticausative.
6.1.2.1 Passive and antipassive
1
We mentioned the passive in 5.6.1. In English, the common passive is formed by
combining the past participle with the auxiliary be. It is only possible with agentive
verbs with a further argument. One of the further arguments becomes the subject
argument, while the subject of the active sentence is either dropped or turned into a
by-PP. (3) gives the passive versions of some sentences.
(3)
a.
active: The boy (AGENT) watched the man (PATIENT).
passive: The man (PATIENT) was watched [by the boy (AGENT)].
b.
active: She (AGENT) threw the letter (THEME) into the shredder (GOAL).
passive: The letter (THEME) was thrown into the shredder (GOAL) [by her
(AGENT)].
c.
active: She (AGENT) taught Dutch (THEME) to the immigrants (RECIPIENT).
passive: Dutch (THEME) was taught to the immigrants (RECIPIENT) [by her
(AGENT)].
d.
active: She (AGENT) taught the immigrants (RECIPIENT) Dutch (THEME).
passive: The immigrants (RECIPIENT) were taught Dutch (THEME) [by her
(AGENT)].
The three-place verb teach alternates between two constructions that correspond to
different passive constructions. The application of passive has two effects: a former
non-subject argument NP is ‘promoted’ to the privileged function of subject. At the
same time, the former subject is ‘demoted’; it is either dropped altogether or degraded
to an optional oblique argument, peripheral to the predication. Thus, the number of
1 There is also the so-called get-passive as in He got injured in the fight, which will not be discussed
here.
Verbs
137
core arguments is reduced by one. In German, the number of core arguments can be
reduced to zero, if the verb has an AGENT as its only argument. There are sentences like
those in (4); in English the same effect can only be achieved with impersonal active
constructions using they, one, you and other impersonal expressions as subject. This
is a secondary way in English to demote the subject argument.
(4)
a.
Hier wird gearbeitet.
here work.PASS.PRS.3SG
lit. ‘here is worked’ = ‘people/they/you work here’)
(German)
b.
Gestern wurde gestreikt.
yesterday strike.PASS.PAST.3SG
lit. ‘yesterday was striked’ = ‘they went on strike yesterday’
(German)
The antipassive in English consists of demoting the direct object argument by
omitting it. It removes the THEME/PATIENT argument from the set of core arguments.
English does not mark the antipassive by a different form of the verb, but there
are languages that do. Thus, the antipassive in English is a matter of alternation of
certain verbs. We already looked at one example of this in the previous chapter: the
use of eat without a direct object:
(5)
active:
antipassive:
She (AGENT) ate her cereals (THEME).
She (AGENT) ate.
The omitted THEME/PATIENT argument is interpreted as ›something/someone‹
depending on the selectional restrictions of the verb for this argument. The
antipassive alternation with eat is responsible for the inconclusive status of the
object argument (recall the discussion in 5.3.3 on the number of arguments of
the verb). In fact, eat can be used in different voices. In the active voice, AGENT and
THEME are both obligatorily specified, by subject and direct object, respectively;
in the antipassive voice, only the THEME is specified, by the subject; in the passive
voice, the THEME obligatorily appears as subject, while the AGENT may optionally be
specified with a by-phrase. Thus, the number of arguments of a verb is not just a
matter of the verb as it is lexicalized, but a matter of the verb and the voice in which
it is being used.
The antipassive is not possible with all transitive verbs. A minimal pair often
quoted is eat vs devour, which does not allow omission of the object.
6.1.2.2 Causative and anticausative
The causative diathesis has the opposite effect of passive: it adds a core argument.
The causative of drink, for example, means ›let drink‹ or ›have drink‹. It introduces
a new AGENT, the ‘causer’ and turns the original subject argument into a PATIENT, the
‘causee’. The causer appears as the subject of the sentence, the causee as direct or
indirect object. (6a, b) are examples from Japanese. Japanese has a suffix -(s)aseattached to the stem of the verb which turns it into a causative:
138
(6)
Understanding semantics
a.
b.
kodomo ga
biiru
child
NOM
beer
‘The child(ren) drank beer’.
o
ACC
nonda.
drink.PAST
obâsan
ga
kodomo ni
biiru
grandma NOM
child
DAT
beer
‘Grandma had/let the child(ren) drink beer.’
(Japanese)
o
ACC
nom- ase- ta
drink- CAUS PAST
The causative form in Japanese has a weak reading of allowance and a strong reading
of causation. English has no grammatical causative. There are three options of
expressing causation: (i) using, if possible, the same verb in a causative construction:
intransitive open > transitive open; (ii) using a corresponding lexicalized causative
verb: eat > feed; (iii) combining the verb with auxiliary-like have or let: eat > have eat,
let eat. Neither (i) nor (ii) constitute general options for English verbs. By contrast,
the Japanese causative can be applied to almost all verbs. The third type of English
construction, too, is generally available; let carries the weak meaning, have the
strong meaning, but lexically causative verbs like feed ‘make eat’, teach ‘make learn’,
show ‘make see’, kill ‘make die’ and very many others only have the strong reading of
causation. The same holds for lexically causative verbs in Japanese, for example for
oshieru ‘teach’, korosu ‘kill’, miseru ‘show’, etc.
In Japanese, the passive can be applied to causatives. First, the causative introduces
a causer into the argument structure, then the passive demotes or eliminates the
causer. What remains is the element of causation in addition to the event as such
that is expressed by the original verb. Compare (6c) to the simple active sentence
in (6a):
(6)
c.
kodomo ga
biiru o
nom- asechild
NOM
beer ACC drink- CAUS
‘someone let/had the child(ren) drink beer’
rare-
ta.
PASS
PAST
(Japanese)
The anticausative is a diathesis that removes a causer AGENT from the set of core
arguments. We already encountered the example of intransitive open. Another
example would be break, widen or burn:
(7)
a.
She (AGENT) opened the door (THEME).
anticausative:
The door (THEME) opened.
b.
She (AGENT) burnt his letters (THEME).
anticausative:
His letters burnt well.
Unlike with the antipassive and the passive, the argument omitted is not understood
as still being involved. Along with the causer AGENT, the whole meaning component
of causation is eliminated; (7a) does not mean ›someone/they opened the door‹ –
although we might infer causation from our world knowledge. Likewise, (7b) does
not mean that someone burnt the letters. In German and other languages, including
Spanish, the anticausative is expressed by using a reflexive pronoun as object. The
Verbs
139
reflexive pronoun is a semantically empty grammatical element that just fills the
object position required by the syntax for the originally transitive verb.
(8)
Die Tür (THEME) öffnete sich.
lit. ‘the door opened itself ’
(German)
6.1.3 Levin’s classification of verbs
Levin (1993) used alternations of verbs in English for a systematic semantic
classification of more than 3,000 English verbs. According to her, verbs with the
same patterns of syntactic constructions and alternations among them have certain
meaning components in common, and these are responsible for their syntactic
behaviour. In the first half of her book, Levin describes eighty alternations in English,
where an alternation is defined by a pair of constructions applicable to the same
verb, but differing in the choice and linking of the arguments involved. For example:
active–passive, active–antipassive or transitive–anticausative would be alternations.
In the second half, she defines fifty-seven major verb classes (many with several subclasses) in terms of the alternations they do or do not exhibit. We will take a look at a
few sample classes, mainly concerning verbs we have already encountered.
Open. The verb open belongs to a sub-class of class 45, ‘Verbs of change of state’ (pp.
240 ff.). They share the ‘causative/inchoative alternation’ and the ‘instrument subject
alternation’. Due to the causative/inchoative alternation, there is a transitive verb
version with an AGENT subject and THEME object (she opened the door) along with the
corresponding anticausative ‘inchoative’ variant (the door opened). The instrument
subject alternation allows the use of an INSTRUMENT subject (this key opens the door).
There are further, minor, alternations possible for this class. Other verbs belonging to
the larger class are break, bend and bake.
Eat. Eat and drink form a sub-class of class 39, ‘Verbs of ingesting’ (pp. 213 ff.). Most
of them allow an antipassive use, which Levin captures with the ‘unspecified object
alternation’: the THEME object can be omitted (Kate ate) and the THEME argument is
understood as existing, but not specified. In addition, these verbs exhibit the ‘conative
alternation’:
(9)
a.
Sheila ate the sashimi.
b.
Sheila ate [at] the sashimi. (conative)
The conative variant means that the act of eating is related only to parts of THEME
– Sheila may be just nibbling at the sashimi, while the basic construction expresses
that THEME is affected as a whole. Note that this alternation is impossible with verbs
such as open.
Give. Give is representative of a larger sub-class within class 13, ‘Verbs of change of
possession’ (pp. 138 ff.). Most of them exhibit the ‘dative alternation’ between, e.g.
‘give RECIPIENT THEME’ and ‘give THEME to RECIPIENT’. Similar verbs are pass and offer.
140
Understanding semantics
The dative alternation is also possible for send, but this verb belongs to class 11,
‘Verbs of sending and carrying’ (pp. 132 ff.).
Shave. Shave belongs to class 41,‘Verbs of grooming and bodily care’. It has a transitive
AGENT–PATIENT use and an intransitive use (he shaved), where the PATIENT is understood
to be the AGENT him- or herself. Levin calls this the ‘understood reflexive object
alternation’. Like the antipassive, this alternation eliminates the direct object from
the construction, but its semantic effect is different (note that Kate ate does not mean
‘Kate ate herself’). In languages with a more extensive use of reflexive pronouns or
verb forms, the equivalent alternation would make use of a reflexive pronoun object:
(10) a.
b.
transitive: Das Kind (AGENT) wusch
the child
washed
reflexive:
Das Kind (AGENT) wusch
the child
washed
die Puppe (PATIENT). (German)
the doll.
sich.
REFL
This group includes verbs such as wash, dress or comb.
Marry. Marry is a ‘verb of social interaction’ (class 36, pp. 200 ff.). There are two
characteristic alternations; some verbs of social interaction exhibit the first, others
the second:
(11) simple reciprocal alternation
a.
Eunice and Eugene hugged/kissed/married.
b.
Eunice hugged/kissed/married Eugene.
understood reciprocal object alternation
c.
Eunice and Eugene quarrelled/fought/agreed.
d.
Eunice quarrelled/fought/agreed with Eugene.
In (11a and c), the subject is an NP conjunction that refers to two arguments;
alternatively the subject may make use of a collective noun (the team, the pair, the
family). The persons referred to inhabit the same role in the situation; there is no
asymmetry as between the usual AGENT and PATIENT, where the AGENT is in an active
role and the PATIENT in a passive one. The construction in (11b) is asymmetric, and
it does allow for an interpretation where the subject referent plays a more active role.
(11d) is syntactically asymmetric, but the understanding of the two roles is reciprocal.
6.2 SITUATION STRUCTURE
Levin’s classification does not relate to the temporal structure of the situations
expressed by the verbs. The temporal or situation structure of the situations
expressed by verbs is by far not as manifold as Levin’s classes. There are a number
Verbs
141
of terms for what is captured with this kind of classification of verbs: there is the
traditional term ‘aktionsart(en)’ (German, lit. ‘kind(s) of action’), others talk of
‘inherent aspect’ or ‘aspectual class’. We will use the term ‘aspectual class’. All these
terms are motivated by the fact that these classes are relevant for understanding the
interaction of verb meaning and aspect. Some distinguish only three or four, others
more, up to a magnitude of twenty (Russian ‘aktionsarten’). Thus, classifications of
verbs according to situation structure are much more general than Levin’s classes.
The main aspects of situation structure are whether a verb is stative or dynamic,
whether it conceives of the situation as simple or complex, and whether the verb
concept provides a criterion for the completion of the event denoted. I will introduce
five classes in this section, and take a look at their interaction with aspect in 6.3.
When talking of aspectual classes, it is most important to realize that the
classification applies to linguistic expressions and their meaning, NOT to types of
situations in the world, such as events, activities, processes, states, etc. Unfortunately
the literature in this field is often inconsistent in this regard. You will find formulations
like ‘work is an activity’. This is a confusion of semiotic levels; it should be: work is an
activity VERB and its meaning an activity CONCEPT.
6.2.1 Accomplishment terms
Accomplishment expressions conceive of a situation as a process or activity that
leads to a specific result. The process or activity is conceived of as uniform and
as involving some sort of constant change; it ‘culminates’ in the result specified.
Accomplishment verbs, i.e. verbs that by their very lexical meaning express an
accomplishment, are hard to find in English. But accomplishments can be easily
expressed by the combination of certain transitive verbs of action with an object NP:
eat an apple, write a paper, build a house, drive to the station.
Let us take a closer look at eat an apple. The expression depicts a situation denoted
as a particular activity of the AGENT (let us assume for the sake of illustration that the
AGENT is a person) and that they perform their activity on the THEME: the AGENT bites
off pieces of the THEME, chews them and swallows them, making the THEME smaller
and smaller. This ongoing activity can be referred to in the progressive mode: She
is eating an apple. Eventually, the AGENT may have reached the end of their activity
by having eaten as much of the THEME as there is to be eaten (people differ in their
eating habits as to which parts of the apple they actually eat – some peel it, some
leave the core, some eat everything except the stalk – but this does not matter). Thus,
apart from the THEME chosen for the example, the concept ›eat an apple‹ specifies two
ingredients of the situation expressed: (i) the kind of process or activity going on and
(ii) a criterion for its culmination and completion. These ingredients constitute an
accomplishment concept.
In Fig. 6.1, the horizontal arrow represents the time line; there is a certain period,
or stage, of a uniform activity or process in time represented by the grey rectangle
covering a section of the time line; the little triangles indicate the uniform and
culmination-oriented dynamicity of this phase; the bold vertical stroke signifies the
culmination point. The broken-line ellipsis encircles the whole situation. This is what
142
Understanding semantics
Figure 6.1
Structure of an accomplishment concept
an accomplishment term refers to: the dynamic activity or process together with its
culmination.
It is very important to realize that situations in the world are not accomplishments
by themselves, or of any other aspectual class. When you read this text, there is a
situation in which you are looking at your screen or tablet or holding a copy of this
book, looking at it, reading the words, trying to make sense of them. Describing
the situation, you might opt for putting it as an ongoing accomplishment, by saying
(12a), or you might choose to say (12b), putting it as just an ongoing activity without
formulating a description that provides a criterion of completion for the ongoing
situation.
(12) a.
b.
I’m reading the section on accomplishment terms.
I’m reading in the verbs chapter.
Thus the difference between an activity as part of an accomplishment and the mere
activity does not lie in the actual events; it lies in the way one chooses to describe
them, and in the conceptual distinctions which a speaker chooses to apply.
It must be cautioned that the term ‘accomplishment’ is not restricted to human or
animate actions, i.e. to situations with an AGENT argument. There are also concepts
for non-agentive processes with a specified result, but these verbs are rare; evaporate
would be one. As it happens, however, the term ‘accomplishment’ is so solidly
established in the literature that a more neutral term never came up.
6.2.2 Activity and process terms
Activity concepts lack a criterion of culmination. We will use the more general
term ‘process’ in order also to capture expressions that lack an AGENT argument. The
structure of a process concept is depicted in Fig. 6.2.
Figure 6.2
Structure of a process concept
Verbs
143
Typical process terms are intransitive verbs that denote some manner of acting by
the AGENT – eat, work, sleep, sing, read, walk, jog – or a process involving the THEME,
e.g. intransitive verbs such as glide, flow, buzz, vibrate. Due to the process component
of the concept, process terms, too, can be used in the progressive.
Accomplishment and activity predications with the same verb. A very
important point to observe with verbs of activity like eat, write, read, build is that
the specification of the object will determine whether they yield an accomplishment
or an activity predication. If the activity is such that you perform it only once on
a given object (like eating or writing, but not necessarily watching or singing),
specification of a quantity of the object argument (‘quantized object’) will lead to
an accomplishment predication. But bare plurals or bare mass nouns do not fix a
quantity and therefore provide no criterion for the activity being completed (recall
4.4.5). Thus, only sentences like those in (13a) can be taken as accomplishment
predications, while those in (13b) express activities.
(13) a.
b.
quantized THEME
non-quantized
she ate an apple/the apple/three apples/all the apples …
she ate apples/soup
In addition, it must be cautioned that some activity verbs with a quantized object
may be taken in a conative sense like in (9b). If so, they take on an activity meaning,
because the conative voice excludes the culmination of the activity. Thus the
conative voice can be understood as removing the culmination component from an
accomplishment concept.
For a large group of verbs, the accomplishment character with quantized objects
(in non-conative voice) depends to some degree on context. These are verbs for
actions which are connected to specific aims, where the aim may be accomplished
or not, or be accomplished more or less, when the action is performed. Consider the
transitive wash the clothes, or tidy the bedroom or the implicitly reflexive wash. These
activities are aimed at achieving certain results, like the THEME or AGENT’s body being
clean or the THEME being tidy. The intended results are part of the meaning, because
the kind of activity these verbs refer to is largely left open in the verb meaning except
for the criterion that they are apt to lead to the result to be achieved. (Just think of
the completely different activities involved with washing clothes by hand or by using
a washing machine – both are ‘washing clothes’). If someone says ‘I’ve washed/I’ve
washed the clothes/I’ve tidied my room’, these will by default be taken as reports of
accomplishments, as activities which reached their goal. But the same expressions
can also be used for referring to the mere activities which may or may not have
succeeded.
There is a standard diagnostic for distinguishing the accomplishment sense
from the mere activity sense. With accomplishment predications one can add an
in-adverbial for specifying the time it took to complete the situation:
(14) a.
b.
c.
He ate an apple in 30 seconds.
He washed in 3 minutes.
He tidied his desk in half an hour.
144
Understanding semantics
For mere activities, one can add a for-adverbial that specifies the length of time
during which the activity was carried out, leaving open whether the aim was
achieved or not.
(15) a.
b.
He washed for 3 minutes (but it didn’t make much of a difference).
He tidied his desk for half an hour (but it didn’t make much of a
difference).
For accomplishment predications, for-adverbials are impossible – or the predication
must be shifted to a different reading. Consider (16a) and (16b):
(16) a.
Sheila drank a cocktail for two hours.
b.
Sheila drove to the bar for 20 minutes.
(16a) coerces a conative activity reading. (16b) is odd if it is to be taken in the sense
that Sheila engaged for 20 minutes in the driving-to-the-bar activity. The sentence
does have an acceptable reading if the adverbial for 20 minutes is taken to specify
the duration of Sheila’s stay at the bar. But then, the for-adverbial relates not to the
situation referred to, but to the state resulting from it. Thus, we have to be careful
when applying the for-test: the adverbial must be taken to apply to the situation itself.
Conversely, an in-adverbial coerces an accomplishment reading on an activity verb:
(17) She jogged in ten minutes.
This sentence would be understood as expressing that the AGENT did her standard
quantum of jogging in ten minutes; ›do one’s standard quantum of jogging‹ is an
accomplishment concept.
Let me sum up what we have said about accomplishment and process terms. Both
refer to a situation as having a dynamic process component. The process component
is a uniform dynamic situation of some kind that continues in time. In addition,
accomplishment concepts contain a criterion of culmination. We mentioned three
linguistic characteristics which can be used as diagnostics. Both, accomplishment
and activity terms, can be used in the progressive, relating to the ongoing process.
Accomplishment terms can be combined with in-adverbials that specify the amount
of time consumed till the situation is completed; process terms cannot. Process
terms can be combined with for-adverbials that specify the overall duration of the
process; accomplishment terms cannot. Application of the ‘wrong’ adverbials either
does not relate to the situation itself or coerces a meaning shift of the VP.
6.2.3 Simple change and achievement terms
Many verbs express just a simple change: the THEME or AGENT is in a certain condition
before, and in a different condition after, the event. Let us refer to these conditions as
the initial condition and the resultant condition. (18) lists a few examples:
Verbs
145
Predication
Initial condition
Resultant condition
a.
She entered the room.
she is not in the room
she is in the room
b.
She turned the TV on.
the TV is not on
the TV is on
c.
She stopped.
she is moving
she is not moving
d.
She started reading.
she is not reading
she is reading
e.
She left.
she is here
she is not here
f.
She died.
she is alive
she is not alive
g.
The door opened.
the door is not open
the door is open
h.
The letter arrived.
the letter is on its way
to its destination
the letter is at its
destination
i.
She reached the top.
she is climbing to the top
she is at the top
(18)
Simple change predications always presuppose that the initial condition holds prior
to the change (recall 4.7.2.2); otherwise this kind of change would be logically
impossible. The initial condition may be a dynamic process, as in (18c, h, i), or a
state, i.e. a constant condition that does not involve change. The same applies to the
resultant condition: in (18d) it is dynamic, in the other cases a state. In the literature,
often all simple change terms are called achievement terms. I prefer to reserve the
term for predications with a dynamic initial condition, i.e. to terms that denote the
culmination of a process, such as arrive or reach. Usually, one of the two conditions is
defined as the negation of the other. In most cases, it is the resultant condition that is
defined positively; we will then have an ‘ingressive’ or ‘inchoative’ verb which is used
to express the beginning of a state or process. If the initial condition is the one which
is positively defined, as in stop, halt, end, leave, die we have an ‘egressive’ verb that
describes a situation as the ending of a state or process. Simple change verbs can be
schematically illustrated as in Fig. 6.3; the initial condition is marked with horizontal
stripes, the resultant condition with vertical stripes. Figure 6.4 displays the special
case of an achievement concept.
Figure 6.3
Structure of a simple change concept
Accomplishment terms and achievement terms differ in one point only: the
referent of accomplishment terms includes the culmination point plus the whole
146
Understanding semantics
Figure 6.4
Structure of an achievement concept
process leading to it; achievement terms only refer to the culmination point, while
the existence of the preceding process is presupposed. Consider the following pair
of sentences:
(19) a.
b.
Sheila drove to the supermarket.
Sheila arrived at the supermarket.
The first is an accomplishment sentence. It refers to a situation which comprises
both the drive and the arrival. If the sentence is negated (Sheila didn’t drive to the
supermarket), it is denied that Sheila undertook the whole action of driving to the
supermarket. Compared to this, the second sentence only refers to reaching the
endpoint of going to the supermarket. If the sentence is negated (Sheila didn’t arrive
at the supermarket) it is understood, that Sheila was on her way to the supermarket,
but for some reason did not complete going there.
The concept of a change term refers just to the change; the change is the transition
from the initial condition to the resultant condition. A simple change concept does
not contain information about the manner of transition, or its duration. In this sense,
the change is ‘simple’. Simple change verbs are often said to be punctual, but this is
misleading because it evokes the wrong impression that the situation in the world
that is referred to with a simple change verb must not have any temporal extension,
or at least be very short. These concepts are ‘punctual’ in the abstract sense that they
do not contain any information about the transition event itself; to these concepts the
changing event might as well be just a ‘point’ without temporal extension. To illustrate
the point, let me mention some Japanese examples: iku (‘go’), kuru (‘come’) or otiru
(‘fall, drop’) are simple change verbs. They cannot be used in the progressive in order
to express that the AGENT/THEME is on their way between A and B.
Simple change verbs are not used in the progressive, because there is no process
defined which the progressive might state as going on. They do not take foradverbials, or if they do, the for-adverbial relates to the resultant condition (she
opened the window for ten minutes). If the initial condition is not a process, they do
not take in-adverbials either, except with special contextual support. For example, she
opened the window in ten minutes would require a context in which her opening the
window was due for some reason. Simple change verbs are fine with at-adverbials
– provided the time specified by the adverbial is long enough to cover the actual
transition. For example, she changed to Beijing University denotes a change that
actually would take a couple of days, at least. Combining this statement with at 2:43
would make no sense, but at the beginning of the next term would be fine.
Verbs
147
6.2.4 Simple occurrence terms
Verbs like click, hit or knock are similar to simple change verbs in not claiming
temporal extension of the situation referred to. They differ in not expressing a
change between an initial and a resultant condition. The world is the same before
and after a click, a hit or a knock. These simple occurrences are conceived of as just
interruptions. Schematically, they can be represented as in Fig. 6.5. The term ‘simple
occurrence’ is introduced here. In the literature they are sometimes referred to as
‘semelfactives’; other classifications do not distinguish this class.
Figure 6.5
Structure of a simple occurrence concept
Characteristically, simple occurrences can be repeated because the conditions
which make a first occurrence possible will not change with it. If you have just
knocked, you can knock again (actually, one usually knocks several times when one
knocks). This is different with change terms: for example, if you have just opened
a window, you cannot open it again, you would have to close it first. By applying
repetition, simple occurrence predications can be turned into predications about
uniform, dynamic processes and then lend themselves to the progressive and to foradverbials (she knocked for 30 seconds, she is presently knocking at his window). Thus,
the progressive and for-adverbials coerce a repetitive reading of these verbs. In their
single occurrence reading, they are not combinable with the progressive, nor with
for- or in-adverbials. Of course, they are fine with appropriate at-adverbials.
Before we proceed to the aspectual class of states, which is very different from
the classes discussed so far, let me sum up the four dynamic classes and their
respective diagnostics. In Table 6.1, a class receives a negative entry for a given test,
if its application is necessarily connected with a meaning shift such as coercing
simple occurrence predications into process predications or applying the conative
alternation to accomplishment terms. Among simple change verbs, achievement
terms allow for in-adverbials while the others do not.
Table 6.1
Diagnostics for the dynamic aspectual classes
Aspectual class
progressive
in-adverbial
for-adverbial
at-adverbial
accomplishment term
+
+
–
–
process term
+
–
+
–
simple change term
–
–/+
–
+
simple occurrence term
–
–
–
+
148
Understanding semantics
While these are just diagnostics, the four classes can be distinguished semantically
by the following properties of the underlying concepts: (i) presence of a process
component as part of the situation referred to; (ii) specification of a resultant
condition; and (iii) conceptual ‘punctuality’. For accomplishment predications, the
resultant condition is the state reached with the culmination of the situation: the
apple is finished, the paper is written, the house is built, the drive has reached its
destination. The distribution of the three properties is displayed in Table 6.2. They
will play an important role for the interaction of verb meaning and aspect.
Table 6.2
Semantic properties of aspectual classes
Aspectual class
process component
resultant condition
‘punctuality’
accomplishment term
process term
+
+
–
+
–
–
simple change term
–
+
+
simple occurrence term
–
–
+
6.2.5 State terms
State terms denote a condition which is, at least temporarily, constant. States are
predicated of times, as we will see in the next section. States may result from a
change, but they do not themselves involve change. For example, the dynamic verb
memorize may lead to a state of knowing which once established is conceived of as
constant. State predications can be schematically illustrated as in Fig. 6.6. There
are not so many state verbs in English, mainly verbs such as like, love, hate, know,
understand, want or have, belong, contain, etc.; some verbs are state verbs only in
certain readings, e.g. cost, weigh, taste, mean in uses like the ones in (20). Other verbs
are ambiguous between a state reading and an ingressive change reading; these
include understand as well as sit and lie.
Figure 6.6
Structure of a state concept
(20) a.
A flight to Glasgow costs 19 euros.
b.
She weighs 59 kilos.
c.
This cake tastes awfully sweet.
d.
Japanese uni means ‘sea urchin’.
Verbs
149
State predications as such are not used in the progressive. Progressive with a state
expression such as she is being rude/silly coerces a meaning shift from state to activity,
in this case to ‘be acting in a rude/silly manner’. The states expressed by state verbs
may last for a long or short period of time, they may be temporary or permanent,
they may have a beginning and an end or not. For example, the state expressed by be
dead has a beginning, but not an end; it is what is called an irreversible state. With
conditions like these, it depends if a state expression can or cannot be combined with
a for-adverbial or with an at-adverbial. Thus the applicability of temporal adverbials
yields an inconclusive overall picture for state terms.
There is one point in which state predications differ clearly from all the dynamic
aspectual classes considered above: when used in an episodic sentence (i.e. applying
to a particular concrete situation, recall 4.6 on genericity) with the present tense, it
is only state predications that can be used for genuine present time reference, i.e.
reference to the time of utterance. With dynamic predications, present tense use
with the plain (non-progressive) form of the verb either is impossible in episodic
sentences or yields future time reference. Consider the following contrast:
(21) a
State
I know the answer.
b.
State
She hates me.
c.
Accomplishment
I write a paper. (??)
d.
Process
I walk. (??)
e.
Simple change
I switch on the TV. (??)
f.
Simple occurrence
I knock. (??)
Dynamic verbs in their plain (i.e. non-progressive) form can only be used with the
present tense in special cases.
(22) a.
I promise I will wait for you.
b.
Now Khedira passes the ball towards Özil, but …
c.
In 1815, Napoleon loses the battle of Waterloo and is banished to Saint
Helena.
d.
I commute to work by subway.
In (22a), promise is used as a ‘performative verb’, i.e. a verb which indicates the
speech act that is performed by using it: saying ‘I promise …’ constitutes a promise.
(22b) is only possible as a running commentary. (22c) is in the ‘historic present’,
referring in the present tense form to past events and situations and thereby
simulating a witness perspective. (22d) is a habitual statement, i.e. a generic sentence
about a situation that recurs regularly; a single such situation need not be in progress
at the time of utterance. (22a) and (22b) constitute, or at least claim, coincidence
of the event expressed with the utterance itself; historic present refers to the past,
150
Understanding semantics
habitual predications do not refer to single events. We will return to these special
cases in 6.4.2. Apart from these exceptions, for dynamic verbs, the present tense with
reference to the time of utterance is only possible in the progressive – and this is
restricted to accomplishment and process predications:
(23) a.
I am writing a paper.
b.
I am walking.
c.
I am switching on the TV. (??)
d.
I am knocking. (repetitive reading only)
The classification presented here ultimately goes back to the very influential work
on English aspectual classes in Vendler (1957/67); many call aspectual classes
simply ‘Vendler classes’. He introduced the four classes of accomplishment, activity,
achievement and state terms and used the diagnostics of progressive and for-, in- and
at-adverbials. His proposal was later generalized and elaborated, to be extended to
more general and additional types of verbs, as well as to different languages. The
classification presented here can be considered essentially common sense in the field.
6.3 ASPECT
As to the description of aspect and tense, there are a lot of half-truths to be met in
the literature, in particular in descriptive grammars. In the following two sections, I
2
will try my best to give a clear and consistent picture of the matter.
6.3.1 The central distinction: imperfective vs perfective aspect
Talking of ‘aspect’, the fundamental distinction is between imperfective and
perfective aspect. It can be illustrated with a minimal pair such as this:
(24) a.
b.
imperfective:
I was watching TV [when the door bell rang].
perfective:
I watched TV for a while [then went to bed].
The two sentences might be used with reference to the very same instance of the
speaker’s watching the TV. The distinction between perfective and imperfective
aspect is not to be found in the way things are in the world. Rather, it is a matter of
the way in which the speaker chooses to talk about the world.
The imperfective variant is a predication about a given time; in the case of (24a),
the time is determined by the punctual event of the door bell ringing (a simple
occurrence expression) and, less specifically, by the past tense. What the main clause
of the imperfective sentence says is that, at the time considered, the speaker was in
2 The approach taken here owes much to the theory of aspect developed in Galton (1984).
Verbs
151
the state of watching TV. An imperfective predication is about a contextually given
time; it tells what kind of situation is given at that time. It does not tell anything
about the situation before or after the time referred to. Therefore people often say that
imperfective aspect relates to uncompleted situations (in fact the term imperfective
is motivated by this view); but this is not really adequate – since imperfective aspect
leaves open what happens after the time referred to, the situation may be later
completed or not. Very often you will find the definition that with imperfective aspect
the situation is ‘seen from within’. This is not a very clear notion. The crucial point
is that predications in the imperfective aspect are predications about a given time
which describe the situation at that time. If the situation extends beyond the time
predicated about, the time can be considered embedded in the total situation, and
therefore the total situation is in some sense ‘seen from within’.
The time referred to may be a point in time or an extended period. In any event,
the state expressed by the imperfective predication must not change during this time.
This is a consequence of the presupposition of indivisibility (4.7.2.2) which applies
to every predication whatsoever, including the predication about time expressed by
an imperfective predication. In this sense, the time referred to is LIKE a point, i.e. not
divided. It is the same sense of punctuality which we encountered with ‘punctual’
verbs. The time referred to is located by the tense of the verb: in (24a) it is a past
time, i.e. located before the time of utterance. Thus a tensed imperfective sentence
constitutes a twofold predication about a given time: a temporal localization by
means of grammatical tense and a specification of the conditions at that time by the
rest of the predication.
The perfective sentence in (24b) does not predicate about a time, it predicates
about an event – the referential argument of the verb. In this case, the predication
describes what the speaker did: watching TV (and then going to bed). The event is,
in addition, located in time, expressed by the past tense: it occurred at some time
before utterance time. Thus, tensed perfective sentences, too, provide a twofold
predication, but this time the two predications are about an event – by tense about
its temporal location, and by the rest of predication, about the referential argument
of the verb. Due to the presupposition of indivisibility, the two predications must
apply to the event argument as a whole: the event, or situation, referred to must be
of the kind the sentence describes and it must be wholly located in the interval of
time referred to. The perfective aspect is therefore often described as ‘referring to an
event as a whole’. You may also find the more traditional description that perfective
aspect refers to a completED event. This is wrong, for the simple reason that it is
possible to use perfective aspect referring to future events which, of course, would
not be completED yet. Perfective aspect deals with complete events, not necessarily
with completED events.
Figure 6.7 illustrates the difference between imperfective (left) and perfective
(right) aspect. It is a matter of perspective. Imperfective aspect views the situation
from a given time, stating that the world is, at this time, in a certain state. Being
about one time, the predication can only be a state predication because any kind of
change can only be stated with respect to at least two times. Perfective aspect views
the situation from an event in the world, stating that and when it happens. Thus,
152
Understanding semantics
Figure 6.7
Imperfective aspect and perfective aspect
Imperfective aspect
Perfective aspect
event
world
state
time
world
interval
time
time
imperfective aspect ‘looks’ from a time into the world, and perfective aspect ‘looks’
from an event in the world onto the time line. These are two basic ways of correlating
situations in the world with the time line.
The difference, and the interaction, of perfective and imperfective aspect can be
illustrated with narrative texts, such as stories, tales, reports or novels. The following
is the beginning of the translation of Grimm’s fairytale of the frog prince. The tensed
predications are marked for imperfective (ipf) and perfective (pf) aspect. The
perfective predications are put in italics.
(25) One fine evening a young princess put on her bonnet and clogs (pf), and went out
to take a walk by herself in the wood (pf); and when she came to a cool spring of
water (pf), that rose (ipf) in the midst of it, she sat herself down to rest a while
(pf). Now she had a golden ball in her hand (ipf), which was her favourite
plaything (ipf); and she was always tossing it up into the air (ipf), and catching
it (ipf) as it fell (pf). After a time she threw it up so high (pf) that she missed
catching it again (pf) as it fell (pf); and the ball bounded away (pf), and rolled
3
along upon the ground (pf), till at last it fell down into the spring (pf).
The narration starts with a time adverbial One fine evening that defines the temporal
frame for the series of events described in the following. All the events mentioned in
the passage are located within this time interval. It corresponds to the marked time
zone in the diagram for perfective aspect. The first two events are told chronologically
(putting on bonnet and clogs and then going out). The first event is located
indeterminately within that evening, setting a temporal starting point for the story.
The second event is taken as following next; narration time advances a bit. Some time
will have elapsed until the next event mentioned takes place, her coming to the spring
in the wood. The narration here inserts an imperfective predication about the location
of the spring. This is background information predicated about the time reached in
3 Quoted from Grimm’s Fairy Tales, Amazon Kindle edition.
Verbs
153
the narration. The spring, of course, has existed much longer, but this is irrelevant at
this point; what matters is the state of the world in the situation described. Now she
sits down – the next event in the series. By the following two imperfective statements
we learn that at this moment she has a golden ball in her hand and that it is her
favourite plaything. There follow two predications in the progressive; progressives are
imperfective. They express continued uniform action as a temporary state. We will
relate the total predication to some stretch of time, because the repetition of tossing
and catching the ball necessarily consumes time; we will locate this interval after she
has had her rest. The text then describes a series of events in perfective aspect, one
following the other, carrying narration time further to the situation where the frog
will now appear. What we can see looking at this example is the dynamic function of
perfective aspect – it takes the story forwards – and the static function of imperfective
aspect which gives information about a given situation at a given time.
6.3.2 Perfective aspect
Among the aspectual verb classes introduced above, state terms, when used with
the plain form of the verb, yield imperfective predications; dynamic aspectual
types yield perfective aspect with the verb in its plain form and in episodic use.
If the perfective aspect is applied to state predications, they are turned into event
4
predications. One way to turn a state concept into a simple occurrence concept is
by adding an adverbial that provides the state with a temporal delimitation. The
resulting event consists of a state that pertains for the time specified:
(26) I was in the garden for a while and then washed the dishes.
Another common strategy is to use a state term to refer (metonymically) to the
beginning of the state; this turns a state predication into an ingressive simple change
predication:
(27) a.
b.
I’ll be with you in a minute.
She had a ticket within half an hour.
6.3.3 Imperfective aspect
While state predications immediately yield imperfective aspect, event predications
must be turned into states for enabling imperfective aspect. The most prominent
means is applying the progressive. The progressive is a variant of the imperfective
aspect; it expresses the state of a dynamic condition that continues uniformly, such
as some activity being done or some process going on. In English, the progressive
is possible with accomplishment and process predications, as well as with simple
occurrence predications (knock, beat) in a repetitive interpretation. As mentioned
4 In the discussion of aspect, I will use the term event predication for dynamic predications in
general.
154
Understanding semantics
above, certain states (be silly, be rude, be polite) can be used in the progressive, taking
on the meaning of the activity of acting in a certain way.
Another way of using event predications in imperfective aspect is the habitual
mode of speaking. Habitual predications express actions and events that are repeated
regularly; the repetitions form what is called a ‘serial state’:
(28) a.
b.
This shop opens at 9:30 a.m.
She listens to her mp3 player on the subway.
While these are examples where the plain verb is used, there are also special forms
in English for the expression of habitual predications, but these are restricted to the
past tense:
(29) a.
b.
This shop used to open at 9:30 a.m.
She would listen to her mp3 player on the subway.
6.3.4 Perfect aspect
In addition to perfective and imperfective aspect, we will discuss two more aspects:
perfect and progressive. The perfect aspect, too, yields a state predication about a
given time. The state is a state that results from a previous event. Consider a simple
example:
(30) I have written my term paper.
The sentence predicates about the present time that it is located in the state resulting
from a past event of the type ‘speaker write a term paper’. While reference is
primarily to the time predicated about, the event from which the current state results
is also referred to as having occurred before the time referred to. The perfect aspect
can be schematically depicted as in Fig. 6.8.
Figure 6.8
Perfect aspect
event
world
time
time
Verbs
155
In English, the perfect aspect is expressed by using the auxiliary have with the
past participle. It lends naturally to all classes of predications that provide a resultant
condition: accomplishment and change predications (recall Table 6.2). The perfect
is possible for non-resultative verbs, too. In this case, context must provide a sense
in which the situation at the given time results from the former event. For example,
consult in (31) is a simple occurrence verb that does not come with a specified
resultant condition. The sentence would be read in the sense that the speaker’s
consulting the dentist has led to some result, like now knowing better what to do
with their aching molar.
(31) I’ve consulted my dentist.
One variant of perfect, the experiential perfect takes any occurrence of an event as
resulting in a change. With this variant one would express that one has had a certain
kind of experience:
(32) Have you ever had renal colic?
English has a special use of the perfect, called the perfect of persistent situation.
It is used to express that a certain state has prevailed up to the time referred to. In
German, for example, this would be expressed in plain present tense.
(33) a.
b.
We’ve lived here for six years now.
Wir wohnen [present tense] hier jetzt seit sechs Jahren.
lit.: ‘We live here now since six years.’
(German)
It is very important to realize that the relation between the former event and the
subsequent state is not just temporal precedence, but a causal relation: the former
event is the cause of the subsequent state. It is this causal relation which connects the
time referred to to the former event. The causal connection is indicated by the curved
arrow from the event to the state in Fig. 6.8. Often this connection is described in
terms of ‘relevance’: the former event is relevant to the subsequent situation.
6.3.5 Prospective aspect
Prospective aspect is the temporal mirror of perfect aspect. It refers to a state that
leads up to a future event:
(34) a.
I’m leaving!
b.
It’s going to rain.
c.
The train is about to leave.
Be going to and be about to are usually cited as ways of expressing prospective
aspect: the event expressed is scheduled or on its way at the time referred to. The
156
Understanding semantics
be V-ing form normally used for the progressive, however, can also be considered
an expression of prospective aspect if it is combined with a punctual verb. (With
non-punctual verbs, the form inevitably yields the progressive reading.) This is the
case in (34a), taken as an announcement of an imminent action; and this is the way
in which we are arriving or he is dying would be interpreted. These constructions do
not express that a situation as described by the verb is going on; rather they describe
a situation that is causally linked to a future event of the kind – the situation is such
that this kind of event is likely to arise from it. Figure 6.9 illustrates the prospective
aspect.
Figure 6.9
Prospective aspect
event
world
time
time
While the prospective be V-ing form is restricted to punctual verbs, to be going to
V is possible with all aspectual classes, including state terms. The construction to be
about to V is restricted to prospective states of imminent events and hence restricted
to event predications.
Concluding the section on aspect, we find that the perfective aspect constitutes a
predication about an event, while the other three aspects – imperfective, perfect
and prospective – are stative. They all express a state predication about a time: the
given time is related to either the active phase of the event itself or to a state the
event led to or to a state that is likely to lead to such an event. I will call the event
referred to in the perfective aspect and the time referred to in the stative aspects as
the tense argument of the predication. In the aspect diagrams it is represented by
the symbol 䡵.
As we will see in 6.5, and have partly already demonstrated in this section,
aspect is not always marked explicitly. For example, I went to work on the subway
can be interpreted as an episodic perfective predication about a particular event
or as a habitual imperfective statement. Many languages do not mark perfective
vs imperfective aspect; the Romance languages mark it only in the past tense and
standard German does not mark it at all. Nevertheless, if we interpret a given
sentence with its central predication, we have to make up our mind whether we are to
take it as stating that an event of a certain kind occurred, and if so when (perfective),
or whether we are to take it as a description of a certain state that existed at a given
Verbs
157
time (imperfective). Since these are predications about quite different things, there is
no way of leaving this issue open. Therefore, whether marked for aspect or not, every
sentence will receive some aspectual interpretation or other.
We can sum up the considerations on aspect in the following definition:
DEFINITION 1 Aspect
The aspect of a verbal predication concerns the way in which the situation
expressed and the time referred to are related to each other. Aspect determines
whether the predication is about an event in time (perfective aspect) or about
a time related to a situation (imperfective, perfect and prospective aspect).
6.4 TENSE
Aspect concerns the way in which a given time is aligned with a state (imperfective,
perfect, progressive) or how an event is aligned with time (perfective), but aspect
does not locate the situation in time. This is the function of tense.
DEFINITION 2 Tense
Tense locates the situation expressed in time.
Thus tense and aspect are independent, but – as we will see immediately – they
interact. We will confine the discussion in this book to ‘absolute’ tenses, i.e. tenses
which relate to the time of utterance: past, present and future tense. So-called relative
tenses primarily relate to a ‘reference’ time, which itself may be related to the time
of utterance. The combination of relative tense with absolute tense yields temporal
relations like past-in-the-past (pluperfect), past-in-the-future (‘future past’) or
future-in-the-past.
6.4.1 Interaction of tense and aspect
As a start, let us have a look at the three standard absolute tenses – present, past and
future tense. In principle, they can be combined with all aspects:
(35)
perfective
imperfective
perfect
prospective
Past tense
I promised
I was reading
I had finished
I was leaving
Present tense
I promise
I’m reading
I have finished
I am leaving
Future tense
I will promise
I will be reading
I will have finished
I will be leaving
While this all looks very smooth, there is a problem with the second case in (35),
the combination of perfective aspect with present tense, the ‘present perfective’ for
158
Understanding semantics
short. As we saw in connection with the examples in (21) and (22) above, dynamic
predications can be used with present time reference only in some very special cases.
Recalling the remarks in 4.3 on temporal deixis, the reference of present, past and
future tense is determined by the temporal relation to the time of utterance, where
the time of utterance is the time which the event of uttering the sentence takes.
Figure 6.10 illustrates the definition of ‘past’, ‘present’ and ‘future’ TIME, to which past,
present and future TENSE refer, respectively. The utterance U is an event in the world.
It defines a time on the time axis, the ‘time of utterance’, abbreviated TU. This is the
present time; past is prior to TU, and future subsequent to it.
Figure 6.10
Past, present and future time
U
world
time
past
future
TU
present
Now, what is it that tense relates to TU? Grammatical tense is applied to a
predication, the main predication of the sentence. The predication is interpreted
as carrying a certain aspect, and it is the tense argument of the predication that is
related by tense to TU. Let us consider the past tense examples in (35). I promised is
a past perfective; it predicates about an event as a whole; the past tense of the verb
locates this event in the past. Consequently, the event is completed by the time of
utterance. This is small wonder with a punctual verb like ‘promise’, but it also holds
for accomplishment and process verbs denoting events of arbitrary duration: she
built a house means that the house is finished.
The past imperfective I was reading predicates about a contextually given time;
this time lies completely within an ongoing process of the speaker reading; past
tense places this TIME in the past – not the state in its total temporal extension. The
situation of reading may be continued after the time referred to; it may even extend
into the future. Thus, with imperfective predications, it is just one time in the past
that is located within the situation.
The past perfect I’ve finished is similar to the past imperfective; it differs in that the
state predicated of the time is defined as resulting from a previous event of finishing.
Again, this state can extend into the future.
The past prospective I was leaving or I was about to leave predicates of the past
time referred to that a later event of leaving was being envisaged or prepared at that
time. The event may later actually take place or it may fail to come about; it may take
Verbs
159
place in the past or in the future (cf. for instance I was about to leave the day after
tomorrow). The three cases of past perfective, past imperfective and past perfect are
illustrated in Fig. 6.11.
Figure 6.11
Past perfective, imperfective and perfect (right)
event
U
U
state
U
TU
TU
TU
time
past perfective
I jogged
event
past imperfective
I was jogging
time
past perfect
I had jogged
The diagrams are obtained by adding the U–TU component of Fig. 6.10 to the basic
pictures of perfective, imperfective and perfect aspect. For past tense, U–TU is added
to the right of the tense argument 䡵. Placing U–TU to the left of 䡵, yields the pictures
for future tense. The situation with the present tense is more complex; we will turn
to it in a minute.
6.4.2 Past, present, non-past and future tense
Past tense. There is little to say about the past tense beyond what was stated in
the previous section. The past tense places the tense argument before utterance
time. In actual discourse, the tense argument will be temporally located with much
more accuracy – recall the interpretation of the narrative passage in (25). This is
a consequence of interpretation in context: we will inevitably try to connect the
event or time referred to to other events and times in order to come up with a
coherent interpretation of the given sentence in its context. Except by tense, the
tense argument may be more precisely located by temporal adverbials such as the
at-adverbials discussed above or expressions like in a minute, soon, yesterday, this
morning, etc. This is, of course, also possible for present and future time reference.
In languages without grammatical tense, this is the only means of explicit temporal
location of the tense argument.
Present tense. Applying the present tense to the stative aspects can be done simply
by identifying the tense argument with TU. The result is a predication saying that the
state described applies at TU.
In order to align a perfective statement with U–TU one would temporally have to
match two events: the tense argument event and the utterance event. For ‘normal’ events
160
Understanding semantics
this is impossible. Consider ‘AGENT-eat-an-apple’ as a tenseless perfective predication.
Application of the present tense would mean to identify the event of the AGENT eating
the apple – which of course would take some time – with the event of uttering, for
example, ‘I eat the apple’. This is impossible, because the two events do not coincide
temporally. There are basically two scenarios in which coincidence can be considered
as given; both were mentioned in connection with the examples in (22). The first case is
performative verbs. They refer to an event which comes about by making an utterance
which describes this event. Thus, in the case of performative verbs used in this way (in a
‘performative utterance’), the event of utterance indeed coincides with the event referred
to of performing the speech act expressed. The second case is running commentaries,
like a live soccer game commentary on radio or TV. In this style of report, simultaneity
of commentary and events described is claimed and tried to be achieved – and accepted
by the hearer as a manner of speaking, although, strictly speaking, the events reported
usually minimally precede the corresponding verbal reports.
The other two cases are different. Historic present is an established non-literal
as-though use of present tense. The effect of quasi-witnessing the situations
described comes about because the present tense in its literal use relates to really
present situations. It would be wrong to conclude from the existence of historic
present tense that relation to a past time is just another meaning of the present tense,
or that the present tense relates indiscriminately to past and present time. If this
were the case, the historic present would not have its witnessing effect. It is crucial to
observe that this is a SHIFTED meaning of the present tense.
Habitual predications with event verbs in their plain form (I go to the university by
subway) involve an aspectual shift from singular events to a serial state. As we saw
above, they are a variant of imperfective predications, and hence not relevant here.
There are state predications such as ducks are birds and two plus two equals four
which do not refer to temporary conditions. In many descriptions of the present
tense you will find examples such as these cited as evidence for a ‘timeless’ or
‘eternal’ variant of the present tense. Ascribing cases like these to the semantics of
(an alleged) variant of present tense is inadequate. In such cases, too, the meaning
of the present tense is the same as with all stative predications. Present tense places
the time of utterance within the state described. Strictly speaking, ducks are birds is
a predication only about the time of utterance. It lies, however, in the nature of the
predications expressed by such sentences that they apply forever if they ever apply.
Therefore, it FOLLOWS from the fact that if such a predication is true of any arbitrary
time when it is uttered then it is always true. There is no need to claim a special
meaning variant of the present tense for these cases.
Non-past tense. Many languages have a non-past tense (usually called ‘present
tense’). Non-past includes relation to the present and to the future. Therefore, with
the non-past tense no problem arises with perfective aspect. A non-past perfective
will be taken as referring to a future event, while a non-past imperfective will be
taken to relate to the time of utterance, unless the context indicates reference to
a future time. In German, the so-called present tense is in fact a non-past tense.
Consider the following sentences:
Verbs
(36) a. Es regnet.
lit. ‘It rains’ = ‘It is raining.’
161
(German)
b. Morgen regnet es.
(German)
lit. ‘tomorrow rains it’ = ‘It will be raining tomorrow’ or ‘it will rain tomorrow’
c. Ich fahre nach Brüssel.
lit. ‘I go to Brussels’ = ‘I’m going to Brussels.’
(German)
They are all in the plain ‘present’ tense form. (36a) is taken in an imperfective
progressive reading (note that there is no progressive form in standard German). In
(36b), the verb has the same ‘present’ tense form; the sentence can be used either as
a future imperfective or a future perfective; future time reference is only indicated
by the adverb meaning ›tomorrow‹. (36b) is taken as referring to a future event; it is
a non-past perfective. English is held by some scholars to have no future tense, but
only a non-past tense. While this view is controversial (and will not be adopted here),
English uses ‘present’ tense forms for future time reference in certain subordinate
clause, for example in before-clauses:
(37) Before I leave, let’s fix a date for our next meeting.
Future tense. Future tense relates the tense argument to the future. For perfective
aspect, the event referred to is completely located in the future; for the stative
aspects, the time referred to lies in the future, but this does not mean that the
state that applies at this time is confined to the future. There is no contradiction
in saying: ‘I’m sick today and will be sick tomorrow’, meaning that the state of
being sick will not be interrupted between today and tomorrow. It is often argued
for English, German and other languages that what is traditionally called ‘future
tense’, like they will work, sie werden arbeiten (German), is not a real tense, but the
expression of some kind of modality, like probability, conjecture, etc. One argument
against considering these forms as a real future tense is the fact that they also have
non-future readings. Both, they will work and sie werden arbeiten can also be used
with present time reference for uttering a conjecture, as in (38a) and its German
equivalent (38b).
(38) a.
b.
What do you think they are doing? – I think they will be working.
Was, glaubst du, machen sie gerade? – Ich denke, sie werden arbeiten.
This, however, is just a matter of polysemy. The forms have two different functions;
one is the modality of conjecture with present or non-past tense, and the other is
future tense. Another argument against the future tense analysis of such forms is that
future time reference in both languages can also be expressed with non-past ‘present’
tense; in fact, in German using the non-past forms is the normal way of referring to
future time. What this shows is that the so-called present tense is in fact a non-past
tense. Existence of a non-past tense does not preclude existence of an additional
162
Understanding semantics
future tense proper. Tense systems are not necessarily such that the different tenses
available mutually exclude each other. To sum up, English and German can well be
considered to have a future tense.
6.4.3 The verbal onion
Similar to the noun phrase, the verb forms an ‘onion’ with several structural layers.
The innermost is the verb stem; it is followed by one or more diatheses. These two
layers determine the argument structure. They are followed by the layer of aspect; it
determines the tense argument. It is followed by tense, which predicates about the
tense argument. The general structure is given in Fig. 6.12. This hierarchy is directly
reflected in the order of affixes in Japanese verbs as shown in (39a). Two more layers
can be added – which we do not deal with: mood/modality (39b) and sentence
type (39c):
(39) a. mat- aserarete i
stem diath 1 diath 2 aspect
wait- CAUS
PASS
CONTINUATIVE
‘[someone] is having [her] wait’
ru
tense
(Japanese)
PRESENT
b. mat- aserarete i
ru
darō
stem diath 1 diath 2 aspect
tense
modality
wait- CAUS
PASS
CONTINUATIVE
PRESENT TENTATIVE
‘I think [someone] is having [her] wait’
c. mat- aserarete i
ru
darō
ka
stem diath 1 diath 2 aspect
tense
modality sentence type
wait- CAUS
PASS
CONTINUATIVE
PRESENT TENTATIVE QUESTION
‘Will [someone] be having [her] wait ?’
Figure 6.12
The verbal onion
(((
( verb stem) diatheses )
passive
causative
etc.
aspect
perfective
imperfective
perfect
progressive
)
tense
)
past
present
future
The tense argument is the central object of anchoring when it comes to establishing
reference for a sentence. If the situational element is an event, the event is anchored
along with its participant arguments – not only in time, but also within the facts
applying at that time. Similarly, if the tense argument is a time, the predication with
its arguments is located at this time and within the given circumstances.
Verbs
163
6.5 SELECTED TENSE AND ASPECT SYSTEMS
We will conclude this chapter by having a brief look at four tense and aspect systems;
the description will be confined to the major tenses and aspects introduced above.
English. The English system has two aspectual distinctions. The first distinction is
expressed by the plain form vs the form be + gerund. The latter expresses progressive
in most cases (and in some cases prospective). For the majority of dynamic verbs,
this form can be considered as marking imperfective aspect. The imperfective
aspect is not formally marked with state verbs. The perfective aspect goes unmarked
throughout. The second aspectual distinction is between perfect and non-perfect.
The perfect aspect is expressed by have + past participle; non-perfect is not marked.
These two distinctions can each be combined with three tenses, past, non-past and
future tense. Future tense is (almost) obligatory in main clauses with future time
reference.
German. The German verbal system is formally similar to the English one, except
5
that it lacks a distinction corresponding to the English progressive form. German
has one aspectual distinction, perfect vs non-perfect. Perfect is expressed by the
auxiliary haben ‘have’ or sein ‘be’ + past participle. The non-perfect (plain) forms
can be used indiscriminately for imperfective as well as perfective predications.
As in English, there are three tenses, past, non-past, and future tense, which can
be combined with the non-perfect and the perfect aspect. Complicating the matter,
the present perfect forms have a second function of simple past; thus the so-called
present perfect (‘Perfekt’) in German is either a non-past perfect or a past nonperfect. In colloquial German, the present perfect forms are replacing the genuine
past forms more and more as the standard means of relating to the past. The nonpast forms are the unmarked way of relating to the future, but genuine future tense
is also possible.
Russian. Russian distinguishes between imperfective and perfective verbs. Almost
all verb stems are imperfective; by using a rich set of prefixes, they are turned into
perfective verbs. Prefixation also affects meaning; for some verbs a certain prefix
yields a neutral aspectual counterpart with the same meaning except for inherent
aspect, but for most verbs there is no neutral counterpart. For example, imperfective
čitat’ (read) has the perfective derivations pročitat’ (read, recite), počitat’ (read a
while), načitat’ (read a lot) and dočitat’ (finish reading). Thus, the distinction between
the imperfective and perfective aspect is inherent to the verb meaning and lexicalized
rather than a matter of the grammatical form. In addition, there is ‘secondary
imperfectivization’: adding a suffix to a perfective verb yields an imperfective
predication. This instrument is a means of grammatical aspect, like the progressive
and perfect forms in English. Russian does not have a distinct perfect form. There is a
5 This applies to standard written German. Colloquial German and most, if not all, German dialects
have an equivalent of the English progressive form: am + infinitive + sein (inflected): cf. ich bin am
lesen ‘I am reading.’
164
Understanding semantics
past tense that can be applied to all verbs. The formal present tense, however, functions
as a present tense with imperfective verbs and as a future tense with perfective verbs.
There is no present tense with perfective verbs; all the exceptional cases of present
perfective mentioned above – performative verbs, running commentaries, historic
present and habituals – are expressed with imperfective verbs. Thus, the formal
present tense is a real future tense with perfective verbs, not just a non-past tense.
Present tense with imperfective verbs is restricted to present time reference. For
future time reference with imperfective verbs, there is a special future tense which is
morphologically different from the future tense of perfective verbs.
(40) (Russian)
Past tense (fem. sing.)
Imperfective verb
6
Present tense (1st sing.)
Future tense (1st sing.)
Perfective verb
ita-la
pro ita-la
ita-ju
---
budu itat’
pro ita-ju
Japanese. Japanese has two tenses: non-past tense marked with the ending -(r)u and
past tense marked with the ending -ta. It has a major aspectual distinction between
perfective (plain form of the verb) and ‘continuative’, marked with a suffix -te i(gerund + be) before the tense ending. For taberu ‘eat’ the four forms are:
(41)
non-past tense
past tense
Plain form
tabe-ru
tabe-ta
Continuative
tabe-te i-ru
tabe-te i-ta
Similar to English, the plain forms yield the imperfective aspect with state verbs.
State verbs do not take the continuative form; there are only very few, including i‘to be’ which appears in the continuative form. With dynamic verbs, the plain form
expresses perfective aspect. The continuative form yields a progressive reading with
accomplishment and process terms and with simple occurrence verbs in repetitive
use; with accomplishment and simple change terms, the form yields a resultative
perfect. Thus, for accomplishment verbs such as kiru ‘put on’, the continuative form
has two readings – kimono o [ACC] ki-te i-ru can mean ‘[she] is putting on a kimono’
and ‘she is wearing a kimono.’ In addition, the continuative form can express general
perfect for all dynamic verbs.
6.6 CONCLUDING REMARK
Taking a closer look at the general semantics of verbs, we are now able to recognize
a very important trait of human language – the way in which a situation in the
6 Historically deriving from adjectival past participles, Russian past forms are inflected for number
and gender, but not for person.
Verbs
165
world can be put into words to describe it is by no means determined by the
way the situation ‘is’. First, of course, it is up to the speaker which aspects of the
situation they want to talk about. But even if the speaker chooses to talk about a
very particular situation, say a ‘read’ situation by a particular person of a particular
book, there are many different ways to do so. The speaker can choose to mention
the AGENT or not and to mention the THEME or not; if the speaker mentions both,
they can manipulate the structure of the sentence, choosing the active voice for
highlighting the AGENT or the passive voice for promoting the THEME. The speaker
can choose between various aspectual variants of ‘reading’, describing the situation
as an activity or an accomplishment. The reader can choose to use the sentence as
a description of the situation at the given time (imperfective), or as an event that
happened (perfective), or as a situation resulting from a reading event (perfect), or
leading to it (prospective). Finally, the speaker can place the situation in time relative
to when they utter the sentence. All these choices can be applicable to the very same
‘objective’, ‘real’ situation. Thus, the semantic distinctions we encountered in this
chapter are not distinctions that apply to the situations in the world; rather they
are conceptual distinctions among the ways in which we may cast reality when we
describe it by means of words and grammar.
EXERCISES
1. Describe in your own words the effect on the argument structure of the diatheses
a. passive b. antipassive
c. causative
d. anticausative
2. For each of the following alternations, find five more verbs that exhibit it. Do not
consult Levin (1993).
a. the causative/inchoative alternation
b. the dative alternation
c. the simple reciprocal alternation
3. Define in your own words the five aspectual classes introduced in 6.2:
accomplishment, process, simple change, simple occurrence and state terms.
4. Use the diagnostics mentioned in 6.2 (progressive, for-, in-, at-adverbials, present
time reference) to determine the aspectual classes of the following predications;
consider the whole phrases. Set up a table.
a. say
b. dress
c. exceed
d. hammer
e. melt (intransitive)
f. snow
g. write e-mails
h. grow
166
Understanding semantics
5. Describe in your own words the distinction between the imperfective and
perfective aspects.
6. What is common to the
a. perfective and perfect aspects
b. imperfective and perfect aspects
7. What is the aspect of the following sentences:
a. She baked cookies.
b. They tasted awfully sweet.
c. I tasted them.
d. The dog was sick on the floor.
e. I’ll never have any of those again.
8. Put the predication ‘Sheila-mow-her-lawn’ into the following tense-aspect
combinations:
a. present progressive
b. past habitual
c. past perfect
d. future perfective
e. present prospective
9. Discuss the problem of the present perfective.
10. There are four different uses of would + verb in English. Try to describe them in
terms of aspect, tense and mood/modality.
FURTHER READING
Argument structure. On passive in English: Huddleston & Pullum (2002: ch. 16.10
by Ward, Birner & Huddleston); on passive and other diatheses in English Givón
(1993: ch. 8). Levin (1993) on her classification. On causative and passive in Japanese,
Shibatani (1990: ch. 11.4.1–11.4.2).
Aspect and tense. Comrie (1976) and Galton (1984) on aspect and the interaction of
aspect and tense, Comrie (1985) on aspect. On English tense and aspect, Huddleston
& Pullum (2002: ch. 3.3–3.5 by Huddleston), Givón (1993: ch. 4.1–4.3). In WALS
(Dryer and Haspelmath, eds, 2011, http://wals.info), you will find four chapters by
Östen Dahl and Viveka Velupillai on ‘Perfective/imperfective aspect’ (ch. 65), ‘The
past tense’ (ch. 66), ‘The future tense’ (ch. 67) and ‘The perfect’ (ch. 68).
7
Meaning and logic
The logical approach to meaning is a first step into the investigation of meaning
relations. Taking up the notions of truth and reference from chapter 2, we will
consider sentences from the perspective of their truth conditions. The logical view
allows the introduction of basic concepts such as logical consequence (entailment),
logical equivalence and incompatibility. In 7.5 these notions are applied to words.
7.1 LOGICAL BASICS
7.1.1 Donald Duck and Aristotle
Let us start with the provocative (and highly important) question: ‘Is Donald Duck a
duck?’ Suppose you are one of those who answer spontaneously: ‘Why, of course!’ In
that case you would subscribe to the truth of (1):
(1)
Donald Duck is a duck.
Well, ducks are birds, and birds are animals. Would you also say that (2) is true?
(2)
Donald Duck is a bird.
And how about (3)?
(3)
Donald Duck is an animal.
It would not be surprising if you were less sure about the truth of (2) and would not
subscribe to the truth of (3). But, if (1) is true, (2) is true, and if (2) is true, so is (3).
Hence, if (3) is false, there must be something wrong: (2) must be false as well and,
consequently, (1) cannot be true either. That is logic: if Donald is a duck, then he is a
bird. If he is not a bird, he cannot be a duck.
Well, then, let us take a second look at the original question. Why is it that we are
inclined to say that Donald is a duck? Well, it is a fact that his name is Donald ‘Duck’
and that Donald looks like a duck, at least roughly, i.e. if we ignore his having arms
with hands instead of a duck’s wings. But names are just names, and beyond his
looking like a duck there is little to be said in defence of Donald’s duckness. Does he
quack rather than talk? Does he swim or fly like a duck? Would we expect him to dive
168
Understanding semantics
for food as ducks do? No. As far as we know, Donald Duck behaves, feels, and thinks
in every respect like a human being. So, let us try this:
(4)
Donald Duck is a human being.
Have you ever seen a human being with a duck’s body, with feathers, a beak and duck
feet? Is he not much too short for an adult man? Could he run around without his
pants all the time if he were not a duck? If we are serious about the initial question,
we have to admit that (4) is not true either:
(5)
Donald Duck is neither a duck nor a human being.
But if we decide to take this stand, we are throwing out the baby with the bath
water. According to (5), Donald could be anything except a duck or a human. This
is certainly not what we want to say. If anything, Donald is a duck or a human.
Somehow, he’s both at the same time:
(6)
Donald Duck is both a duck and a human being.
He is a duck that behaves like a human, and he is a human being in a duck’s guise. If
we take (5) and (6) together, we get (7) and (8):
(7)
Donald Duck is a duck and he isn’t.
(8)
Donald Duck is a human being and he isn’t.
What does logic say about this? That is very clear: (6) contradicts (5), and (7)
and (8), as they stand, are each self-contradictory. This cannot be: (6), (7) and (8)
cannot be true. Therefore, Donald Duck cannot exist, or (6), (7) and (8) are false.
The consequence is acceptable, in a sense. In a world where ducks are ducks and
cannot be human, and vice versa (e.g. in what we consider the real world), we would
not accept that something or someone like Donald Duck really exists. We would not
accept that (6), (7) and (8) are true of anything that really exists.
The underlying principle goes back as far as Aristotle. In his work Metaphysics, he
formulated the following fundamental law of logic, and not only of logic, but of truth
in general (Aristotle, Metaphysics, 1005b, p. 262):
DEFINITION 1 Law of Contradiction
The same attribute cannot at the same time both belong and not belong to the
same subject in the same respect.
Meaning and logic
169
Aristotle assumes that, basically, every sentence says that some attribute – the
predicate of the sentence – belongs to some subject. What the principle says is
simply this: a sentence, in a certain reading, cannot be true and false at the same
time. Our reasoning about Donald Duck with the outcome of (5), (6), (7) and (8)
violates this law. If (5) is true, (6) must be false, and vice versa. So, if (5) and (6) are
both true, they must also be both false. (7) says that (1) is both true and false, and
so does (8) for (4).
But, seriously, is not there something to be said in favour of the truth of (5), (6),
(7) and (8)? Yes, there is. And if we take a closer look at Aristotle’s law, we realize how
our findings about Donald Duck can be reconciled with logic: we have to relate the
categorization of poor Donald to different ‘respects’. The apparent contradictions can
be resolved if we replace (5), (6), (7) and (8) by the following:
(5’)
Donald Duck is neither a duck nor a human being in all respects.
(6’)
Donald Duck is a duck in certain respects and a human being in others.
(7’)
Donald Duck is a duck in certain respects, but he isn’t in others.
(8’)
Donald Duck is a human being in certain respects, but he isn’t in others.
or more explicitly:
(5”) Donald Duck doesn’t behave like a duck and doesn’t look like a human being.
(6”) Donald Duck looks like a duck but behaves like a human being.
(7”) Donald Duck looks like a duck but doesn’t behave like one.
(8”) Donald Duck behaves like a human being but doesn’t look like one.
These sentences are no longer contradictory. They are, however, still not compatible
with our experience of the real world. Hence, if we accept the truth of these sentences,
we have to assume a different world for Donald Duck to exist in – and that, of course,
is what we do.
As you might have noted, interpreting the original sentences (5) to (8) in the more
explicit way given in (5’) to (8’) or (5”) to (8”) is an instance of what we introduced
in 3.4.3 as ‘differentiation’. The expressions be a duck and be a human being are
interpreted in a more specific reading than their literal meaning. The process is
triggered by the need to make things fit into their context, i.e. by applying the
Principle of Consistent Interpretation.
170
Understanding semantics
7.1.2 The Principle of Polarity
The basic notion of all logical considerations is truth. As was stated in 1.1.2 above,
truth is not a property of sentences as such, although we mostly talk that way.1 The
question of the truth or falsity of a sentence arises only if the sentence is related to
a certain CoU. Since the CoU may vary, sentences are true in some CoUs and false in
others. Truth and falsity underlie the following fundamental principle:
DEFINITION 2 Principle of Polarity
In a given CoU, with a given reading, a declarative sentence is either true or
false.
This principle too goes back to Aristotle. It entails (for the notion of logical entailment
see the next subsection) the Law of Contradiction since the formulation ‘either true
or false’ is taken in the exclusive meaning of either–or: ‘either true or false, but not
both’. The ‘but not both’ part is the Law of Contradiction. The Principle of Polarity
adds to the Law of Contradiction the condition known as the Law of the Excluded
Middle (Latin Tertium non datur, which means ‘there is no third [possibility]’): there
are only these two possibilities, truth or falsity, and no others, i.e. no in-between, no
both-true-and-false, no neither-true-nor-false. Later in 11.5 we will see how this
principle is reflected in the structure and use of natural language.
The principle as it stands disregards presuppositions. As we saw in 4.7, a sentence
usually carries presuppositions, and if the context fails to fulfil these conditions,
the sentence will indeed be neither true nor false. For the time being, we will just
assume that the sentences we are dealing with are only used in CoUs where their
presuppositions are fulfilled. This is the usual way in standard logic to circumvent
the problems caused by presuppositions. Actually, logic becomes much more
complicated if presuppositions are properly taken into account. We will come back to
this issue at the end of this chapter.
In order to have a neutral term for being true or false, one speaks of the truth value
of a sentence, a notion introduced by Frege. A sentence has the truth value TRUE if it is
true; it has the truth value FALSE if it is false. In general, the truth value of a sentence
becomes an issue only when it is used in a certain CoU. The truth conditions (2.2.2)
of the sentence define the circumstances under which a sentence is true, i.e. in which
CoUs. For example, (9) is true if, in the given CoU, the cat is in the garden, and (10) is
true if there’s milk on the floor:
(9)
The cat is in the garden.
(10) There’s milk on the floor.
1 When we talk of sentences in this chapter, it is tacitly understood that we talk of declarative
sentences. The question of truth or falsity does not immediately apply to interrogative sentences,
imperative sentences or other non-declarative types (recall 2.2.3 on sentence types).
Meaning and logic
171
In a given CoU, referring to a certain cat, a certain garden and a certain floor (of a
certain room), the two sentences may both be true, or both be false, or one may be
true, and the other one false. Their truth conditions are independent of each other.
7.1.3 Negation
Due to the Principle of Polarity, any declarative sentence, for example (11a), when
uttered in a particular CoU, is either true or false. The same holds for its negation, in
this case (11b):
(11) a.
b.
John knows the solution.
John does not know the solution.
By asserting (11a), I not only express that I think that John knows the solution, but
I deny that John does not know the solution. Conversely, if I utter (11b), I deny that
(11a) is true. Any statement that one can express in a natural language is, in this
sense, polarized – it is the result of a decision between just two opposite possibilities:
yes or no, true or false. Polarization pervades language totally and the Principle of
Polarity captures this trait of human verbal communication. As we shall see in 7.7,
this is not relativized by the existence of presuppositions.
It is no surprise then that all languages have systematic means of expressing
the polar contrary of a sentence: negation. Negation reverses the truth value of
a sentence; it makes a true sentence false and a false sentence true. We discussed
negation in 4.5.3 and gave a simple method of determining the negation of a
declarative sentence: transform the sentence into a yes–no question, answer it with
‘No’ and form the sentence that expresses exactly what the negative answer says. In
English, negation is usually achieved by negating the finite verb of the sentence, using
the auxiliary do if the verb is not an auxiliary itself.
(12) a.
b.
c.
John knows the solution.
John will die.
John is clever.
negation:
negation:
negation:
John doesn’t know the solution.
John will not die.
John isn’t clever.
In other cases, negation is formed by negating other parts of the sentence, for
example quantifiers such as all, every, some, always and the like, or by replacing them
with appropriate negative expressions (for quantificational NPs, recall the discussion
in 4.5.4):
(13) a.
b.
c.
d.
e.
Mary is already here.
Everybody knows her.
She’s always late.
She sometimes apologizes.
Only John knows the reason.
negation:
negation:
negation:
negation:
negation:
Mary is not here yet.
Not everybody knows her.
She’s not always late.
She never apologizes.
Not only John knows the reason.
172
Understanding semantics
For the present purposes, we need not be concerned with the exact rules and
subtleties of negation in English. Let us simply define the negation of a sentence
as follows
DEFINITION 3 Negation
If A is a declarative sentence that is not negated itself, its negation is a sentence
that
(i) is true whenever A is false and false whenever A is true and
(ii) is formed out of A by a standard grammatical procedure such as
– adding not to the verb itself, if it is an auxiliary (e.g. was not, cannot)
– adding the auxiliary do to the verb phrase and not to the auxiliary
(e.g. did not know)
– adding not to a quantifier expression (e.g. not every)
– substituting a positive expression by its negative counterpart (e.g. some
by no).
There are only a handful of expressions that require negation by substitution:
quantifier expressions containing some (some – no, somewhere – nowhere, etc.) and a
couple of particles such as already and still (negation: not yet and no more). Negation
by substitution is only relevant if regular syntactic negation with not is impossible.
Table 7.1 displays some examples of exceptional negations. Usually, a positive
sentence has exactly one negation, and this is what will be assumed throughout this
chapter. For the sake of convenience not-A will be used as an abbreviation of the
negation of A.
Table 7.1
Types of exceptional negations in English
Sentence
Negation
He’s always late.
He’s not always late.
Everybody knows that.
Not everybody knows that.
He sometimes apologizes.
He never apologizes.
He was already there.
He was not yet there.
You must insist.
You need not insist.
Only Claude knows why.
Not only Claude knows why.
7.2 LOGICAL PROPERTIES OF SENTENCES
Given the notion of truth conditions, a couple of basic logical properties of sentences
can be defined. A ‘normal’ sentence will be sometimes true and sometimes false.
Meaning and logic
173
This property is called contingency. There are two kinds of sentences that are not
contingent: sentences that are always true and those that are always false. 23
DEFINITION 4 Logical properties of sentences
A sentence (in a given reading) is contingent iff 2 it is true in some CoUs and
false in others.
A sentence (in a given reading) is logically true iff it is true in all CoUs.
A sentence (in a given reading) is logically false iff it is false in all CoUs.3
The three logical notions can be alternatively defined by means of tables displaying
which truth values are possible or impossible in the respective cases (cf. Table 7.2).
As is common practice in logic, 1 is used for TRUE and 0 for FALSE. Note that the
‘impossible’ entries in the tables for logical truth and logical falsity imply that the
other truth value is possible since a sentence cannot have no truth value at all. As this
follows automatically, there is no need to write ‘possible’ into the free slot.
Table 7.2
Logical properties of sentences
A is contingent
A
A is logically true
A
1
possible
1
0
possible
0
A is logically false
A
1
impossible
impossible
0
The sentences in (14) are logically true:
(14) a.
Either Donald Duck is a duck or he is not a duck.
b.
Every duck is a duck.
c.
Ducks are birds.
d.
Two times seven equals fourteen.
(14a) is true in every CoU due to the Principle of Polarity. We might replace Donald
Duck by any other subject and is a duck by any other predicate. It is the sentence
pattern ‘either x is p or x is not p’ which makes the sentence true, independent of any
contextual conditions. Likewise, (14b) is invariably true due to the structure of the
sentence, and (14c) is true because the words duck and bird mean what they mean,
i.e. due to the semantic facts of English. (14d) is a mathematical truth. Two, seven
and fourteen when used as NPs like here always refer to the same abstract objects,
2 Iff is commonly used as an abbreviation of if and only if; it connects two conditions that are
equivalent. If and only if constructions are the proper form of precise definitions.
3 In other terminologies, logically true sentences are called tautologies (tautological) and logically
false sentences, contradictions (contradictory). Informally, logically false sentences were referred to
as ‘self-contradictory’ above.
174
Understanding semantics
the numbers 2, 7 and 14; their referents cannot vary with the choice of a CoU, nor
can the outcome of multiplying 2 with 7. In philosophical terminology, sentences
such as (14c) and (14d) are called analytic(ally true), while the notion of logical truth
(and falsity) is reserved for cases such as (14a) and (14b) which owe their truth (or
falsity) to specific rules of logic. From a linguistic point of view, there is no essential
difference between the type of truth represented by the four sentences: all four of
them are true due to their structure and the meanings of the words they contain.
The following sentences are logically false:
(15) a.
Donald Duck is a duck and Donald Duck is not a duck.
b.
Donald Duck is neither a duck nor is he not a duck.
c.
Ducks are plants.
d.
Two times seven is twenty-seven.
(15a) violates the Law of Contradiction, (15b) the Law of the Excluded Middle. (15c)
violates the semantic rules of English and (15d) the rules of mathematics.
The examples show that even logical truth and falsity rest on some basic
assumptions:
∑ the Principle of Polarity
∑ the semantics of the language
These assumptions are absolutely indispensable. The Principle of Polarity is at the
very heart of the notions of truth and falsity. The semantic rules of language are
necessary to make it possible to deal with questions of truth at all. If the sentences
and the words they consist of did not have their proper meanings, there would be no
point in asking any logical or semantic questions.
The logical properties of a sentence are connected to the information it is able to
convey. Contingent sentences may be true or false. Thus, when they are actually used
for assertions, they convey information about the situation referred to: it must be
such that the situation expressed does in fact pertain. If John tells Mary there is beer
in the fridge, she learns something about the situation referred to (provided John
is not lying and Mary believes him). If John uttered a logically true sentence like
Ducks are birds, Mary would not learn anything about the given situation. She would,
at best, learn something about English. Similarly, she would ask herself what John
wants to tell her by saying Either Donald Duck is a duck or he is not a duck. Taken as a
MESSAGE, as information about the world, it would be uninformative. Only contingent
sentences can convey information about the world.
If logically true or logically false sentences are actually used to say something
about the situation in the given CoU, application of the Principle of Consistent
Interpretation (3.4.3) leads to a reinterpretation. Logically false sentences are freed
from their inner contradictions and can then be interpreted as contingent statements.
For example, the compositional meanings of the logically false sentences in (6), (7)
and (8) can be shifted, by means of differentiation, to yield the contingent readings
in (6”), (7”) and (8”). Logically true sentences, too, can be shifted to contingent
Meaning and logic
175
readings. For instance, the logically true sentence Ducks are birds can be used to
point out the fact that ducks are able to fly; this might be relevant in some contexts,
e.g. when the question is discussed whether or not Donald Duck is a duck. From this
interpretation of the sentence, the contingent implication would derive that Donald
could fly (if he were a duck).
7.3 LOGICAL RELATIONS BETWEEN SENTENCES
7.3.1 Logical entailment
Recall our discussion above: necessarily, (16B) is true, if (16A) is true. This is an
instance of a logical relation between sentences called ‘logical entailment’.
(16) A
Donald Duck is a duck.
B
Donald Duck is a bird.
The relation of logical entailment is defined by one crucial condition: it must be
impossible that B is false if A is true. (This is the case with A and B in (16): it is
impossible that Donald is not a bird (= B false), if he is a duck (= A true).)
DEFINITION 5
A
B
A logically entails B /
B logically follows from A /
AfiB
1
1
0
1
0 impossible
1
iff : necessarily, if A is true, B is true.
0
0
Two arbitrary sentences A and B may be independently true or false. This yields four
possible combinations of truth values. Logical entailment rules out one of these four
combinations, A-true-B-false. If A entails B, the truth values of A and B depend on
each other in a particular way: B cannot be false if A is true, and A cannot be true if B
is false. Thus, logical entailment results in a certain link between the truth conditions
of the two sentences. If A entails B, then A is called the premise of the entailment,
and B the conclusion.
The definition of entailment does not specify whether A and B are actually true. A
and B might both be true or both be false, or A might be false and B true. Entailment
means: if A is true, then B is true. But it does not stipulate whether A is true and
therefore one does not know if B is true, either.
If A entails B, it may or may not be the case that B also entails A. If it does, we
talk of mutual entailment. Mutual entailment means that not only A-true-B-false
is impossible, but also A-false-B-true, i.e. B-true-A-false. This results in a second
‘impossible’ entry, in the third row. Alternatively – and this is the normal case – B
176
Understanding semantics
does not entail A. In this case, A unilaterally entails B; the combination A-falseB-true is therefore possible. Thus, entailment is either unilateral or mutual because
A-false-B-true is either possible or impossible, respectively. Table 7.3 shows the
general condition for entailment along with the conditions for the two subcases of
unilateral and mutual entailment.
Table 7.3
Logical entailment and special cases
General entailment
A B
1
1
1
0
0
0
Unilateral entailment
A B
Mutual entailment
A B
1
1
1
1
1
0
impossible
1
0
impossible
1
0
1
possible
0
1
impossible
0
0
0
0
0
impossible
(16) is a unilateral entailment because Donald need not be a duck if he is a bird.
Likewise, the following entailments are unilateral:
fi B
(17) A
It’s raining heavily.
(18) A
Ann is a sister of my mother. fi B
(19) A
Today is Monday.
fi B
It’s raining.
Ann is my aunt.
Today isn’t Wednesday.
If it is raining, it need not be raining heavily; Ann could be my aunt because she is
my father’s sister; it may not be either Wednesday or Monday. (20) is an example of
mutual entailment.
(20) A
Today is Monday. fi
‹ Tomorrow is Tuesday.
Mutual entailment will be called ‘logical equivalence’ in the next subsection. Since
there is a term of its own for this special case of entailment, the term ‘entailment’ is
normally taken as denoting the unilateral case.
There is one way of reversing an entailment: if A entails B, then, if B is false, A is
necessarily false; if B could be false and A not, the entailment would not hold. Thus,
if A entails B, then not-B entails not-A. We can swap the two sides of an entailment if
we negate both sides. Therefore, the converse holds, too: if not-B entails not-A, then
A entails B.
(21) Entailment reversal
A fi B iff not-B fi not-A.
Table 7.4 shows how the truth values of not-A and not-B co-vary with those of A
and B.
Meaning and logic
177
Table 7.4
Entailment reversal
A entails B = not-B entails not-A
A
B
1
1
1
0
not-B not-A
impossible
0
0
1
0
0
1
0
1
0
0
1
1
Let us now take a look at a few examples which are not instances of logical entailment,
although in each case sentence B would, under normal circumstances, be inferred
from A. What matters, however, is whether the consequence is really necessary or
whether it is based on additional assumptions.
(22) A
Mary is John’s mother.
fi
/ B
Mary is the wife of John’s father.
(23) A
John said he is tired.
fi
/ B
John is tired.
(24) A
The beer is in the fridge. fi
/ B
The beer is cool.
There are no logical reasons for drawing these conclusions. It is logically possible that
parents are not married, that John was lying, and so on. In most cases we draw our
conclusions on the basis of our world knowledge, i.e. of what we consider as given,
as normal, plausible or probable. The notion of logical entailment does not capture
all these regularities and connections. It merely captures the really ‘hard’ cases of an
if-then relation, those based on the Principle of Polarity and the semantic facts alone.
Other, more mundane inferences like those in (22) to (24) could be turned into valid
entailments too – if all the tacit premises were added.4 For instance, in the case of
(24) we would have to add that the fridge was on and functioning normally, that the
beer was in it long enough, that it was not in a vacuum flask, etc. If you come to think
of it, it is like a never-ending story of additional premises. In everyday life, when we
actually say things like ‘If the beer is in the fridge, then it is cool’ we rely on a rich
body of shared knowledge and assumptions which need not be made explicit but
are nevertheless necessary for drawing the conclusions we draw. Thus, an ordinary
if-statement involves a huge complex of implicit premises. Logical entailment strips
an inference of all implicit assumptions by making explicit what the premise is (or
the premises are) and relating only to these.
What does logical entailment mean for the meanings of A and B? If A and B are
contingent and A unilaterally entails B, both sentences contain information about
the same issue, but the information given by A is more specific than the information
given by B. The truth conditions that B imposes on the situation are such that they
4 The notion of entailment can be straightforwardly generalized to include more than one premise.
Also, more than one premise can be made into one by connecting all with and into one sentence.
178
Understanding semantics
are always fulfilled if A is true. Therefore, the truth conditions of B must be part of
the truth conditions of A. In general, if no further logical relation holds between A
and B, A will impose additional conditions on the situation referred to. In this sense,
A contains more information; it is more informative and more specific, than B. The
situation expressed by A is a special case of the situation expressed by B. As we shall
see in 7.3.5, this does not hold if A and/or B are not contingent.
One further property should be noted here: logical entailment is what is called
a transitive relation. The general property of transitivity5 is defined as follows: a
relation R is transitive iff ‘x is in relation R to y’ and ‘y is in relation R to z’ entails ‘x is
in relation R to z’. Applied to entailment, this means the following:
(25) Transitivity of the entailment relation
If A fi B and B fi C, then A fi C.
For example, Donald is a duck fi Donald is a bird; Donald is a bird fi Donald is an
animal; hence Donald is a duck fi Donald is an animal. The property of transitivity
immediately follows from the way entailment is defined. Suppose A fi B and B fi C;
then if A is true, necessarily B is true; if B is true, necessarily C is true, hence: if A is
true, necessarily C is true, i.e. A fi C.
7.3.2 Logical equivalence
DEFINITION 6
A and B are logically equivalent,
A¤B
iff:
necessarily, A and B have equal truth values.
A
B
1
1
0
1
0 impossible
1 impossible
0
0
Equivalence means having identical truth conditions. Like entailment, equivalence
is a transitive relation, but unlike entailment it is a symmetric relation. Since
the combinations A-true-B-false and A-false-B-true are both ruled out, the table
combines the conditions for A fi B and B fi A: equivalence is mutual entailment.
Thus, if A and B are contingent, A must contain all the information B contains and
B must contain all the information A contains. In other words, the sentences must
contain the same information. Let us consider a few examples:
(26) A
He is the father of my mother. ¤ B
(27) A
Today is Monday.
¤ B
Yesterday was Sunday.
(28) A
The bottle is half empty.
¤ B
The bottle is half full.
(29) A
Everyone will lose.
¤ B
No-one will win.
He is my maternal grandfather.
5 There is no connection between the notion of a ‘transitive relation’ and the syntactic notion of a
‘transitive verb’.
Meaning and logic
179
The equivalence in (26) is due to the definition of maternal grandfather. For (27), we
have to assume that every Monday is necessarily preceded by a Sunday, an assumption
that may be taken for granted for the point to be made here. The equivalence in (29)
holds if we assume a reading of lose and win in which lose means ›not win‹. Given
these assumptions, all four cases rest merely on the semantic facts of English.
7.3.3 Logical contrariety
DEFINITION 7
A is logically contrary to B /
A logically excludes B / B is incompatible with A
iff:
necessarily, A and B are not both true.
A
B
1
1
0
1 impossible
0
1
0
0
The defining condition is that A and B cannot both be true. Obviously, this is a
symmetric relation. If A and B are contraries, then A entails not-B and B entails
not-A: the ‘impossible’ entry in the first row yields both. Other common terms for
contrariety are logical exclusion and incompatibility. Examples are:
(30) A
It’s cold.
B
It’s hot.
(31) A
Today is Monday.
B
Today is Tuesday.
(32) A
Ann is younger than Mary.
B
Ann is older than Mary.
If A and B are incompatible, it may or may not be possible that both are false. Usually,
they may both be false. In (30) to (32) this is the case: it may be neither cold nor
hot, neither Monday nor Tuesday, and Ann and Mary may be the same age. The two
contraries do not form an exhaustive alternative – there are more possibilities. If,
however, two contraries cannot both be false, they form an either-or-alternative. This
special case of contrariety is the last of the four logical relations introduced here.
7.3.4 Logical contradiction
DEFINITION 8
A and B are logical contradictories
iff:
necessarily, A and B are not both true.
A
B
1
1
0
0
1 impossible
0
1
0 impossible
The definition of contradiction adds to the definition of contrariety the condition
that A and B cannot both be false. If A and B are contradictories, then in every CoU
180
Understanding semantics
either A is true and B is false or B is true and A is false. The classical case of logical
contradiction is formed by a sentence and its negation (33); (34) and (35) show,
however, that there are other cases as well:
(33) A
It’s late.
B
(34) A
Today is Saturday or Sunday. B
Today is Monday, Tuesday, Wednesday,
Thursday or Friday.
(35) A
Everyone will win.
Someone will lose.
B
It’s not late.
Although the B sentences in (34) and (35) are not the negations of the A sentences,
they are nevertheless logically equivalent to the respective negations, today is neither
Saturday nor Sunday and not everyone will win. A sentence and its negation are by
definition always logically contradictory.
Logical contradiction of A and B is logical equivalence of A and not-B, or of B and
not-A. Hence, contradiction too can be defined in terms of entailment.
(36) The logical relations in terms of entailment
A and B are contraries
iff A entails not-B
A and B are equivalent
iff A entails B
and
A and B are contradictories iff A entails not-B and
B entails A
not-A entails B
Table 7.5 displays the crucial conditions for the four logical relations we introduced.
Each ‘impossible’ entry corresponds to an entailment relation. The table also shows
that entailment subsumes equivalence, and contradiction is a special case of
contrariety.
Table 7.5
Logical relations
A
B
1
1
1
0
0
1
0
0
A¤B
A¤B
impossible
impossible
contraries
contradiction
impossible
impossible
impossible
impossible
7.3.5 Logical relations involving logically true or false sentences
According to the definition, there is one simple condition for A logically entailing B:
A-true-B-false is impossible. This condition is automatically given if either A-true or
B-false is impossible. Consequently, A generally entails B if A is logically false or if
B is logically true. More specifically: if A is logically false, entailment holds for any
Meaning and logic
181
conclusion whatsoever, and if the conclusion is logically true, it logically follows from
any premise whatsoever. For logically false A, we even have A entails not-A, and for
logically true B, not-B entails B. This is, however, no contradiction. In both cases, the
premise is never true and hence the entailment is never carried out. The two cases,
A logically false and B logically true, each combined with an arbitrary partner B and
A, respectively, yield the picture in Table 7.6. In mathematical jargon, these cases are
called ‘pathological’.
Table 7.6
Pathological cases of entailment
A is logically false
B is logically true
A B
A B
1
1
impossible
1
1
1
0
impossible
1
0
0
1
0
1
0
0
0
0
impossible
impossible
With logically false premises or logically true conclusions, we get entailments such
as the following. In (37a), A is logically false, in (37b) B is logically true. Mary is tired
is contingent. Putting (37a) and (37b) together, we obtain (37c) by transitivity of
entailment:
(37) a.
A Ducks are dogs. fi B Mary is tired.
fi B Ducks are birds.
b.
A Mary is tired.
c.
A Ducks are dogs. fi B Ducks are birds.
The point with these examples is that they are cases which one would never actually
state as entailments. If the premise cannot be true at all, there is no point in stating
that if it were true then something else would follow.6 Likewise, if the conclusion is
true anyway, it makes no sense to point out that it is true assuming that some premise
holds. This is another difference between natural language ‘if ’ and logical entailment.
The use of if–then statements in actual communication is subject to pragmatic
conditions: they normally make no sense if the if-clause could not be true or the thenclause not false. Logical entailment, however, is a logical relation that simply holds
or doesn’t. If it holds, it does so regardless of its pragmatic relevance. The notion
that these cases are ‘pathological’ reflects the fact that stating them violates the usual
pragmatic conditions.
When you learnt about logical entailment in the preceding sections, you will
have thought about possible relevant applications of the notion. In doing so you
6 We do, however, exploit this case rhetorically when we say things like: ‘I’ll eat my hat if [A]’ in order
to express that A is impossible.
182
Understanding semantics
naturally did not think of the pathological cases, but of instances where premise
and conclusion are contingent. This led you to forming the intuition that there is
a connection in meaning between premise and conclusion of a valid entailment.
And if you really captured the notion, your understanding will have formed that the
connection is a purely semantic one, not involving any further implicit premises.
However, the examples in (37a) and (37b) show that one may have logical
entailment between sentences even if they lack any semantic connection. The same
holds for the logical relations of equivalence, contrariety and contradiction. One
can easily find these relations in semantically unrelated sentences, if one or both
are not contingent. For example, all logically true sentences are equivalent, and so
are all logically false sentences. All logically false sentences are incompatible and
contradictory with all logically false sentences and vice versa. Therefore – and this is
very important to realize – logical relations are not meaning relations. They may hold
good independently of the meaning of the sentences involved, and even despite the
meaning relations between them, as in the case of (37c).
The pathological cases yield not only unexpected logical relationships, but also
unexpected combinations of relationships. For example, if A is logically false and
B an arbitrary sentence, then A entails B and A and B are contraries – due to the
‘impossible’ entries both in row 2 and row 1, respectively.
As we will see in a minute, all this does not mean that our intuitions about a
connection between logical relations and meaning have to be thrown overboard
altogether. If we consider only contingent sentences, a connection does exist. But for
the moment, it is important to realize that logical relations do not warrant a meaning
relation, let alone a particular meaning relation.
7.3.6 Logical relations between contingent sentences
Let us assume that A and B are both contingent. This has far-reaching consequences
for the significance of the logical relations. First of all, we may fill in the entry ‘possible’
into many cells of the defining tables. For example, the definition of entailment rules
out A-true-B-false. The assumption that A is contingent allows us to fill in ‘possible’
in row 1 because otherwise A would turn out to be logically false. Likewise, we can
make the same entry in row 4: A-false-B-false must be possible because otherwise B
would be logically true. Table 7.7 displays the resulting picture for the four relations.
You can easily figure out the other entries yourself.
Table 7.7
Logical relations between contingent sentences
A
B
1
1
1
0
0
1
0
0
AfiB
A¤B
contraries
contradict
possible
possible
impossible
impossible
impossible
impossible
possible
possible
impossible
possible
possible
possible
possible
impossible
Meaning and logic
183
The restriction to contingent sentences renders the four relations much more
specific. Compared to the original definitions in Table 7.5, the relations here all carry
two ‘possible’ entries in addition to the defining ‘impossible’ entries. For example,
the general definition of equivalence leaves open whether the cases A-true-B-true
and A-false-B-false are possible or not. Fixing these issues makes equivalence
between contingent sentences a more specific relation than equivalence in general.
As a consequence, the more specific relations can no longer co-occur more or less
freely. This is so because ‘possible’ and ‘impossible’ entries in the same row are
incompatible, while ‘impossible’ and no entry are compatible. Therefore entailment
in general is compatible with contrariety. But for contingent sentences, entailment
and contrariety have incompatible entries in row 1 and hence exclude each other.
More generally, entailment and equivalence between contingent sentences are
incompatible with contrariety and contradiction.
Still, some cells remain open. The open cell in row 3 for entailment makes
equivalence compatible with entailment, and the open cell in row 4 makes contrariety
compatible with contradiction. We have already made these two observations above.
They reflect that equivalence is a special case of entailment, and contradiction of
contrariety.
Within the domain of contingent sentences a further logical relation can
be introduced, the relation of non-relatedness, as it were. It is called logical
independence and holds between two sentences iff all four truth value combinations
are possible. This amounts to A entailing neither B nor not-B and B entailing neither
A nor not-A, or in other words: the truth value of A does not determine the truth
value of B or vice versa.
When one tries to find examples of pairs of contingent sentences that are related by
one of the logical relations (except independence), one will realize that, now indeed,
this is only possible if the sentences bear some meaning connection. For example, if
two sentences have the same truth conditions and are hence logically equivalent, then
they must have closely related meanings, because it is the meanings that determine
the truth conditions. It cannot be formally proved that a logical relation between
contingent sentences is always due to some meaning connection. But the assumption
is one of the most important working hypotheses for semantics. It can be formulated
as follows:
Working hypothesis
If two contingent sentences exhibit the relation of logical entailment,
equivalence, contrariety or contradiction, this is due to a particular way in
which their meanings are related.
The restriction to contingent sentences does not impose any serious limitation on
the field of semantic research. Therefore, logical relationships are very valuable
instruments for the investigation of meaning relations not only of sentences but also
of words (to which we will turn in 7.5).
184
Understanding semantics
7.4 SENTENTIAL LOGIC
7.4.1 Logical connectives
Sentential logic7 (SL, for short) is a simple formal system with rules for combining
sentences. The sentences are usually simply represented by variables. They are
combined by means of certain basic connectives such as and and or. Needless
to say, the Principle of Polarity is assumed to hold: every simple or complex SL
sentence is either true or false. The only connectives to be considered are those
whose meaning can be exhaustively described in terms of the truth values of the
sentences they are applied to. (This rules out connectives such as because, when,
but, nevertheless, therefore, etc.) We will only introduce three such connectives: Ÿ for
›and‹ (conjunction), ⁄ for ›and/or‹ (disjunction) and Æ ›if … then‹ (subjunction);
in addition we introduce negation, written with the ‘negator’ ÿ.
DEFINITION 9
Negation in sentential logic
If A is an SL sentence, then ÿA is, too.
ÿA is true if A is false, and false if A is true.
Conjunction in sentential logic
If A and B are SL sentences, then (AŸB) is, too.
(AŸB) is true if A and B are both true, and false if either or both are false.
Disjunction in sentential logic
If A and B are SL sentences, then (A⁄B) is, too.
(A⁄B) is true if A and/or B are true, and false if A and B both are false.
Subjunction in sentential logic
If A and B are SL sentences, then (A→B) is, too.
(AÆB) is true if A is false and/or B is true, and false if A is true and B is false.
The subjunction symbol is usually read ‘if … then’, but its real meaning in SL is only
loosely related to natural language ›if‹. We will come back to this issue in a minute.
The truth conditions of complex SL sentences can be given in truth tables (Table
7.8). The right column shows the resulting truth value of the connection, given the
truth values of A and B in the first columns. Thus, for example, if A is false and B is
true (line 3 in the tables), then (AŸB) is false, (A⁄B) is true and so is (A→B).
7 Sentential logic is also called propositional logic and statement logic. We prefer the term sentential
logic, because the units of the system are sentences, rather than statements or propositions. It is
sentences which are connected by connectives, and it is sentences for which the logical notions are
defined.
Meaning and logic
185
Table 7.8
Truth tables for logical connectives
Negation
Conjunction
Disjunction
Subjunction
A
ÿA
A
B
(A Ÿ B)
A
B
(A ⁄ B)
A
B
(A → B)
1
0
1
1
1
1
1
1
1
1
1
0
1
1
0
0
1
0
1
1
0
0
0
1
0
0
1
1
0
1
1
0
0
0
0
0
0
0
0
1
With these four definitions, we can form complex expressions such as:
(38) a.
ÿÿA
b.
(A Ÿ ÿA)
c.
(A ⁄ ÿA)
d.
ÿ(A Ÿ ÿB)
e.
(A ⁄ ÿB)
f.
(A → (A ⁄ B)) etc.
It follows directly from the definition that certain complex sentences are logically false
or logically true due to their form. For example, all sentences of the form (A Ÿ ÿA)
are logically false: A and ÿA necessarily have opposite truth values; therefore they
can never be both true, and so (A Ÿ ÿA) is never true. This fact implements the Law
of Contradiction. Among the logically false sentences in (15), (15a) has this form.
The other four cases in (15) call for different explanations. Similarly, the Principle of
Polarity is reflected in the fact that (A ⁄ ÿA) is logically true: since A has to be either
true or false, either A is true or ÿA; it follows that (A ⁄ ÿA) is always true (cf. lines 2
and 3 in the truth table for disjunction).
7.4.2 Logical connectives and their natural language counterparts
It must be cautioned that the logical connectives do not necessarily coincide with
the natural language correspondents that are used for reading the formulae. Usually,
natural language negation does not negate the sentence as a whole, but only a certain
part of it. To see this, look at the grammatical side of negation (see Definition 3
in 7.1.3): it is only the finite verb or a quantifier that is negated, while parts of the
sentence remain unaffected. As we saw in 4.7, all presuppositions of a sentence are
preserved under negation. For example, negation of the simple sentence Mary is
sleeping (which would be Mary is not sleeping) does not affect the subject Mary:
negation does not deny that it is Mary whom the sentence is about.
The logical connectives Ÿ and ⁄ can only be used to connect sentences, but their
natural language counterparts and and or can also connect single words or phrases
such as adjectives, verbs, VPs and NPs.
186
Understanding semantics
(39) a.
Warsaw is [the capital of Poland and Poland’s biggest city].
b.
[John and Sheila] will marry.
c.
Mary [fell and broke her collar bone].
In (39a, b), and connects two NPs, in (39c), it is two VPs. Sometimes a sentence with
and connecting less than whole sentences can be transformed into an equivalent
conjunction of two sentences. This is the case with (39a), but look what happens if
we apply this procedure to the other two sentences:
(39) a’.
Warsaw is the capital of Poland and Warsaw is Poland’s biggest city.
b’.
John will marry and Sheila will marry.
c’.
Mary fell and Mary broke her collar bone.
The most plausible reading of (39b) is that John and Sheila marry each other
(‘reciprocal’ reading), but (39b’) would rather be used if each of them marries
independently. (39b) may also be used in the sense of (39b’), but (39b’) cannot have
the reciprocal reading. In (39c) we would naturally assume that there is a causal
connection between Mary falling and breaking her collar bone: she broke her collar
bone when she fell. By contrast, (39c’) sounds like relating to two independent events.
Disjunction by natural language or is mostly used in the sense of ›either … or‹
and not in the sense of logical ›and/or‹. Also, there are constructions such as the one
in (40), where or is rather equivalent to Ÿ . Imagine Sheila is asked what kind of ice
cream she would like, and she answers (40a). This is not the same as (40b), which
sounds as though she could not make up her mind. Rather (40a) amounts to (40c).
(40) a.
Pecan or chocolate is fine.
b.
Pecan is fine or chocolate is fine.
c.
Pecan and chocolate are fine.
In the case of subjunction, the reading of → as ‘if … then’ is actually strongly
misleading. Natural language if A then B is only used if there is a connection between
A and B. For example, if one says (41), it will be interpreted as saying that there is
some regularity to A and B being both true; for example, Paul regularly drinks on
weekends and will have a hangover on Mondays. At least we will assume that it being
Monday and Paul having a headache is something of a general observation – for
whatever reason.
(41) If it is Monday, Paul has a headache.
Subjunction, by contrast, does not express any regularity that would connect B to
A. It just says that the truth value constellation of A and B is such that we do not
have A-true-B-false in the given CoU. Thus, subjunction is not really the logical
Meaning and logic
187
counterpart of natural language if. I would rather recommend reading (A→B)
simply as ‘A arrow B’.
There is a straightforward connection between subjunction and logical entailment:
if (A→B) is true in any CoU whatever, then A entails B. This is so because if (A→B)
is necessarily true, we never have A-true-B-false, whence A entails B. Conversely, if
A entails B, then the subjunction is true in every CoU. Thus, subjunction is a much
weaker statement than entailment because it does not claim anything beyond the
particular CoU given. As we have seen in 7.3.1, even entailment is usually a weaker
relation between two propositions than is expressed by natural language if.
7.5 LOGICAL RELATIONS BETWEEN WORDS
The logical relations between sentences can easily be applied to establish
corresponding relations between lexemes and other expressions below sentence level.
To be precise, this is possible for all predicate expressions (chapter 5). To establish
logical relations between two expressions, we insert them into an appropriate test
sentence with variables representing their arguments and check the resulting logical
relations. In order to be able to represent the arguments with variables, the test
sentences for nouns and adjectives must be predicative constructions. Appropriate
test sentences are illustrated in Table 7.9.
Table 7.9
Test sentences
Test word
Test sentence
count noun
car
x is a car
mass noun
mud
x is mud
adjective
dirty
x is dirty
intransitive verb
smell
x smells
transitive verb
sell
x sells y
7.5.1 Logical equivalence
Let us first consider the case of equivalence. Examples are hard to find, but here are
two:
(42) A
x is a female adult ¤ B x is a woman
(43) A
x costs a lot
¤ B x is expensive
What follows from these equivalences for the meaning of the expressions? (42)
means that whatever can be called a woman can be called a female adult and vice
188
Understanding semantics
versa. More technically: the potential referents of woman and female adult are the
same, i.e. the expressions have the same denotation. Similarly, due to (43) cost a lot
and be expensive are true of the subject referent under the same conditions. Rather
than introducing a new term, we will extend the notion of logical equivalence to
words and complex expressions such as female adult and cost a lot.8
DEFINITION 10 Logical equivalence of predicate terms
Two predicate terms (in given readings) are logically equivalent iff they
necessarily yield the same truth value for the same arguments.
In terms of the denotations of A and B, A and B are logically equivalent iff their
denotations are identical (Fig. 7.4).
7.5.2 Entailment and logical subordination
Let us assume the test sentences for two expressions result in entailment:
(44) a.
b.
A x is a duck
fi B x is a bird
A x shrinks y fi B x changes y
According to (44a), whatever can be called a duck can be called a bird. Put more
technically, the denotation of duck, the more specific term, is included in the
denotation of the more general term bird. Due to the second entailment, the
denotation of shrink is part of the denotation of change. Every act of shrinking
something is an act of changing it.
The resulting relation between a general term and a specific term will be called
logical subordination (subordination for short):
DEFINITION 11 Logical subordination
A predicate term A is a logical subordinate of B (and B a logical superordinate
of A) iff B is true of some arguments whenever A is true of them.
In terms of the denotations of A and B, A is a subordinate of an expression B, iff the
denotation of A is included in the denotation of B (Fig. 7.4). If A is a subordinate
of B, B is called a superordinate of A. Alternatively, the term ‘entailment’ can also be
applied to verbs, adjectives and nouns. In set-theoretical terms, A is a subordinate of
8 Some authors consider logical equivalence a variant of synonymy, for example Cruse (1986: 88),
who uses the term cognitive synonymy for the same relation. Synonymy and equivalence must,
however, be distinguished, a point we will come back to in 7.6.1.
Meaning and logic
189
B iff the denotation of A is a subset of the denotation of B. In the cognitive terms to
be introduced in chapter 11, the denotation of a subordinate term is a subcategory of
the denotation of its superordinate terms.
Figure 7.1
Denotations of duck and bird
denotation of the more general term bird
denotation of the more
specific term duck
7.5.3 Logical incompatibility
Usually, a superordinate expression does not have just one subordinate, but a set
of co-subordinates. For example, all the other terms for types of bird, such as owl,
pigeon, penguin, sparrow, swan, are co-subordinates of duck. In addition, they are
mutually exclusive: x is a swan logically excludes x is an owl, and so on for all other
pairs. Two terms A and B will be called logically incompatible iff their denotations
have no elements in common, or equivalently:
DEFINITION 12 Logical incompatibility
Two predicate terms A and B are logically incompatible if they cannot both be
true of the same arguments.
The denotation of swan could be represented by the hatched circle in Fig. 7.1: an area
included within the area for the denotation of bird and not overlapping with the area
symbolizing the denotation of duck.
The representation of lexemes in hierarchy trees such as in Fig. 7.2, a hierarchy for
linguistic terms, is based on the two relations of subordination and incompatibility.
Since trees are used for representing different relations and structures (e.g. syntactic
trees for the syntactic structure), it is important to realize what the arrangement in
a given kind of tree signifies. In trees that depict lexical hierarchies, the vertical lines
express logical subordination. Co-subordinates are arranged at the same level and
understood to be mutually incompatible.
190
Understanding semantics
Figure 7.2
Hierarchy of word classes
word
noun
count
noun
verb
mass
noun
transitive
verb
adjective
intransitive
verb
The tree in Fig. 7.2 is incomplete in several respects. Further subordinates of word
could be added, e.g. article or preposition. We could also expand the tree by adding
subordinates of adjective. Further subdivisions would be possible below the lowest
level, distinguishing subclasses of count nouns, intransitive verbs, etc. In contrast, the
small tree in Fig. 7.3 is, in a sense, complete. In English, there are only two specific
terms for siblings: no further co-subordinates can be added to sister and brother. Also
there are no English words for subordinates of brother and sister. (Other languages,
e.g. Hungarian, Turkish and Japanese, have different terms for elder and younger
sisters and brothers.)
Figure 7.3
Hierarchy of sibling terms
sibling
sister
brother
7.5.4 Logical complementarity
The subordinates in Fig. 7.3 are not only incompatible but form an exhaustive
alternative, a strict either-or constellation. The corresponding test sentences x is a
sister and x is a brother are logical contradictories – provided we presuppose that x is
a sibling. This meaning relation is called logical complementarity: two terms A and B
are logically complementary iff their denotations have no elements in common and
together exhaust the set of possible cases. Equivalently:
DEFINITION 13 Logical complementarity
Two predicate terms A and B are logically complementary if they necessarily
have opposite truth values for all arguments.
Meaning and logic
191
The notion of complementarity is always relative to a given domain of relevant cases.
Absolute complementarity does not occur in natural languages. Take any ordinary
noun – for example, banana; try to imagine an absolute complementary, say, nonbanana. The denotation of non-banana would have to exclude bananas, but include
everything else that could be denoted by any noun whatsoever plus all those things
for which we do not have any expressions at all. A word with such a meaning is hard
to imagine. Good examples for complementaries are member–non-member (domain:
persons), girl–boy (domain: children), child–adult (domain: persons), indoors–
outdoors (domain: locations). A survey of the logical relations at word and sentence
level is given in Table 7.10.
Table 7.10
Logical relations between words
Word relation
Corresponding
sentence relation
Example
equivalence
equivalence
woman
–
subordination
entailment
bird
– duck
incompatibility
contrariety
duck
–
complementarity
contradiction
member –
female adult
swan
non-member
Logical relations between content words result in certain relations between their
denotations. They can be depicted with the Venn diagrams known from elementary
set theory in Figure 7.4.
Figure 7.4
Denotation relationships
equivaleent
sub
bordinate
incompatib
ble
compleementary
We emphasized above that the logical relations between sentences do not constitute
meaning relations. The same holds for the logical relations between words.
7.6 LOGIC AND MEANING
It will now be demonstrated why logical relations must not be confused with
meaning relations such as synonymy (two expressions have the same meaning) and
hyponymy (the meaning of an expression contains the meaning of a superordinate
expression). These and other meaning relations will be introduced in chapter 8.
192
Understanding semantics
7.6.1 The semantic status of logical equivalence
It is tempting to assume that logical equivalence means identity of meaning, and in
fact this is often done in the literature.9 However, if meanings are considered concepts,
this turns out to be wrong. All logical notions are based on truth conditions and
denotations. As we saw in 2.2.2, truth conditions and denotations are determined by the
meaning of an expression, but its meaning is not exhausted, or fully captured, by them.
Apart from that, logical properties and relations only concern descriptive meaning;
other dimensions of meaning such as social meaning, expressive meaning or the
meaning of sentence type are disregarded and therefore irrelevant for logical notions.
7.6.1.1 Truth conditions and non-descriptive meaning
As far as the restriction of logical notions to descriptive meaning is concerned, let
us recall the criteria for correct use with respect to descriptive, social and expressive
meaning that were stated in Table 2.5. If, for example, one describes the truth
conditions of the sentence this is a barbecue, one has to describe the denotation of
the word barbecue. This, in turn, indirectly says something about the descriptive
meaning of barbecue, because the descriptive meaning determines the denotation. In
this sense, truth conditions relate to descriptive meaning. But they have nothing to
do with social meaning and expressive meaning. For example, the German sentence
and the English sentence in (45) differ in the meanings of the pronouns Sie and you
(cf. 2.3.1; we take the rest of the sentences to be equivalent, in particular the verbs
verhaften and arrest).
(45) a.
b.
Ich werde Sie verhaften.
(German)
I will arrest you.
The German pronoun of address Sie has the same descriptive meaning, but in
addition a social meaning indicating a formal relationship between speaker and
addressee(s). The difference, however, has no bearing on the truth conditions. If
the speaker of the German sentence used the informal pronoun of address instead,
the resulting sentence would have exactly the same truth conditions, although it
might be socially inappropriate. Similarly, expressions with the same descriptive but
different expressive meanings do not differ in truth conditions. Opting, for example,
for (46b) rather than (46a) is not a matter of objectively given facts but of subjective
preference.
(46) a.
b.
John didn’t take his car away.
John didn’t take his fucking car away.
Consequently, it may happen that words and sentences are logically equivalent, but
differ in non-descriptive meanings. We will now see, that logical equivalence does not
even mean equal descriptive meaning.
9 For example in Lyons (1995: 63).
Meaning and logic
193
7.6.1.2 Logical equivalence and descriptive meaning
As we saw in 7.2, all logically true sentences have identical truth conditions. Hence
they are all logically equivalent. Clearly, logically true sentences may differ in
descriptive meaning (cf. the examples in (14) and (15)). The same, of course, holds
for logically false sentences. Thus non-contingent sentences provide a particularly
drastic class of examples of logically equivalent sentences with different meanings.
But even for contingent sentences, equivalence does not mean that they have the
same descriptive meaning. To see the point, consider once more sentences (27) to
(29), here repeated for convenience:
(27) A
Today is Monday.
¤ B
Yesterday was Sunday.
(28) A
The bottle is half empty. ¤ B
The bottle is half full.
(29) A
Everyone will lose.
¤ B
No-one will win.
In all three cases, A and B intuitively do not have the same meaning, but somehow
they amount to the same. They express the same condition in different ways. It is
part of the meaning of (27A) that the sentence refers to the day that includes the
time of utterance, and part of the meaning of (27B) that it refers to the immediately
preceding day. (28B) highlights what is in the bottle, and (28A) what is not. (29A)
is about losing, (29B) about winning. What determines the situation expressed by
a sentence, i.e. its proposition, are the elements of the situation and how they are
interlinked. The situation expressed by (27A) contains the day of the utterance as an
element and specifies it as a Monday. The situation expressed by (27B) is parallel, but
different. More than simply defining truth conditions, a natural language sentence
represents a certain way of describing a situation which then results in certain truth
conditions. Whenever we speak, we make a choice among different ways of expressing
ourselves, different ways of putting things; we are not just encoding the facts we want
to communicate. There is usually more than one way to depict certain facts.
Although less obvious, the analogue holds for words. For example, in German the
big toe is called either großer Zeh (›big toe‹) or dicker Zeh (›thick toe‹) or, by some
people, großer Onkel (›big uncle‹). The Croatian term is nožni palac ›foot thumb‹.
These would all be terms with different descriptive meanings because they describe
what they denote in different ways. More examples of logically equivalent expressions
with different descriptive meanings can be easily found if one compares terms from
different languages which have the same denotation. English has the peculiar term
fountain pen for what in German is called Füllfederhalter (›fill feather holder‹, i.e. a
‘feather holder’ that can be filled) or just Füller (›filler‹, in the meaning of ›something
that one fills‹); in Japanese, the same item is called mannenhitsu, literally ›tenthousand-years brush‹. For a bra, German has the term Büstenhalter, ›bust holder‹;
the French equivalent soutien-gorge literally means ›throat(!) support‹, while Spanish
women wear a ›holder‹ (sujetador) or ›support‹ (sostén) and speakers of the New
Guinea creole language Tok Pisin put on a ›prison of the breasts‹ (kalabus bilong susu).
194
Understanding semantics
Of course, the meaning of mannenhitsu is not literally ›ten-thousand-years
brush‹ and kalabus bilong susu does not literally mean ›prison of the breasts‹.
But a comparison of the examples shows that the respective concepts are based
on different properties of the referents. Different languages may adopt different
naming strategies for the same categories of things. An interesting field is terms for
technical items. The English term power button (e.g. of an amplifier) rests on the
concept ›button‹, which is borrowed from the domain of clothing, and connects it
in an unspecific way with the concept ›power‹, again a metaphor. The equivalent
French term is interrupteur d’alimentation, literally ›interrupter of supply‹; the object
is primarily named after its function of interrupting the current – a somewhat
arbitrary choice, since the power button can also be used for switching the device
on; the second part d’alimentation specifies what is interrupted, namely the
‘alimentation’, a metaphorical term for power supply, originally meaning ›feeding‹,
›nourishing‹. German has yet a different solution for naming that part: Netzschalter,
›net switch-er‹, where Netz is the mains.
Thus, let us fix the following important point:
Logical equivalence is not a sufficient criterion for having the same meaning,
or even the same descriptive meaning.
The converse is, of course, true: since the descriptive meaning determines truth
conditions and denotations, two expressions with the same descriptive meaning
necessarily are logically equivalent. Employing the semiotic triangle in a somewhat
distorted form, Fig. 7.5 displays a configuration of two equivalent expressions with
different descriptive meanings.
Figure 7.5
Logically equivalent expressions with different descriptive meanings
expression1
descriptive
meaning1
expression2
denotation/
truth conditions
descriptive
meaning2
Logically equivalent expressions have the same logical properties: they share the
same denotations or truth conditions. Logically equivalent expressions entail the
same expressions and they are contrary or complementary to the same expressions.
In terms of reference and truth conditions, the use of logically equivalent expressions
Meaning and logic
195
amounts to the same, but the objects and situations referred to may be being described
in conceptually different ways, as is illustrated by the examples in (27) to (29).
7.6.2 The semantic status of logical entailment
The fact that the denotation of a logical subordinate is a subset of the denotation of
the superordinate may be due to a reverse relation between the descriptive meanings
of the two terms. Let us roughly consider the descriptive meaning as a set of
conditions that a potential referent must fulfil. Then a ‘duck’ must fulfil all conditions
a ‘bird’ must fulfil plus those particular conditions that distinguish ducks from
other sorts of birds. The descriptive meaning of duck in this sense contains [all the
conditions which make up] the descriptive meaning of the superordinate term bird.
This is what we intuitively mean when we say that the term duck is ‘more specific’
than the term bird. Generally, AfiB may be due to the fact that the meaning of A fully
contains the meaning of B. This is the case, for example, for the sentences in (47): A
has the same meaning as B except for the addition that the beer is cool.
(47) A
B
There’s cool beer in the fridge.
There’s beer in the fridge.
However, since entailment is only a matter of truth conditions, there need not be such
a close connection between sentences or other expressions related by entailment.
(48) is a simple example of two sentences A and B where A entails B, but the meaning
of B is not contained in the meaning of A:
(48) A
B
Today is Sunday.
Tomorrow is not Friday.
The analogue holds for logical subordination. Consider, for example, the expressions
son of x’s mother-in-law and x’s husband. One’s husband is necessarily a son of one’s
mother-in-law. (The converse is not true.) Hence, x’s husband is a subordinate of son
of x’s mother-in-law. But while the meaning of x’s husband is something like ›man x
is married to‹, the meaning of son of x’s mother-in-law is ›male child of the mother of
the person x is married to‹. The latter contains ›mother-in-law‹ as an element of the
resulting concept, but this element is not part of the concept ›husband‹. Therefore,
the meaning of the superordinate is not part of the meaning of the subordinate.
Some authors use the term hyponymy for logical subordination.10 In this volume,
hyponymy will be reserved for the meaning relation that holds between A and B if
(i) B is a logical subordinate of A plus (ii) the meaning of A is fully contained in
the meaning of B. The notion will be formally introduced in 8.2.1. The point to be
10 E.g. Lyons (1977), Cruse (1986).
196
Understanding semantics
stated here, in analogy to the relationship between logical equivalence and identity
of meaning, is this:
Logical entailment and subordination do not necessarily mean that the
descriptive meaning of one expression is included in the descriptive meaning
of the other.
Analogues hold for the other relations of logical incompatibility and complementarity.
For example, as we could see in connection with (34) and (35), A and B can be logical
contradictories without one being the negation of the other.
7.6.3 Logic and semantics
The discussion has shown that logical properties and relations are only indirectly
related to meaning. Rather, they concern denotations and truth conditions, i.e.
aspects of linguistic expressions which are determined by meaning, more precisely
by descriptive meaning. A logical approach to meaning is therefore limited as follows:
∑ It does not capture those parts of meaning that do not contribute to the
determination of truth and reference. Expressions with the same descriptive
meaning but different social or expressive meanings cannot be distinguished by
logical methods.
∑ It does not capture descriptive meaning itself, but only effects of it.
∑ It does not capture differences between the descriptive meanings of expressions
with identical truth conditions or denotations. In particular, it fails to yield any
insights into the meanings of non-contingent sentences.
If the limitations of the logical approach, and the nature of the results it is able to
produce, are carefully kept in mind, it is, however, a powerful instrument for the
semanticist. Due to the fact that truth conditions and reference are determined by
descriptive meaning, the logical approach can produce the following kinds of results:
∑ If two expressions are not logically equivalent, their meanings are different.
∑ If an expression A does not entail an expression B, the meaning of B is not part
of the meaning of A.
∑ If one of the logical relations holds between two expressions, their descriptive
meanings must be closely related.
The logical approach therefore provides us with simple instruments to test expressions
for their meaning relations with others. This makes logical relations very important
data for semantic analysis. If two expressions are equivalent, or if one entails, or
excludes, the other, this is a fact which semantic analysis has to account for.
Meaning and logic
197
7.7 CLASSICAL LOGIC AND PRESUPPOSITIONS
7.7.1 Presuppositions and the Principle of Polarity
Sentential logic is based on the assumption that language is ruled by the Principle
of Polarity: there are just two possibilities for any sentence in any context – truth
and falsity. Adopting the Principle of Polarity, we can define logical properties,
relations and basic connectives in terms of truth and falsity. All these notions are
straightforward. The result is a clear and rigid – and very simple – basic system for
thinking about the meaning of natural language sentences. However, the Principle of
Polarity is not as innocent as it might appear. Every natural language sentence comes
with presuppositions. Thus, for every natural language sentence, the case that it is
neither true nor false may in fact occur.
This does not mean that the Principle of Polarity is simply wrong: presupposition
failure is not a third possibility on a par with a sentence being true or being false.
Rather it leads to a truth-value gap, and there being a truth-value gap does not
amount to a third truth value. In order to take care of presuppositions, the Principle
of Polarity has to be remedied as follows:
DEFINITION 14 Principle of Polarity (revised)
In a given CoU where its presuppositions are fulfilled, with a given reading, a
declarative sentence is either true or false.
Thus, the Principle of Polarity does hold – but it only holds given that the
presuppositions are fulfilled. Still, ‘there is no third’, in the sense of ‘no third on a par
with truth and falsity’. This is in accordance with the revision of the definition of
truth conditions which we undertook in 4.7.3: presuppositions are not part of the
conditions for being true or false, but preconditions for having a truth value.
In accordance with the Principle of Polarity, the main predication of a sentence is
always presented as a true-or-false matter. Usually, the main predication is expressed
by the VP. Every verb carries presuppositions, namely the selectional restrictions on its
arguments. In addition, change-of-state verbs carry presuppositions concerning the
initial states preceding the event denoted. On top of these verb-specific presuppositions,
there is the general Presupposition of Indivisibility carried by any predication
whatsoever (4.7.2). In fact, this presupposition makes sure that the outcome of a
predication cannot be ‘split’ into truth and falsity. Subject and object NPs, too, often
carry presuppositions, if they are definite or quantificational. This is how every
sentence comes with presuppositions. These presuppositions are not just a nuisance,
but rather a very forceful device for communicating the type of context which serves
as the background for a true-or-false alternative. Compared to the whole world, such
contexts are extremely narrow spaces of possibilities and it is only this heavy cutting
down on possibilities and focusing on very particular aspects of the world that makes
the true-or-false mode of verbal communication possible and meaningful.
198
Understanding semantics
Let me illustrate this with a very simple example. Assume you say ‘I’m tired.’ By
using the 1st person singular pronoun as the subject of the sentence, you cut down
the zillions of possible referents in the universe, and among them the billions of
people on earth, to just one case – yourself. The rest of the universe fades into
oblivion. Then, by using the present tense, you select a very limited time slice of your
existence. The Presupposition of Indivisibility restricts it to a time span where you
are either tired or not tired throughout, thus not more than a couple of hours. Finally,
the predication be tired selects one very specific aspect of your physical condition out
of the unlimited number of aspects concerning your person at that time. This is how
a normal everyday sentence zooms in on some very particular aspect of our world.
The focusing is mainly achieved by presuppositions which drastically narrow down
the context of any utterance to a very specific true-or-false affair. In addition to the
effect of presuppositions, any actual utterance is further massively focused by the
mutual assumptions of background knowledge.
Figure 7.6 is an attempt to illustrate the difference between two basic strategies:
disregarding presuppositions, as in traditional logic, or including them into one’s
semantic considerations. If presuppositions are disregarded, utterances are treated
as conveying an unconditioned polar alternative, represented by the schema V. With
presuppositions, an utterance presents certain things as given and uncontroversial
– the presuppositions – and on this ground presents a polar alternative; this is
symbolized by the Y schema. The Y schema holds independently of the particular
CoU because it is implemented by the mere semantic content of the utterance. When
a sentence is uttered in a concrete CoU, its presuppositions are embedded into the
vast complex of background assumptions with which the interlocutors operate. This
is the third, Y-in-the-swamp picture. Traditional logic treats sentences according
to the V schema, neglecting the actual basis for polarization. A logic adequate for
capturing the logical structure of natural language would actually have to cope with
the Y picture of sentences.
Figure 7.6
Binary, presuppositional and contextual polarity
S
not-S
S
not-S
S
not-S
PRESUPP
PRESUPP
BACKGROUND
Meaning and logic
199
7.7.2 Radical negation
In 4.7 and throughout this chapter, including the definition of negation, it was
assumed that (normal, regular) negation does not affect the presuppositions of a
sentence, and this will be assumed in the rest of the book. There are, however, cases
where one wants to deny a presupposition of an utterance by someone else and uses
negation for that purpose. Imagine the following dialogue:
(49) A:
B:
The Turks are celebrating their victory.
Oh, no. The Turks are NOT celebrating their victory – they didn’t win, you
know.
By repeating A’s statement and negating it, B first complies with A’s perspective which
apparently presupposes that the Turks have won. Knowing better, B contradicts, by
negating the sentence. However, this is not a standard negation. Had B simply replied
(50), with normal intonation, B would have committed to the same presuppositions
as A did and denied that the Turks are celebrating.
(50) The Turks are not celebrating their victory.
Instead, B puts a heavy accent on the negation and adds a rectification of A’s
presupposed assumption that the Turks won. This reply amounts to a refutation of
one of the presuppositions of A’s utterance. It may involve refusing that the Turks are
celebrating at all, but it need not; it may also be used to convey the notion that the
Turks are actually celebrating something else.
This kind of negation is called radical negation. Radical negation always comes
with an extra emphasis on the negation word. It can be used to deny that one or the
other presupposition is fulfilled. In order to know which aspects of the sentence are
subject to negation, an explicit rectification is necessary, like the second sentence in
(49B). Without these heavy cues, negation will be taken not to affect presuppositions.
Figure 7.7 illustrates the polarity schema for radical negation. The positive alternative
is formed by the original predication plus those presuppositions claimed to fail; the
Figure 7.7
Radical negation polarity
S
not-S
PRES
200
Understanding semantics
negative alternative negates this package. However, the radical negation also rests
on some presuppositions. Trying to negate absolutely all presuppositions would not
result in a meaningful utterance. Radical negation needs a rectification in order to be
interpretable, and any rectification sentence inevitably carries some presuppositions.
7.7.3 The projection problem of presuppositions
Integrating presuppositions complicates logic considerably. One fundamental
problem is the determination of the presuppositions of complex sentences. The
problem is known as the projection problem of presuppositions. If A carries the
presuppositions P and B carries the presuppositions Q, what are the presuppositions
of (AŸB), (A⁄B) or (A→B) – let alone all the sorts of combinations and embeddings
of sentences that are available in natural language? If there is presupposition failure
for either A and B, will the presuppositions fail for the whole? Let us consider two
examples concerning conjunction:
(51) Greg started drinking and Sue stopped smoking.
If the conjunction is interpreted as the logical Ÿ, the two conjuncts Greg started
drinking and Sue stopped smoking are taken as two independent sentences with
the same CoU as background. (51) can only be true if for each conjunct its
presuppositions are fulfilled in the given common CoU; (51) would be false if at least
one of the conjuncts is false, which again requires its presuppositions to be fulfilled.
Hence, the presuppositions of both conjuncts are presuppositions of the whole. Now
consider (52):
(52) Helen married Greg and got divorced only three years later.
The sentence is uttered incrementally: the first conjunct first and then the second
conjunct. The first conjunct carries the presupposition that before the time
considered Helen had not been married. Uttering Helen married Greg changes the
CoU: it now contains the fact that Helen married Greg. The second conjunct shifts
the time three years ahead and it finds its presupposition – that Helen had been
married – in the context provided by the first conjunct. What this shows is that we
cannot always assume that the presuppositions of a conjunction as a whole are just
the presuppositions of the conjuncts taken together. Sentence (52) as a whole does
not presuppose that Helen had been married to Greg.
Similar problems arise with logical disjunction and subjunction and with natural
language or and if. The problem of presupposition projection is very complex and
there is a lot of literature about it. For the purpose of this book, this much may suffice
on this topic.
Meaning and logic
201
EXERCISES
1. For the following sentences S, which one is the proper negation, A or B? Check the
truth conditions of A and B: which one is necessarily true if the positive sentence
is false?
a. S It sometimes rains here.
A It never rains here.
B It doesn’t always rain here.
b. S She is cold and sick.
A She is neither cold nor sick.
B She either isn’t cold or isn’t sick.
c. S Everybody helped me.
A Nobody helped me.
B Not everybody helped me.
d. S She’s already here.
A She’s not yet here.
B She isn’t here anymore.
2. Which of the following statements are true, which ones are false?
a. If A is contingent, then not-A is, too.
b. If A is not logically true, then not-A is.
c. If not-A entails B, then not-B also entails A.
d. Not-A never entails A.
e. A cannot entail both B and not-B.
3. Apply the formal definition of entailment to show that contradiction entails
logical contrariety.
4. Check the following pairs of sentences A and B: which truth-value combinations
are possible? Do they represent cases of logical entailment, equivalence,
contrariety, contradiction or none of these relations? In case of entailment, check
both directions.
a. A The door is open.
B The door has been opened.
b. A The room is not large.
B The room is small.
c. A She never empties the ashtray. B She often empties the ashtray.
d. A She’s sometimes late.
B Sometimes she is not late.
e. A By midnight, half of the
B At midnight, only half of the
guests had left.
guests were left.
5. What is the logical relationship between A and B, in terms of entailment(s), if
they cannot both be false?
6. Show that two logically false sentences are both equivalent and contraries.
7. Which logical relation applies to the following pairs of words?
a. book, textbook
b. textbook, dictionary
c. free, occupied
d. young, old
e. receive, send
f. left, right
8. Which dimensions of meaning are not captured by the logical method and why
are they not?
202
Understanding semantics
9. Discuss the limits of the logical method for the investigation of descriptive
meaning.
10. Discuss the ways in which the logical method is useful for the investigation of
word meaning.
FURTHER READING
Cruse (1986, ch. 4) for logical relations between words. Partee et al. (1993, ch.6) for
definitions of the logical properties and relations with respect to logic.
8
Meaning relations
It is very difficult to describe the meanings of lexemes or sentences explicitly. In
fact, semanticists do not even agree as to what kind of entities meanings are. The
conceptualist point of view taken in this book – which regards meanings as concepts
– is the prevalent view, but by no means uncontroversial. It is, however, relatively easy
to establish meaning RELATIONS between sentences or expressions. One influential
traditional position, the structuralist position, holds that the description of meaning
is exhausted by the description of meaning relations (chapter 9). We do not share this
view. This notwithstanding, meaning relations are most important semantic data.
The logical relations that were introduced in the last chapter are defined in
terms of relations between the denotations of expressions, or truth conditions of
sentences; for example, two content words are logically equivalent iff they have
identical denotations. Consequently they are relations concerning the extra-linguistic
correlates of expressions (i.e. entities at the right bottom corner of the semiotic
triangle). Meaning relations, by contrast, are defined in terms of conceptual relations,
between meanings themselves (i.e. entities at the left bottom corner of the semiotic
triangle). For example, two expressions are synonymous iff they have the same
meaning. While logical relations may be indicative of meaning relations of some
kind, it is important to realize that logical relations and meaning relations are defined
in a fundamentally different way.
8.1 SYNONYMY
The meaning relation simplest to define is (real, not approximate) synonymy:
DEFINITION 1 Synonymy
Two expressions are synonymous iff they have the same meaning.
Synonymy in the strict sense, also called total synonymy, includes all meaning
variants for two, possibly polysemous lexemes and it includes all meaning dimensions,
including descriptive, social and expressive meaning. Two polysemous expressions
are synonymous iff they have the same set of meaning variants; and expressions with
204
Understanding semantics
descriptive meaning and other meaning dimensions are synonymous iff they have
the same descriptive as well as the same non-descriptive meaning. Examples are
very hard to find; good candidates are pairs of expressions with the same meaning
but different regional preference, such as the two German expressions for Saturday
Samstag and Sonnabend; the latter is preferred in Northern Germany, but both belong
to common usage. Trivial cases of synonyms are pairs of abbreviations and their full
forms, e.g. the English words perambulator and its shortened form pram, compact
disc and CD or the linguistic terms NP and noun phrase. These cases are trivial in that
the two forms can be considered alternative variants of the same lexemes.
While total synonymy is rare or trivial, there are many cases of partial synonymy.
Two lexemes may have one or more meaning variants in common. Examples would
be English hide and conceal, which are very similar in meaning but, at a closer look,
1
not identical. A large group of cases is constituted by pairs such as mail and e-mail,
disc and compact disc: a compound lexeme is synonymous with its second part (its
so-called ‘head’, see 8.2.1) when taken in a specific reading. The specific reading is
a differentiation (recall 3.4.3.1) of the more general reading in which, for example,
mail may also refer to a letter on paper or a parcel or to the institution of mail. Other
instances of partial synonymy are English spectacles and glasses which may both
denote the same sort of objects that persons wear on their noses to look through;
but glasses may also just be the plural of glass in one of its other meaning variants.
In everyday use, the term ‘synonymy’ is applied more broadly to pairs of
expressions which can mean the same in certain contexts. This non-technical notion
of synonymy is not based on the real meanings of the expressions, but just on their
use, i.e. on their denotations. In our terminology, such ‘synonyms’ are equivalents
in certain contexts rather than having a similar conceptual meaning. If candidates
for synonymy are tested in various contexts, they will most probably turn out to be
matching in some, but not in others. For example, it is easy to perceive of contexts
where the word glasses can be exchanged for spectacles; however, of course, there are
also cases where this is impossible (e.g. she drank two glasses … but not: … two
spectacles of orange juice).
Having identical meanings, synonymous expressions are logically equivalent, since
meaning determines denotation and truth conditions. The converse is not true, as
we saw in 7.6.1: logically equivalent expressions may have different meanings. A
remarkable group of such cases is constituted by pairs consisting of a new expression
introduced for the sake of political correctness and the expression it is meant to
replace. Consider, for example, the expression blind and its replacement visually
impaired, introduced in order to avoid discrimination. The new ‘politically correct’
expression is (meant to be) denotationally equivalent, but it is chosen because it does
not possess the same meaning. Language provides many different ways of denoting
the same thing, and the alternative expressions may well differ in meaning.
1 See Cruse (1986: §12.1) for a discussion of this and similar cases.
Meaning relations
205
8.2 HYPONYMY
8.2.1 Hyponymy
The meaning relation of hyponymy can be defined as follows: 2
DEFINITION 2 Hyponymy
An expression A is a hyponym of an expression B – and B a hyperonym of
A – iff
(i) the meaning of B is a proper part of the meaning of A and
2
(ii) A is a logical subordinate of B, i.e. A entails B.
The meaning of A must contain further specifications in addition to the meaning
of B, rendering the meaning of A, the hyponym, more specific than the meaning of
B. Hyponymy is a relation between lexemes – or more complex expressions – that
results from a relation between their MEANINGS and gives rise to a relation between
their DENOTATIONS (cf. condition (ii)). These relations are shown in Fig. 8.1 which
integrates the semiotic triangles for a hyponym and its hyperonym.
Figure 8.1
Hyponymy
duck
hyponym
meaning of hyponym
meaning of
hyperonym
bird
hyperonym
denotation
of hyperonym
denotation
of hyponym
In the semantic literature, hyponymy is often not distinguished from logical
subordination; condition (ii), alone, is regarded sufficient. This is the result of a
tendency to restrict semantics to the investigation of descriptive meaning and this, in
2 Recall 7.5.2 for the definition of subordination. Strictly speaking, the definition has to be related
to single meaning variants of expressions. For example, the word compound is a hyponym of
expression (in one of its readings) if it is taken in its linguistic sense, but a hyponym of substance
(again, only in one of its meanings), if taken in the sense of chemical compound. Thus, the
omnipresence of polysemy complicates the matter of meaning relations considerably. For the sake
of simplicity, polysemy will be ignored throughout this chapter unless indicated otherwise.
206
Understanding semantics
turn, to the logical aspects of meaning. In this book, however, the distinction between
logical and conceptual notions is considered a point central for the understanding of
semantics and linguistic meaning.
Examples of hyponymy are countless. For hyponymy among lexemes, consider
pairs of subordinates and superordinates in hierarchies like those in Figs 7.2 and
7.3. Here we are entitled to assume the crucial meaning relation (i) in addition to the
logical relation (ii) because, for example, it is part of the meaning of verb that verbs
are words; thus the meaning of word must be contained in the meaning of verb. Other
examples would be beer (hyponym) and beverage (hyperonym), or car and vehicle,
respectively.
A very common type of complex expressions consists of one expression, the ‘head’,
combined with another expression, called the ‘modifier’. Modification occurs in
syntax, e.g. adding an adjective to a noun, and in word formation. The modifier adds
a specification to the meaning of the head. Examples would be compounds such as
book + shop (recall 1.2.5) or syntactic combinations such as red cap. If such complex
expressions are interpreted compositionally, they are generally hyponymous to their
mere heads. Condition (i) is fulfilled since the meaning of the modifier adds some
specification to the meaning of the head. As to condition (ii), usually the denotation
of the whole will be part of the denotation of the head: a book shop is a shop; a red
cap is a cap. We called compounds of this kind ‘regular compounds’ in 1.2.5; they will
be treated in more detail in the next subsection.
In connection with synonymy, cases were mentioned where simple lexemes are, in
a special reading, synonymous with more specific compounds (e.g. disc and compact
disc). Since the special reading is a conceptual differentiation of the more general
reading, it adds content to the latter. In general, if the meaning of an expression A
is a differentiation of the meaning of B, then A is a hyponym of B. A similar case
concerns certain alternations: one use of a verb may be hyponymous with another:
if I eat an orange, I eat, if I open the door, the door opens; thus, transitive verbs like
open are hyponymous with their intransitive antipassive variant and with verbs of
change like open, the causative variant is hyponymous with the inchoative variant
(recall 6.1.2, 6.1.3).
As was pointed out in 7.6.2, not all subordinates are hyponyms; two expressions
may fulfil condition (ii) without fulfilling (i), for example mother-in law’s son and
husband. There are also pairs of expressions which fulfil condition (i), but not (ii). In
barfly the head is to be taken metaphorically, whence a barfly is not a fly. Similarly,
Blue Helmet metonymically refers to certain persons wearing blue helmets; thus a
Blue Helmet is not a helmet. In yet a different type of compounds, represented e.g. by
blue-green, the meaning of the second part (or the first part, likewise) is clearly part
of the meaning of the whole, but the whole is not subordinate to one of its parts: bluegreen is not a special shade of green, but a colour between blue and green, which is
3
neither blue nor green. Apart from such compounds, there are also combinations of
3 In a sense, compounds such as blue-green or Blue Helmet are regular, too. They are just less
frequent, and their formation and interpretation obey different rules than the type represented by
bookshop.
Meaning relations
207
adjectives and nouns that are not hyponymous with the noun itself: an ‘alleged thief ’
4
is not necessarily a thief, a ‘former butcher’ usually not a ‘butcher’.
8.2.2 Regular compounds
A regular compound has two parts, the modifier and the head; the modifier adds
a specification to the meaning of the head noun. For example, in apple juice, the
modifier apple adds the specification ›produced from apples‹ to the concept ›juice‹.
Similarly, park bench means ›bench in a park‹, headache pill ›pill for curing a
headache‹, company flat ›flat belonging to a company‹, flower shop ›shop selling
flowers‹, and so on. The interpretation of regular compounds ensures that the
compound is a hyponym of its head.
The meaning of the modifier is also bound into the meaning of the compound, but
not in the same way as the meaning of the head. Therefore, the resulting meaning
relation between the compound and the modifier is not hyponymy (apple juice is not
an apple, a flower shop not a flower, etc.). Rather, the meaning relation depends on the
way in which the meaning of the modifier is built into the meaning of the whole. For
example, the meaning relation between flower and flower shop is as follows: referents
of the first are sold in referents of the second. Special though this meaning relation
is, it provides a basis for a general pattern of composition (cf. analogous compounds
such as stationery shop, ticket counter, hot dog stand). The other examples mentioned,
park bench, company flat and apple juice represent different patterns. More will be
said about the relations between the modifier concept and the head concept in 12.3.4.
Figure 8.2 illustrates the meaning relations within regular compounds in general,
using apple juice as an example.
Figure 8.2
Meaning relations between a regular compound and its components
apple
juice
MODIFIER
HEAD
(various)
hyponymy
apple juice
COMPOUND
4 Recall 5.4.2 on non-predicational adjectives
208
Understanding semantics
8.3 OPPOSITIONS
8.3.1 Examples
If people are asked to specify the opposites of words like old, buy and aunt, they will
answer spontaneously. The answers will vary, however, because words may have
different kinds of opposites. The intuitive notion of opposites covers a variety of
meaning relations. In this section, we will take a closer look at them. The technical
term for such relations is oppositions.
This is what people answer if they are asked to give the opposite of old, buy and
aunt:
(1)
old:
(i) new
(ii) young
buy:
(i) sell
(ii) steal, borrow, rent
aunt:
(i) uncle
(ii) nephew, niece
The two opposites of old reveal that old is polysemous. Old as an opposite of young
relates to age, primarily of living creatures. The opposites old and young can also
be used for inanimate entities that are metaphorically conceived as subject to
developments such as growth, ripening and ageing (cf. a young nation, an old wine, a
young language). In such cases, young and old denote early and late stages of internal
development. However, old can also be used for cars, books, buildings, words. Old
then means that the respective object has been in use for a long time (probably
resulting in wear and tear, in becoming out of date, etc.); new is the opposite of this
meaning variant. As to their relation to old, both opposites are of the same type:
old denotes one extreme on a scale, while young and new denote the other extreme.
The difference between new and young and the two corresponding meanings of old
concerns the underlying scale.
What counts as the opposite of buy is less clear. If we take a look at the second set
of verbs (steal, borrow, rent), we realize that all these verbs denote alternative ways of
acquiring things. Borrow and rent contrast with buy in that the transfer of possession
is temporary. This meaning element is absent in the cases of buy and steal. Buy and
rent involve a compensation with money, steal and borrow do not. The three opposites
under (ii) are of the same kind: they are alternatives in certain respects, but they do
not denote an opposite EXTREME on a scale.
The opposite sell, however, is completely different. There are two ways of viewing
the relation between buy and sell. First, any event of buying is at the same time an
event of selling: if x buys something from y, then y sells that thing to x. To conceive
of an event as an instance of selling rather than buying is just a converse way of
conceptualizing this transfer: an object z changes from the possession of y into the
possession of x, while x pays y an equivalent amount of money in exchange for z. Buy
describes the event as an action on the part of the new possessor x, sell describes it as
an action on the part of the original possessor y. Buy and sell express the same with
reversed roles. Second, we can compare x buying something with x selling it. Then,
Meaning relations
209
buy and sell are opposed in the sense that buy means ‘moving into’ the possession
of the object and sell ‘moving out’ of it. Viewed in this way, the two verbs express
opposite directions of transfer.
Whenever we think of oppositions in the intuitive sense, we try to find something
common to the meanings of the two expressions and a crucial point in which their
meanings differ. In the case of the opposition aunt/uncle, the point in common is
the kind of kinship relation: being a sibling of a parent, the spouse of a sibling of
a parent or something similar, a relative one generation back but not in direct line.
What differs is the sex of the relative. If we consider the pairs aunt/nephew and aunt/
niece, we encounter the same kind of relation as between buy and sell under the first,
reversed-roles, perspective: if x is an aunt of y, then y is a nephew or niece of x. The
kinship link between x and y is the same, but aunt refers to one side of the link and
nephew/niece to the other. The relation is not strictly reversed, since the terms aunt,
nephew and niece also contain a specification of sex. We cannot say that if x is an aunt
of y then y is a nephew of x, because y could also be a niece of x. Nor can we say that
if x is a nephew of y then y is an aunt of x, because y could also be an uncle.
8.3.2 Antonyms
DEFINITION 3 Antonymy
Two expressions are antonyms iff they express two opposite extremes out of a
range of possibilities.
The prototypical examples of antonymous expressions are pairs of adjectives such
as old/young, old/new, big/small, thick/thin, good/bad, light/dark, difficult/easy.
Their meanings can be illustrated by means of a scale of age, size, diameter, quality,
brightness, difficulty, etc. which is open on both sides. One might object that scales
such as the scale of size are not open at the lower end, i.e. that zero size delimits the
scale at its bottom. However, in our imagination, things can be smaller and smaller;
for every small thing there can be something smaller. There is no smallest size and
hence no lower bound of the scale of size. The scale of size asymptotically approaches
zero size, but does not include it. (Note that we cannot say something is ‘small’ if it
has no size at all.) One of the adjectives denotes the upper section, the other one
the lower section of the scale. There may be a neutral middle interval where neither
expression applies. Every value on the scale that lies above the neutral interval counts
as, for example, ‘big’; every value below the neutral interval counts as the opposite,
e.g. ‘small’ (cf. Fig. 8.3).
Antonyms are logically incompatible, but not complementary (7.5). If something
is big, then it is not small, and vice versa. But if something is not big, it need not
be small, and vice versa – it can be in the neutral range of the scale, i.e. neither big
nor small. Words for the neutral case are rare and rather artificial recent inventions
(middle-aged, medium-sized). Normally, the neutral case can only be expressed by
210
Understanding semantics
Figure 8.3
Antonymy of big and small
← small
SMALLER
big →
neutral
BIGGER
such cumbersome expressions as neither difficult nor easy. Antonymous adjectives
of the type illustrated in Fig. 8.3 are called scalar adjectives. They are ‘gradable’ and
therefore allow for the full range of adjectival forms and constructions: comparative
(bigger than), superlative (biggest), equative (as big as), or modifications such as very
big, big enough and too big.
In very many cases, the antonym of an adjective is formed by prefixing un- or its
Latin-origin equivalent in-/im-/ir-/il-: pleasant/unpleasant, likely/unlikely, adequate/
inadequate, probable/improbable, rational/irrational, logical/illogical. In each case,
the unprefixed adjective is felt to be positive, not in the sense of an expressive
meaning component, but as intuitively denoting the section at the upper end of the
scale, while the prefixed opposite is the negative counterpart at the lower end of the
scale. We shall see below that pairs of this form (X vs un-X, etc.) are not necessarily
antonyms. The pattern also occurs with pairs of expression that form different types
of opposition.
Antonymy is not restricted to adjectives. There are antonymous pairs of nouns
such as war/peace, love/hate and some antonymous pairs of verbs: love/hate or
encourage/discourage. The pair all/no(thing) is antonymous, as are pairs of adverbs
such as always/never, often/seldom, everywhere/nowhere.
8.3.3 Directional opposites
Pairs such as in front of/behind, left/right, above/below have much in common with
antonyms. For each such pair there is a point of reference from which one looks in
opposite directions on a certain axis. Imagine yourself standing in a normal upright
position, your head not turned or bent. Then the direction in which you look is what
is denoted by in front of (you) and behind (you) denotes the opposite direction. The
two directions are distinguished in several ways: the body has a front, which includes
the face, the breast, etc., and it has a back; when one walks in the usual way, one walks
into the direction that is in front of one. Thus, the body defines an axis in space, the
front–back axis, or primary horizontal axis. Another axis defined by the body is
the vertical head–feet axis. The extension of the direction from the body centre to
the head defines the direction above; the opposite direction is denoted by below.
Finally, the body with its two eyes, two ears, two arms, two legs and its general lateral
symmetry defines another horizontal axis, the left–right axis. The three axes and
how they relate to the corresponding expressions are depicted in Fig. 8.4.
Meaning relations
211
Figure 8.4
Directions
above
behind
left of
right of
in front of
below
DEFINITION 4 Directional opposition
Two expressions are directional opposites iff they express opposite cases with
respect to a common axis.
The type of opposition represented by in front of/behind is called directional
opposition. Directional opposites are related to opposite directions on a common
axis. Further examples that involve the vertical axis are: top/bottom, high/low, up/
down, upstairs/downstairs, uphill/downhill, rise/fall, ascend/descend and many more.
Examples related to the primary horizontal axis are forwards/backwards, advance/
retreat.
A similar axis is the time axis. We talk about things happening ‘before’ vs ‘after’ a
certain time, or ‘later’ vs ‘earlier’. Thus, pairs such as before/after, past/future, since/
until, yesterday/tomorrow, last/next, precede/follow are directional opposites related
to the time axis. A further case is provided by past tense and future tense (recall
6.4). Also related to time are pairs of accomplishment verbs or verbs of change with
opposite resultant conditions (recall 6.2), such as tie/untie, pack/unpack, wrap/
unwrap, dress/undress, put on/put off, get on/get off, arrive/depart, switch on/switch
off, embark/disembark, charge/discharge, enter/leave, begin/stop, start/finish, fall
asleep/wake up, appear/disappear, open/close and many more. One member denotes
the coming about, or bringing about, of a certain state, while the other member
denotes the reversed process or action. Among the examples in (1), the pair buy/sell
in its second interpretation is of this type.
212
Understanding semantics
8.3.4 Complementaries
The type of opposition represented by aunt/uncle, buy/rent or buy/steal is known as
complementary opposition.
DEFINITION 5 Complementary opposition
Two expressions are complementaries iff they express an either-or alternative
in some domain.
Complementary opposites are logically complementary (7.5.4): the negation of one
term is equivalent to the other term, e.g. not even (of numbers) means the same as
odd. Each expression denotes one out of the only two possibilities in some domain
of cases. In this domain, complementary opposites represent an either-or alternative.
Complementary adjectives are not gradable; they do not permit the comparative,
5
superlative, equative or modification with very, etc. Examples of complementaries are
married/unmarried, female/male (as a specification of sex), even/odd (of numbers),
possible/impossible, free/occupied. As some of the examples show, prefixation with
un- or in- is also used for the formation of complementary opposites.
Complementarity more typically occurs with nouns, e.g. pairs of terms for persons
of opposite sex, or pairs such as member/non-member, official/non-official. The
meanings of two complementaries are identical except for one crucial feature in
which they differ (cf. the examples aunt/uncle or buy/rent). This feature has to be a
binary, yes-or-no, feature that leaves just two possibilities, e.g. female or male sex.
8.3.5 Heteronyms
DEFINITION 6 Heteronymy
Expressions are heteronyms iff they denote alternatives in some domain of
more than two possibilities.
Heteronymy is a matter of more than two expressions. A typical example is the set of
terms for the days of the week, the set of basic colour terms (for more on colour terms,
see 10.4) or terms for kinds of animals, plants, vehicles, etc. A set of heteronymous
terms jointly covers a wider range of possibilities. Logically, heteronyms are mutual
contraries: if x is Sunday, it cannot be Monday, Tuesday, etc.; if x is a dog, it cannot be
5 To be sure, adjectives such as married actually can be used in the manner of a grading adjective.
Someone might choose to say, e.g., Mary is more married than John is. In such a case, however,
the adjective married is coerced by the comparative form into a gradable property that is
metonymically connected to marriage, for example ‘feeling married’ or ‘acting like being married’.
This contextual meaning shift is triggered by the comparative.
Meaning relations
213
a duck, a cow or any other species. But unlike antonymy, heteronymy is not related to
scales; heteronyms are not opposite extremes, but just members of a set of different
expressions which have a common hyperonym.
There are large fields of heteronymous terms, such as the sets of co-hyponyms
in larger taxonomies (see 8.4.3): terms for plants, flowers, animals, birds, breeds of
dogs, kinds of clothing, drinking vessels, food, vehicles, musical instruments, etc.
Sets of more than two co-hyponyms occurring at the same level in a lexicalized
hierarchy always form a set of heteronyms. Apart from nouns, there are many fields
of heteronymous verbs, such as the different verbs of motion (walk, run, fly, swim),
and verbs denoting human activities such as eat, work, sleep, dance, etc. Further
examples of heteronymy are colour terms or number words.
8.3.6 Converses
The pairs buy/sell and aunt/niece-nephew represent what are called converses. This
kind of relation is restricted to predicate expressions with two or more arguments.
Such terms express a relation in the widest sense, e.g. an action involving two
participants, a comparison, or a kinship relation. Converses are defined as follows:
DEFINITION 7 Converses
Two expressions are converses of each other iff they express the same relation
between two entities, but with reversed roles.
Examples are above/below, before/after, borrow/lend, parent/child, as well as some of
the technical terms introduced here: entail/follow from, hyponym/hyperonym. These
pairs of expression give rise to the following logical equivalences:
(2)
above/below :
x is above y
y is below x
before/after :
x is before y
y is after x
borrow/lend :
x borrows z from y
y lends z to x
parent/child :
x is a parent of y
y is a child of x
entail/follow from :
x entails y
y follows from x
hyponym/hyperonym :
x is a hyponym of y
y is a hyperonym of x
Converses differ from other types of opposites in that they do not correspond to a
uniform logical relation. The pairs above/below, before/after, borrow/lend, hyponym/
hyperonym are contraries (for example, x is above y logically excludes x is below y).
Parent and child, too, are contraries. Entail and follow from are logically independent
because if A entails B it may or may not be the case that A also follows from B (i.e.
B also entails A, recall 7.3.1). Some expressions could even be considered their own
214
Understanding semantics
converses, namely all terms which express symmetric relations (different from, sibling
of, married to) or events that involve two arguments in the same role (meet, fight,
agree, recall the verbs of social interaction mentioned in 6.1.3): x is different from y/a
sibling of y/married to y means the same as y is different from x/a sibling of x/married
to x; similarly x meets/fights/agrees with y has the same meaning as y meets/fights/
agrees with x. Such expressions are called ‘reciprocal’.
One major group of converses is provided by the comparative forms of antonymous
adjectives:
(3)
thicker/thinner:
x is thicker than y
For transitive verbs, passive constructions with explicit
general means of conversion:
(4)
watch/be watched by:
x watches y
y is thinner than x
AGENT
provide another
y is watched by x
Related to the passive conversion is a productive pattern for pairs of converse nouns
in English: noun derivations from verbs with the suffixes -er and -ee. For example, x
is an employee of y means the same as y is an employer of x.
As some of the examples may have shown, one and the same pair of expressions
can constitute opposites of more than one type. Buy/sell and also pairs of spatial
prepositions and adverbs such as above/below can be considered converses as well
as directional opposites.
Table 8.1 shows a survey of the meaning relations treated here.
Table 8.1
Types of oppositions
Examples
Type
Characterization
Logical relation
big/small
war/peace
to love/to hate
antonyms
opposite extremes
on a scale
contraries
above/below
before/after
lock/unlock
directional
opposites
opposite directions
on an axis
contraries
even/odd
girl/boy
voter/non-voter
complementaries
either-or alternatives
within a given domain
complementaries
Monday/Tuesday/…
red/green/blue/…
heteronyms
more than two alternatives
within a given domain
contraries
buy/sell
wife/husband
bigger/smaller
employer/employee
converses
the same with
reversed roles
(relations only)
(various logical
relations)
Meaning relations
215
8.4 LEXICAL FIELDS
8.4.1 The notion of a lexical field
Most lexical items form groups with other lexemes. Antonyms and complementaries
belong together as opposites, as do pairs of words such as father and mother or adult
and child, or sets of words like the terms for the days of the week, for colours, for
numbers, for pieces of furniture or other kinds of thing within one superordinate
category.
Semantic theories of different orientations have tried to capture this phenomenon
by the notion of a lexical field. The literature offers very different definitions. Here,
the following informal characterization will be adopted.
DEFINITION 8 Lexical field
A lexical field is a group of lexemes that fulfils the following conditions:
∑ the lexemes are of the same word class
∑ their meanings have something in common
∑ they are interrelated by precisely definable meaning relations
∑ the group is complete in terms of the relevant meaning relations
Lexical fields are notated like sets by including a list of lexemes in set brackets { … }.
For example, the field of the terms for the days of the week would be written {Sunday,
Monday, …, Saturday}.
The meaning relations that constitute a lexical field may be rather particular (e.g.
sex and age oppositions). The six prepositional expressions in Fig. 8.4, above, below,
in front of, behind, right of, left of, form a field of three opposite pairs relating to three
axes in space that are orthogonal to each other. The names for the seven days of the
week are not merely heteronyms but are in addition interrelated by a cyclic order
Sunday Monday Tuesday Wednesday Thursday Friday Saturday Sunday … The relation of one name to the others in terms of this cycle is part of
its meaning. This is the reason why sentences such as (5) are logically true.
(5)
If today is Monday, tomorrow is Tuesday.
A similar relation of order interrelates the numerals. In addition, the field of numerals
is structured by specific meaning relations that capture the arithmetic properties of
numbers; for example the relation that thirty denotes a number that is three times
the number denoted by ten.
8.4.2 Small fields
Some fields are quite small. For example, each pair of antonyms, such as thick and
thin, forms a lexical field of two members. The meanings of the two antonyms have
216
Understanding semantics
in common that both relate to an end section of the same scale, they are related by
antonymy, and the group is complete because there are no other adjectives relating to
the same scale. Polysemous words with two different antonyms, such as old with its
antonyms new and young, belong to two different fields, {old1, young} and {old2, new}.
Strictly speaking, it is not words that belong to lexical fields but words in certain
meaning variants.
Complementary expressions such as girl and boy also form a field of two, viz. the
field of terms for children specified for sex. If we add the sex-neutral expression child
(in its non-relational meaning ›non-adult‹, not in its relational meaning ›offspring
of‹) to this field, we obtain the field of general terms for children. This field is
structured by two different meaning relations: hyponymy between child and boy/girl
and opposition in terms of sex between girl and boy. There are other lexical fields of
three elements with the same structure, e.g. adult/woman/man or horse/mare/stallion.
We can unite the field of general terms for children with the field for general terms for
adults. The resulting six expressions form a field of six words linked by the additional
meaning relation of opposition in terms of age that relates the members of the
pairs child/adult, girl/woman and boy/man. The result is shown in Fig. 8.5. Different
meaning relations are symbolized by different kinds of arrow.
Figure 8.5
The lexical field of general person terms
adult
woman
man
child
girl
boy
8.4.3 Taxonomies
Terms for animals, plants, food or artefacts such as furniture, vehicles, clothes,
musical instruments, etc. form lexical fields of considerable size. Their underlying
structure is a hierarchy with two or more levels: a topmost hyperonym like vehicle,
a level of general terms such as car, bicycle, boat, aeroplane, and further levels of
Meaning relations
217
more specific heteronymous terms for kinds of cars, bicycles, boats, planes, etc. Such
systems represent a special type of hierarchy called taxonomy:
DEFINITION 9 Taxonomy
A set of expressions is a taxonomy iff:
(i) they form a conceptual hierarchy in terms of hyponymy
(ii) hyponyms denote sub-kinds of what their hyperonyms denote
The hierarchy of word class terms in Fig. 7.2 is a taxonomy: nouns, verbs and
adjectives are kinds of words, count nouns and mass nouns are kinds of noun,
etc. The hierarchy of the three sibling terms sibling, sister and brother in Fig. 7.3,
however, is not a taxonomy: brothers and sisters are not kinds of siblings. Referents
of co-hyponyms in taxonomies differ in many properties. Referents of co-hyponyms
of small, often binary, hierarchies such as sibling/sister/brother differ only in one
property, e.g. in sex. Figure 8.6 contains a section of the animal taxonomy. We will
come back to taxonomies and their properties in 11.3.
Figure 8.6
An incomplete taxonomy of animal terms
animal
horse
cow
husky
pig
dog
terrier basset
cat
elephant tiger crocodile
collie
8.4.4 Meronymies
Many objects in the world are conceived as a whole consisting of different parts. And
correspondingly, our concepts for complex objects contain specifications of these
parts as elements. One of the best examples of a complex object is the human body
with its various parts, their subparts, and so on. The respective terms form what
is called a mereology, a system of terms for a whole and its parts, the parts of the
parts, and so on. Parts in a mereology are not only pieces of the body, like a butcher
might cut out of the body of a slaughtered animal. They also exhibit a certain degree
of autonomy; they serve a distinct function – it may be possible to move them
independently of other parts; in the case of artefacts, they may be replaceable; and
the parts may in turn be composed of other parts. For example, the head is the part
218
Understanding semantics
6
of the body that carries the most important sense organs, viz. eyes, ears, nose and
tongue; it contains the brain, etc. The face forms the front side of the head; it contains
the eyes, the nose and the mouth; it serves facial expression. Within the face, the
mouth is used for eating, drinking, speaking, breathing, biting, kissing, smiling, etc.;
the tongue serves for articulation, tasting, licking, swallowing and other purposes.
The items that form a mereology are necessarily relational nouns. In many
languages, body part terms are used with inalienable possessive constructions
(recall 5.4.1.2). A small section of the system of body part terms in English is given
in Fig. 8.7.
Figure 8.7
An incomplete meronymy of body part terms
body
neck
head
ear
belly
face
eye
mouth
lip
trunk
leg
arm
breast
hand
finger
forearm
palm
tongue
A system of this type is not to be confused with a hierarchy based on hyponymy.
The vertical lines stand for the part–whole relation between single potential referents
of the terms. For example, a potential referent of face is always part of a potential
referent of head. By contrast, the vertical lines in a logical hierarchy represent a
part–whole relation between the total denotations of the terms involved: the set of
all ducks is a subset of the set of all birds, but, of course, a single duck is not part of a
single bird. Hyponymy is based on a third kind of part–whole relation: the meaning
of the ‘higher’ (!) term, the hyperonym, is part of the meaning of the lower term, the
hyponym.
6 The word body is ambiguous because it can denote either the trunk or the trunk plus the head and
the limbs; the term body is used in the latter meaning here.
Meaning relations
219
The technical term for the constituting meaning relation is meronymy (from
ancient Greek meron ‘part’; the term partonomy is also used). A system based on
meronymies is called a mereological system, or mereology.
DEFINITION 10 Mereology, Meronymy
A set of expressions forms a mereology iff they form a hierarchy in terms of
holonyms and meronyms, where A is a meronym of B, and B a holonym of A,
iff A denotes constitutive parts of the kind of things that B denotes.
By a ‘constitutive part’ we mean a part that necessarily belongs to the whole and
contributes to making the whole what it is. Thus, the meaning of the meronym is
a concept that describes its potential referents as belonging to this kind of whole;
conversely, the meaning of the holonym describes its potential referents as having
this kind of part(s). For example, the concept ›hand‹ describes hands as having
fingers, and the concept ›finger‹ describes fingers as being certain parts of hands.
Meronymies involving nouns can be spelt out with the help of sentences of the
form ‘an A has a B’, e.g. a head has a face, a face has a mouth, a mouth has a tongue, a
tongue has a tip, etc. This is, however, not a reliable test for meronymy. Every person
has a mother, a weight, an age, and so on, but these are not parts of the person.
There is a further important difference between lexical hierarchies and meronymic
systems: unlike hyponymy and subordination, meronymy is not necessarily a
transitive relation. Although in Fig. 8.7 the term eye is linked to face and face is linked
to head, this does not result in the same kind of link between eye and head. If ‘part’ is
taken in the sense of ‘constitutive part’, i.e. something that essentially contributes to
making up the whole, then an eye is not a part of the head, as the tongue is not a part
of the face, or the face a part of the whole human body.
This much on meaning relations may suffice for the chapters to come. Meaning
relations will play a prominent role in the next chapter on meaning components.
EXERCISES
1. What are meaning relations? How do they differ from logical relations?
2. If A is a hyponym of B, what does this mean for the way in which
a. the meanings
b. the denotations
of A and B are related to each other?
3. Try to define the meaning relations between the modifier and the whole
compound for the following cases; find two analogous compounds each:
a. wood house
b. gold digger
c. soup spoon
220
Understanding semantics
4. Explain the difference between hyponymy and meronymy.
5. Which special feature distinguishes taxonomies from arbitrary hierarchies based
on hyponymy?
6. Discuss the meaning relations that hold between the following pairs of words; it
may be necessary to distinguish different meaning variants:
a. marry – divorce
b. red – orange
c. somewhere – nowhere
d. clean – dirty
e. friend – enemy
f. violin – string
7. The adjective light has several antonyms. Try to determine the pairs of opposites
and the meaning light has in each case. Consult the Oxford online dictionary for
meaning variants of the adjective light, http://oxforddictionaries.com/ (note that
there are three separate entries).
FURTHER READING
Cruse (1986) for an in-depth treatment of ‘sense relations’. For a recent survey
see Cann (2011). Foley (1997, ch. 6) on kinship terms from an anthropological
perspective (lexical fields of kinship terms were extensively studied in anthropological
linguistics.)
9
Meaning components
The assessment of semantic phenomena such as ambiguities and logical or meaning
relations forms a very important step of developing a semantic theory. However,
theorizing itself only starts when one tries to EXPLAIN the phenomena. For example,
the senses of uncle and aunt are related in the same way as those of sister and brother,
and Violetta is an aunt of Christopher logically entails Christopher is a nephew of
Violetta (provided Christopher refers to a male person and Violetta to a female one).
But how can this be derived from the meanings of the relevant terms? Likewise, verbs
carry selectional restrictions for their arguments (5.7). How can this be accounted
for in a principled way?
There are many different semantic theories. Each one defines the notion of
meaning differently and uses different means for the representation of meanings,
some formal, some informal. But almost all of them share a basic strategy of
explaining the semantic data, a strategy that ties in with our intuitive notion
of meaning: they assume that the meanings of most words are complex, composed of
more general components. The meaning of a lexical item is analysed by identifying
its components and the way in which they are combined. In a sense, the process
of semantic analysis is the converse of the process of composition by which we
determine the meaning of a complex expression on the basis of its components and
its structure. Therefore analysis into meaning components is called decomposition.
A decompositional theory allows the pursuit of the following objectives:
(M) Meaning. Providing models for word meanings.
What kind of entities are lexical meanings? What is their structure? How are
meanings to be represented?
(B)
Basic meanings. Reducing the vast variety of lexical meanings to a smaller set
of basic meanings.
Are there lexical items which are semantically basic? How can the meanings of nonbasic items be built up from more basic ones?
(P)
Precision. Providing a means of representation that allows a precise analysis of
lexical items.
What exactly are the components of lexical meanings? How can they be determined
and how can they be described in a precise way?
(R)
Meaning relations. Explaining meaning relations within and across lexical
fields.
Understanding semantics
222
How can meaning relations be accounted for on the basis of decompositional
meaning descriptions?
(C)
Composition. Explaining the compositional properties of lexical items.
With which kinds of expression can a given item be combined? How does its
meaning interact with the meanings of the expressions it is combined with?
Selectional restrictions represent one kind of phenomena relevant to the last point.
A more ambitious decompositional semantic theory would further strive to
develop an analytical apparatus that can be used for different languages. With such
an apparatus it is possible to compare the meanings of expressions from different
languages.
(L)
Language comparison. Determining the semantic relations between
expressions of different languages.
Are there expressions with the same meaning in other languages? If not, how are
the meanings of similar expressions related to each other?
For example, English and Japanese terms for ‘brother’ can be described as in Table
9.1. Japanese distinguishes between elder and younger siblings. Furthermore, there
are pairs of formal and informal terms with the same descriptive meaning. If the
relative is treated as superior, the formal terms (onīsan and otōtosan) are used.
Obviously none of the four Japanese terms is equivalent to the English word brother;
they all relate to brother like hyponyms.
Table 9.1
Meaning components of English and Japanese terms for ‘brother’
Meaning component specifying the referent’s
..kinship relation
to EGO
..sex
..age relation
to EGO
social
meaning
E
brother
›sibling‹
›male‹
–
–
J
ani
›sibling‹
›male‹
›elder‹
informal
J
onı-san
›sibling‹
›male‹
›elder‹
formal
J
oto-to
›sibling‹
›male‹
›younger‹
informal
J
oto-tosan
›sibling‹
›male‹
›younger‹
formal
Four different approaches to decomposition will be introduced in this chapter.
Each of them adopts a different conception of meaning with its own merits and
shortcomings. The first one to be introduced is the structuralist approach, now
a century old, which has exerted an enormous influence on the development of
modern linguistics and will therefore be treated in greater detail and without
restriction to semantics.
Meaning components
223
9.1 THE STRUCTURALIST APPROACH
9.1.1 Language as a system of signs
The Swiss linguist Ferdinand de Saussure (1857–1913) is generally credited with
the approach known as structuralism. The outlines of his approach appeared in
1916 under the title Cours de linguistique générale (‘Course of General Linguistics’), a
compilation of material from lectures Saussure held around 1910.
To Saussure, a language is an abstract complex system of relations and rules that
underlies all regularities to be observed in actual language use. The system is formed
by signs which are related in multiple ways. A sign, e.g. a word, consists of two parts.
One part is its sound form. The other part is its meaning. The association between
form and meaning of a sign is fixed by conventions of language use. The association
is ARBITRARY, i.e. any word could just as easily have a different meaning, and the
meaning could be associated with a different expression – provided the conventions
of the language were different. Saussure emphasizes that the sign and its parts are
real entities in the minds of the speakers – mental or cognitive entities in modern
terminology.
What distinguishes the structuralist approach is the method of analysis and the
resulting concept of linguistic signs. Saussure argues that language is to be studied
exclusively ‘from within’. The language system is a very complex structure formed by
units at different levels: single sounds, syllables, words, syntactic phrases, sentences.
Each unit is related to the smaller units it consists of and to the larger units it is a
part of, and also to all sorts of other units which are similar in some respect or other.
A sign constitutes a sign only as part of a system – only insofar as it is related to, and
different from, the other signs of the language. This holds for both the form and the
meaning of a sign. Form and meaning are only defined ‘negatively’, i.e. by differing
in specific ways from the other signs. They have no positively defined ‘substance’
and therefore cannot be described just for themselves; they can only be analysed by
describing their relationships to other signs. Neither the form nor the meaning of a
sign could exist independently of the language system it belongs to.
Let me illustrate these abstract considerations with a concrete example. The
1
French word rouge, standard pronunciation [], means ›red‹. Its sound form
is a string of three sound units (phonemes), /r/, /u/ and //, for which there are
certain conventions of articulation. Each of the phonemes allows for considerable
phonetic variation; /r/, for example is standardly pronounced []; but it can also
be pronounced as in Italian (rolled [r]) or even as [x] in German Buch. As long as
it can be distinguished from the other sounds of French, in particular /l/, its actual
articulation may vary. Japanese /r/ can even be pronounced [l] because Japanese
does not have two phonemes /l/ and /r/ but just one of this type (so-called liquids).
1 See Gussenhoven and Jacobs (2011, ch.1) for an explanation of the sounds. It is common practice
in linguistics to write the phonetic pronunciation in square brackets […] and with the symbols of
the IPA (International Phonetic Association) alphabet, while the phonological form and phonemes
(sound units of the language system) are written in slashes /… / using standard letters if possible.
224
Understanding semantics
Thus, Saussure argues, a sound unit of the system is not primarily defined by its
articulation or acoustic quality but by its relation to the other sound units.
The other component of the sign, the meaning of French rouge, is similar to the
meaning of English red or German rot (but see the next chapter for other languages).
The exact range of colours that the word denotes depends on which other colour
terms compete with rouge: orange (›orange‹), rose (›pink‹), violet (›purple‹) and brun
(›brown‹). If, for example, French lacked a term for purple, the respective range of
colours would be divided between the words for red, brown and blue, resulting in an
extension of the range of rouge. A sign, according to structuralism, IS CONSTITUTED BY
being different from the other signs of the system. Its form and its meaning are the
sum of their differences and relations to the other forms and meanings of the system.
The structuralist notion of meaning is thus radically relational. Strictly speaking, it
implies that we cannot determine the meaning of a lexeme independently, but only its
relations to the meanings of other lexemes. This notion of meaning is not commonly
accepted. While it is adopted by some semanticists (e.g. in Cruse 1986), alternative
theories have been developed, in particular cognitive semantics (chapter 11). The
cognitive approach does view the meaning of a lexeme as something that can in
principle be investigated and described independently. In this theory, meanings are
considered integral parts not of a system of linguistic meanings but of our overall
cognitive system, which can be studied by the methods of cognitive psychology.
In terms of the semiotic triangle, structuralism focuses exclusively on the left
side of the schema (Fig. 9.1). Semantics is not a matter of reference and denotation.
Rather it concerns the sign, i.e. a pair of form and meaning, and how it is related to
other signs.
Figure 9.1
The focus of structuralist semantics
expression
descriptive
meaning
denotation
9.1.2 Syntagmatic and paradigmatic relations
All linguistic units – sounds, syllables, words, phrases, sentences – can be combined
into more complex units. In the original terminology of structuralism, such a
Meaning components
225
complex is called a syntagm (from ancient Greek syntagma ›compound‹). A syllable
is a syntagm of phonemes, a word taken as a sound unit is a syntagm of syllables;
as a semantic unit, a word may be a syntagm of morphemes (e.g. un-natur-al, see
next section); a syntactic phrase like the NP a lucky decision is a syntagm of words,
a sentence is a syntagm of phrases. For each kind of syntagm there are specific
formation rules and within each syntagm the constituents are related to each other
in specific ways. The general term for the relations within a syntagm is syntagmatic
relations. The syntagmatic relations that a given unit bears to other constituents
are determined by its combinatorial properties. For example, an English NP may
take the form article + noun but not noun + article. Thus English articles have the
combinatorial property of preceding the noun.
Let us consider the syllable /set/ as an example (Fig. 9.2) of a syntagm of phonemes.
Figure 9.2
Three paradigms defined by the syntagm /set/
onset
/b/
bet
/p/
pet
/s/ /e/ /t/ set
/m/
met
/w/
wet
…
…
nucleus
/i/
/æ /
/s/ /e/ /t/
/u:/
/ae/
…
sit
sat
set
soot
cite
…
coda
/k/
/p/
/s/ /e/ /t/
/n/
/l/
…
sec-tion
sep-tet
set
cen-tury
cell
…
The syllable consists of the three phonemes /s/, /e/ and /t/, that occupy the
three basic positions of a syllable, commonly called onset, nucleus and coda. The
syntagmatic relations are first of all those of order: the onset precedes the nucleus
and the nucleus precedes the coda. In addition, nucleus and coda form a unit within
the syllable, the so-called rhyme. Therefore the relation that holds between nucleus
and coda is closer than, and different from, the relation between nucleus and onset.
For each position we can determine which units can occur in this position.
2
The set of all alternatives is called a paradigm (from ancient Greek paradeigma,
›something shown side by side‹). The onset position of the syllable /set/ can be
filled by certain, but not all consonant phonemes (e.g. //, the last sound of sing,
is not possible here) and by certain consonant clusters such as /fr/, which we will
disregard here. The nucleus position defines a paradigm that consists of all vowel
and diphthong phonemes. The coda paradigm consists of all consonant phonemes
that can occur in syllable-final position (excluding, for example, /h/, /y/ and /w/).
The units within a paradigm exhibit relations of difference and similarity. These
are called paradigmatic relations. For example, the opposition voiced vs voiceless
2 This is a special use of the term paradigm in structuralist terminology.
226
Understanding semantics
distinguishes /b/ and /p/ within the onset paradigm. Let us call the corresponding
properties of the units, e.g. voice (being voiced), their contrastive properties.
Table 9.2 displays the basic structuralist concepts. They allow the definition of
basic notions such as syntactic category (a paradigm of syntactic units), word class or
grammatical category (a paradigm of lexical units, 3.1) or lexeme (a unit with certain
paradigmatic and syntagmatic properties).
Table 9.2
Basic structuralist concepts
Structuralist concept
Definition
syntagm
complex unit
syntagmatic relations
relations between the constituents of a syntagm
combinatorial properties
syntagmatic properties
properties of a unit that determine its syntagmatic relations:
how it can be combined with other units
paradigm
set of all elements that can fill a certain position in a syntagm
paradigmatic relations
relations between the elements of a paradigm
contrastive properties
paradigmatic properties
distinctive properties of the elements of a paradigm
9.2 APPLYING THE STRUCTURALIST APPROACH TO MEANING
9.2.1 Semantic units: morphemes and lexemes
For applying the structuralist approach to meaning, one has to determine what the
semantic units of a language are, and describe them in terms of their combinatorial
and contrastive properties. Words may share meaningful parts with other lexemes.
Therefore, words (or lexemes) are not the smallest semantic units of a language.
Consider for example:
(1)
a.
meaningful, beautiful, harmful, successful, useful, sorrowful
b.
ugliness, cleverness, baldness, usefulness, carelessness, laziness, fitness
c.
clearing, hearing, cleaning, thinking, meeting, brainstorming
The meanings of these and many other words can be derived from the meaning
of the first part and the meaning of the suffix: -ful, -ness, -ing. The suffix -ful can
be attached to a large number of nouns and yields an adjective that, very roughly,
means ›having much of N‹ or ›being full of N‹. Thus, the suffix -ful has a meaning
of its own. Likewise, the meanings of the other two suffixes can roughly be defined
as ›the property of being A‹ for -ness attached to an adjective A, and ›situation
where someone V-s (something)‹ for words of the form V-ing. It should be noted,
though, that along with these examples where the meaning can be straightforwardly
Meaning components
227
decomposed into the meaning of the suffix and the meaning of the first part, there
are quite a number of lexemes that contain such a suffix without allowing for
straightforward decomposition. For example, meaning does not mean a situation
where somebody means something, wilderness does not denote the property of being
wilder and lawful does not mean ›full of law‹.
The two components of cleaning, the verb clean and the suffix -ing, cannot be
further divided into meaningful parts. Such minimal semantic units are called
3
morphemes. Clean can be used as a word on its own. It therefore constitutes what
is called a free morpheme. The suffix -ing is a bound morpheme as it can only occur
as a part of a word. The bound morphemes in (1) serve the derivation of words from
others. A second class of bound morphemes, the grammatical morphemes, is used
to build the different grammatical forms of words. Among the comparatively few
examples in English are the 3rd person singular present tense ending (eat-s), the
noun plural ending (horse-s), the possessive ending (king-’s), the past tense ending
(wash-ed), the comparative ending (high-er) and the ending -ing used to form the
progressive form of verbs (be wash-ing), the present participle (the ring-ing bell) and
the gerund (I hate iron-ing).
9.2.2 Paradigmatic and syntagmatic semantic relations
When meaning relations were discussed in chapter 8, we assumed that the related
items were of the same word class. Word classes are essentially paradigms in the
structuralist sense. Hence, meaning relations such as the oppositions, synonymy,
hyponymy, meronymy and the more specific relations structuring certain lexical
fields (8.3) are paradigmatic relations. For example, all terms for the days of the
week can be inserted into the empty position of the syntagm today is ___. Within the
resulting paradigm, we can assess the meaning relations between the terms.
Syntagmatic meaning relations hold between the constituents of a syntagm. They
consist in the way in which the meanings of the constituents are combined to yield
the meaning of the whole syntagm. Let us consider two simple examples. In (2) a
transitive verb is combined with a subject NP and an object NP to form a sentence:
(2)
Mary seized the bottle.
The syntagmatic meaning relation that holds between the subject NP and the verb
is that of the NP specifying the verb’s agent argument. The object NP is related to
the verb as its theme specification. The semantic relations of the NPs to the verb are
indicated by the linking rules (5.6). There is no direct semantic relation between the
subject and the object NP.
(3)
the red balloon
3 See Haspelmath (2010) on the notion of morpheme and other basic concepts of morphology.
228
Understanding semantics
The combination red balloon in (3) describes the referent of the NP as a red balloon.
Both the adjective and the noun are one-place predicates about the potential referent
of the NP (5.4.2.1). Thus the syntagmatic meaning relation between adjective and
noun is one of sharing their argument. This results in a conjunction of the two
predications. In 5.4.2.3 we encountered examples of non-predicational adjectives:
combinations such as alleged terrorist, future neighbour, atomic war, educational
disaster or doctoral dissertation. These adjectives interact with the noun in a different
way.
Predicate terms carry selectional restrictions. These conditions constrain the
choice of expression that can be meaningfully combined with the predicate terms
(5.7). Thus they constitute syntagmatic, or combinatorial, semantic properties of
verbs, adjectives and nouns. In addition to selectional restrictions, which impose
logical conditions on argument specifications, lexemes may have combinatorial
properties that restrict their usage further. For example, German has different
systems of terms for animals and people. Where people ‘essen’ (eat), animals ‘fressen’,
drinking of people is ‘trinken’, of animals ‘saufen’. People have a ‘Mund’ (mouth), but
4
animals a ‘Maul’, ‘Schnauze’, etc. Such distinctions give rise to language-specific
combinatorial meaning properties.
9.3 SEMANTIC FEATURES
9.3.1 Binary semantic features
One of the areas where the structuralist approach proved particularly fruitful was
phonology. As early as in the 1920s and 1930s the Prague School of structuralism
developed a phonological theory where the phonemes were described by sets of
distinctive binary features corresponding, for example, to the mode and place of
5
articulation. The use of features not only allowed the explanation of contrastive
and combinatorial properties of the phonemes within one language. Features could
also be established across languages, enabling a comparison of the sound systems of
different languages. Because of its great success, the feature approach was adopted in
other areas of linguistics, including semantics.
In the variety of approaches subsumed under the label of feature semantics,
semantic features are used as meaning components. For instance, the lexemes in (4a)
would receive a common semantic feature [FEMALE], those in (4b) receive the feature
[MALE], while neither feature is assigned to the expressions in (4c). (It is common
practice to write features within square brackets and with small capital letters.)
4 This only holds when the animal-related terms are used in their primary sense. Not surprisingly,
these expressions can also be used derogatively for people. For example, ‘saufen’ applied to people
means drinking greedily, excessively, etc. or heavily drinking alcohol. Halt dein Maul! (‘Shut up!’, lit.
‘hold your [animal] mouth’) has an extra negative expressive quality compared to the phrase Halt
deinen Mund! with the neutral term for mouth. (Both expressions are rude.)
5 See Gussenhoven and Jacobs (2005, ch. 5) for a phonological feature analysis.
Meaning components
(4)
229
a.
girl
woman
sister
wife
queen
[FEMALE]
b.
boy
man
brother
husband
king
[MALE]
c.
child
person
sibling
spouse
monarch (sex not specified)
The features [FEMALE] and [MALE] are not just different, but complementary. We
can therefore replace them with one binary feature, either [FEMALE] or [MALE], that
assumes the value + or the value –. For the sake of brevity and male chauvinism,
6
the [MALE] variant is chosen. The general feature [MALE] is written [±MALE] or
just [MALE], the fact that the feature has the value + or – is indicated by [+MALE]
and [–MALE], respectively. The term feature is used for both features without a value
([MALE]) and features with a value ([–MALE]).
The words in (4) share the property of denoting persons. This can be captured by
another feature [+HUMAN] that distinguishes them from the terms in (5). (We assume
the primary animal readings of the terms.)
(5)
a.
mare
bitch
cow
ewe
[–HUMAN]
[–MALE]
b.
stallion
cock
tomcat
bull
[–HUMAN]
[+MALE]
c.
horse
pig
sheep
fish
[–HUMAN]
With [HUMAN], [MALE] and an additional feature [ADULT], the meanings of the six
general person terms (8.4.2) can be described as in Table 9.3. The right side is a
so-called feature matrix.
Table 9.3
Features and a feature matrix for the six general person terms
Features and their values
[HUMAN]
[ADULT]
+
–
[MALE]
child
[+HUMAN]
[–ADULT]
girl
[+HUMAN]
[–ADULT]
[–MALE]
+
–
–
boy
[+HUMAN]
[–ADULT]
[+MALE]
+
–
+
adult
[+HUMAN]
[+ADULT]
+
+
woman
[+HUMAN]
[+ADULT]
[–MALE]
+
+
–
man
[+HUMAN]
[+ADULT]
[+MALE]
+
+
+
A feature with the value + or – constitutes a one-place predicate about the
potential referents of the lexeme. For example, if a noun carries the feature [+ADULT],
6 In the view of more recent feature theories, the choice is a matter of markedness. Of the two
alternatives, usually one is marked, or special, and the other unmarked, or normal, with clear
linguistic criteria for markedness. Features and feature values are then chosen in the way that [+a]
stands for the marked variant. In English, there is a certain tendency of marking expressions for
female persons or animals (cf. steward vs steward-ess) by adding a special morpheme. This would
provide a criterion for using the feature [±FEMALE] instead of [MALE].
230
Understanding semantics
its referents have to fulfil the condition of being adult; if it carries the feature
[–ADULT], they have to be not adult; if it has neither feature, no condition with respect
to being adult is imposed on its referents. Thus, in feature semantics word meanings
are considered combinations of a number of one-place predications. For example,
according to the analysis in Table 9.3, the meaning of the word boy is ›human and
male and not adult‹.
9.3.2 Application to paradigmatic relations
Binary features are directly related to certain logical relations and the corresponding
meaning relations (cf. Table 9.4). Two expressions with opposite values for some
feature [α], e.g. boy and mare, are logically incompatible, regardless of the rest of
their meanings. If the meanings of two expressions differ ONLY in the value of one
feature (boy vs girl), they are complementary opposites; the rest of their meaning
defines a semantic domain within which they are contradictory. Hyponymy holds
if two expressions have the same meaning except for an additional feature [+α] or
[–α] for one of them (e.g. child and boy). As will be argued later, hyponymy (of this
type) and complementary opposition are the only meaning relations that can be
captured with binary features.
Table 9.4
Binary features and logical relations
Meaning of A
Meaning of B
Logical relation
Meaning relation
X and [+a]
boy
vs.
Y and [–a]
mare
incompatibility
(undetermined)
X and [+a]
boy
vs.
X and [–a]
girl
incompatibility
complementary opposition
X and [+a]
X and [–a]
girl, boy
vs.
X
child
subordination
hyponymy
9.3.3 Application to combinatorial meaning properties
Features can be used to formulate selectional restrictions. For example, if a verb
(e.g. vaccinate) carries the selectional restriction ›human‹ for its subject referent, the
restriction can be described by the feature [+HUMAN]. It must, however, be carefully
observed what this is a feature of: it is a feature of a complement argument of the
verb, not a feature of the verb itself. If it were the latter, it would have to relate to the
referent of the verb, i.e. the event expressed (5.3.2). But the event, e.g. of vaccinating,
is not a human being; rather the agent is that performs the action. Conditions on
complements of the verb can, in fact, be captured by a decompositional analysis
of verb meanings, as we shall see in section 9.4.2, but not with the binary feature
approach.
Meaning components
231
9.3.4 Ideal properties of semantic features
So far only single features have been extracted from the meanings of lexemes. The
natural next step in feature analysis is the attempt of decomposing lexical meanings
COMPLETELY into binary features. The result for a particular lexeme would be a finite
list of features that captures all relevant properties of its potential referents and
differs from the feature list of any other non-synonymous lexeme in at least one
feature. This, however, has proved very difficult. It is by no means obvious how, for
example, the set of co-hyponyms of large taxonomies such as the animal or plant
taxonomy should be distinguished systematically by binary features.
Formally, a trivial feature analysis is always possible. We could, for example,
distinguish the animal terms bear, pig, rabbit, tiger, donkey, kangaroo, etc. by simply
introducing as many binary features [±BEAR], [±PIG], [±RABBIT], etc. and assigning to
each animal term a positive value for its ‘own’ feature and a negative value for each
other feature. Tiger would be [–BEAR], [–PIG], [–RABBIT], [+TIGER], [–DONKEY] and
[–…] for all the other features. The result is shown in Fig. 9.3.
Figure 9.3
A trivial feature matrix for co-hyponyms in the animal taxonomy
mammal
bear pig
[±BEAR]
[±PIG]
[±RABBIT]
[±TIGER]
[±DONKEY]
[±KANGAROO]
[±…]
rabbit
tiger donkey
kangaroo
…
+
–
–
–
–
–
–
–
+
–
–
–
–
–
–
–
+
–
–
–
–
–
–
–
+
–
–
–
–
–
–
–
+
–
–
–
–
–
–
–
+
–
–
–
–
–
–
–
…
This approach would capture in an extremely plain way one major aspect of
the structuralist notion of meaning: that the meaning of a unit is the sum of
its differences from the meanings of the other units of the system. But, clearly,
such an ‘analysis’ would not explain anything beyond the mere incompatibility
of the different animal terms (and this only if we tacitly assume that the features
themselves are mutually incompatible, i.e. if we already presuppose what should be
explained). The main point of decomposition is, of course, the REDUCTION of meaning
to more basic and ultimately minimal components. At least we would expect all the
expressions in Fig. 9.3 to carry a common feature [+ANIMAL]. The use of this feature
232
Understanding semantics
would then lead to the employment of features less specific than [TIGER], etc. This
leads us to a first important postulate for any feature approach, and in fact for any
decompositional theory:
∑ Features should be primitive, i.e. not further analysable.
The 1:1 analysis into as many features as there are expressions to be distinguished
suffers from a further principal drawback: the features are very special. For example,
the feature [±PIG] would only be of use when it takes the value +, i.e. with terms for
more specific kinds of pigs or for male, female or baby pigs. The feature value [–PIG],
though a necessary component of the meanings of all other animal terms contributes
almost no information (I do not know much about the meaning of rat if I know that
rats are [–PIG]). A feature system should provide distinctions that can be used more
widely, e.g. [±DOMESTIC]. Features must not be ‘ad hoc’, i.e. created just in order to
provide a solution to a particular problem, while disregarding more general aspects.
∑ Features should be general.
Of course, generality is a matter of degree. But what is certainly ruled out are features
that are distinctive for a handful of lexemes only.
9.3.5 Linguistic motivation
If we try to find more general features that allow us to structure the taxonomy of
animals, we will soon fall back upon our knowledge from biology lessons, trying to
apply the criteria of scientific classification such as the distinction between rodents
and other orders. But, as will be argued more thoroughly in 11.6, a distinction must
be drawn between semantic knowledge, i.e. the knowledge of meanings, and world
knowledge, i.e. knowledge about the entities we have words for. Semantic analysis has
to be based on LINGUISTIC data, namely the meaning relations observable within the
language system. After all, we are not doing the work of biologists, etc., but of linguists.
∑ Features should be linguistically motivated.
What would constitute relevant linguistic data that justify the assumption of
semantic features? So far we have only made use of instances of complementary
opposition or hyponymy. There are various other phenomena; examples will be taken
from the realm of animal terms. Of course, the phenomena are not restricted to this
part of the lexicon.
Selectional restrictions of verbs. There are specific verbs for the sounds certain
animals produce: e.g. miaow, bark, chirp, neigh or words for modes of running such
as gallop, trot, etc. The selectional restrictions of such verbs give rise to corresponding
features in the meanings of animal terms.
Verbs with implicit arguments. The German verb reiten (a cognate of the English
verb ride) is interpreted as ›to ride a horse‹ if the respective argument is not explicitly
specified otherwise. This justifies the assumption of a meaning component ›horse‹.
It is not clear, however, how in this approach a feature [+HORSE] can be built into the
Meaning components
233
meaning of the verb, since it is not the referent of the verb (i.e. the riding event) to
which the feature applies. Similarly there might be adjectives which are only used
when referring to properties or conditions of certain kinds of animals.
Lexicalizations of distinctions within certain kinds or groups. For some kinds of
animal there are general terms along with terms for the sexes (horse vs mare, stallion)
or their young (kitten, chicken, puppy, calf, lamb, cub, foal, etc.). There are specific
terms for horses of certain colours, e.g. sorrel for a horse of a light reddish-brown
colour. The more specific terms (mare, foal, sorrel) require features that they share
with the more general terms (horse). In other cases, the same word may be used to
form expressions for the males or females of different kinds of animal, e.g. hen and
cock for most kinds of bird. The terms for these animals too must share a feature.
Mereological relations. For many kinds of animal there are mostly analogous
meronymies (8.4.4) due to their similar anatomies. But often different terms are used
for corresponding body parts, for instance snout, muzzle, trunk or beak for mouth
and nose. Thus, animal terms can be grouped together according to the expressions
used for their body parts: terms for animals with a snout, animals with a beak, etc.
In such cases, a feature is linguistically motivated, e.g. [+BEAK] for animals with a
‘beak’. We also may distinguish terms for animals with special body parts, e.g. with
‘claws’, ‘hooves’, ‘fins’, ‘feathers’, ‘scales’, with ‘fur’ or a ‘shell’, or with a ‘tail’, a ‘mane’,
an ‘udder’ and so on. Although features like these border closely on extralinguistic
scientific characterizations, they nevertheless are linguistic features. We cannot
explain the meaning of fin, feather, udder, etc. without relating the meaning of the
words to the meanings of the words fish, bird, cow, etc. or hyperonyms of them.
There is a conceptual connection between, for example, ›bird‹ and ›feather‹: the fact
that birds have feathers is part of the meaning of bird as well as part of the meaning
of feather. A feature for the property of having feathers is therefore linguistically
motivated for the word bird as well as for all terms for kinds of birds with feathers.
Note that such a criterion is different from saying that, for example, seagulls can be
distinguished from other kinds of bird by the fact that their eggs are a certain colour:
there is no lexical relation between the word seagull and the respective colour word.
Metonymical relations. The semantic relations based on mereology are a special
case of metonymical relations. These apply more generally between terms for animals
or other kinds of objects and terms for things that belong to them. Examples are the
terms spawn for the eggs of fish or certain other water animals, nest for many animals,
or web for spiders. Between fish and spawn we have a meaning relation in which
referents of one term belong to referents of the other, i.e. a metonymical relation as
introduced in 3.4.2. The same kind of general relationship links the terms for certain
animals and for their meat when used as food: pig–pork, cow–beef or sheep–mutton.
Among the desirable properties of semantic features, or meaning components in
general, linguistic motivation is the most important. Unlike the other three properties
mentioned, it is absolutely necessary: only linguistic motivation guarantees that
the conditions are really meaning components, not just parts of our general
world knowledge. There may be a conflict between linguistic motivation and the
234
Understanding semantics
requirements that features be primitive and general. For example, the feature [HORSE]
is linguistically motivated due to the existence of words such as mare, stallion, colt or
neigh, but it is neither primitive nor general.
There is a linguistically motivated meaning relation between lexical items
whenever it is possible to form logically true sentences (7.1.4), called meaning
postulates, that express the relation.
(6)
a.
A mare is a female horse.
b.
A kitten is a baby cat.
c.
Pork is meat from pigs.
d.
Roe is the eggs of fish.
e.
Birds have feathers.
f.
Birds have a beak as their mouth.
9.3.6 Types of features
Although such linguistic data will take us a bit further along the way to an
exhaustive analysis, it is by no means clear whether it is possible to find a set
of appropriate features for distinguishing the great number of animal terms or
other large taxonomies. In fact, it is commonly accepted that for many lexemes
an exhaustive analysis into binary features that are at the same time primitive,
general and linguistically motivated is impossible. Features which exhibit all three
ideal properties are called classemes, or markers. It is markers which allow an
explanation of the combinatorial properties as well as the more general contrastive
properties. Markers may also play a role in grammar. For example the feature
[±HUMAN] is responsible for the choice of the pronouns who vs what, or [±MALE]
for the choice of he vs she. Ideally, markers also fulfil a further important condition,
related to the aim of language comparison:
∑ Features should be universal.
Indeed, features such as [±MALE], [±HUMAN], [±ANIMATE], etc. play an important role
in the structure of the lexica of very many languages.
For most lexemes, however, there will remain a residue of meaning that cannot
be captured by markers. For such cases, the notions of semes and distinguishers
were introduced. Semes serve for specific, more or less unique, distinctions within
particular lexical fields. For example, in the lexical field of the verbs buy, rent, take,
borrow, etc. one can use the semes [±PERMANENT] or [±FOR MONEY]. Semes are
primitive and linguistically motivated, but they are neither general nor, in most
cases, expected to be universal. Distinguishers are used for whole meaning residues.
If the meaning of mare were analysed just into the two features [–MALE] [+HORSE],
[–MALE] would be a marker and [+HORSE] a distinguisher. Distinguishers are neither
primitive nor general; they are improbable candidates for universality, but they may
be linguistically motivated.
Meaning components
235
9.3.7 Evaluation of the binary feature approach
While the analysis of lexical meanings into a set of binary features can be used for
explaining some semantic data, the limitations of the binary feature approach (BFA)
are in fact severe.
As stated above, binary features are essentially one-place predicates about the
referent of the lexeme. For example, the decomposition of the meaning of woman
into the three features [+HUMAN], [–MALE], [+ADULT] is equivalent to the following PL
definition (cf. 5.5):
(7)
x is a woman = human(x) Ÿ ¬male(x) Ÿ adult(x)
‘x is human and not male and adult’
The PL notation shows that any possible decomposition of meaning into binary
features has a uniform and very simple structure: a conjunction of a finite set of
one-place predicates, possibly negated, if the feature takes the value ‘–’ (minus). All
predicates have the same argument, namely the potential referent of the lexeme.
∑ In BFA there is only one kind of meaning component and only one way of
composing meanings out of them.
Consequently, the potential meaning components can only combine to form one
type of meaning: one-place predicates, since the conjunction of one-place predicates
always yields a one-place predicate.
Therefore, most verbs cannot be analysed along the approach of BFA, nor does it
apply to relational nouns or multi-place adjectives. Some aspects of the meanings
of multi-place verbs can be captured; for example, the difference between ›buy‹ and
›rent‹ by means of a feature [±PERMANENT]. But the core of the meaning, common
to both verbs, is a three-place concept about a change of possession which involves
the former and the future possessor and some object of possession. This three-place
predicate can never be composed out of one-place predicates.
The restriction to one-place predicate expressions is a severe limitation of BFA. In
addition, the approach is limited even in the case of one-place predicate expressions,
due to the principal difficulties of total decomposition discussed above.
∑ BFA is only applicable to a limited range of lexemes.
A further restriction results from the fact that the features of BFA can only relate to
the potential referents of a lexeme.
∑ In the case of multi-place predicate terms, the BFA can only capture conditions
concerning the potential referents of the lexeme.
As we have seen, certain phenomena, such as selectional restrictions, require the
possibility of imposing conditions on arguments of predicate terms. Similarly, we
might for example want to distinguish terms for items of clothing with respect to
the sex of the wearer, or terms for body parts of either female or male persons. But it
would be inappropriate to assign a feature [–MALE] or [+MALE] to the words bikini or
236
Understanding semantics
prostate: it is not the bikini or the prostate that is female or male, but the persons for
which bikinis are designed or which have this type of organ.
As is shown in Table 9.4, among the basic meaning relations only semantic
complementarity and certain cases of hyponymy can be explained in terms of
binary features. Conversion cannot be captured in principle, because this relation
involves two-or-more-place predicate terms. The BFA also has difficulties with other
meaning relations, e.g. antonymy. For example, in order to distinguish big and small,
one would have to use two features such as [± HIGH] and [± LOW] relating to the
7
intervals of the scale of size they denote (cf. Fig. 9.3). This is not particularly elegant.
More specific meaning relations that underlie certain lexical fields can provide
huge if not insurmountable problems for BFA, among them mereological relations,
the cyclic order of terms for the days of the weak or the months of the year, or the
various relationships between head and modifier of a regular compound (5.2.2). It is
impossible to capture the semantic relations between, say, three, four, twenty, hundred
8
and three hundred and twenty-four with binary features for these five numerals.
Thus, the last severe shortcoming of BFA is as follows:
∑ Most kinds of meaning relation cannot be accounted for with BFA.
To sum up, BFA is much too simple an approach to be able to meet the complexities
of semantic phenomena in natural languages. The use of binary features may be
appropriate for dealing with the small and limited areas of phonology or syntax,
where only a comparatively small set of units or categories has to be distinguished
and analysed with respect to their relations within the system. But the lexicon of a
language is of a complexity incommensurable to that of the sound system or system
of word classes or forms. It forms a large part of our utterly complex cognitive
system. It is not plausible to assume that the decomposition of word meanings is
restricted to structures as simple as those provided by BFA. It should, however, be
emphasized that the failure of BFA in semantics does not mean that the structuralist
approach as such has failed. Its strong emphasis on linguistic data and the systematic
relations within languages helped linguists to develop their theoretical methods of
today. The basic notions and methods of structuralism are compatible with semantic
theories quite different from BFA. In principle, it would be possible to develop
theories of lexical meaning within the framework of structuralism which are much
more adequate and powerful than BFA.
In the following, we will have a look at three approaches to decomposition that
provide alternatives to the structuralist tradition.
7 Big is assigned the feature [+HIGH], and small [+LOW], while not big would express [–HIGH]. In
addition, one would have to stipulate that [+HIGH] entails [–LOW] and [+LOW] [–HIGH]. However,
entailments should be derivable from the decomposition rather than being stipulated somewhere
else. The logical relationship between [HIGH] and [LOW] proves that these two ‘meaning components’
have common content; thus a decomposition in terms of [HIGH] and [LOW] is not of much
explanatory value.
8 These relations consist, among others, in the fact that three hundred and twenty-four denotes the
sum of the referents of three hundred and of twenty-four, three hundred denotes the product of the
referents of three and of hundred, and so on.
Meaning components
237
9.4 SEMANTIC FORMULAE
One phenomenon that gave rise to the development of alternative decompositional
approaches is the existence of systematic meaning relations between verbs, or
between verbs and adjectives, as given in (8). We encountered the phenomenon
above, when dealing with verb alternations in 6.1.
(8)
stative
x is open
x is hard
x is dead
x has y
inchoative
x opens
x hardens
x dies
x gets y
causative
y opens x
y hardens x
y kills x
z gives y to x
The verbs and adjectives in the first column denote a certain state of x. The inchoative
verbs in the second column express the coming about of this state of x. The third
column consists of causative verbs, used to express that someone (or something),
y (or z), brings about the state of x. While the stative and inchoative expressions
have the same number of arguments, the causative verbs have an additional causer
argument. There are hundreds of such triplets in English, and in addition many pairs
of verbs, or pairs of an adjective and a verb, which would fill two of the three places
in a row with the third term lacking in the lexicon, for example, wet/–/wet, –/lie down/
lay down. In some cases the expressions that enter this scheme have the same form
(but note that they represent three different lexemes as each belongs to a different
grammatical category), as in the case of open/open/open and harden/harden. In
others, the terms are related in form: hard/harden, and historically also dead/die. But
the forms of the terms may as well be morphologically unrelated (small/–/reduce).
There is a systematic meaning relation between the three members of each triplet. It
can be captured by logical entailments, such as:
(9)
a. y hardens x
fi x hardens and
x hardens
fi x will be hard
b. z gives y to x
fi x gets y
x gets y
fi x will have y
and
Clearly, a BFA approach is not able to explain these phenomena. One could, for
example, assign a common feature [+OPEN] to the three terms in the first group,
and distinguish the three elements by two additional features, say [–CHANGE] for the
adjective and [+CHANGE] for the two verbs and [+CAUSE] for distinguishing the third
item from the other two. This would help to keep the three elements of a group apart
and mark the terms of different groups in the same row as terms with common
semantic characteristics. But it would not allow the explanation of the exact meaning
relations.
9.4.1 Dowty’s decompositional semantics
A classical, though not the earliest, decompositional approach to this phenomenon
is Dowty (1979), a theory developed in the framework of Montague Grammar, which
238
Understanding semantics
will be introduced in chapter 13. Dowty uses decomposition in order to model
aspectual classes (6.2). His method is to give a general formula for each aspectual
class. The general formulae have a fixed structure in which individual elements are
inserted for individual verbs. The formulae are defined in predicate logic (5.5).
Dowty analyses the adjective openA as a stative expression not further decomposed
into meaning components. Using the notation of predicate logic, he represents the
meaning of x is open by a simple formula in which the state term open is used as a
one-place predicate constant:
x is openA:
(10)
open(x)
The meaning of the intransitive verb openIV is then represented by means of an
operator BECOME which can be applied to any stative formula:
x opensIV:
(11)
BECOME(open(x))
The component BECOME is used to express a change from a situation where the stative
formula is false to a situation where it is true. If ‘BECOME(open(x))’ is true at a certain
time t, then ‘open(x)’ is false before t and true afterwards. It thus follows from the
meaning representations of openA and openIV that x opensIV entails x will be openA.
For the causative verb openTV , Dowty employs an operator CAUSE which takes
two argument terms, an individual term for the causer, and a formula for the event
caused. The meaning of y opensTV x is represented as in (12) and interpreted as: ›y
causes x to become open‹. Hence, y opensTV x logically entails x opensIV.
y opensTV x:
(12)
9
CAUSE (y, BECOME(open(x)))
Crucially, the meaning of openTV does not simply contain the three meaning
components CAUSE, BECOME and open in the way the meaning of woman is thought
to contain the three meaning components [+HUMAN], [–MALE] and [+ADULT] in BFA
(three one-place predicates linked by conjunction). Rather, the three components
combine with each other in a specific way. The component CAUSE is a two-place
predicate; its first argument is the agent argument of the verb; its second argument
is BECOME(open(x)), a complex containing the two meaning components BECOME and
open. BECOME is used as a one-place predicate with a (stative) situation argument.
This argument, in turn, consists of the one-place predicate open applied to the theme
argument of the original verb.
An analogous analysis can be given for the other groups in (13). For example,
(13) a.
x has y:
have(x,y)
b.
x gets y:
BECOME(have(x,y))
c.
z gives y to x:
CAUSE (z, BECOME(have(x,y))
9 The formulae used here are slightly simplified versions of the original ones given in Dowty (1979).
Meaning components
239
The explanatory potential of the approach is obvious. There are hundreds if not
thousands of stative, inchoative and causative verbs and adjectives whose meanings
can be decomposed in this way. The decomposition completely uncovers the meaning
relations that hold within and across the groups. This particular analysis is less
concerned with the question whether meaning components such as open or have are
primitives. It might well be the case that some of them can be further decomposed.
But this would not reduce the explanatory power of this kind of approach.
9.4.2 Jackendoff’s Conceptual Semantics
10
Jackendoff ’s Conceptual Semantics is another theory that deals with semantic
phenomena concerning verbs (but not exclusively). The foundations of his theory
are quite different from Dowty’s. Unlike Dowty, Jackendoff assumes that meanings
are concepts represented in our mind. To him, basic semantic components are at
the same time conceptual components out of which the human mind is capable of
composing all kinds of concepts it operates with. The way in which meanings are
decomposed in Jackendoff ’s approach is however similar to Dowty’s. We cannot go
deeper into the foundations of his theory nor enter into the details of his semantic
representations, but will confine the discussion to a few examples that give an idea of
one more way in which meanings can be decomposed.
With tense neglected, the sentence John went home would be analysed as (14):
(14) John went home:
[Event GO ([Thing JOHN], [Path TO ([Place HOME])])]
This looks more complicated than it actually is. Let us first omit the square brackets
together with their subscripts. We then obtain a structure of the same kind as those
used by Dowty:
(15) John went home:
GO(JOHN, TO(HOME))
is a primitive concept, and meaning component, for a movement of an object
along a path. Correspondingly, the concept GO has an object argument and a path
argument. In (15) the primitive concept JOHN fills the first argument position. The
second position is occupied by TO(HOME), a specification of the path in terms of its
goal. The one-place operator TO is another primitive concept, which takes a (concept
for a) place as its argument, in this case HOME, and yields a concept for a path to this
place. If we now return to the more complex notation in (14), we see that each pair
of square brackets [ ] encloses one concept; the subscript at the left bracket assigns
the concept to one of a few basic categories: Event, Thing (including persons), Path
and Place, others being Action, State, Time, Amount or Property. From (14) we can
abstract a semantic representation of the verb go by omitting the arguments of the
GO concept, replacing them by empty slots.
GO
10 The following examples are taken from Jackendoff (1990).
240
Understanding semantics
go: [Event GO ([Thing __ ], [Path __ ])]
(16)
It could be argued that (16) does not really constitute an instance of decomposition
since the meaning of go is essentially represented by a conceptual primitive GO. This
is not much better than representing the meaning of pig by a single binary feature
[+PIG]. Jackendoff, however, argues that the concept GO happens to be a conceptual
primitive. This makes the verb go a semantic primitive. The conceptual primitive GO
is a component of concepts for all kinds of transitions. (Jackendoff ’s is a so-called
localistic theory. Localism assumes that the linguistic concepts of space and motion
are basic and serve as patterns for non-local concepts.) For instance, John gave Bill
$5 is represented as in (17a), brackets and subscripts omitted. The meaning of give
itself would be (17b), with three empty argument places to be filled with concepts for
the giver, the given and the recipient:
(17) a.
b.
John gave Bill $5:
CAUSE(JOHN, GO($5, TO(BILL)))
give:
CAUSE( __ , GO( __ , TO( __ )))
Embedded within the CAUSE concept, which has the same structure as the predicate
cause in Dowty’s analyses, we find the concept GO used to represent a transition of
some object into the possession of Bill: the concept of motion in space is applied
to a transfer of possession. The example may convey a first impression of the
kind of generalizations a decompositional approach like this allows. In fact, it
appears possible to reduce the meanings of thousands of verbs in English or any
other language to a very limited number of general patterns composed out of a
comparatively small set of basic conceptual components.
Again, the main asset of this approach lies in the possibility of forming complex
meaning representations by way of embedding components into each other. Let me
give you two more examples. The first one illustrates how selectional restrictions can
be accounted for. The verb drink requires the object to be liquid. Jackendoff gives the
analysis in (18a), simplified in (18b):
(18) a.
drink:
[Event CAUSE ([Thing __ ]i , [Event GO ([Thing LIQUID]j ,
[Path TO ([Place IN ([Thing MOUTH OF ([Thing __ ]i)] )] )] )] )]
b.
drink:
CAUSE ( __ i , GO (LIQUIDj , TO (IN-MOUTH-OF __ i )))
The structure of the meaning of drink is very similar to the structure of give in
11
(17b). The place for the ‘moving’ thing, i.e. the object that is drunk, is not just left
open and categorized as ‘thing’ but specified by the concept LIQUID. The subscript ‘j’
11 The subscripts i and j are so-called referential indices. Identical indices indicate identical referents.
Thus, the indexing in (18a) indicates that the drinking agent causes the object to ‘go’ into their
own mouth. According to this analysis, the verb drink cannot be used for the action of pouring
something into someone other’s mouth, which is correct.
Meaning components
241
indicates that the argument is not implicit but has to be specified in the sentence.
The specification of the object term is then ‘fused’ with the specification LIQUID (recall
5.7.2 for the process of fusion).
The other example is Jackendoff ’s analysis of the verb butter. The verb is obviously
semantically related to the noun butter: it means ›spread (something) with butter‹.
In this concept, ›butter‹ is an implicit argument of an underlying predicate (›spread
with‹). Jackendoff gives the following analysis for butter in a sentence like Harry
buttered the bread, simplified in (19b):
(19) a.
butter:
[Event CAUSE ([Thing __ ]i ,
[Event GO ([Thing BUTTER], [Path TO ([Place ON ([Thing __ ]j )] )] )] )]
b.
butter:
CAUSE ( __ i, GO ( BUTTER , TO (ON ( __ j ))))
The sentence Harry buttered the bread can then be represented as in (20):
(20)
butter:
CAUSE (HARRY, GO ( BUTTER , TO (ON (BREAD))))
There are very many noun-verb pairs of this type in English. Of course, not all such
pairs can be analysed in exactly the same way as butter/butter: think of hammer/
hammer or host/host; obviously there are various patterns of decomposition and
relations between the noun and the verb.
This may suffice as a sketch of the basic mechanisms and the explanatory power
of what I would like to call ‘formula’ approaches to lexical decomposition. They
represent lexical meanings by means of predicate logic formulae which are able to
combine various types of meaning components in different ways. Both Dowty’s
and Jackendoff ’s approaches are parts of elaborated theories dealing with much
more ambitious questions than just lexical decomposition. In both theories,
decomposition serves as a means for explaining the combinatorial properties of the
analysed expressions: how they can be combined with other elements in a sentence
and how their meanings interact. For example, Jackendoff ’s approach also allows
a derivation of the thematic roles of verbs from their meaning descriptions and a
theory of linking.
Although their respective theoretical foundations and objectives are different,
they use similar decompositional structures. Both work with abstract semantic
primitives that have one or more arguments and thereby allow the embedding of
other components. It is therefore possible to build up complex structures capable
of capturing the meanings or multi-place predicate expressions. Unlike in feature
combinations in BFA, these structures can express conditions on arguments other
than the referential argument.
An apparent restriction of the two approaches is the fact that they are only applied
to verb meanings. There are, however, theories in a similar vein that deal with other
lexical categories (cf. Bierwisch and Lang 1989 for adjectives, and Pustejovsky 1995
for his unified ‘qualia’ approach to nouns and verbs). What all formula approaches
242
Understanding semantics
have in common is a formal apparatus (technically, a ‘formal language’) with a
vocabulary of primitives and a precise syntax and semantics of its own that allows
the formation of formulae as meaning representations.
9.5 SEMANTIC PRIMES: WIERZBICKA’S NATURAL SEMANTIC
METALANGUAGE
The last approach to decomposition to be discussed was developed by Wierzbicka.
Her aim is the development of a system that allows the description of all meanings in
all languages by means of a strictly limited set of semantic primitives (or ‘semantic
primes’, like ‘prime numbers’, in her terminology). She calls this system ‘Natural
Semantic Metalanguage’, or NSM for short. NSM consists of a set of semantic
primitives and a syntax for their combination. Let me give you her definition of the
meaning of envy as a first example (Wierzbicka 1996, p. 161):
(21) X feels envy =
sometimes a person thinks something like this:
something good happened to this other person
it did not happen to me
I want things like this to happen to me
because of this, this person feels something bad
X feels something like this
The defining part contains the semantic primitives SOME, PERSON (= SOMEONE), THINK,
(SOME)THING, LIKE, THIS, GOOD, HAPPEN, OTHER, NOT, I, WANT, BECAUSE, FEEL, BAD. They are
combined into simple sentences following the syntactic rules of NSM.
For Wierzbicka and her followers the careful choice of semantic primitives is of
central concern. An item qualifies as a semantic primitive if
∑ it is undefinable, i.e. its meaning cannot be defined in terms of other expressions
∑ it is universal, i.e. it is lexicalized in all natural languages
Both requirements can only be met approximately. For example, many concepts
appear to be interdefinable, for example ›eye‹ and ›see‹ (›eye‹ = ›part of the body
used for seeing‹ and ›see‹ = ›perceive with the eyes‹). Which one is primitive can only
be decided in practice. Universality is, of course, not verifiable. But the researchers
adopting the approach did investigate a great number of languages, from Europe,
Africa, Asia and Australia. So the primitives they sorted out on this basis are good
candidates for universal primitives. Wierzbicka (1996, pp. 35f., 73f.) lists fifty-five
primitives, here rearranged; the list has been constantly growing:
(22)
I, YOU, SOMEONE (PERSON), SOMETHING (THING), PEOPLE; WORD
THIS, THE SAME, OTHER, PART (OF), KIND (OF)
ONE, TWO, MANY (MUCH), MORE, VERY, ALL, SOME (OF)
Meaning components
243
THINK, KNOW, WANT, FEEL, SEE, HEAR, SAY
GOOD, BAD; BIG, SMALL
DO, HAPPEN; MOVE, THERE IS, (BE) ALIVE
WHEN, BEFORE, AFTER; A LONG TIME, A SHORT TIME; NOW
WHERE, UNDER, ABOVE; FAR, NEAR; SIDE; INSIDE; HERE
NOT, CAN; IF, BECAUSE, LIKE, IF … WOULD, MAYBE
To the extent that the primitives are universal, every language is supposed to
possess a corresponding set of lexemes, for example the words I, you, someone, etc.
in English. If these lexemes are polysemous, they have to be taken in the relevant
reading. For example, the English verb want, when considered the lexicalization of
the NSM primitive WANT, must be taken as meaning ›wish‹, not ›lack‹. In addition,
every language is supposed to possess syntactic rules for combining the lexical
counterparts of the semantic primitives to the same effect as the primitives are
combined in NSM. Thus the NSM meaning definitions can be translated into
particular languages.
A few more examples may serve to show the kind of decompositional analysis
possible in this framework. ABOVE and FAR are primitives that allow, among others,
the definition of the meaning of sky:
(23) sky (Wierzbicka 1996, p. 220)
something very big
people can see it
people can think like this about this something
it is a place
it is above all other places
it is far from people
Once this concept is available, one can proceed to define the meaning of sun and then
of blue:
(24) sun (Wierzbicka 1996, p. 220)
something
people can often see that something in the sky
when this something is in the sky people can see other things because of this
when this something is in the sky people often feel something because of this
(25) X is blue = (Wierzbicka 1996, p. 309)
at some times people can see the sun above them in the sky
when one sees things like X one can think of the sky at these times
Let me finally illustrate how the NSM approach works for the kind of examples
treated by Dowty and Jackendoff. Goddard gives the following NSM definition of
causative break, a verb that is of the same type as openTV, discussed above.
244
Understanding semantics
(26) xPerson break(s) y (e.g. Howard broke the window) =
a.
x does something to y
b.
because of this, something happens to y at this time
c.
because of this, after this y is not one thing any more
12
Along the lines of Dowty and Jackendoff the verb would be analysed as in (27a) and
(27b), respectively:
(27) a.
x breaks y:
CAUSE(x, BECOME(broken(y)))
b.
x breaks y:
CAUSE(x, GO(y, TO(BROKEN)))
The CAUSE component relates to an action of x that causes the change of y into the
state of being broken. (The predicate broken/BROKEN in the two formulas is chosen
just for convenience; for a non-circular definition it would have to be replaced
appropriately.) This action, and the fact that it affects y, is expressed by the first two
clauses in (26). The change of state corresponding to BECOME(broken(y)) in (27a)
and GO(y, TO(BROKEN)) in (27b) is expressed in a different but equivalent way in the
third clause of the NSM definition. Thus the analyses are very similar, although they
look different.
Unlike the other approaches to decomposition, Wierzbicka’s is not restricted to
the analysis of certain word classes. It even offers meaning descriptions of emotion
words like envy or colour terms or interjections (like yuk!), which other theories
are unable to decompose. It is able to capture all sorts of meaning relations such
as those that make up taxonomies (captured by means of the primitive KIND OF),
or meronymies (captured by PART OF). More than any other approach, it allows the
definition and comparison of word meanings in different languages. But the NSM
method has its shortcomings too. First, it must be stated that NSM descriptions
suffer from a certain lack of precision, as the primitives of NSM are very general and
the definitions therefore rather vague. For example, the definition for envy given in
(21) would also apply to self-pity or feeling neglected, etc. and the definition given
for break also covers cut, split, shatter and similar verbs of destruction. For many
NSM definitions, it is questionable whether they are sufficiently specific to capture
the meaning of the terms defined. While this deficiency might be overcome by more
elaborate definitions, a further drawback cannot be remedied: the approach is unable
to explain meaning relations that hold between the primitives such as the antonymy
of big and small, good and bad, far and near, or the directional opposition between
after and before, or under and above. Their meaning relations are ‘non-compositional’
in this theory, i.e. they cannot be explained on the basis of meaning definitions in
NSM, because these expressions have the status of undefinability.
12 The example is taken from Goddard (1998, p. 281). The original definition is in the past tense (‘X
broke Y …’). The subscript PERSON on the variable X is used to distinguish this variant of the verb
from the one in which the subject is an event, as in the storm broke the window.
Meaning components
245
9.6 SUMMARY AND EVALUATION OF THE APPROACHES TO
DECOMPOSITION
In this chapter we have taken a look at different theoretical approaches to meaning,
in particular lexical meaning. First, essentials of the structuralist approach were
presented. Although no longer a leading paradigm in present-day linguistics, it still
is very important. The method of analysing language as an autonomous system,
closed in itself, helped to establish linguistics as a science of its own. It also provided
students of language and languages with the basic technique of determining the
contrastive and combinatorial properties of units at different levels (from phonemes
to sentences) in a systematic way. These methods are still valid.
Four approaches to decomposition were then introduced: BFA, which was developed
in the tradition of structuralism, the formula approaches of Dowty and Jackendoff,
and Wierzbicka’s NSM.
BFA, which is still enjoying some popularity, is the one with the most severe
limitations. Being confined to a very simple pattern of decomposition (i.e.
conjunctions of one-place predicates), it is unable to account for all other types of
meaning structures and, consequently, unable to capture meaning relations other
than simple incompatibility, complementarity and hyponymy. Due to its insufficiency
for semantic analysis in general it is of equally limited use for language comparison.
The two formula approaches perform equally well, due to their rich structures of
meaning representations. They do not lay too much emphasis on the basicness of their
primitives. Nevertheless, central components of their theories, such as CAUSE, BECOME
and GO, are plausible candidates for universal meaning components. Their strength
lies in the explanation of the compositional properties of verbs. Wierzbicka’s is the
only approach that makes reduction to basic meanings a central issue and focuses on
language comparison. It is, however, less clear how the somewhat amorphous NSM
definitions can be used for rigidly deriving meaning relations between lexical items,
including the basic logical relations, and it remains open to what extent it is able to
account for the compositional properties of lexical items. These drawbacks are all
due to the lack of precision of the meaning descriptions.
There are, of course, a number of other approaches to decomposition in addition to
those discussed here. One approach which has gained some influence is Pustejovsky’s
‘Generative Lexicon’. I consider the general meaning format of this theory too
complex for a discussion in this book. In a later chapter, we will discuss a proposal
which was developed in cognitive psychology – ‘frames’ in chapter 12. The general
format of frames is at the same time simple, general and very powerful. I will not
introduce it yet, as it requires a certain background in cognition.
246
Understanding semantics
EXERCISES
1. In which way is a linguistic unit integrated into, and defined by, the language
system? What kinds of relations and properties determine the character of a
linguistic unit?
2. What is the meaning of a lexeme according to the structuralist doctrine?
3. What is meant by the terms syntagmatic relations and paradigmatic relations?
What, in particular, are syntagmatic and paradigmatic meaning relations? What
do selectional restrictions have to do with these relations?
4. What is the difference between a lexeme and a morpheme?
5. Give a feature analysis of the nine members of the system of German personal
pronouns (Table 4.1). Describe the properties of the referent which the features
you use are to capture. If a feature is not relevant for a certain case, leave the feature
open; the value ‘–’ is reserved for those positively negative cases where ‘+’ is to be
excluded. Reread the section on the meaning of personal pronouns in 4.1.1.
6. How many and which morphemes make up the following words? For each
morpheme (i) determine its category (N, V, A) if it is basic, or its input and
output category if it is an affix and (ii) find two more words that contain the same
morpheme in the same function. Indicate the structure of the word by bracketing.
a. localization
b. unsuccessful
c. regrettable
7. Which types of semantic features are distinguished in BFA? How do they differ?
Try to give examples of expressions for which the different kinds of features can
be considered meaning components.
8. Find ten more pairs or triplets of stative, inchoative and causative verbs (or
adjectives, for the stative part). Check the entailments illustrated in (9).
9. Assign the three types of verbs in Dowty’s examples to the aspectual classes
mentioned in 6.2.
10. Discuss the advantages of formula approaches over BFA.
11. Discuss the differences between NSM and other approaches to decomposition.
FURTHER READING
Saussure himself on structuralism (= Harris 1983), also Lyons (1977, ch. 12);
more comprehensively Matthews (2001). Foley (1997, ch. 4) on structuralism and
its relation to linguistic anthropology. Dillon (1977) for an application of feature
semantics. Lyons (1977, ch. 9.9) for a criticism of feature semantics. Dowty (1979)
requires a background in Montague Grammar, Jackendoff (1990) is somewhat
easier to read, but cannot be fully understood without knowledge of Chomsky-type
Meaning components
247
language theory and grammar. For a short introduction, see Riemer (2010: 8.1); more
elaborately Jackendoff (2011). Goddard (1998) gives a survey of decompositional
approaches and a very readable introduction into NSM analysis. Wierzbicka (1996)
offers the essentials of NSM theory and its relation to semantic and cultural studies.
For recent surveys on decomposition see Engelberg (2011a, 2011b). Cann (2011)
offers a survey of structuralist semantics.
10
?
Meaning and language comparison
If structuralism is right in assuming that every individual language is a system of
its own, we should expect that languages can be very different, that they have terms
for different things or different ways of naming the same things and that similar
situations are expressed in different ways and from different perspectives. But
how big are the differences between languages? Are all languages essentially alike,
allowing the expression of the same thoughts, the communication of the same things,
only differing in their respective ways of formulation? Or are languages different to
an extent that it may be altogether impossible to express in one language what can
be expressed in another? Are the semantic systems completely arbitrary, or are they
constrained by universal principles?
We will start with simple examples that illustrate problems of translation. A closer
look at different ways of expressing that someone has a headache will take us a bit
deeper into the matter of semantic comparison. The second part will be devoted
to studies of colour term systems in various languages. These studies played an
important role not only in the understanding of the relationship between languages
but also in the development of semantics in general.
10.1 TRANSLATION PROBLEMS
Everybody who tries seriously to learn a foreign language will sooner or later realize
how different languages are, not only with respect to grammar and pronunciation
but also with respect to their vocabulary and how it is organized. When we naively
begin to learn a new language, we will probably start with the unconscious working
hypothesis that for each word in our native tongue there is a corresponding word
in the target language. But the more we get into the new language, the more this
hypothesis will crumble. People who have managed to master a foreign language to
a degree close to native-speaker competence would probably say that the semantic
equivalence of two lexemes is the exception rather than the rule.
There are different types of mismatch. One language may have two or more words
where another language has only one.
(1)
English
Japanese
mouse rat
nezumi
finger toe
yubi
water
mizu
yu
›cold w.‹
›warm w.‹
Meaning and language comparison
249
But the relations can be more complicated. Consider the following lexical field in
English and Japanese, and German for comparison:
(2)
German
English
Japanese
Wald
wood1
mori
Holz
wood2
ki1
Baum
tree
ki2
English has a count noun wood1 and a mass noun wood2. A wood1 consists of trees,
while wood2 is the substance trees largely consist of. Hence, the two meanings can
be clearly distinguished on linguistic grounds. Likewise, the Japanese meanings of
ki1 and ki2 are clearly different. It follows that neither of the two English expressions
matches with either of the two Japanese words. They only match in one meaning
variant, respectively. In general, the ubiquitous polysemy of the vast majority of
words suggests that (with the exception of number terms) there are almost no two
expressions from different languages that have the same overall meaning (cf. 8.1 for
the parallel problem of total vs partial synonymy within one language).
In (2), the meanings of the English, German and Japanese terms at least (roughly)
match in some of their possible readings. But in many cases, correspondences turn
out to be more complicated. Even apparently basic and universal concepts such as
›eat‹ and ›drink‹ are differently lexicalized. A first glance at dictionaries will give us
the following correspondences:
(3)
German
English
Japanese
essen
eat
taberu
trinken
drink
nomu
But as was mentioned in 9.2.2, the German terms essen and trinken are reserved
for people, unlike their English and Japanese counterparts. Even if we restrict the
verbs to human agents, they do not match perfectly. In English and German, a
(thin) soup can be either ‘eaten’ (essen) or ‘drunk’ (trinken), depending on whether
a spoon is used or the soup is directly taken into the mouth. Japanese use the verb
nomu regardless of whether the soup is drunk (which is the traditional technique)
or consumed with a spoon. One might now think that the crucial criterion for using
nomu is that the object be liquid. But this is not the case. The verb is also used for oral
medicine, including pills. The crucial point appears to be that the object is directly
swallowed without being chewed. This is a different criterion from the one that
regulates the use of English drink or German trinken. We cannot ‘drink’ a pill and we
can eat a soup even if we do not chew it.
Still, these are cases where for every context appropriate translations are possible.
But often a certain term in one language does not correspond to any term at all in the
other language. This may be due to the fact that the things the term refers to simply
do not exist where the other language is spoken, e.g. plants or animals or meals or
artefacts or social institutions. But it also happens in semantic areas that are shared.
Consider the area of working. English has a verb to work which, like German arbeiten,
covers a broad range of activities, paid work as well as unpaid work (e.g. in one’s own
250
Understanding semantics
kitchen or garden). The work may be physical or intellectual work such as reading
this book, learning German or studying linguistics. Japanese has no term that covers
all these activities. The closest equivalent to the verb work is hataraku. But it does
not cover intellectual work. On the other hand, Japanese has a verb asobu that can
be used for any kind of activity that is not work: playing, any kind of entertainment
or leisure including doing nothing at all. No simple verb with this meaning exists in
English or German.
An insurmountable problem for adequate translations can be posed by differences
in social meaning. Imagine an American movie of the type ‘boy meets girl’. John, a
yuppie real estate agent, meets Mary, a tough customer, and falls in love with her. If
the dialogues are translated into German, the translators have to decide whether John
and Mary address each other with Sie or du (2.3). They will have to start with Sie,
because that would be adequate, to a German audience, for the business relationship
between them, and will end up addressing each other with du once they have become
intimate. The problem for the translator will be: when do they have to switch from
Sie to du? As German personal pronouns of address, unlike English, carry social
meaning, speakers of German are forced to express distinctions which English does
not provide.
With Japanese, things are much more complex. First, there are several expressions
available to render English I and you, respectively. Each one has a different social
meaning, i.e. indicates different social relations. Second, in very many cases the use
of pronoun-like expressions for self-reference and address is altogether inadequate
in Japanese. For example, in a dialogue between mother and son, the son would
normally address his mother with ‘Mrs Mother’ (okāsan), not with any variant of
‘you’. Referring to himself he would either use a variant of ‘I’ or his own name (!),
but not ‘Mr Son’. His mother would refer to herself preferably as ‘Mrs Mother’ and
address her son with a variant of ‘you’ or his name. For example, little Taro could ask
his mother something of the form (4a) and his mother could answer (4b).
(4)
a.
‘Will Mrs Mother buy Taro an ice-cream?’
b.
‘Mrs Mother will buy Taro an ice-cream.’
Compare these forms to the adequate ways of expression in Western languages with
the characteristic switch between the pronouns I and you for the same arguments of
the predicate.
‘Will you
b.
‘
I
buy me
an ice-cream?’
⇔
a.
⇔
(5)
will buy you an ice-cream.’
For Western ears, the sentences in (4) are formulated as though they were said about
third persons. In fact, (4a) can as well be used, for example, by Taro’s elder sister for
asking a question of her father ABOUT Taro and her mother, and her father could use
(4b) as an answer.
Meaning and language comparison
251
Similar rules govern the forms of address and self-reference outside the family.
Students would normally address their teachers as ‘Teacher’ (sensei) or ‘name +
Teacher’ (e.g. Yamamoto Sensei), but not with ‘you’. A further crucial difference
between English and Japanese is the fact that in Japanese pronouns are normally
completely omitted. For example, in informal speech the question ‘Will you buy me
an ice-cream’ would be formulated as ‘Ice-cream buy-give?’ Omitting pronouns in
subject position is not allowed by the grammar of standard English. Due to all these
fundamental differences between Japanese and English, a direct, literal translation
of Japanese dialogues into English would be inadequate, or even ungrammatical. If,
however, the translation were in accordance with the natural means of expression
in English, it would not convey the ways in which social interaction is handled in
Japanese by choosing terms of address and self-reference.
10.2 HEADACHE, INTERNATIONAL
10.2.1 Grammatical constructions
Deeper in the language system we find differences in grammar. Examples are
abundant once we compare historically unrelated languages such as English and
Japanese. In terms of structural differences, Japanese is not as ‘exotic’, in comparison
with English, as many other languages, for example the native languages of the
Americas, Australia or Papua New Guinea. But with respect to some semantic
phenomena Japanese grammar differs considerably from English or other Western
languages.
For an illustration, we will compare the ways in which having a headache is
expressed in some European languages and in Japanese. A headache situation
involves three ingredients: (i) an experiencer E who feels the headache; (ii) the
sensation S, a pain that E feels; (iii) the body part B that aches: the head. Languages
express the situations by a variety of grammatical patterns in which these three
ingredients figure in different ways:
(6)
a.
English:
1
I have a headache
The English construction ties S and B, the pain and where it is felt, together into
one concept ›headache‹. What is expressed by have (6a) is that the headache is in
some way associated with E. Note that the relation would be much more specific if
one were to say I feel a headache instead. The English construction is paralleled in
German, with the slight difference that the sensation is in the plural, but there is also
the synonymous ich habe Kopfweh (singular, mass). The standard French phrase too
is similar.
1 This is the most common construction. Alternatively, headaches can also be expressed by
the variants my head is aching or my head is hurting me, which are similar to the Hungarian
construction (6f) and German 2 (6d), respectively.
252
(6)
Understanding semantics
b.
German 1:
ich
hab-e
1S.NOM
have-1S.PS
lit. ‘I have headaches’
Kopfschmerz-en
headache-PL
c.
French:
j’
ai
have.1S.PS
1S.NOM
lit. ‘I’ve pain at the head’
mal à
pain at
la
DEFART
tête
head
French has a possessive construction with three argument terms, E in the subject
position, S in the object position and B in a locative PP.
There is an alternative German construction that is similar to the standard Russian
way of expressing the situation.
(6)
d.
German 2:
mir
tu-t
der
1S.DAT ache-3S.PS DEFART
2
lit. ‘[to] me aches the head’
Kopf
head.S.NOM
e.
Russian:
u
menya
at
1S.GEN
lit. ‘at me aches head’
golova
head.SG.NOM
bolit
ache.3S.PS
weh
ache
The main predicate, i.e. the verb, directly expresses the sensation itself, not some
abstract association of E with S (the German verb wehtun ‘ache/hurt’ splits into
two parts in this sentence type). The subject specifies the aching part B, while
the experiencer term is put in an indirect object or PP position. Thus the head is
what the statement focuses on, not the experiencer. A similar variant is used in
Hungarian:
(6)
f.
Hungarian: fáj
a
fej-em
3
DEFART
head-POSS1S, SG
aches
lit. ‘aches the head-of-mine’
Here the verb is used as a one-place predicate with B as its only argument. B, in turn,
is specified with the possessive suffix as the B of E. B and E are tied into one.
The usual Japanese equivalent of the sentences in (6a–f) is (7a):
(7)
a.
Japanese
4
atama
ga
ita-i
head
NOM
feel hurting-PS
lit. ‘head feels hurting’
2 The two German constructions are not completely equivalent, the latter being more generally
applicable to any kind of pain felt in or at the head, but that does not matter for the present
discussion.
3 -em/-am is a noun suffix that indicates possession of the speaker and singular number of the noun.
4 In order to keep the grammatical structures of the sentences as transparent as possible, Japanese
examples are given in the so-called plain style lacking all formality markers (cf. 2.3.2).
Meaning and language comparison
253
Itai is a so-called verbal adjective (VA). Verbal adjectives carry a tense ending and
can form the VP in predicative use. They function like a copula plus predicative
adjective in English. Interestingly, the VA itai has only one argument, B, specified by
the subject of the sentence. But what about the experiencer E? How do I know from
a sentence like (7a) whose head it is that hurts? The really interesting thing about
Japanese is that one does know from the sentence, and this is a point where Japanese
differs radically from English, German, French, Russian, Hungarian and many other
languages.
The answer is simple: the experiencer in the case of (7a) is the speaker. There are
only two exceptions. (i) The sentence could, with appropriate intonation, also be
used as a question; it would then mean ›Do you have a headache?‹. (ii) In a literary
text with a central protagonist, through whose eyes and mind the narration is
told, it could relate to that person. The crucial point is that itai means a subjective
sensation, not an objective phenomenon. Subjective sensations, like the feelings
of pain, itching, appetite, grief or longing, only exist for the experiencer. (7a) says
something like ‘the head feels aching’. E is the only one who can state this or answer
the corresponding question. A specification of E can be explicitly added to the
sentence, as in (7b, c), but it has to be semantically redundant. The experiencer
specification is added as a topic of the sentence, not as a complement of ita-i.
(7d) is unacceptable, except for the special literary reading mentioned. (TOP is the
Japanese topic marker, which is hard to translate. It means something like ‘as for …’,
but is less heavy.)
(7)
b.
c.
d.
Watashi
I
TOP …
wa atama ga itai.
Anata
you
TOP …
wa atama ga itai?
*Mary wa atama ga itai.
‘As for me, head feels hurting’
‘As for you, head feels hurting?’
5
‘As for Mary, head feels hurting.’
Sentence (7a) is syntactically complete. Words such as itai have an implicit
argument for the direct experiencer. Thus Japanese represents the third logical
possibility of tying together two of the three ingredients, this time S and E. It
is worth noting that, wherever two ingredients are tied together, one of them
is dominant, and the expression denotes the dominant one. Headache and
Kopfschmerzen denote S and locate it at B; fejem denotes B and assigns it to E; itai
denotes S and assigns it to E.
The five patterns are compared in Fig. 10.1. A rectangle indicates the central
predicate expressed by the verb, circles represent its arguments; the subject is shaded.
Where two ingredients are tied together, the dominating one is indicated by larger
type.
5 The sentence may, perhaps, be read as meaning something like ‘when I think of Mary, I’m getting
a headache’, but it cannot mean ‘Mary has a headache’.
254
Understanding semantics
Figure 10.1
Headache constructions
English, German 1
E
have
French
E
have
S-B
S
LOC
B
S
B
Hungarian
S
B-E
Japanese
E-S
Russian, German 2
E
B
These are not all the types that are logically possible. For example, our sample
contains no constructions with B or E in predicate position. These, however, are
unlikely constructions. Given the typical function of verbs as denoting events
and states of temporary nature as opposed to nouns that tend to denote the more
permanent objects of experience, we will expect that neither E nor B will figure as the
verb of the construction. But there remain further plausible possibilities. Instead of
a verb of the type have there might be constructions using be (‘a pain is in my head’)
that put the sensation into the subject position. Also there may be languages where
the pain is expressed by a two-place predicate with E in subject position (‘I ache-feel
my head’). And there are yet other languages that tie all three ingredients together
into one complex predication of type S-B-E.
10.2.2 Evidentials
As we have seen, the Japanese construction does not provide a syntactic position for
the specification of the experiencer. How, then, is it possible to express that someone
else has a headache? In the other languages considered, we can simply replace the
experiencer ‘I’ by an appropriate NP and change the verb form if necessary. The
resulting expressions are perfectly analogous to the 1st person variants (with the
exception that the Hungarian construction needs an additional dative position for
the possessor NP if it is not pronominal; the result is a construction very similar to
German 2).
Meaning and language comparison
(8)
a.
English:
Ken has a headache.
d.
German 2:
Ken
Ken.DAT
tut der Kopf weh.
aches the head
f.
Hungarian:
Ken-nak
Ken.-DAT
fáj a fej-e.
aches the head-POSS3SG,SG
255
Strictly speaking, Japanese has no way of saying this. But of course it has means
of talking about third persons’ headaches. The ways in which it can be done are
interesting because they reflect a point that usually goes unnoticed to native speakers
of languages such as English, Russian or Hungarian: we can never really KNOW that
someone else has a headache, not in the sense of knowing as the experiencers
themselves do. Every person has their own body and their own sensations which
cannot be immediately perceived by anybody else. Knowledge about someone else’s
sensations is necessarily second-hand and of a different quality than knowledge from
sources that are immediately accessible. Consequently, statements that are based on
second-hand knowledge are less reliable. In Japanese, this difference is obligatorily
expressed; second-hand knowledge is always marked as indirect evidence. (9) shows
the three most common possibilities for the case of others having a headache:
(9)
a.
b.
c.
John
wa
atama
J.
TOP
head
‘John says he has a headache’
John wa atama ga
…
‘John seems to have a headache’
ga
NOM
ita-i-sō
ache-PS-EVD1
ita-sō
ache-EVD2
da.
COP.PS
da.
COP.PS
John wa atama ga
ita-gat-te i-ru.
…
ache-EVD3-PROG-PS
‘John is displaying symptoms of having a headache’
In each case, the NP John is added to the basic construction as a topic and/or subject
marked with the topic particle wa. In (9a) the nominal suffix -sō plus the copula
verb da is added to the full form of the adjective itai, including the tense ending
-i, to render the sense of John saying he has a headache. In (9b) the nominal suffix
-sō plus copula is added to the bare stem ita- of the adjective, meaning something
like the English seem to construction. The third pattern puts the experiencer into a
more active role of displaying symptoms of, or behaving like, having a headache. The
construction involves a verbal suffix -garu that turns a verbal adjective into a verb. It
is here used in its progressive present tense form -gatte iru.
The grammar of Japanese strictly distinguishes between statements that we
are entitled to make on the basis of first-hand knowledge, and statements for
which we rely on second-hand evidence. The so-called evidential constructions
in (9) are widely used. There are about seventy adjectives for emotions, feelings,
sensations and mental states that are used in the same way as itai, e.g. for feeling
256
Understanding semantics
happy (ureshii), sad (kanashii), lonesome (sabishii), cold (samui) or hot (atsui), for
liking (suki) and hating (iya, kirai), for being afraid of something (kowai), finding
6
something cute (kawaii) or delicious (oishii). One can only say of something that it
is ‘oishii’ if one has actually tasted it. Something that appears delicious would be said
to be ‘oishi-sō’ (cf. (9b)). This type of adjectives also includes the volitional forms
of verb that are formed with the suffix -tai; it turns verbs into verbal adjectives
rendering the meaning ›want to V‹. One can say ‘I want to go’ (iki-tai, from iku ‘go’),
but not with the same construction ‘John wants to go’ (*John wa iki-tai). Again, one
would have to use an evidential construction meaning ‘John seems to want to go’
(John wa iki-ta-sō da).
The comparison of the way in which bodily sensations, feelings and wishes are
expressed offers an insight into the nature of knowledge. Languages such as Japanese
force upon the language user the distinction between first-hand and second-hand
evidence. This insight could not be obtained from the analysis of languages such as
English, which treat feelings of oneself and the feelings of other persons alike and in
the same way as any kind of objective observation, e.g. John has red hair.
10.3 RELATIVISM AND UNIVERSALISM
How different are languages? The possible positions regarding this question range
between two extremes. One is called universalism. According to the universalist
position, all languages obey the same principles. The structure of every language is
some variant of universal grammar. And universal grammar is part of the human
genetic equipment. Likewise, the cognitive system is genetically determined. For
biological reasons, all human beings perceive themselves and their environments
essentially in the same way; they form the same kind of concepts and organize them
into the same kind of complex model of the world. Consequently, languages can
only differ within a limited range of variation. Considerable differences between
languages do exist. The environment in which a linguistic society lives and the
culture it has developed will be reflected in the language. For example, each language
community will have a particularly elaborate vocabulary in those areas of life that
are of central importance. But ultimately, the universalist would argue, all languages
make use of the same mental resources. They differ only in the way in which these
resources are used.
The opposite position is known as linguistic relativism. To an extreme relativist,
each language is radically different. Due to its unique grammar and its uniquely
structured lexicon, it represents a unique way of talking about the world and
corresponds to a particular way of thinking. Each language represents, and creates,
a worldview of its own. The relativist position is connected with the names of two
American linguists who worked on indigenous North American languages in the
first half of the twentieth century, Edward Sapir (1884–1939) and Benjamin Whorf
(1897–1941). The following passage from Whorf is often quoted as a formulation of
6 See Martin (1975: 361) for a survey.
Meaning and language comparison
257
the so-called Sapir-Whorf hypothesis (Whorf 1956: 212–14, the original date of
publication is 1940):
Formulation of ideas is not an independent process, strictly rational in the old
sense, but is part of a particular grammar, and differs, from slightly to greatly,
between different grammars. We dissect nature along lines laid down by our
native languages. The categories and types that we isolate from the world of
phenomena we do not find there because they stare every observer in the face;
on the contrary, the world is presented in a kaleidoscopic flux of impressions
which has to be organized by our minds – and this means largely by the linguistic
systems in our minds. We cut nature up, organize it into concepts, and ascribe
significances as we do, largely because we are parties to an agreement to organize
it in this way – an agreement that holds throughout our speech community and is
codified in the patterns of our language. The agreement is, of course, an implicit
and unstated one, but its terms are absolutely obligatory; we cannot talk at all
except by subscribing to the organization and classification of data which the
agreement decrees.
Applied to the Japanese evidentials, this view would mean that the members of the
Japanese speech community implicitly agree that second-hand evidence of emotions
and perceptions is to be distinguished from first-hand knowledge. This hypothesis
is plausible, and there are many other phenomena where linguistic structures can be
linked to the culture of a speech community. Take, for example, the Japanese terms
for siblings which, in usual colloquial Japanese, force the language user to distinguish
between younger and elder siblings. This trait of Japanese is not accidental. Japanese
society is extremely hierarchical. No two persons that are in some social relation to
each other are of the same rank. For example, those who are older rank higher than
those who are younger and men rank higher than women. The ranking rules are
deeply rooted in social behaviour AND in language use and structure. For example,
while siblings in Western societies address each other mutually by their first names,
the mode of address between Japanese siblings of different age is asymmetric. Elder
siblings are addressed by their title, as it were, namely the polite versions of ane and
ani (onê-san and onî-san, respectively, the latter literally meaning ›Mr elder brother‹),
but younger siblings are just called by their names.
The domain of sexuality provides another case of correspondence between culture
and language. Due to a long history of sexual taboo enforced by the Christian church,
European languages exhibit remarkable lexical peculiarities. The taboo is reflected
in the massive use of circumscriptions, i.e. indirect ways of expression. The words
for ›bra‹ mentioned in 7.6.1 may illustrate the point. The French expression soutiensgorge, literally meaning ›throat support‹, avoids mentioning what is really ‘supported’
by referring instead to a neighbouring part of the body that is not taboo. The English
term bra is even more indirect. It is an abbreviation of brassiere (a French loan word
deriving from bras ›arm‹) and originally means a short vest or top. The Spanish term
sujetador (lit. ›subjugator‹) reflects the sexual taboo in not mentioning the object of
subjugation and in a negative attitude towards it as something to be subjugated. By
contrast, the Tok Pisin notion ›prison of the breasts‹ (kalabus bilong susu) mentions
258
Understanding semantics
the breasts directly and reflects the loss of freedom that came with this particular
7
item of Western culture and the foreign taboo that made it necessary.
Such phenomena, however, do not directly support the strong relativist position. Is
it really language that forces the Japanese into their hierarchical social thinking, or
Westerners into observing sexual taboos? Rather, it appears, that language REFLECTS
social structure and cultural standards. It is true that it is not possible to talk directly
about sexuality if the language does not provide the appropriate words. But the
so-called sexual revolution in Western culture in the last decades of the twentieth
century has shown that speech communities rapidly develop socially acceptable
expressions when the taboo loses force.
It must also be observed that particular grammatical traits of a language need
not influence the worldview of its users. For example, many European languages
have grammatical gender. In French and Spanish each noun is either masculine or
feminine, in German and Russian nouns are either masculine, feminine or neuter.
Gender determines, among other things, the form of the definite article (le/la in
French, der/die/das in German). It would, however, be absurd to conclude that gender
classification imposes any kind of different worldviews upon the user. The fact that
the German nouns for ‘government’, ‘street’ and ‘banana’ (Regierung, Straße, Banane)
are of feminine gender does not make speakers think of these things as having
8
anything in common, in particular not female sex.
Although these observations point towards a differentiated position somewhere
between universalism and relativism, the central questions remain. How different
are languages? How deep are these differences? What do the observable differences
indicate? Do they correspond to different ways of thinking? The questions are far
from being settled. We will now turn to one field where extensive comparative
research has been done: the field of colour term systems.
10.4 BERLIN AND KAY’S INVESTIGATION OF COLOUR TERMS
The spectrum of colours, the same for all human beings with normal vision, forms
a continuum with no natural boundaries between the colours: red shades into
orange, pink, purple, brown; orange into yellow and red; and so on. Thus it is to be
expected that languages cut the colours up in different ways, into a different number
of colour terms that denote different parts of the colour space. In fact it was held that
9
languages lexicalize colours arbitrarily. It was the investigation reported in Berlin
and Kay (1969) which changed the scene dramatically.
The study covered approximately a hundred languages from all over the world. For
twenty languages native speakers were interviewed. For the rest, Berlin and Kay drew
7 For many more examples of prudery in language see Bryson (1990, ch. 14) on English.
8 There are slight gender effects for speakers of gender languages. For example, it has been shown
that subjects can remember male proper names for objects better, if the noun used for them is
masculine. But findings such as these can hardly be considered evidence for a different worldview
(see Deutscher 2010, ch. 8).
9 See Berlin and Kay (1969, n. 1, p. 159f.) for quotations on this position.
Meaning and language comparison
259
on previous studies, grammars and dictionaries. They investigated how many colour
terms these languages possess and to which range of colours each term refers. Such
an investigation is only possible if it is restricted to what Berlin and Kay called basic
colour terms (BCT, for short). English, for example, has hundreds of lexicalized
colour terms (think of the vocabulary for the colours of lipsticks, eye-shadows,
fashion accessories, cars, etc.). Only a few of them are basic. The main criterion for
basicness is:
∑ BCTs are not subordinates of other colour terms.
This excludes terms like olive (a shade of green), crimson (red) or chocolate (brown).
In addition, Berlin and Kay used the following criteria: BCTs are simple words, not
compounds or derivations (like greenish, dark green, mint-green, blue-green); BCTs
are not restricted to a narrow class of objects (cf. blond for hair); BCTs (not the
colours they denote) are ‘psychologically salient’, i.e. they ‘[tend] to occur at the
beginning of elicited lists of color terms’; their use is stable across informants and
10
across occasions and they are used by all informants. Berlin and Kay also ruled
out recent loan words (such as aubergine) and colour terms that are derived from
the name of an object characteristically having that colour, such as gold, silver, olive,
etc. Orange originally belonged to this class, but it qualifies as a genuine BCT since
it fulfils the main criterion of being a term on a par with red and yellow rather than
a subordinate of either. As we will see, these criteria yield a set of eleven BCTs for
English.
Berlin and Kay had their subjects perform two tests. First, they elicited the set
of BCTs in their respective native languages. Then the informants were shown a
chart of 329 colour chips (from the Munsell Color Company), 9 for black and white
and different shades of grey, and 320 others, arranged in a field of 8 by 40 chips,
that horizontally range from red via orange, yellow, green, blue, purple to red and
vertically from near-white to near-black (see Fig. 10.2). The original colour chart is
depicted in Berlin and Kay 1969 and in Palmer 1996; it is online at http://www.icsi.
berkeley.edu/wcs/data.html. For each BCT the subjects were asked (a) to point out
the focal colour chip, i.e. the chip that was considered the best example for this BCT,
and (b) to indicate the outer boundary of the range of colours denoted by the BCT.
It turned out that informants of the same language narrowly agreed on the focal
colours, but indicated the boundaries differently. The main result, however, was this:
11
the languages vary in the number of BCTs from two to eleven. The focal colours for
the respective BCTs are not distributed randomly within the field of colours; rather
they concentrate on just eleven colours which were indicated by the informants
consistently, and with very little variation, as the best examples for the BCTs.
Apparently, speakers of different languages, with different numbers of BCTs, agree
what is the best ‘red’/‘rouge’ (French)/‘akai’ (Japanese), etc. These eleven colours
are focal (i.e. ‘pure’, ‘real’, ‘typical’) white, grey, black, red, orange, yellow, green, blue,
purple, pink and brown (indicated by the dots in Fig. 10.2). These are the foci of the
10 See Berlin and Kay (1969, p. 6) for the details.
11 The maximum number was later corrected to twelve (see below).
260
Understanding semantics
English BCTs white, grey, black, red, orange, etc. and of the corresponding words in all
other languages with eleven BCTs. What is much more important is the finding that
the foci of BCTs in languages with fewer than eleven are also among this set of eleven
focal colours. Apparently these colours form a set of fixed points of orientation for
the denotation of BCTs.
Fig. 10.2
Munsell colour chart used by Berlin and Kay
white
yellow
grey
pink
pink
orange
green
red
blue
brown
puiple
red
black
The second important finding is this: languages with the same number of BCTs
have basic colour terms with the same foci, and if a language has n + 1 BCTs, it has
BCTs with all those foci that are denoted in languages with n BCTs, plus an additional
BCT with a new focal colour. In this way the existing systems of BCTs form a sequence
where a BCT for a new focus is added step by step; at the final stage there are eleven
BCTs that have their focus on the eleven focal colours. The first foci in this sequence
are focal white and black (Stage I); the third focus added is red (Stage II); red is
followed by green and yellow in arbitrary order (Stages III and IV), then blue (V),
then brown (VI). For the remaining four foci, grey, pink, purple and orange, there
is no fixed order. According to this order of appearance of BCTs, there will be no
language that, for example, has a BCT for brown, but not for blue and green.
Of course, if a language has just three BCTs with focus on focal white, black and
red, the red term does not have the same narrow denotation as the red term in a
language with eleven BCTs such as English. The smaller the number of BCTs in a
language, the larger are the respective ranges of colours they denote. For example,
the red term in a language with five BCTs (with focus on black, white, red, yellow and
green) will also cover the reddish parts of the denotations of English orange, brown,
pink and purple because there are no more specific BCTs for these hues; lighter
orange and yellowish brown will fall under the yellow term; blue and bluish purple
will be covered by the green term or the black term, if the colours are very dark; a
very light pink will be called ‘white’ (for details see charts in Berlin and Kay 1969).
Thus the range of colours covered by a BCT with a given focus depends on how
many other BCTs the language has and what their foci are. A black term in a system
of Stage I does not have the same meaning as a black term in a system of another
Meaning and language comparison
261
stage; therefore the total number of BCTs is added to the names of the colour terms
in Table 10.1. Berlin and Kay distinguish seven stages of BCT systems; the last stage
includes all languages with eight to eleven BCTs; for BCTs of languages in this last
stage, ‘8’ is added to the BCT names. For example, there are no languages with three
BCTs that focus on yellow, red and blue which might make sense; let alone such
unlikely variants as foci on pink, purple and turquoise, seemingly the most popular
colours with children.
Table 10.1
The seven stages of colour term systems in Berlin and Kay (1969)
Stage
BCTs
I
BLK/2
WHI/2
II
BLK/3
WHI/3
RED/3
IIIa
BLK/4
WHI/4
RED/4
YEL/4
IIIb
BLK/4
WHI/4
RED/4
IV
BLK/5
WHI/5
RED/5
YEL/5
GRN/4
GRN/5
V
BLK/6
WHI/6
RED/6
YEL/6
GRN/6
BLU/6
VI
BLK/7
WHI/7
RED/7
YEL/7
GRN/7
BLU/7
BRN/7
VII
BLK/8
WHI/8
RED/8
YEL/8
GRN/8
BLU/8
BRN/8
GREY/8
or
ORANGE/8
or
PINK/8
or
PURPLE/8
Focus
black
white
red
yellow
green
blue
brown
grey etc.
From the distribution of BCT systems, Berlin and Kay drew the conclusion that
the possible systems form evolutionary stages. The two BTCs of Stage I systems,
WHITE/2 and BLACK/2, are terms for light/warm colours (white, yellow, red) and dark/
cool colours (black, blue, green), respectively. In the transition to Stage II, WHITE/2
splits into WHITE/3 for light colours (white) and RED/3 for warm colours (red, yellow),
while BLACK/3 continues to denote dark/cool colours. When we proceed to Stage III,
in one variant yellow splits from RED/3 to form a category of its own, resulting in a
system with WHITE/4 for light (white), RED/4 for red, YELLOW/4 for yellow, and BLACK/4
still covering all dark/cool colours. Alternatively, green and blue together split from
BLACK/3, forming a new colour category often called ‘grue’ that focuses on either focal
green or focal blue (but never on some blue-green midway between focal blue and
focal green). In the resulting system, yellow is still mostly denoted by RED/4. In the
transitions to later stages up to V, the denotations of multicolour BCTs split further,
e.g. grue into blue and green. The higher BCTs with foci on brown, grey, orange, pink
and purple are not the result of splittings but of establishing terms for hues that
lie between the foci of the six terms of Stage V: grey between white and black, pink
between white and red, orange between yellow and red, purple between red and blue,
and brown between yellow and black.
262
Understanding semantics
The findings of Berlin and Kay triggered a great number of follow-up studies, in
particular the ‘World Color Survey’ where Berlin and Kay’s experiments were applied
12
to 110 unwritten languages. As a result, their findings were slightly revised. The
larger categories of the early stages may have more than one focus, e.g. white, yellow,
and red for WHITE/2 or green and blue for GREEN/4. It turned out that there are certain
types that do not fit into the sequential chain, e.g. languages of Stage III with a BCT
for grue but yellow still going with white, while in Stage II systems yellow is already
separated from white. In addition the order of appearance of the ‘higher’ colours
is less strict: a BCT for brown may appear before a term for blue. Some languages
13
have twelve BCTs. But these are minor modifications. The main tendencies were
14
confirmed:
∑ The denotations of BCTs are best described in terms of focal colours.
∑ The range of colours that a given BCT denotes depends on the total number of
BCTs in the language and on the foci they have, respectively.
∑ There is a limited universal set of eleven (or twelve) focal colours. The best
representatives of BCTs invariably are among these focal colours.
∑ The possible types of system form a sequence – starting with a contrast between
WHITE/WARM and BLACK/COOL – in which the higher systems are extensions of the
lower systems.
10.5 CONSEQUENCES
What follows from the findings on colour terms for the structuralist view of
language? The results certainly prove that the arbitrariness of lexicalization can be
constrained. Still, languages differ largely in their colour terminology. Also, another
doctrine of structuralism has been confirmed: the denotations of terms within the
same field are interdependent and delimit each other: the range of a BCT with a given
focus depends on how many other BCTs compete with it.
And what do the findings mean for the debate between relativism and universalism?
Originally, the results were considered evidence against the Sapir-Whorf hypothesis.
But are they really? It was claimed (Kay and McDaniel 1978) that the universal
constraints on colour vocabularies are directly rooted in the nature of human colour
perception. Even if this is right, this cannot be generalized to most other semantic
15
domains. Most of the things languages have words for are not like colours. For
12
13
14
15
See the World Color Survey homepage www.icsi.berkeley.edu/wcs/.
Russian has two BCTs for lighter and darker blue, Hungarian has two for red.
See Kay and Regier (2003).
Meanwhile, there is evidence that even for colour terms there are what are called ‘Whorfian effects’.
According to Drivonikou et al. (2007), language influences the discrimination of colours: similar
colours represented to the right eye (which is connected to the left hemisphere where the linguistic
lexicon resides) are more easily discriminated if there are two different BCTs for the colours in the
native language of the subjects.
Meaning and language comparison
263
example, we can see animals, but we do not have specialized dog cells and cat cells.
Animal concepts, e.g. the meaning of the English word pig, are not based on sense
data alone such as visual appearance – although visual shape is certainly part of the
concept. In addition, animal concepts are based on cultural aspects. For example,
the English ›pig‹ concept will reflect that pigs are domesticated and eaten (witness
the meaning relation between the words pig and pork). The concepts are related to
the culture, i.e. to the way speakers of a language ‘look at’ things. For other semantic
domains, the findings on colour terms and colour perception have no significance at
all. Many words denote things that cannot be immediately perceived with our senses;
it would not be possible to ask subjects to point to the best examples from a selection
of cases. Consider a word like mistake. The notion ›mistake‹ appears quite natural
and basic to us. But it presupposes complex cultural systems of rules that determine
what is right and what is wrong and in which regard. Such concepts are completely
culture-dependent.
Relativism is certainly right in emphasizing the differences between languages.
They do exist, they are the rule rather than the exception, they are deep – and they
are fascinating. A relativist attitude is absolutely necessary for all who seriously try
to understand other languages. It provides the only chance to escape the natural
tendency of applying one’s old views and categories to new things. Only when we
expect other languages to be different from our native language, will we be able to
recognize and understand the differences. Thus, relativism is the better working
hypothesis. Once we are aware of the differences between languages, we may set
out to satisfy the universalist by trying to determine the common denominator that
makes a comparison possible. After all, it must exist, because otherwise we would
not have been able to understand our native as well as another language with our
one and only mind.
EXERCISES
1. When words are imported from other languages, their meaning is often changed
more or less. For example, the German word Arbeit (›work‹) was borrowed
into Japanese, where it was shortened to baito. Japanese baito has a narrower
meaning, denoting part-time students’ jobs such as giving private lessons.
a. Try to find three words in your native language that are borrowed from
other languages. Consult dictionaries to determine their meanings in both
languages.
b. Try to find an explanation why the meanings of loanwords so often differ
from their origins – despite the fact that they are apparently borrowed for
their meaning.
2. Try to determine the meaning relations that hold between the following pairs of
words. First look up the translation of the English term in a bilingual dictionary,
then crosscheck with a reverse dictionary:
a. English cook (verb) vs
German kochen
b. English hot
vs
German heiß
264
Understanding semantics
4. Ask three people to spontaneously list twenty or more colour terms in the order
in which they come to their minds. Compare the results. Are the eleven BCTs
among the words listed? Do they appear towards the beginning of the list? Which
are the first five items?
5. Given that a BLACK/3 BCT covers not only black but also blue, green and other
colours, why is it considered a case of BLACK/3 rather than, say, BLUE/3?
6. What are the essentials of the relativist and the universalist position? Which
position do you think is realistic? What is your own position in view of your
personal experience with foreign languages?
FURTHER READING
Suzuki (1978, ch. 5) for Japanese terms of address and self-reference; Kuroda (1973)
for Japanese evidentials; Palmer (2001, ch. 2.2) on evidentials in general. Whorf
(1940) for a basic outline of the relativist position. Salzmann (1993, ch. 8) for a
discussion of the Sapir-Whorf Hypothesis and the relation between language and
culture. Palmer (1996, ch. 6) on language and worldview. Berlin and Kay (1969) for
the original report of their investigations of colour term systems; Kay and McDaniel
(1978) for an explanation of the universals in terms of colour perception; Wierzbicka
(1996) for an alternative account in the framework of NSM. Foley (1997, ch. 7) for a
more recent discussion of the research in colour terminology, ch. 6 on kinship term
systems in the context of the relativism vs universalism debate. Lee (1996) offers
a careful reconstruction of Whorf ’s original theory and its fate in the relativistuniversalist debate. For general discussions of the relationship between language and
thought see Lucy (1992) and Deutscher (2010); both contain a thorough discussion
and evaluation of the research about colour terms; on the latter, also see Dedrick
(1998). For methods in cross-linguistic semantics see Matthewson (2011).
11
Meaning and cognition
Over the last couple of decades, the development of a new branch of science, cognitive
psychology, or more generally cognitive science, has given important fresh impulses
to linguistics in general and semantics in particular. Cognitive science is concerned
with how the human mind works, how it receives information from the environment
via the senses and processes this information, recognizing what is perceived,
comparing it to former data, classifying it and storing it in memory. It tries to account
for the complex ways in which the vast amount of information is structured in our
minds, how we can operate with it when we think and reason. Language plays a
central role in these theories. On the one hand, speech perception and production
and the underlying mental structures are major objects of investigation. On the other
hand, the way in which we use language to express what we ‘have in mind’ can tell
much about how the human mind is organized.
The importance of cognitive psychology for semantics lies in its emphasis on the
exploration of the concepts and categories we use. While semantics in the wider
tradition of structuralism aims at a description of meaning relations, the cognitive
approach focuses on meanings themselves. It tries to provide positive descriptions
of the word meanings, and thereby to account for why and how words denote what
they denote. In terms of the semiotic triangle, a cognitive approach to semantics
can be characterized as focusing on the base of the triangle, the meaning and how it
determines the denotation (Fig. 11.1).
Figure 11.1
The focus of cognitive semantics
expression
meaning
denotation
266
Understanding semantics
11.1 CATEGORIES AND CONCEPTS
The fundamental notion of cognitive science is categorization. In his introduction
to cognitive psychology, Barsalou (1992a: 15) describes it as follows:
Upon walking into a home, people know instantly what is present. They recognize
chairs, stereos, plants, dogs, friends, guacamole, wine, and just about anything
else they happen to perceive. […] When people interact, they recognize friends,
facial expressions, actions, and activities. When people read, they categorize
letters and words. Categorization occurs in all sensory modalities, not just vision.
People categorize the sounds of animals and artifacts, as well as the sounds of
speech; they categorize smells, tastes, skin sensations, and physical movements;
and they categorize subjective experiences, including emotions and thoughts.
Categorization provides the gateway between perception and cognition. After a
perceptual system acquires information about an entity in the environment, the
cognitive system places the entity into a category.
Categorizing something that we perceive (or imagine, or remember) means to
perceive it as something of a kind. If we look at a photograph of a person, we will
categorize it as an object of the kind ‘photograph’ that displays something of the
kind ‘person’, and if we recognize the person, we will categorize him or her as that
person. We assign everything that enters our minds to one or more categories. The
category DOG (we will use SMALL CAPITALS for categories) consists of all those things we
would categorize as dogs. It includes not only the real existing dogs but also former
and future dogs or fictitious dogs in novels, jokes, etc. Entities belong to different
categories at the same time. Our common friend John belongs to the categories
PERSON, MAN, FICTITIOUS, BICYCLE OWNER and maybe (we just do not know) LOUSY DANCER
along with countless other categories. The single entities that belong to a category are
called exemplars or members of the category. (John is an exemplar of the category
BICYCLE OWNER.) Larger, more general categories include smaller, more specific ones.
These are subcategories of the former: MAN and WOMAN as well as BICYCLE OWNER are
subcategories of the category PERSON. All members of a subcategory are also members
of the more general category. Note that the subcategories of a category are not
members of it. Rather both share members. If you are familiar with set theory, it may
help to realize that categories are sets, subcategories subsets and members elements.
We encountered categories before: denotations (2.2.2) are categories. The
descriptive meaning of a word defines a category, the set of all its potential referents.
When we refer to Donald using the term duck, we treat him as a member of the
category DUCK.
Categorization is only possible if the respective categories are in some way
available in the cognitive system, i.e. in our mind. In the terminology of cognitive
science, categorization requires mental representations of the categories. There are
various theories concerning the nature of category representations. Here the view
will be adopted that categories are represented by concepts for their exemplars. The
Meaning and cognition
267
1
category DOG is represented in the mind by the concept ›dog‹. When we encounter
an object, our cognitive apparatus will produce a preliminary description of it, which
consists of what we perceive of the object, e.g. its size, shape, colour, smell, etc. The
description will be compared with the concepts we have in our mind and if the
description happens to match with the concept ›dog‹, the object will be categorized
as a dog.
Two things must be kept in mind when we think of word meanings and concepts.
First, word meanings do not coincide with our concepts for actual categories. For
example, the meaning of the word bicycle is a relatively abstract concept that is just
rich enough for defining the category of bicycles. But you will probably have a richer
concept depending on your individual knowledge of bicycles and your personal
experiences. Thus, the bicycle concept that constitutes the meaning of the word is
only part of the concept that in your mind defines your personal category of bicycles.
The meaning of the word must be a much leaner concept that is shared by all English
speakers who know the word and its meaning.
Second, we do not have a word for every category we have a concept for. Trivially
there are infinitely many concepts which can be expressed only by complex
expressions, e.g. the concept ›expensive sushi restaurant closed on Saturdays‹ – the
syntactic possibilities of language reflect this potential. But there are also categories
and concepts which cannot be verbalized at all, or only insufficiently. Many concepts
for bodily sensations, feelings, emotions, for facial expressions and physiognomies,
for flavours and odours, for melodies and harmonies, etc. can hardly be put into
words, if they can at all. For example, a verbal description of a face can never be as
accurate as a photograph. Words can never fully describe the taste of an orange, the
smell of carnations or the sound of a violin. In fact, it is plausible to assume that only
the lesser part of our concepts can be expressed in language.
The system of lexical meanings is only part of the overall system of concepts. And
a lexicalized concept is only part of the richer concept that defines the category we
actually connect to the word. In this regard, a distinction will later be drawn between
semantic concepts (word meanings), cultural concepts (richer concepts shared by a
cultural community) and personal concepts (cf. 11.6).
11.2 PROTOTYPE THEORY
11.2.1 The traditional model of categorization
The traditional view of categorization is shaped by the model of ‘necessary and
sufficient conditions’ (NSC model for short) that goes back to Aristotle. According to
the NSC model, a category is defined by a set of necessary conditions, which together
are sufficient. For example, if we assume that the category WOMAN is defined by the
1 Henceforth ›…‹ quotes are used not only for meanings but for everything on the conceptual level,
i.e. meanings, concepts and components of concepts.
268
Understanding semantics
three conditions of being human, female and adult, each one is necessary. If someone
is not human or not female or not adult, he or she is not a woman. On the other
hand, the conditions of being human and female and adult are jointly sufficient for
membership in the category WOMAN. It does not matter what other conditions someone
or something may fulfil. Being a woman or not depends on these three conditions.
As you will have noticed, the NSC model of categorization, also called the checklist model, is implemented by the BFA notion of meaning (9.3). The binary features
that make up the meaning of a word according to BFA are jointly sufficient necessary
conditions of the NSC model.
The Aristotelian model can be characterized by the following points:
∑ Categorization depends on a fixed set of conditions or features.
∑ Each condition is absolutely necessary.
∑ The conditions are binary (yes-or-no) conditions.
∑ Category membership is a binary (yes-or-no) issue.
∑ Categories have clear boundaries.
∑ All members of a category are of equal status.
That categories have clear boundaries is a direct consequence of the fact that the
defining conditions are binary. Everything either fulfils this set of conditions or it
does not. If it does, it belongs to the category, otherwise it does not. As a consequence,
categories have clear boundaries, and within their boundaries all members enjoy the
same status of full members. Each single point of the NSC model was challenged in
prototype theory, the result of the first extensive studies in categorization undertaken
by cognitive psychologists and semanticists.
11.2.2 Prototypes
The findings on colour terms (10.4), which at the same time were findings on colour
categorization, did not seem to fit the NSC model at all. Clearly, the subjects did not
categorize the colour chips by checking a list of binary features. The basic colour
categories are primarily defined by focal colours. A definite boundary between
neighbouring categories cannot be drawn.‘Category boundaries […] are not reliable,
even for repeated trials with the same informant’ (Berlin and Kay 1969: 15). When
one moves from focal red to focal brown in a system that has BCTs for both, one
somewhere leaves the range of red hues and enters that of brown hues. Furthermore,
the categories RED and BROWN intuitively overlap. Brownish red is predominantly
red but still more or less brown, and reddish brown is vice versa. Thus, it appears, a
colour belongs to a category if it is sufficiently similar to the focal hue, i.e. the best
example. Since, however, similarity is a matter of degree, category membership too is
a matter of degree, rather than a yes-or-no issue.
Linguists and psychologists set out to investigate other areas for similar
phenomena. They found that for many different categories best examples can be
empirically established. These come to mind first and are consistently considered
Meaning and cognition
269
‘better members’ than less typical cases. Such central examples came to be called
prototypes. Experiments were carried out to examine whether category membership
in general is a matter of similarity to the prototype and hence a matter of degree;
whether categories have fuzzy boundaries; whether the defining conditions, or
features, are binary and always strictly necessary.
It soon turned out that, apparently, many categories have a ‘graded structure’.
They contain prototypical exemplars that represent the category best and other
exemplars that do it to a lesser extent but are still good examples, while others only
enjoy a marginal status. For example, in a famous study cited in every textbook on
cognitive semantics, Eleanor Rosch, whose name is inextricably linked with the early
development of prototype theory, established a ranking within the general category
BIRD. The subjects participating in the study were asked to rate on a scale from 1
(for best) to 7 the ‘goodness-as-example’ of different kinds of birds. The results
were surprisingly consistent. Robins were considered the best examples, followed
by doves, sparrows and canaries; owls, parrots, pheasants and toucans occupied a
medium position, ducks and peacocks were considered even less good examples,
while penguins and ostriches ranked lowest. Similar results were obtained for the
categories FURNITURE, FRUIT, CLOTHING, etc. Furthermore, the rankings were consistent
with other findings, such as the time it took the subjects to answer questions of the
type ‘Is a penguin a bird?’, ‘Is an eagle a bird?’, etc.; the less typical the examples were,
the longer the reaction time.
Since prototypical examples are what we think of first, we will exclude other cases
when a category is mentioned, as long as there is no reason to do otherwise. Thus
prototypes play an important role in what is called default reasoning, i.e. reasoning
in terms of assumptions which replace specific actual information as long as none is
provided. For example, when Mary tells John,
(1)
Look, there’s a bird on the window sill.
John will think of a prototypical bird, not of an owl, a condor, an ostrich or a penguin.
If someone mentions a ‘car’, we will not think of a truck or a veteran car. It is therefore
misleading to use the general terms for non-prototypical cases, misleading – but
not semantically incorrect. Penguins are birds and we can refer to them as birds in
appropriate contexts. For example, (2a) and (2b) are perfectly acceptable:
(2)
a.
The only birds that live in the Antarctic are penguins.
b.
Penguins come ashore to nest. The birds lay one to three eggs.
11.2.3 Fuzzy boundaries
Other experiments were performed in order to assess the fuzziness of category
boundaries. The linguist William Labov presented pictures similar to those in Fig.
11.2 to students and asked them to name the objects depicted.
270
Understanding semantics
Figure 11.2
Cups, bowls, vases
1
2
8
3
4
5
9
10
7
6
The subjects categorized objects like 3 as a cup, 10 as a vase and 6 as a bowl, but
produced inconsistent answers for vases or bowls with a handle – 1 and 5 – or
exemplars with intermediate height-width ratio such as 2, 4, 7 and 9. The subjects
were also asked to imagine that the objects were filled with coffee or standing on a
shelf with cut flowers in them, etc. This had a strong influence on the categorization
(for example, more objects were categorized as cups when they contained coffee). The
experiment showed that prototypical cups are about as high as they are wide, have a
handle and are used for drinking hot drinks such as coffee, while prototypical vases
are higher than they are wide, lack a handle and are used for putting flowers in. If we
encounter an object where these features are shuffled into a different combination,
the criteria for categorizing it as a cup, a vase or a bowl may come into conflict
with each other. Is object 1 a cup because it has a handle or is it a vase because it is
much higher than wide? Likewise the criterion of the height-width ratio may cause
problems if an object’s ratio lies somewhere between that of cups and vases (2 and
9) or between that of cups and bowls (4 and 7).
11.2.4 Family resemblance
The prototype phenomena cast a new light on the features that define a category. If
a category is primarily defined through its prototype, the conditions that define the
prototype need not be necessary conditions for the rest of the category. (If they were,
all members of the category would be equally prototypical.) For example, members
of the category CUP may lack a handle although the prototype has one; the prototype
of the category BIRD is small and able to fly, but other members may be unable to fly
or much bigger. In the case of cups, let us assume that they are distinguished by (i)
being used for drinking coffee or other hot drinks, (ii) having a handle, and (iii) a
balanced height-width ratio. In Fig. 11.3 the ten objects are arranged in accordance
with these features.
If we accept that all objects except 6 and 10 (the prototypical bowl and vase) are
in some sense cups, the resulting category is not defined by any common condition.
Some of the objects fulfil one condition, others two, prototypical cups all three. What
ties the objects together in one category is what is called a family resemblance.
Meaning and cognition
271
Figure 11.3
Distribution of cup features and family resemblance
balanced ratio
used for drinking coffee
1
2
3
4
5
10
9
8
7
6
handle
The philosopher Wittgenstein (1958) introduced the notion in connection with his
famous example of the category GAME. He argued that there is no property shared by
all games. Rather, some games share properties with certain others, e.g. the property
of being played on a board or being competitive. Others may share no properties at
all, but the whole domain of games is interconnected by similarities between the
members. Applied to our simple example, we see that object 1 resembles object 2 in
having a handle, object 2 resembles object 9 in that we would drink coffee from it. But
object 1 and object 9 share none of the three properties, yet they belong to the same
category, due to the linking element 2.
11.2.5 Graded membership
Still, the ten objects do not seem to be simply cups or not. We may count object 1
as a cup, but it certainly resembles a vase more than a cup. Other objects such as 7
seem to fall right between the categories CUP and BOWL. The many cases of uncertain
category membership and the ranking results in the tests mentioned suggested that
category membership is not a yes-or-no question but a matter of degree. Degrees of
membership can be determined by ranking experiments. Or they could be calculated
on the basis of the prototypical properties, weighting the features and reducing the
degree of membership if a certain property is missing. For example, one might rate
the prototypical cup 3 as a 1.0 cup, and 2, 4 and 8 as 0.8 cups (= pretty good cups).
Cups 7 and 9 might be 0.5 cups (half-cups), and 1 and 5 0.2 cups (very marginal
cups). But the prototypical bowl 6 and the prototypical vase 10 would be 0.0
members of the category, i.e. definitely non-cups.
11.2.6 The prototype model of categorization
The characteristics of the resulting new model of categorization can be summarized
as follows:
∑ Graded structure
The members of a category are not of equal status.
272
Understanding semantics
∑ Prototypes are best examples
There are prototypical members that are consistently considered the best
examples of the category.
∑ No set of necessary conditions
Category membership is not a matter of a fixed set of necessary conditions. The
prototype of a category may be defined by properties absent with less typical
examples.
∑ Family resemblance
Category members are connected by family resemblance.
∑ Prototypes are reference points
Prototypes serve as reference points for categorization.
Category membership is a matter of similarity to the prototype.
∑ Graded membership
Category membership is a matter of degree.
∑ Fuzzy boundaries
Categories have fuzzy boundaries.
As we shall see in the next section, most of these claims are not unproblematic. But
when they were first established, they appeared suggestive and plausible and proved
very influential. In fact they still are, despite the fact that researchers like Rosch
herself soon revised her initial views. Before we turn to a critique of PT (henceforth
short for prototype theory), we will take a closer look at the notion of the prototype.
In order to make the argumentation in the following more transparent, Fig. 11.4
displays the central points and how they are interconnected in PT.
Figure 11.4
Central claims of Prototype Theory and how they are connected
There are
best examples:
the prototypes
therefore
therefore
Graded structure:
members differ
in status
Prototypes are
the reference points
for categorization
therefore
therefore
Graded membership
Fuzzy boundaries
According to PT, prototypes have two crucial properties. First, they are the best
examples for their category. This point is due to the graded structure of categories,
i.e. the existence of better and poorer examples. Conversely, the existence of best
Meaning and cognition
273
examples, along with less good examples, entails graded structure. Second, PT
claims, prototypes serve as reference points for categorization, the crucial criterion
for membership in the category being similarity to the prototype. The two properties
are seen as mutually dependent: prototypes are the best examples because they serve
as reference points, and they do so because they are the best examples. Similarity
itself is a matter of degree. Hence the second property of prototypes implies graded
membership: category membership is not a yes-or-no matter but a matter of
degree. As a consequence of graded membership, categories have fuzzy boundaries.
Conversely, fuzzy boundaries are understood as giving rise to graded structure: the
clear members of a category are better members than those that form the fuzzy
boundary area of the category. Graded structure and graded membership are seen
essentially as two sides of the same coin. If something is a poorer member, it is less
similar to the prototype and hence belongs to the category to a lesser degree, and
vice versa. For the sake of simplicity, family resemblance and the role of necessary
conditions are not included in the schema.
In the following, the network of assumptions and claims of PT will be subjected
to a critical evaluation from a semantic point of view. As it will turn out, the picture
gives rise to a couple of serious questions.
11.2.7 What kinds of entity are prototypes?
The informal definition of prototypes as the ‘best examples’ suggests that prototypes
are distinguished individual members of the category. This makes sense in some
cases, for example for the colour categories. The prototype of GREEN is one particular
colour, i.e. one member of the category. But if prototypes are to be not only best
examples but also reference points for categorization, then for most categories
they cannot be exemplars. If, for example, a certain dog exemplar had to serve as
the prototype of the category DOG, one would have to know this particular dog in
order to be able to categorize anything as a dog, because potential members would
have to be compared with it. Exemplar prototypes may work in very early stages of
language and concept acquisition where we first learn what a dog is by being shown
a particular dog. But later we will replace the representation of the first dog we knew
by a more abstract general representation of ‘the’ prototypical dog.
In the ranking experiments by Rosch and others, the subjects were asked to
judge the representativity not of exemplars but of subcategories of a given category.
For example, they were asked to rank subcategories of birds. But identifying the
prototype with a particular subcategory does not work either. First, the problem
is only shifted. If ROBIN is the prototype of BIRD, what is the prototype of ROBIN?
Most of us do not know subcategories of robins, but certainly there are better and
poorer examples of robins too. Also, the question arises as to whether all robins
are prototypical examples of the category BIRD or only prototypical robins. If the
latter, the prototype of ROBIN would be identical with the prototype of BIRD and the
categories would coincide. If we decide instead that any robin whatsoever counts as
a prototypical case of a bird, we have to include extraordinary big fat monster robins
unable to fly among the prototypical birds. Second, robins exhibit distinctive features
274
Understanding semantics
such as having a red breast which we would not include among the features of the
prototypical bird.
It makes much more sense to assume that the prototype is an abstract case defined
by a concept that fixes certain features and leaves others open. The concept for the
prototypical bird contains a specification of its appearance and anatomy and of the
most prominent features of its behaviour. But other traits may be unspecified, such
as their diet, the exact shape of the beak, the colour of the feet or the markings of
the feathers. Such a concept is different from the concept for an individual bird,
which would have to be specified for all features. Therefore, the prototype is not an
exemplar. It is not a subcategory either, because these would also be specified for
some of the features left open, e.g. the four mentioned. An individual bird will count
as prototypical if it exhibits all the features that make up the prototype. Prototypical
birds may differ in the features left open. If a certain subcategory, e.g. ROBIN or
SPARROW, matches the concept of the prototype for its normal members, it may be
roughly identified with the prototype but, strictly speaking, it is something different.
It must be noted, however, that for more abstract categories, such as the biological
category ANIMAL that includes not only mammals, prototypes cannot be defined in
2
this way. There are certainly best examples for the category, probably cats and dogs.
But dinosaurs, birds, fish, insects, worms, coral polyps and amoebae are also animals.
The category is much too general to fix any concrete anatomical or behavioural
properties of the kind that make up the prototype. Rather, for such categories the
choice of best examples is not a matter of exemplariness but of familiarity. It is
therefore questionable if this kind of prototype can possibly serve as a reference
point for categorization with respect to such categories.
11.2.8 Which features make up the prototype?
One is tempted to answer: the typical features. Some features of birds are intuitively
considered essential and typical features of birds, e.g. having wings and feathers
and being able to fly as well as the properties of singing, having a beak, laying eggs
and hatching them in a nest. Others, like having a particular colour, a certain weight
and size may be characteristic for certain kinds of birds but not for birds in general.
Having feathers is a ‘good’ feature because it well distinguishes birds from non-birds.
The feature is therefore said to have a high cue validity for the category BIRD which
means that it applies to a high proportion of members and to a low proportion of
non-members. The features of having wings, being able to fly and laying eggs have
lower cue validity because they are shared by other kinds of animal, e.g. most insects.
They are, however, of higher cue validity within the narrower domain of vertebrates.
Here, the feature of having wings and being able to fly is very distinctive, since except
for birds only bats share it. Thus, a prototype must be defined by a combination of
features that together maximize the cue validity. Features of low cue validity that
2 Along with this category, there is a much narrower category ANIMAL, that equates roughly to
QUADRUPED and does not include birds, fish, snakes, etc., let alone insects and amoebae. Here and
below, the category ANIMAL is not meant in the narrow sense.
Meaning and cognition
275
distinguish birds and other vertebrates from insects have to be included in order to
enhance the cue validity of the features such as having wings and being able to fly.
11.2.9 Similarity
Prototype theory claims that membership in a given category is a matter of similarity
to the prototype. While this seems a clear and simple criterion, a closer look
shows that it is anything but that. When thinking of applications, we will think of
comparing, for example, various kinds of birds to a prototypical bird with respect to
physical appearance, primarily shape, size and colour. But, in fact, similarity in this
concrete sense of resemblance is not relevant in most cases. Consider, for example,
a wolf, a husky and a poodle. Wolves and huskies are physically very similar and
both very different from poodles. Yet the husky and the poodle belong to the same
category DOG and the wolf does not. The categorization of dogs and wolves is not only
a matter of physical similarity. Membership in the category DOG must be a matter
of similarity to the prototypical dog in some other respects. The question, then,
arises: in which regard is a would-be member of a given category to be similar to
the prototype? Which of its properties are relevant for the comparison? The answer
is far from trivial.
Consider a second case. Mary has two cousins, Marc and Tom (now you know).
They are brothers and physically very similar. Let us assume we know that Marc is
a fairly prototypical member of the category MAN. The physical resemblance of Tom
would then entitle us to categorize him too as a man. But if Marc is a butcher, Tom’s
physical similarity to Marc does not entitle us to conclude that he has the same
profession. Rather, the relevant aspect of Marc and Tom would be the kind of job they
have or they are qualified for. Categorization in the category MAN requires similarity
to the prototype in quite different respects from categorization in the category
BUTCHER, or in the category BACHELOR or NEIGHBOUR or HIP-HOP FREAK, for that matter.
These considerations show again that in order for prototypes to serve as reference
points of categorization they must be defined by a set of crucial features. And it is
these features that must be checked in order to judge the similarity to the prototype.
The features may be of different weight or interdependent. For example, for birds the
feature of being able to fly may depend on the body weight and the form and relative
size of the wings. In this sense, the resulting model of categorization is not the same
as the simple checklist model of NSC, where each feature is supposed to be necessary,
independent of the other features. But the difference is not as radical as PT makes it
appear.
Another difficult question concerns the relevant scale of similarity on which the
degree of membership depends. Let us assume similarity is measured in values
between 0 and 1. Given a prototype and the set of criteria relevant to the comparison,
it is clear which cases receive the value 1 on the scale of similarity: all those that
completely agree with the prototype in all relevant aspects. But in the other direction
the scale is undefined and, in fact, indefinable. Consider again the case of the category
DOG. Since wolves are definitely non-members of the category DOG, they must be
assigned the value 0.0 on the scale of similarity. But some of the crucial defining
276
Understanding semantics
properties of the prototypical dog must also be properties of wolves. Certainly
wolves are more similar to dogs than cows. For that reason, the categories WOLF and
DOG belong to the same superordinate category CANINE while COW does not. It then
appears we should assign cows the value zero and wolves some value reasonably
greater than 0.0. But, of course, this kind of adjustment of the scale can be repeated
arbitrarily. Cows are more similar to dogs than crabs are, because both cows and
dogs are mammals; crabs are more similar to dogs than potatoes, potatoes more
similar than stones, and stones more similar than, say, French personal pronouns.
The more kinds of cases we include in the assignment of similarity values, the
greater the value of wolves becomes and the more similar (in terms of the scale) they
become to dogs. It is just impossible to fix, for a given category, a general zero point
of the scale and degrees of membership for potential and actual members – except
for uncontroversial members. Apparently the degree of similarity and membership
depends on the given context, namely on the range of rival categories.
The considerations about the nature of the prototype, its defining properties and
the notion of similarity show that PT is not as unproblematic as it appears at first
sight. The claim that categorization is a matter of similarity to the prototype raises a
lot of non-trivial questions.
11.3 THE HIERARCHICAL ORGANIZATION OF CATEGORIES
11.3.1 The basic level
Look at Fig. 11.5.
Figure 11.5
What do you see? Your spontaneous answer will probably be, ‘A lion!’ Although
3
your answer is not hard to guess, the fact that it is not is far from trivial. The reason
why the unanimous answer is not trivial is that ‘A lion’ is by no means the only
possible answer to the question. You could have categorized the entity referred to
more specifically as a ‘male lion’, an ‘adult male lion’, a ‘prototypical male lion’ or a
‘marching male lion with a big mane’, ‘a marching big lion in side view’ and so on.
Alternatively it would be correct to use more general terms such as feline, mammal,
animal, living thing or entity. In the given context, the question can also be taken as
referring to the picture rather than what it depicts. So what you see when you look
3 You can test it yourself: Show twenty people a toothbrush, ask them to answer spontaneously to the
question ‘What is this?’, and twenty people will answer ‘A toothbrush.’
Meaning and cognition
277
at Fig. 11.5 is also a ‘picture’, a ‘picture of a lion’, the ‘MS Office™ ClipArt’s lion in
black and white’, an ‘illustration’, an ‘example’, etc. Even if we disregard the latter
kind of answer, the phenomenon remains that an object can always be categorized
at different levels of generality. What the little experiment shows is that there is
obviously one level of categorization that we prefer, an intermediate level somewhere
between very specific and very general categorization.
Berlin and others studied large plant taxonomies in a Mexican language and
managed to establish this preferred intermediate level. Rosch applied this kind
of research to other areas such as clothing and furniture. The main result is this:
there is an intermediate level, the so-called basic level which is in several regards
distinguished and privileged. Fig. 11.6 shows a minor part of the taxonomy of
English terms for musical instruments.
Figure 11.6
Categories of musical instruments – basic level
MUSICAL INSTRUMENT
basic level
TRUMPET
SAXOPHONE
JAZZ
BACH
TRUMPET TRUMPET
SOPRANO
SAX.
ALTO
SAX.
PIANO
…
GRAND
PIANO
UPRIGHT
PIANO
The basic level is occupied by the categories TRUMPET, SAXOPHONE, PIANO (as well
as all those not depicted here, such as VIOLIN, FLUTE, GUITAR, DRUM, ORGAN, CLARINET,
etc.). The subordinate level is by no means complete. Of course, it is not the lowest
level that exists. On the second subordinate level there are special categories of jazz
trumpets or soprano saxophones, and so on. But it should be noted that there is a
limit to levels of categorizations as far as lexicalized categories are concerned. The
subordinate level displayed here is probably the lowest level of categories in nonexpert terminology. Table 11.1 gives some further examples of the basic level.
Table 11.1
Examples for the basic level
Level
superordinate
basic level
Categories
GARMENT
VEHICLE
TROUSERS,
SKIRT, SHIRT,
DRESS, BRA,
JACKET,
subordinate
ANIMAL
COLOUR
BICYCLE, CAR,
DOG, CAT,
WHITE, BLACK,
BUS, TRUCK,
HORSE BEAR,
RED, YELLOW,
PIG, RABBIT,
GREEN, BLUE,
MOTORBIKE
…
…
BLUE JEANS
…
TIGER, LION,
RACING BIKE
…
COLLIE
…
…
BROWN,
…
TURQUOISE…
278
Understanding semantics
11.3.2 Properties of the basic level
11.3.2.1 The psychological aspect: properties of basic level categories
The basic level (BL) is privileged in the cognitive system in various ways. It operates
faster than higher and lower levels: in psychological experiments, response times
for BL categorizations are the shortest. Apparently it is the level at which most of
our general knowledge is organized. For example, the knowledge of what people
normally wear is in terms of BL garment categories, the composition of a jazz band
is in terms of BL instruments, and so on.
The BL is the highest level at which category members have a similar overall shape.
(Trumpets have a common shape and saxophones have, but not musical instruments
in general.) The overall shape is not only a matter of the visual appearance. An
important aspect of shape is the characteristic parts the object consists of. Most BL
terms have their own mereology (8.4.4). For example, a ‘piano’ has a ‘keyboard’; the
keyboard in turn consists of white and black ‘keys’ which each serve to move a felt
‘hammer’ inside the piano that hits one to three ‘strings’ thereby producing a tone of a
certain pitch. Similarly, the category of birds is characterized by having ‘feathers’ and
a ‘beak’ among other parts of their anatomy. No such common parts necessarily exist
for higher level categories such as ANIMAL or MUSICAL INSTRUMENT.
In the case of visible categories, it is possible to draw a general picture of the
members of BL and lower categories, e.g. a picture of a pig, a piano or a car, but not
so for any higher level categories.
BL categories are partly defined in terms of the ways in which we interact with
their members. For example, a trumpet, a piano and a harp are each played in their
own way and used for different kinds of musical activity. Shirts are put on and worn
in particular ways that distinguish them from coats or socks. For artefacts, the
characteristic parts (e.g. keys, wheels, sleeves) are intrinsically connected to the ways
in which we use the objects or in which they function. (Therefore not any arbitrary
piece of an object counts as a part of it, but only such that are linked to the way in
which the object functions or is handled. A key of a piano is a part of it, but not the
middle part of its lid.) Likewise, members of natural categories play a different role
in human life. In Western culture, pigs are held for slaughter, and cats as pets. Tigers
mostly live in the wild and if we encounter one, we will act differently than if facing
a reindeer, an ant or a snail.
11.3.2.2 The linguistic aspect: properties of basic level terms
The preference for BL categories in our cognitive system is reflected by the terms we
have for BL categories and the privileged role they play in communication. They are
mostly simple, short, native and old (except for terms for new categories of artefacts).
If BL terms are complex, long or of foreign origin, they are likely to be shortened and
assimilated (note the origins of bra < brassiere, bus < omnibus, piano < pianoforte,
bike < bicycle). In contrast, lower level category terms are often composite (cf.
wonder-bra, jazz trumpet, racing bike, olive green) and sometimes borrowed from
other languages. BL terms are learned first. And they are the ones used most
Meaning and cognition
279
frequently. Their use complies with one of the fundamental rules of communication,
the demand to be as informative as required and not to be more informative than
4
necessary.
11.3.2.3 The basic level and prototypes
The concept of prototypes works best for BL categories. The common overall shape
and the uniform ways in which we interact with the members of BL categories
make up the greater part of the prototype. BL categories combine a high degree of
distinctness from others with a rich specification of characteristics, thus exhibiting
a high cue validity.
11.4 CHALLENGES TO PROTOTYPE THEORY
In this section and the next, some issues will be pointed out that present challenges to
PT. The discussion offers the opportunity of addressing a number of basic questions.
The critical points are those of the right part of the schema in Fig. 11.4:
∑ the role of prototypes as reference points for category membership and,
consequently, the abandonment of necessary conditions
∑ the notion of graded membership
∑ the fuzziness of category boundaries
It will be argued that these points are not as firmly established and as closely
interconnected as was suggested in early PT. The phenomena PT addresses allow for
alternative interpretations and explanations. It should be emphasized, however, that
the critique does not concern the results concerning the hierarchical organization of
categories.
11.4.1 Graded membership vs graded structure
Some experiments of cognitive psychologists produced evidence in conflict with
PT. One such finding is that there are best examples even for categories with clear
boundaries and binary membership. For example, it was proved by experiments
that for the category ODD NUMBER the smaller odd numbers 1, 3, 5, 7 and 9 were
considered the best examples. But there are no borderline cases, and in no sense is
18764098376542141 less of an odd number than 3. There are two defining conditions:
an odd number must be a natural number > 0, i.e. a member of the set {1, 2, 3, 4,
…} and it must not be divisible by 2. Both conditions are necessary. Hence there is
no room for more or less similarity to the prototype(s). The number 4 is somehow
similar to an odd number in that it is not divisible by 3 (instead of 2), but this does
not count. Likewise, 12.999999 is almost identical with an odd natural number, but it
4 The so-called ‘Maxim of Quantity’, one of four conversational ‘maxims’ from Paul Grice. See chapter
1, Further reading, for references.
280
Understanding semantics
is no more of an odd number than 2 or π. Thus, the concept of similarity boils down
to the trivial borderline case of absolute equivalence. And as for that, any member
of the category would obviously do as an object of comparison. The main point, of
course, is that categorization in this case does not involve comparison at all. If we
want to decide whether a number is odd or even, we usually think of its decimal form
and check if the last digit is 1, 3, 5, 7 or 9. Thus, these five cases do play a crucial role
in categorization but not the role PT provides for them, i.e. not the role of serving as
reference points of comparison. The best examples of the category ODD NUMBER may be
considered prototypes (they certainly come to mind first and represent the category
best) and consequently the category itself has a graded structure. But, if these cases
are prototypes, they do not play the role of reference points for categorization. And
although the category has a graded structure, it has clear boundaries.
Similarly, in the case of the much-cited category BIRD one must admit that
penguins, although certainly odd birds, nevertheless are birds, not quarter-birds or
something. Likewise, for superordinate categories such as the broad category ANIMAL,
membership cannot be a matter of comparison with the prototype. There is no
similarity between an amoeba and a dog, not in any intuitive sense, yet an amoeba is
100 per cent an animal. For such categories, membership is a matter of definition, i.e.
of necessary conditions, although there are prototypes in the sense of best examples.
In general, the prototype model of categorization seems to work at the basic level and
below, but not above it. As a consequence, we have to state the following qualifications
of PT:
∑ A category may have prototypes, but they need not be reference points for
categorization.
∑ Graded structure is not necessarily linked with graded membership.
∑ Category membership is not necessarily a matter of similarity to the prototype.
∑ Category membership may be a matter of necessary conditions as assumed in
the NSC model.
The case of categories such as ODD NUMBER and ANIMAL shows that the two criteria
for prototypes – being the best examples and serving as reference points for
categorization – are, in fact, independent. Consequently, graded structure and
graded membership are not two sides of the same coin. Even if some members of
a category are more representative than others, all members may be 100 per cent
members separated from non-members by a well-defined boundary. Figure 11.7
schematically illustrates the case of a category with graded membership on the left,
which is necessarily accompanied by graded structure and a fuzzy boundary. The
right shows a category with yes-or-no membership and a resulting clear boundary,
yet with graded structure (such as ODD NUMBER or BIRD).
11.4.2 Fuzzy boundaries
From a semantic point of view, the question of fuzzy category boundaries and graded
membership is more important than the existence, nature and role of prototypes.
Meaning and cognition
281
Figure 11.7
Graded membership vs. graded structure
NO
YES
NO
graded membership
membership
NO
←
YES
→
NO
graded structure
It will be argued in 11.5 that for semantic categories, i.e. word and sentence
denotations, the assumptions of graded membership and fuzzy boundaries are
generally unacceptable. Since all examples hitherto mentioned (BIRD, CUP, etc.) are
semantic categories, this is indeed a strong claim. We will therefore take a closer look
at this part of PT.
There seem to be two reasons for the wide acceptance of the fuzziness claim.
(A) The claim appears to be in accordance with our intuitive impression that
category boundaries very often are in some sense fuzzy, unclear, variable or fluid. The
impression has several sources and, in fact, concerns different phenomena, which do
not all bear on the issue in the same way. The sources include:
∑ variation of word meanings within a language community (meaning variation)
∑ partial knowledge of word meanings (ignorance)
∑ sloppy use (pragmatism)
∑ inherent flexibility of word meanings (vagueness)
(B) The second reason is, of course, the evidence produced in experiments such as
the cups test by Labov. The evidence, however, is not as conclusive as it appears at
first sight. First, the outcomes of the experiments are in part a consequence of the test
design. Second, the results allow for alternative interpretations.
11.4.2.1 Sources of fuzziness
Meaning variation. As you will certainly have experienced countless times, people’s
use of terms actually differs. For example, one might have an argument with others
about where blue ends and green begins. Disagreement is even the rule for all
subjective categories such as BEAUTIFUL, BORING or FUN, which depend on personal
attitudes and evaluations. The application of personal standards for what is boring,
fun, etc. does, however, not mean that the resulting individual categories are fuzzy.
Apart from such subjective notions, there are words with which people connect
similar but different concepts. Think of categories such as ADULT. For some people,
282
Understanding semantics
age is the main criterion for adulthood, for others maturity, for yet others economic
independence.
Meaning variation plays a role in our intuitive notion of fluid category boundaries,
because in addition to the knowledge of word meanings we have a ‘theory of common
usage’, i.e. a theory of how other people use the words and how their uses vary. Such
a theory is an important part of our communicative competence, because in order
to make ourselves understood we must be able to anticipate how other people will
interpret what we say.
Ignorance. We all use so many words whose meaning we do not really understand. But
obviously, the resulting kind of uncertainty about the exact denotations of words is not
at issue in a general theory of categorization. PT is meant as a model for the categories
we have, not for those we have not. Nevertheless, the fact that there are so many
words we do not really understand contributes strongly to our subjective notion that
categories are fuzzy, and hence to the plausibility of the general fuzziness claim of PT.
Pragmatism. Whether something belongs to a category is one question, whether
it can be successfully referred to with the corresponding word is another. The first
issue concerns the order of things in one’s mind, the other is a matter of successful
communication. Since we do not have categories and words for everything we
may happen to want to talk about, we often find ourselves in need of referring
to something outside our system of semantic categories. If we then choose an
expression that in the strict sense does not apply to the intended referent but yields
a description that fits it approximately, this may well pass as regular language use. In
actual communication, we can expand the range of application for a given word well
beyond its original limits, provided we do not come close to the denotations of other
words. We often choose our terms rather sloppily. Thus, the practical denotation of
a word, i.e. the range of objects it can actually be used for, depends on the CoU and
5
is wider than the denotation defined by its lexicalized meaning.
Vagueness. If all these sources of fuzziness are excluded, the question remains of how
the proper source of fuzziness, variability of a particular, fully known, uncontroversial
category is to be modelled. PT has provided one model: categorization in terms of
graded similarity to the prototype. In 11.5.3 we will see that there is an alternative
explanation – vagueness – which is reconcilable with simple yes-or-no membership.
11.4.2.2 Test design and interpretation of the experiments on category fuzziness
In part, the results of the crucial experiments were predetermined by the very test
designs. When subjects are asked in a scientific experiment to assign membership
ratings to different types of furniture, they will produce a scale of the kind they
feel is required. The results will, however, not so much reflect their individual
categorizations as their willingness to compromise (pragmatism) and their theories
5 For a much more elaborate and systematic account of the ways in which we make use of our
linguistic abilities in actual communication, see the discussion of ‘variability’, ‘negotiability’ and
‘adaptability’ in Verschueren (1999), ch. 2.
Meaning and cognition
283
of common usage. Such an experiment is not suitable for testing whether for the
subjects membership is a yes-or-no matter or a matter of degree.
Another criticism concerns the nature of the attributes chosen for the tests. It is
questionable if the shape criterion which played such a prominent role in Labov’s
cup experiments is of much importance for the categorization of cups, bowls and
vases. Human artefacts are created for certain purposes. Cups, for instance, are made
for drinking coffee, tea, hot chocolate and other hot drinks. This is the central and
crucial feature of cups: ›for drinking hot drinks‹. From this single feature all other
characteristics of cups can be derived. As these drinks are hot, the material must
be sufficiently heat-resistant and insulating. A handle is a good idea if one does
not want to burn one’s fingers. The normal quantity of such beverages consumed
at a time determines the size of cups. The wish to have the drink while it is still
hot restricts the diameter of the container, and so on. The derivative features follow
from practical experience. Their presence optimizes the suitability for drinking hot
drinks from the objects. That is why they are features of the prototypical cup. None
of the derivative features constitutes a necessary condition. And there is no need to
consider any of them an extra part of the concept that defines the category CUP. But
then the question of whether one or more of them are absent has nothing to do with
the categorization. Likewise, the category VASE is defined by the feature that vases
are for putting cut flowers in. Thus, the impression of a continuum of membership
degrees between different categories such as CUP and VASE is an artefact of an
experiment that emphasized the derivative feature of shape rather than the primary
feature of function. The test design also suggested that the categories tested should
be mutually exclusive. But, in fact, the same object can be used both for drinking
coffee and putting flowers in (a fact that was confirmed by Labov’s results). When it
is used for coffee, it is a cup; when it hosts a bunch of flowers, it is a vase. Thus what
the experiments really show is that categorization is flexible and context-dependent
because it is, for things like cups, not primarily a matter of physical features. But they
do not prove that category boundaries are fuzzy.
11.4.3 Summary
What do the considerations in this section amount to? One of the basic observations
of PT cannot be denied. Certainly for many categories there are prototypes in the
sense of best representatives. More generally, many categories have more central and
more marginal members. In this respect the findings on colour terms carry over
to a large range of other categories. In addition, category boundaries are variable
and flexible. But we have to distinguish between variability in different senses.
Meaning variation, ignorance and sloppy use may contribute to our subjective
notion of category fuzziness. But only inherent variability of single categories in our
individual cognitive systems is relevant for a model of categorization. And this type
of variability can be explained differently, as will be argued in the next section.
Thus, while the observations are valid and challenging, the conclusions are not
necessarily those drawn in early PT. The experimental findings do not compel us to
accept that:
284
Understanding semantics
∑ ?? necessary conditions play no role in defining categories
∑ ?? category membership is generally a question of similarity to the prototype
∑ ?? category membership is generally a matter of degree
∑ ?? category boundaries are generally fuzzy.
Meanwhile, many cognitive semanticists have abandoned the original PT model.
Instead of prototypes they talk of ‘prototype effects’ in connection with graded
category structure. Richer and subtler models of categorization account for the
graded structure of categories (cf. the notion of ICM, ‘idealized cognitive model ’ in
Lakoff 1987). The structure then gives rise to prototype effects, for example typicality
of certain members of the category. In this volume, we will not go into these more
detailed developments of cognitive semantics.
11.5 SEMANTICS AND PROTOTYPE THEORY
11.5.1 Cognitive semantics
When we transfer the notions of cognitive science to semantics, we obtain the version
of the semiotic triangle displayed in Fig. 11.8.
Figure 11.8
Cognitive version of the semiotic triangle
expression
a concept
referent/
situation
The proposition of the sentence is a concept for the category of situations
potentially referred to, and the meanings of the referential phrases are concepts for
their potential referents. In relating a sentence to a particular situation and particular
referents, we categorize them accordingly. For example, if Mary says to John,
(3)
The tea is lukewarm.
she categorizes the situation as a the-tea-is-lukewarm situation, the beverage she
refers to as tea, its temperature as lukewarm. In general, each predication contained
in a sentence represents a categorization of its arguments, and the complex
predication expressed by the whole sentence amounts to a categorization of the
situation (recall 5.8).
In a very general sense, cognitive semantics can be used as a cover term for any
semantic approach that adopts this perspective on meaning and reference (i.e. for
Meaning and cognition
285
mentalist semantics in general). The range of cognitive semantics is smaller than the
range of cognitive science because, as was noted above, the categories and concepts
that can be expressed by words are only a subset of the categories and concepts that
form our cognitive systems. It is therefore useful to have specific terms for semantic
concepts, i.e. meanings, and semantic categories, i.e. denotations.
So-called prototype semantics (PS) applies PT directly to semantic categories.
If the notion is adopted that category membership is a matter of degree, then the
suitability of certain words for certain things is also a matter of degree. In the PS
perspective, we cannot say that we can or cannot use the word cup for referring to
some object x, but rather that x can well or hardly be called a cup. Graded membership
carries over to the whole sentence. A sentence is therefore not simply true or false.
If we identify true with 1.0 and false with 0.0, truth values lie somewhere on a scale
between 0.0 and 1.0 (including the two endpoints). For example, if Fido passes for
a 0.7 member of the category MONSTER, the sentence Fido is a monster will receive
a truth value of 0.7. As will be argued, this consequence is unacceptable from a
semantic and pragmatic point of view.
11.5.2 Polarization
Back in chapter 7, the Principle of Polarity was introduced: ‘In a given CoU where
its presuppositions are fulfilled, with a given reading, a declarative sentence is either
6
true or false.’ The principle now appears in a new light. As it stands, it plainly rules
out graded membership. Is it wrong? A simplification? Or is there evidence for it?
Yes, there is evidence. First of all, if there is any semantic universal, then it is
negation. Every language has grammatical and lexical means for negating a given
sentence. Thus, the possible sentences of a language, apart from a few exceptions
which need not bother us here, come in pairs of positive sentences and their
respective negations. As we saw in 7.1.3, determining the respective pairs is not
always trivial; we may tend to mix up negation with contraries, e.g. when we first
think of the negation of always as being never rather than not always. But it is possible
to determine such pairs, if the methods of logical analysis are properly applied. The
availability of negations for all positive sentences has the effect that whatever we say
constitutes a choice between the two alternatives, Yes or No (recall the discussion in
7.1.3 on negation). We called this phenomenon polarization.
Polarization is not restricted to declarative sentences. It also applies to all other
sentence types such as interrogative or imperative sentences:
(4)
a.
Why is there beer in the fridge? vs Why is there no beer in the fridge?
b.
Please, put beer into the fridge! vs Please, don’t put beer into the fridge!
6 The Principle of Polarity, as formulated here, provides for the possibility that a sentence lacks
a truth value because its presuppositions are not fulfilled. As we saw in 7.7.1, lack of a truth
value does not amount to an in-between truth value, but just to a truth value gap. Therefore, the
possibility of truth value gaps does not affect the argument we are going to develop here.
286
Understanding semantics
In this respect, it is even more general than the Principle of Polarity, which only
applies to declarative sentences. Polarization does, of course, not only apply at
sentence level, but it concerns every single predication contained in a sentence.
Imagine, for example, Mary telling John:
(5)
a.
The mouse is eating a big grain of corn.
In doing so, she not only opts for the positive alternative of the sentence as a whole,
but also for the positive alternative of the predications ‘is a mouse’, ‘is eating’, ‘is a
grain of corn’ and ‘is big’. (5) would not be true unless each of the predications is
true. Thus, when using predicate terms in composing a sentence, one puts a plus or
minus sign before each one and as a result also before the total predication expressed
with the sentence.
The general availability of negation is one piece of evidence for polarization. The
other, equally universal, is the possibility of transforming any declarative sentence into
an interrogative sentence. For Donald is a duck there is the interrogative counterpart
Is Donald a duck? and so on. The resulting questions require a simple type of answer:
Yes or No. Since the interrogative sentence has the same proposition as its declarative
counterpart (2.2.3) the fact that the possible answers to such questions are just Yes
or No proves that there are only two truth values for propositions, TRUE and FALSE, and
not a whole scale between them.
Consider now what would happen to polarization if membership in semantic
categories were a matter of degree. For example, let us assume that the creature
referred to in (5) is only a 0.8 mouse, that it is only 0.7 eating and the grain of corn is
only 0.3 big. Weighting the single predications appropriately, we might obtain a value
of, say 0.5, for the whole situation. (In models with graded membership, 0.5 is usually
considered a pretty high value, which justifies the categorization.) But under these
circumstances, we could with the same right claim the contrary of (5), namely that
the mouse is not eating a big grain of corn. Clearly this is not what we mean when we
opt for the positive sentence. We mean that it is true and not its negation. Even if the
situation we refer to is not exactly as we say, the way we put it is as we say it and not
the other way round. And this is what we will be taken to have said and is therefore
to be considered its meaning proper.
Polarization is inescapable. But it would not be, and probably would not even exist,
if membership in semantic categories were graded. It is therefore concluded that
semantic categories are binary.
Two questions arise now: (i) How can we account for flexible category boundaries,
if membership is yes-or-no? (ii) How does language cope with the obvious fact that
the world does not come divided into black and white but rather as a continuum of
phenomena without natural partitions? The answer to the first question is inherent
vagueness; the answer to the second is the availability of various linguistic means of
differentiation.
Meaning and cognition
287
11.5.3 Flexible concepts: vagueness
The crucial point concerning the flexibility of semantic categories is the fact that it
is built into the very word meanings. Let us consider a simple example, the adjective
big. It presents a prototypical case of what was already introduced as vague meaning
in 3.2.4. The vagueness of the adjective has two sources. First, whether we categorize
something as big or not big is a matter of a norm. If we consider the ‘big grain of
corn’ in (5) once more, the object may be called big if it is big for a grain of corn or
big for a mouse to eat or big for a grain of corn for this mouse to eat (the mouse may
prefer grains of corn of a certain size). These are three different norms. Once we have
chosen a particular norm, we next have to fix a borderline that separates big grains
from grains that are not big, i.e. we have to fix the criteria for a categorization by size
in terms of the chosen norm. Size is an attribute of objects that allows ordering on
a scale, with smaller objects ranging lower than bigger objects. Somewhere on this
scale there is a point beyond which everything is big, while everything below it is not.
This is illustrated in Fig. 11.9.
Figure 11.9
The big categorization
d
f
a
b
c
e
not big …
c
a
bigger
←NO
YES→
f
b
e
bigger
d
… big
bigger
Six objects of different size are shown in the upper part. In the lower part, they are
ordered on a scale of increasing size. We may draw the critical line between f and b as
well as at any other point of the scale. This is a matter of contextual appropriateness.
But wherever the line is drawn, it creates a yes-or-no division on the scale. Thus,
semantic concepts may be vague in the sense that the boundaries of the resulting
categories can be fixed in a flexible way. But in every given context they must be fixed
somehow and will then yield a simple yes-or-no categorization.
In the case of colour terms, too, categorization is a matter of norms. In the
neutral case, these are defined by the focal colours, but the norms may differ for
288
Understanding semantics
certain kinds of object (cf. red ears, blue lips, green apples, yellow leather, etc.).
When we use a colour term in a particular context, we choose the appropriate focus
hue and draw a circle, as it were, around it that marks the boundary of the resulting
denotation.
Many concepts owe their vagueness to single vague meaning components. For
example, among the three components of the semantic concept ›boy‹ – ›human‹,
›male‹, ›not adult‹ – the component ›not adult‹ is vague since it requires, and
allows, a line to be drawn within the continuum between youths and adults. In the
case of ›cup‹, the central criterion of suitability for hot drinks is vague, because
the suitability depends on secondary features that may vary considerably. The
mathematical concept ›odd‹ is not vague, while the everyday concept ›odd‹ is. In fact,
probably the majority of semantic concepts are in some way vague. This is not a flaw
of natural language but a great achievement because it allows a very flexible use of
the semantic material.
Vague concepts are not deficient. On the contrary, the meaning of big is quite
precise. Its flexibility is due to its adjustability to context. Like an adjustable lens,
it can be focused for the given purpose. But whatever focus we choose in a given
context, the concept then has a certain focus and polarizes accordingly.
The observation of inherent vagueness allows a different explanation for the
flexibility of category boundaries. The PT model explains it with a fixed prototype
and allowance for more or less similarity to it, with flexible use of fixed means, so
to speak. The inherent-vagueness model explains it as variable use of adjustable
means. A vague semantic concept such as ›big‹ allows us to choose an appropriate
norm in the given context (e.g. the average size of grains of corn or the average size
of things mice eat). Consequently, the concept ›big‹ determines a different category
in each context of use, e.g. BIG FOR A GRAIN OF CORN or BIG FOR A MOUSE TO EAT. Thus,
the concept ›big‹, in practice, provides us not with one category but with clusters
of categories, each cluster corresponding to the choice of a norm and alternative
decisions as to where to draw the exact borderline. Such concepts can cope both with
the requirement of flexibility and of polarization.
The PT model and the vague concepts model may be symbolized as in Fig. 11.10.
Figure 11.10
Prototypes vs. vagueness
category defined by
prototype and similarity
category clusters defined by
an adjustable concept
Meaning and cognition
289
In the PT model (left side), the arrows point to single candidate members, thicker
arrows indicating a higher degree of membership. In the other model, circles and
ellipses represent possibilities of drawing boundaries for the category, reflecting, for
example, different choices of norms for size combined with choices of drawing the
borderline between ‘big’ and ‘not big’.
If PS is to maintain the position of widespread graded category membership, the
phenomenon of polarization presents a serious challenge. PS would have to postulate
and model a general mechanism that turns the assumed underlying fuzzy categories
into binary semantic categories, whenever the respective words are used in a natural
language sentence in a particular CoU. This challenge, however, has not yet been
addressed in PT or later developments.
11.5.4 Means of differentiation
Language has several means to cope with the black-and-white nature of the semantic
category system. Three of them will briefly be mentioned.
Hedges. So-called hedges are often cited in PT as evidence for fuzzy category
boundaries. These are expressions that allow a general modification of a categorization
yielding either an expansion, (6a), or a narrowing of the categorization, (6b):
(6)
a.
A futon is kind of a mattress/ … is something like a mattress.
b.
John is a real bike freak.
Such phenomena do not prove that category boundaries are fuzzy and memberships
graded. (6a) does not simply mean that a futon is a mattress; rather it describes a
futon as something that is not a mattress, albeit similar to one. ‘Kind of a mattress‹ is
a vague concept derived from the concept ›mattress‹.
Lexical differentiation. Apart from these general means of differentiation, a
language community and individual language users have the possibility of adapting
the lexicon to the needs of categorization in communication. For example in the
domain of colours, the system of basic colour terms is supplemented in various ways.
There are:
∑ expressions for categories half-way between others: green-blue, blue-grey, etc.
∑ lexicalized hedge categories: whitish, reddish, bluish, greenish
∑ terms for more special categories: crimson, scarlet, vermilion, fiery red, blood-red
and hundreds more
Quantification. Another general means of differentiation is provided by
quantification (recall 4.5). Let us consider first a simple example without quantification:
(7)
a.
The eggs are raw.
b.
The eggs are not raw.
Understanding semantics
290
Together, the sentences form a clear alternative. The eggs referred to either fall
into the category RAW or they do not (recall the discussion on the Presupposition
of Indivisibility in 4.7.2.2). The alternative represents the simple all-or-nothing
contrast symbolized in Fig. 11.11 (where we assume that there are nine eggs; black
dots represent raw eggs).
Figure 11.11
All-or-nothing contrast
the eggs
are not
raw
the eggs
are raw
0
1
2
3
4
5
6
7
8
9
The alternative of (7a) vs (7b) obviously suits only the cases 0 and 9. Pragmatism
may allow the use of the sentences also for cases like 1 and 8, but this is not what is
literally asserted. The intermediate cases with some of the eggs raw and some not
are simply cut out, as it were. The alternative constitutes a system of two categories
that includes only the extreme cases. These are at the same time the simplest cases
in that all the eggs are alike. Although this may seem rather drastic, the strategy
of focusing on simple phenomena and disregarding the more complex is quite
representative of the way in which we make use of semantic categories. A certain
degree of simplification serves communicative economy.
We are not, however, forced to use this black-and-white mode of speaking. For the
grey shades, we can make use of quantification, by specifying the quantity of raw
eggs, or more generally, the number or portion of cases for which the VP is true.
(8)
a.
Some of the eggs are raw.
b.
Many of the eggs are raw.
(8a) is true if at least one egg is raw and false if none is. (8b) specifies the number as
great. Like big, the adjective many is vague and its truth depends on the choice of a
norm. If we compare the number of raw eggs to the rest, we may fix ›many‹ as ›five
or more‹. In other cases it might be appropriate to draw the division line at a lower or
7
higher number. In any event, two things happen when we quantify. First, the whole
range of possible cases between all and nothing is opened up and arranged on a
scale of quantity. Second, at some point of the resulting scale a division is introduced
between truth and falsity of the new sentence. Thus we receive a binary alternative of
categorizations that together cover the intermediate cases as well. This is illustrated
for many in Fig. 11.12.
7 You may wonder why case 0 is included within ‘not many’ and case 9 in ‘many’. If one knows the
exact number, one would express case 0 with no and case 9 with all. But the fact that it is acceptable
to say ‘not many, if any, …’ and ‘many, if not all, …’ proves that the two extremes no and all are
logically compatible with not many and many, respectively, and hence covered by their meanings.
Meaning and cognition
291
Figure 11.12
Quantification with many
not many
many
0
1
2
3
4
more
5
6
7
8
9
more
Note that the resulting picture is very similar to the one for big in Fig. 11.9. The
location of the crucial line that separates true from false, or membership from nonmembership for the quantifier category, depends on the choice of the quantifier. If
we replace many by some, the borderline would shift to between 0 and 1; for all, it
would have to be placed between the maximum and the one immediately below (8
and 9 here).
Modalization. Categorization is very often fraught with uncertainty. In order to
verbalize conditions in the world, we are often forced to make decisions about
how to categorize something without knowing exactly what we ought to know
for that purpose. Of course, language furnishes us with the means of expressing
degrees of certainty in connection with categorization. If I am not certain about
the appropriateness of putting things this way or the other, there are a lot of ‘modal’
expressions at my disposal: adverbs like perhaps, maybe, probably, certainly or sure;
modal verbs like may, will, could or must; or evidentials that qualify the source of
what I express.
(9)
a.
The T-shirt may be dirty.
b.
The T-shirt is perhaps dirty.
But even sentences containing such modal elements are polarized: it is or it is not
the case that the T-shirt is perhaps dirty, etc.
11.5.5 Summary
The adoption of the cognitive perspective as such is no doubt very useful. However,
one central point of PT, graded membership and the resulting concept of fuzzy
category boundaries is inappropriate for semantic categories. When people speak,
human language forces upon them the application of verbalized categorization in
a binary yes-or-no mode (polarization). Word and sentence meanings nevertheless
allow a highly flexible use because many meanings are vague and therefore adjustable
to the particular CoU. Since flexibility is built into the meanings themselves, variable
boundaries are compatible with yes-or-no membership. Apart from the availability
of expressions with a vague meaning, language has several devices that allow for a
differentiation of available categories.
292
Understanding semantics
Figure 11.13
Revision of the central claims of prototype theory
9
Prototypes are
the best examples
but
NO categorization by prototype
therefore
and
NO graded membership
9 Graded structure
but
NO fuzzy boundaries
Figure 11.13 shows the main results of our discussion of PT. If you compare this to
Fig. 11.4 above, you will realize that the claims of PT represented on the right side of
the figure are questioned. What remains of the original picture is the notion of graded
structure, reflected in the existence of better and less good examples. This observation
is very valuable and its explanation is the objective of much, and promising, research
in the area. What is to be refuted, however, is the idea that the graded structure is a
direct reflection of different degrees of representativity. Furthermore, it is certainly
necessary to provide for more than one model of categorization. There may be
categories for which membership is a matter of matching with the prototype (in
certain, relevant respects). For other categories, the NSC model may be more
adequate. Yet other models may be required for further types of categorys. Likewise,
the notion of fuzziness is not altogether inadequate for semantic categories. Of
course, vagueness and context-dependence give rise to ‘fuzziness’ in some sense.
But, at least for semantic concepts, the variability of category boundaries must be
reconciled with polarization, i.e. with binary membership.
11.6 SEMANTIC KNOWLEDGE
11.6.1 Personal knowledge vs cultural knowledge
In his textbook Language, written long before the emergence of cognitive science,
Sapir characterized the role of semantic categories in language use as follows:
The world of our experiences must be enormously simplified and generalized
before it is possible to make a symbolic inventory of all our experiences of things
and relations and this inventory is imperative before we can convey ideas. The
elements of language, the symbols that ticket off experience, must therefore
be associated with whole groups, delimited classes, of experience rather than
with the single experiences themselves. Only so is communication possible,
for the single experience lodges in an individual consciousness and is, strictly
Meaning and cognition
293
speaking, incommunicable. To be communicated it needs to be referred to a class
[ = category, S.L.] which is tacitly accepted by the community as an identity.
(Sapir 1921: 12f.)
The passage illustrates the necessity of distinguishing between personal knowledge
of a given category and the knowledge that makes up the category for the community
we are a member of. For example, every one of us knows several things about apples.
Some of what we know is personal knowledge, e.g. how the particular apples we ate
tasted, which role which sorts of apples play in our personal diet, where we can buy
apples in our neighbourhood, etc. But part of our knowledge is common property
of the cultural community we belong to. It includes the knowledge that apples are
fruits, what they look like, how they taste, that they contain vitamins, where they
come from, how they are eaten or prepared, how they are traded, what they cost,
that they are used to produce apple juice, etc. Let us call this knowledge cultural
knowledge as opposed to our personal knowledge. Cultural knowledge defines
what can be called cultural categories, the categories which are ‘tacitly accepted by
the community as an identity’. Names of cultural categories will be marked with
a subscript ‘C’. For example, APPLEC is the cultural category of apples, defined by
our present cultural knowledge of apples. Cultural knowledge is not to be taken as
‘everything that is known about the category’. It does not include expert knowledge.
For example, knowledge of the complex biochemistry of apples is not part of the
cultural concept. Personal knowledge usually includes the greater part of cultural
knowledge, depending on the range of experience of the individual. But we are all
some way apart from commanding all cultural knowledge, as for us all there are
8
many areas of everyday life where we have little experience.
Given that the speech community makes use of cultural categories, the question
arises of how much of it constitutes the meaning of words. Is the meaning of the word
apple the concept for the cultural category APPLEC? If so, every detail that belongs
to the cultural knowledge of apples is part of the meaning of the word. If not, the
meaning of a word is only part of the cultural concept. Aspects such as what apples
normally cost in the supermarket, what sorts of apple one can buy there and how
one prepares baked apples might then not be considered components of the word
meaning.
The distinction between semantic knowledge and ‘world knowledge’ (i.e. cultural
and personal knowledge) is a doctrine of traditional semantics. It can be plausibly
argued that we do not need to have the total cultural knowledge about apples,
computers, mice or alcohol in order to know the meaning of the words apple,
computer, mouse and alcohol. Many researchers of the cognitive orientation are,
however, pleading for abandoning the distinction. They argue that our semantic
categories are interwoven with our overall cognitive system in a way that does not
8 We have chosen the term cultural knowledge instead of world knowledge and encyclopedic
knowledge, either of which you may frequently encounter. Often the notion world knowledge is used
indiscriminately for cultural knowledge and personal knowledge, while encyclopedic knowledge is
understood as including expert knowledge.
294
Understanding semantics
allow their separation. Also, they argue, the full meaning of a word can only be
grasped in the context of larger parts of the cognitive system. This view resembles
the structuralist position in emphasizing the interdependence of categories, but
it does not primarily relate to semantic categories. Rather it aims at the way in
which our categories, among them the semantic categories, are integrated into a
full model of the world, the world as we perceive and interpret it, in which we are
placed ourselves, interacting with it. In the following, we will nevertheless argue
that a distinction between cultural knowledge and semantic knowledge is necessary,
important and feasible.
11.6.2 The apple juice question
Ungerer and Schmid (2006), a recent introduction to cognitive linguistics, which
focuses on cognitive semantics, explicitly subscribes to an approach that tries to
capture meaning in terms of the experiential knowledge that laypersons connect
with words. The approach is representative for a position contrary to a distinction
between cultural knowledge and semantic knowledge. Let us call it the ‘cultural
knowledge approach’ (CKA, for short). The authors report on an experiment for
determining what makes up the category APPLE JUICE (Ungerer and Schmid 2006:
92f.). Lay subjects were asked to list features of apple juice. It was then checked
whether these features could also be assigned to either or both or neither of the
categories APPLE and JUICE, in order to determine where the features stem from. The
result is given in Table 11.2. In addition to these properties, ›made from apples‹ is
listed as ‘salient specific attribute’ (Ungerer and Schmid’s term attribute is replaced
by feature in order to keep to the terminology chosen here).
Table 11.2
Features of the category APPLE JUICE according to Ungerer and Schmid (2006: 93)
Salient specific feature: ›made from apples‹
features that the category APPLE JUICE shares with the categories …
1 … JUICE
›liquid‹
›no alcohol‹
›thirst-quenching‹
›supplied in bottles or cartons‹
›served in glasses‹, etc.
2 … JUICE and APPLE
›sweet or sour-sweet‹
›healthy‹
›tastes good‹
3 … APPLE
›yellow or similar colour‹
›fruity‹, etc.
4 with neither
›mixed with soda water‹
›naturally cloudy‹
Meaning and cognition
295
The word apple juice is a regular compound of the two nouns apple and juice.
According to the ‘classical’ analysis briefly sketched in (8.1.2), its meaning is the
meaning of the head noun plus a specification added by the modifier. In favour
of CKA, Ungerer and Schmid argue that their results provide evidence against the
traditional analysis. If it were correct, they claim, most features of the category APPLE
JUICE should stem from the head category, only one should be contributed by the
modifier (›made of apples‹) and no extra features should hold for the compound only.
The latter requirement follows from the fact that the meaning of regular compounds
should completely derive from the meanings of their parts. Ungerer and Schmid
point out that, however, the category APPLE JUICE is linked to the category APPLE
by quite a number of further common features (rows 2 and 3) and exhibits some
genuine features of its own (row 4).
While the results of the experiment are as they are, it must be questioned if the
subjects described what for them constitutes the meaning of the word apple juice.
Apparently they described the cultural knowledge they commanded and held
relevant. However, a critical look at the features listed shows that most of the features
elicited cannot be considered parts of the meaning of the word.
How can we decide if a particular feature of a category is part of the meaning of
the corresponding word? There are (at least) two tests. First we can check if it is
necessary to know that this is a feature of the category in order to know what the
word means. Second, we can carry out the following thought experiment: if the actual
members of the category happened to lack the feature in question, could we still
use the word for referring to them? Such thought experiments are relevant because
the meaning of a word must fit all potential referents. There is no way of exploring
concepts other than by asking oneself what they would possibly cover.
Let us apply the tests to the critical features in Table 11.2, i.e. those in rows 3 and
4, which according to Ungerer and Schmid contradict the classical analysis. For the
thought experiment, imagine the food industry decides that apple juice needs a
radical change in order to sell better. Thus they filter the stuff, add blue colour and
heavy peppermint flavour and promote it as a definitely-no-soda drink. It would
cease to have any of the critical features and would yet be apple juice, because it
would still be made from apples. The result of the experiment is in accordance with
what we find when we ask ourselves what one needs to know about apple juice (the
stuff) in order to be entitled to claim knowledge of the meaning of the word apple
juice. One does not need to know its colour or if it is cloudy. In fact, as we all know,
there is cloudy as well as clear apple juice. Did you know that potato juice from
ordinary potatoes turns reddish in a few minutes? If not, would you say you did not
understand the word potato juice in the previous sentence? The other features in the
table have the same status. One need not know the taste of apple juice in order to
know the word meaning. (Have you ever tasted potato juice?) And one need not know
the soda water thing. There is one feature that really carries the whole load: ›made of
apples‹. The word apple juice means ›juice made of apples‹. Period. If one knows that,
one knows the meaning of the word. Apple juice being quite a common sort of drink,
the other features are likely candidates for the cultural knowledge about apple juice,
but they do not form components of the meaning of the word apple juice.
296
Understanding semantics
Thus the classical semantic analysis of compounds is not disproved by the
experiment. Rather, the discussion proves the importance of distinguishing between
cultural knowledge and semantic knowledge.
11.6.3 Cultural knowledge vs semantic knowledge
From this the following picture emerges. To any word we know (e.g. apple juice), we
connect in our personal cognitive system a semantic concept, its meaning (›apple
juice‹ = ›juice made of apples‹). The meaning determines a fairly broad category, the
denotation of the word. Let us mark denotations, or semantic categories, with the
subscript S for ‘semantic’. The category APPLE JUICES contains all possible variants of
apple juice including the clear, blue, mint-flavoured, no-soda variant of our thought
experiment. The category APPLE JUICES is broader than the presently valid cultural
category APPLE JUICEC which is characterized by additional features such as those
elicited in the experiment reported in Ungerer and Schmid (2006). The cultural
concept that represents APPLE JUICEC is correspondingly more specific than the leaner
concept that constitutes the word meaning. This is an iron law of cognition: the more
specific a concept, the narrower the category it represents, and vice versa. Thus, the
meaning of the word apple juice, the semantic concept ›apple juice‹, is part of the
cultural concept for apple juice, but the cultural category APPLE JUICEC is a subcategory
(!) of the semantic category APPLE JUICES. In Fig. 11.14, the cultural concept and the
cultural category are integrated into the semiotic triangle. Note how the more specific
cultural concept includes the leaner semantic concept, while, conversely, the cultural
category is only a subset of the semantic category.
Figure 11.14
The semiotic triangle integrating cultural knowledge
expression
cultural concept
meaning
determines
cultural
category
represents
The cultural category can be roughly equated with the actual denotation of the
word, i.e. those members of the total denotation that we encounter, or expect to
encounter, in actual life. These form only a subset of those we might conceive of, i.e.
Meaning and cognition
297
all potential referents of the word. Theorists that equate semantic knowledge with
cultural knowledge equate the actual denotation with the total denotation.
An approach that distinguishes between semantic knowledge and cultural
knowledge, or between the meaning of a word and the cultural concept into which it
is embedded, is superior to CKA in offering an explanation for the following points:
∑ the stability of word meanings as compared to cultural concepts
∑ the abstractness of word meanings as compared to cultural concepts
∑ the communicational economy of word meanings
∑ the simplicity of meaning relations between words
Let us conclude this section with a brief illustration of these points.
Stability. Cultural knowledge is subject to constant change. For example, certain
categories of artefacts such as TELEPHONE or COMPUTER have changed to a degree
that people of, say, two generations back, would have been unable to anticipate.
Nevertheless, the telephones and computers of the 1950s are no less members of the
respective semantic categories than today’s telephones and computers. Both are 100
per cent members of the semantic categories TELEPHONES and COMPUTERS. While the
cultural categories have shifted, e.g. with the invention of the mobile telephone or
the personal computer, we do not feel that the words telephone and computer have
changed their meanings. Rather word meanings form a stable core of the changing
cultural concepts. Constant meanings allow for communication with constant means
in an ever-changing world. Of course, meanings are not completely constant. The
lexicon keeps adapting to the changing communicational needs, but at a much larger
temporal scale.
Abstractness. The lack of many concrete features that narrow down the cultural
categories also accounts for the abstractness of semantic concepts, which allows their
application to uncommon cases outside the familiar cultural categories. One example
was the applicability of the semantic concept ›apple juice‹ to exotic fictitious varieties
of apple juice. Similar thought experiments could be conducted for almost all other
semantic categories.
Economy. Although the greater part of cultural knowledge does not form part of
the word meaning, it is nevertheless relevant to communication. In conversation,
we draw heavily on all sorts of common knowledge. This allows us to keep our
messages semantically to a minimum. For example, if John says (10a) to Mary,
indicating thereby that Mary should bring him some apple juice, he will rely upon
his and Mary’s common cultural knowledge about the packaging of apple juice. If
John’s apple juice is usually packed in bottles, he need not say (10b), yet Mary will
understand as much and be able to locate the juice among the things on the balcony.
(10) a.
b.
There is apple juice on the balcony.
There is apple juice in bottles on the balcony.
298
Understanding semantics
The fact that Mary understands something like (10b) when John says (10a) does
not mean that ›is in bottles‹ is part of the meaning of the word apple juice. On the
contrary, the feature need not be part of the word meaning because it is part of the
additional cultural knowledge.
Meaning relations. Last but not least it must be stated that CKA approaches blur
rather than explain meaning relations such as those between a compound and its
parts. The simple meaning relation between the head and the compound is buried
under a heap of cultural features.
11.7 SUMMARY
This chapter took us through a series of fundamental issues. First, the notions
of meaning as a concept and denotation as a category were embedded into the
larger context of cognitive science. Here the most basic issue is the general nature
of categorization. The traditional view, represented, for example, by the BFA,
considers membership in a particular category a matter of a fixed set of necessary
conditions to be fulfilled by its members. The view implies that membership in
a category is a clear yes-or-no issue. Given a particular category, anything either
belongs to it or it does not. Hence, all members of a category enjoy equal member
status. As a consequence, a sharp category boundary separates members from
non-members (11.2.1). This view was challenged by PT. No! PT says, category
boundaries are not sharp, but fuzzy; there are better and poorer members; in
particular there are best examples: the prototypes. On the other hand, categories
may have very poor members. Thus, no clear distinction can be drawn between
members and non-members, rather membership is a matter of degree. As for
the defining ‘necessary’ conditions, even the most typical ones such as ability to
fly in the case of birds may be violated. So: dispense with necessary conditions!
Categorization is not a matter of checking conditions but is carried out by a
comparison with the prototype (11.2).
While all this has its appeal, the PT approach proved problematic under closer
scrutiny. For one thing, its results do not apply to categorization as generally as
was claimed and are not as cogent as it first appeared (11.4). When PT is applied to
semantics, severe problems arise with the notion of graded membership: it is in direct
conflict with polarization, probably the most fundamental semantic phenomenon
(11.5.2). Polarization totally pervades word and sentence meaning. It underlies each
predication – and therefore almost all word meanings. And since each sentence
contains at least one predication (because it must contain a finite verb and all verbs
are predicate terms), each single sentence is subject to polarization. Consequently,
categorization by verbal means is always binary, i.e. a yes-or-no matter. Despite such
problems, PT did contribute valuable results to semantics. The observation that
for most categories we operate with prototypes, for example as default cases when
no further specification is available, is an important result. A further important
contribution of PT is the detection and specification of the basic level (11.3).
Meaning and cognition
299
In tackling the PT claims about fuzzy category boundaries and graded membership
we obtained insights into the nature of word meanings and the ways in which
language, this polarizing (or polarized) cognitive equipment, copes with the
non-binary character of the ‘world’. For one thing, vague meanings enable a
flexible categorization. There is no need for concepts that would provide fuzzy
category boundaries, because the word meanings available enable us to adapt
category boundaries to the requirements of communication (11.5.3). In addition
to flexible word meanings, other linguistic devices such as lexical differentiation or
quantification allow for relativization of what would otherwise be heavy black-andwhite thinking (11.5.4).
No doubt the main merit of PT is its focus on the conceptual level. The other two
mainstreams of semantics, the structuralist tradition and formal semantics (to be
introduced in the final chapter), avoid addressing meaning itself. The structuralist
tradition approaches meaning indirectly via the investigation of meaning relations.
(Recall the claim of structuralism that the meaning of an expression is the sum of its
relations to other expressions within the system.) Formal semantics tries to capture
meaning equally indirectly by investigating reference and truth conditions.
Addressing the conceptual level itself raises another fundamental problem, the role
of linguistic meaning within our overall cognitive system (11.6). Are the concepts
that are assumed to constitute the meanings of words identical with the concepts
that define our everyday categories? Given that language is a general means of
communication within a cultural community, a distinction must be drawn between
the knowledge at our personal disposal and that part of it which we may assume to
be (by and large) shared by the other users of the language. The latter was dubbed
‘cultural knowledge’ as opposed to ‘personal knowledge’. A closer consideration of
regular compounds led to the conclusion that cultural knowledge must be further
distinguished from semantic knowledge. Semantic concepts are much leaner than
the corresponding cultural concepts. Only this assumption allows an account of basic
phenomena such as simple meaning relations and the stability of word meanings.
Within the given limits, it is not possible to present all of the major cognitive
approaches to semantics in a way that would do justice to the whole field. One area
should be mentioned where modern developments of cognitive semantics are very
successful: the investigation of polysemy and, closely related, semantic change.
In particular, it was shown that metaphor and metonymy are general cognitive
mechanisms that play a central role in meaning variation and historical meaning
shifts. In very many cases, meaning variants of the same word are linked by a
metaphorical or metonymical relation (or both).
We focused on the earlier claims of PT because this allowed us to address several
issues of central importance. Meanwhile ‘Cognitive Semantics’ is an established field
of research and theory. But it must be stated that it is still far from a full-fledged
semantic theory. Central fields of semantics have never been elaborated within the
cognitive framework. These include meaning relations and composition. In part this
is a consequence of particular assumptions. The assumption of widespread graded
membership would render an account of the composition of sentence meaning
very difficult. Similarly, the identification of semantic knowledge with cultural
300
Understanding semantics
knowledge impedes the proper treatment of meaning relations (11.6.2). But none
of these assumptions are necessary components of a cognitive semantic theory. In
the next chapter, a different theory of cognitive representation promises to be able
to overcome some of the basic shortcomings of the older approaches discussed here.
EXERCISES
1. Explain the difference and relation between concepts and categories.
2. Explain the difference between mushroom, MUSHROOM and ›mushroom‹.
3. What kind of entity is a prototype?
4. What is the role of prototypes in PT?
5. What is problematic with the notion of ‘similarity to the prototype’?
6. What is meant by ‘polarization’ and why does it pose a serious problem to
prototype semantics?
7. Perform an experiment asking several people a question like the one at the
beginning of 11.3.1. Discuss the result.
8. Consider the following terms. Using the criteria mentioned in 11.3.2, try to assign
the terms to the basic level, to a more general (higher) or to a more specific level.
television set, radio, toaster, washing machine, portable TV, cell phone, iPhone,
household appliance, PC, notebook (electronic), electric appliance
9. Ask three people spontaneously to write down 30 terms for clothing in the order
in which they come to mind.
a. Try to assess which kinds of clothing are prototypical.
b. Try to arrange the terms in a taxonomy. Which ones are basic level terms?
10. Discuss the distinction between personal knowledge, cultural knowledge and
semantic knowledge using the word money as an example. Try to find features
that (a) are part of your personal knowledge about money, but not of the cultural
knowledge and (b) are part of (present) cultural knowledge but not necessarily
part of the meaning of the word money.
FURTHER READING
For a discussion of PT see Aitchison (2012, chs 5, 6); Ungerer and Schmid (2006,
ch. 1); and Lakoff (1987, ch. 2) for its history. On the hierarchical organization of
the category system see Ungerer and Schmid (2006, ch. 3). On the role of metaphor
and metonymy in cognitive semantics, see Ungerer and Schmid (2006, ch. 4), Palmer
(1996, ch. 8), Foley (1997, ch. 9) and the case study on the category ANGER in Lakoff
(1987: 380–415).
12
Frames
In this book, it is assumed that meanings are concepts. We have talked a lot about
the content of these concepts and the relations among them. We have learnt about
general aspects of categorization and concept formation. But up to now, no proposal
has been presented as to what these concepts are like: What is their structure? What
are their conceptual components? This will be finally done in this chapter, though the
answer to these questions is, necessarily, preliminary and incomplete. There is one
theory of concepts in cognitive psychology which has come up with a concrete model
1
of concepts: Lawrence W. Barsalou’s theory of cognitive frames. The theory claims
that all concepts in human cognition are ‘frames’ and presents various empirical
evidence and theoretical arguments in favour of this claim. Barsalou’s theory of
cognitive representations is superior to the binary feature approach (BFA) as well as
to prototype theory. Unlike BFA, it is much less restricted; unlike prototype theory, it
is explicit about the internal structure of concepts. It is very attractive for linguists,
as it offers solutions to many problems of decompositional semantics; it also offers
the possibility of tackling semantic composition. The application of frame theory
in semantics and in linguistics in general is still under development, but there is
enough to be said about this promising theory of cognitive representations in order
to include it here.
12.1 BARSALOU FRAMES
12.1.1 Chunks of knowledge
The notion of ‘frame’ belongs to a family of similar notions, including ‘schema’,
‘scenario’ and ‘script’, that came up in the 1970s in the fields of artificial intelligence
and cognitive science. These were notions for the ready-made complex ‘chunks of
knowledge’ which we apply in everyday life. For instance, there is the schema (or
frame) ‘university’; it consists of the typical ‘ingredients’ of a university, like its
campus, its faculties, its teachers, lectures, students, courses and study programmes,
examinations, etc.; or the ‘family’ frame that defines the members of a family,
the relationships between them, their responsibilities, etc. There is the ‘script’ or
‘scenario’ of a children’s birthday party or of dining at a restaurant, shopping in a
supermarket, taking an exam or getting married. Such a framework is defined by
constitutive ingredients such as persons in specific roles, engaged in particular
1 The theory was introduced in Barsalou (1992a, 1992b).
302
Understanding semantics
interactions, pursuing certain aims, etc.; their actions will involve certain objects
and locations, they may be undertaken in a particular order, etc. Just think about
eating in a cafeteria: we have the kind of location with its fittings and furniture that
would make up a cafeteria, the staff and customers that will be there, the kind of
food and beverages you may find, the procedure applied to get your meal, pay for it,
take a place, eat and drink and dispose of your rubbish. Chunks of knowledge like
the cafeteria schema form our general world knowledge. They are tacitly employed
as background knowledge when we mention visiting a cafeteria or other activities.
There is no doubt that our cognitive systems host a vast number of such chunks of
knowledge which are interwoven which each other and allow us to activate large
networks of knowledge in a very short time.
In the literature on artificial intelligence or cognitive linguistics, no precise
2
formal definitions are provided for a general format of such chunks of knowledge.
Barsalou was the first to come up with a model of considerable precision and with
the claim that this is the actual format in which our minds organize concepts to
represent the categories of things and phenomena in the world. Barsalou’s is not
primarily a theory of semantic representations. Much of his empirical work was
spent on concept formation in particular contexts and experimental settings. He
emphasizes that concept formation is flexible and able to adapt to the needs and
circumstances of a given situation. Concepts can be arbitrarily enriched or thinned
out or otherwise modified. This general flexibility does not apply, though, to stored
lexical meanings, which is what we will apply frame theory to. Lexical meanings
form stable cognitive entries only subject to the general process of life-long
language learning.
Barsalou frames are cognitive representations of single objects, either individual
objects or objects as members of a category. Compared to these frames, the chunks
of knowledge envisaged in artificial intelligence are macrostructures that integrate a
greater number of objects, events, actions, etc. in a much more complex whole.
12.1.2 The passport frame
The general notion of a frame (henceforth ‘frame’ is to be understood as ‘Barsalou
frame’) can be illustrated by means of an example we all know: the information
that is provided in a passport for the identification of its bearer. Passports issued by
a certain country use a fixed paradigm of description for the holder. The example
we will give is based on the European Union passport (as of 2012). It contains the
following data on its bearer: name, given names, nationality, date of birth, place
of birth, sex, residence, height and colour of eyes; these are specified in words. In
addition, there are two non-verbal pieces of information: a token of the bearer’s
signature (this is primarily non-verbal because it’s the mere written shape that
matters, readable or not), and a photograph of the bearer’s face. In the context of
frames, these parameters of description are called attributes; we will use small
2 For example, Kövecses (2006: 64) gives the following working definition of frames: ‘A frame is a
structured mental representation of a conceptual category.’
Frames
303
capitals for attributes. The attributes take values, e.g. the value Angelika for the
attribute GIVEN NAMES or the value ‘female’ for the attribute SEX. The values can be of
various kinds, depending on the attribute. It may or may not be possible to describe
the values in words (a signature or a face cannot be described in words, although it
can of course be referred to). Figure 12.1 gives the data in a fictitious passport in a
so-called frame matrix.
Figure 12.1
Frame matrix for an EU passport bearer (1)
bearer
.NAME
.GIVEN NAMES
.NATIONALITY
.DATE OF BIRTH
.PLACE OF BIRTH
.SEX
.HEIGHT
.COLOUR OF EYES
.RESIDENCE
.SIGNATURE
.[FACE]
: Postowski
: Angelika
: German
: 03.08.1971
: Bottrop
: female
: 178 cm
: blue
: Köln
: (a signature)
: (a photograph)
The bearer appearing as the only entry in the first column is the object that the
whole frame describes; all entries in the second column relate to this item. The
second column of the matrix lists the attributes, the third column the values they
take. Each line represents one piece of information about the bearer; for example
‘bearer.NAME: Postowski’ which is to be read as ‘the name of the bearer is Postowski’,
or ‘bearer.COLOUR OF EYES: blue’. The attribute FACE is given in square brackets because
it does not appear explicitly in the passport.
The matrix contains eleven attributes of the bearer and their respective values,
as far as they can be verbalized. Four of the attributes take a complex formulation:
DATE OF BIRTH, PLACE OF BIRTH, COLOUR OF EYES and SIGNATURE OF BEARER. The last one
makes explicit the possessor of the attribute which is left implicit in all the other
cases. The other three entries can be decomposed into two steps of attributing. The
term ‘colour of eyes’ refers to the colour of the eyes of the bearer; this is, in a first
step, the value of the attribute COLOUR of the eyes of the bearer; in a second step, the
eyes of the bearer can be analysed as the value of the attribute EYES of the bearer.
Thus, the attribute COLOUR OF EYES is the result of a chain of two attributes; it is the
attribute COLOUR of the value of the attribute EYES of the bearer. This decomposition
is given in (1a). The attributes PLACE OF BIRTH and DATE OF BIRTH can be analysed as
involving the attribute BIRTH of the bearer. They can be analogously decomposed as
in (1b, c):
304
(1)
Understanding semantics
a.
bearer.COLOUR OF EYES: blue
= bearer.EYES: [bearer’s eyes].COLOUR: blue
b.
bearer.DATE OF BIRTH: 03.08.1979
= bearer.birth:[bearer’s birth].date: 03.08.1979
c.
bearer.PLACE OF BIRTH: Bottrop
= bearer.birth:[bearer’s birth].place: Bottrop
The attributes BIRTH and EYES open two small subframes for their respective values:
one represents the birth of the bearer with the attributes DATE and PLACE, the other, the
eyes of the bearer with the only attribute COLOUR. These subframes can be unfolded
in the frame matrix as in Fig. 12.2. Square brackets indicate implicit attributes and
values.
Figure 12.2
Frame matrix for an EU passport bearer (2)
bearer
.NAME
.GIVEN NAMES
.NATIONALITY
.[BIRTH]
: Postowski
: Angelika
: German
: [bearer's birth]
.SEX
.HEIGHT
.[EYES]
.RESIDENCE
.SIGNATURE
.[FACE]
: female
: 178 cm
: [bearer's eyes] .COLOUR :blue
: Köln
: (a signature)
: (a photograph)
.DATE
.PLACE
: 03.08.1971
: Bottrop
The example illustrates a very important property of frames: they are in principle
recursive, i.e. frames may contain embedded subframes which add information on
the value of an attribute by adding to it its own attributes and their values.
The basic unit of frame description is the ascription of an attribute and its value
to either the referent of the whole frame, or to the value of an attribute in the frame:
a piece of information of the form ‘_____.ATTRIBUTE: value’. Frames can not only
be represented by matrices, but also by graphs of nodes and arrows. This mode of
representation is more space-consuming, but it is easier to comprehend. In a frame
graph, the object of representation and the values of attributes are represented as
nodes. The node representing the object of description of the whole frame is encircled
with a double line; it will be called the ‘referent node’. Attributes are represented by
arrows pointing to the value; the arrows carry a label that names the attribute. Thus,
the basic unit in a frame graph takes the form shown in Fig. 12.3. The node to which
the attribute is applied is represented by a broken line circle; we will refer to it as the
‘possessor’ of the attribute.
Frames
305
Figure 12.3
Attribute-value element of a frame graph
(possessor)
value
ATTRIBUTE
Figure 12.4 displays the graph equivalent of the frame matrix in Fig. 12.2.
Figure 12.4
Frame graph for an EU passport bearer
NAME
GIVEN NAMES
NATIONALITY
DATE
[BIRTH]
PLACE
bearer
SEX
HEIGHT
[EYES]
COLOUR
RESIDENCE
SIGNATURE
[FACE]
A passport contains not only information about the bearer, but also about the
document itself. It specifies, among other details, a unique passport number, a type
( = ‘P’), dates of issue and of expiry, the issuing authority – and of course the bearer.
The bearer is an attribute of the passport, and thus the whole bearer frame can be
regarded as embedded in a superordinate passport frame. This is depicted in the
frame matrix in Fig. 12.5. The frame box containing ‘[bearer]’ indicates the frame
matrix in Fig. 12.2.
Understanding semantics
306
Figure 12.5
Frame matrix for an EU passport
passport .NUMBER
.TYPE
. DATE OF ISSUE
. DATE OF EXPIRY
. AUTHORITY
.[BEARER]
: xxxxxxxxx
:P
: 16.03.2012
: 15.03.2022
: Stadt Köln
: bearer
….
….
….
As we saw with the examples of DATE OF BIRTH and COLOUR OF EYES, attributes can
form chains; the general structure is as in (3a), graphically as in (3b).
(2)
a.
possessor.ATTRIBUTE1:value1.ATTRIBUTE2:value2.ATTRIBUTE3:value3 …
poss
b.
value1
ATTR1
ATTR2
value2
ATTR3
value3
…
The passport frame is an example of what is called ‘institutional categorization’, i.e.
a socially established standardized way of categorization. There are countless other
examples essentially in frame format, e.g. telephone registries or library catalogues,
where each entry amounts to a frame format description with the same attributes.
We actually used a frame in 3.1 above when we defined the notion of the lexeme. It
can be transformed into a frame matrix such as in Fig. 12.6.
Figure 12.6
Frame matrix for the lexeme ‘teacher’
lexeme
.SOUND FORM
.WRITTEN FORM
.MEANING
.CATEGORY
: ['ti:t]
: 'teacher'
: ›teacher‹
: noun
In this lexical frame, the entry for the value of the attribute MEANING would have to
be filled by the frame representing the complex concept for a teacher.
12.1.3 The basic structure of a frame
A frame is the description of a potential referent. This can be a particular object,
as in the passport example, or an arbitrary member of the category described – if
the frame is not specific enough to single out a particular case. In the applications
Frames
307
considered in the following, frames will always be of the latter kind. The potential
referent of the frame is represented by the referent node in a frame graph or by the
only element in the first column of a frame matrix. The frame contains attributes;
each attribute assigns a value to the potential referent; further attributes may assign
values to these values and so on. The frame structure is subject to a set of uniqueness
conditions; these are the essential conditions that define a frame. For different types
of concepts there may be additional conditions imposed on their structure:
DEFINITION 1 Frame
A frame is a conceptual network of attribute-value assignments that fulfils the
following uniqueness conditions:
UR Unique frame referent
There is a unique element that represents the potential referent of the
frame. Every element in the frame is connected to the frame referent by a
chain of attributes.
UA Unique attributes
For each element in the frame, an attribute is assigned no more than once.
UV Unique values
For each element in the frame, and each attribute applied to it, the
attribute takes a unique value.
If you take a look at the examples, you see that UR is fulfilled: there is a particular
element representing the referent and all other elements are connected to it. In a
frame matrix, the central element is represented by the only entry in the first column;
in a frame graph, the central element is the node marked as a double-lined node,
from which every other node can be reached.
For each element in the frame, the same attribute is never applied more than once
(UA). Also, the attributes are understood as taking a unique value (UV). The value
may be complex; e.g. the value of GIVEN NAMES may be a series of two or more given
names; but then the whole constitutes one value of the attribute. The value of an
attribute need not be specified with ultimate precision – it may even be left open in
certain cases; for example, in a frame for the concept ›person‹, one might provide for
the attribute SEX without fixing its value. However, the attribute must be such that for
a given possessor it returns exactly one value.
12.1.4 Attributes are functional concepts
It follows – and this point is of utmost importance – that ATTRIBUTES ARE FUNCTIONAL
Barsalou is correct in assuming that frames are the format of cognitive
representations, it follows that our mental descriptions of objects and categories are
entirely in terms of functional concepts and the values they take. Functional concepts
correspond to functions in the mathematical sense in that they return a unique value
CONCEPTS. If
308
Understanding semantics
for any given argument. The only difference between attributes and mathematical
functions is that the arguments and values of attributes are not restricted to
mathematical entities such as numbers, sets, functions, etc; arguments and values
of attributes can be of any type whatsoever. Attributes are subject to something like
selectional restrictions: they can be applied only to a certain domain of arguments.
For example, the attribute SEX can only be applied to certain organisms, HEIGHT only
to objects with a vertical extension, NAME only to nameable objects, and so on. In
addition, attributes take certain types of things as values. For example, SEX takes the
values ‘female’ or ‘male’, HEIGHT a measure of length, NAME some verbal expression
subject to complicated social conventions of naming.
We encountered functional concepts when we dealt with functional nouns in
4.4.2.1: the meanings of functional nouns are functional concepts. If you look at
the attribute terms in the passport example, you will realize that, in the readings
used in the passport, these are functional nouns: they are relational in that they
need a possessor argument for determining their referents and they refer to exactly
one referent for a given possessor. The text of the passport contains the functional
nouns name, nationality, birth, date, place, sex, height, colour, residence and signature;
except for signature, they come without a possessor specification because they are
tacitly understood as relating to the bearer of the passport. The noun name in the
combination given names is used as a relational noun; obviously, the plural admits
for more than one referent. However, in the plural, referring to the totality of given
names, it provides a functional concept. Similarly, the relational concept ›eye‹ yields
a functional concept ›eyes‹ when turned into the plural.
Since attributes are the elementary element of frame composition, frames for
attributes themselves are extremely simple. A frame graph for an attribute contains
a single arrow, labelled for that attribute. It originates from a node that represents its
possessor, or argument. Since this argument is to be filled, we represent it not with
a circle, but with an empty rectangle. The attribute links the argument to its value.
The value forms the referent of the frame. This corresponds to the fact that for an
expression such as price of oil, price expresses a functional concept to be applied to
the referent of oil; the whole expression refers to the value of the ‘price’ function for
oil, i.e. to the price of oil. Fig. 12.7 displays the general frame graph for an attribute
concept.
Figure 12.7
Frame for an attribute
ATTRIBUTE
value
The chaining of attributes mentioned above, repeated in (3a), corresponds to the
chaining of the underlying functions (3b); the mechanism of chaining mathematical
functions is known as ‘function composition’ in mathematics.
Frames
(3)
a.
possessor.ATTRIBUTE1:value1.ATTRIBUTE2:value2.ATTRIBUTE3:value3
b.
f3 (f2 (f1(x)))
309
The chaining mechanism of functions and attributes is replicated in language by the
possibility of forming possessive chain constructions as in (4a) and, less elegantly, in
(4b). The order of elements in (4a) corresponds to the order in (3b), the order of the
Saxon genitive chain, to (3a):
(4)
a.
the colour of the eyes of the bearer of the passport
b.
the passport’s bearer’s eyes’ colour.
The passport frame has terms for each attribute (except for FACE) that it specifies,
and these terms are functional nouns or complex expressions with functional
concept meanings, in the readings relevant here. While this is necessary for frames
of institutional categorization, we cannot expect that there will be words to label all
the attributes that figure in all the frames we supposedly employ in our cognitive
systems. On the contrary, it is to be expected that for most, if not almost all, attributes
we do not have a functional noun in our lexicon. As we stated in 11.1, cognitive
representations are not in terms of words.
12.1.5 Types of attributes
Possible attributes in frames can be classified into four basic types according to the
types of values they assign. All four classes are instantiated in the passport frame:
(i) constitutive part attributes, (ii) correlate attributes, (iii) property attributes and
(iv) event attributes.
Part attributes. Frames for the representation of complex things like cats or pianos
have attributes for their constitutive parts. They contain a mereology of the object
(recall 8.4.4). For example, a frame for a person will contain a subframe for the
body, which in turn will contain two embedded subframes for the arms, each with a
sub-subframe for the hand containing the attributes PALM, THUMB, INDEX FINGER, etc.,
each of them with their own mereologies. Due to UA, a frame must not contain the
same kind of attribute more than once for one possessor. Thus, in the mereological
subframe for the body, there must not be two attributes ARM, but different attributes
such as LEFT ARM and RIGHT ARM. The values of the part attributes are the respective
parts of the whole (the value of the attribute HEAD in a cat frame is the head of the
cat). The constitutive parts of a whole do not exist independently; if the whole is
given, its parts are too. In the passport frame, there are two part attributes, the
attributes EYES and FACE of the bearer.
Correlate attributes. Correlate attributes specify things of independent existence
to which the referent of the concept is uniquely related. For pianos this may be
attributes such as PRODUCER or OWNER; for cats it might be MOTHER or OWNER. The
passport example contains six correlate attributes: BEARER, NUMBER and AUTHORITY for
the passport; NAME, GIVEN NAMES and RESIDENCE for the bearer.
310
Understanding semantics
Property attributes. This group is instantiated in the passport frame by the
attributes TYPE of the passport, NATIONALITY, SEX and HEIGHT of the bearer and COLOUR
of the bearer’s eyes. The values of property attributes are of an abstract nature. They
may be located on a scale, such as in the case of HEIGHT, LENGTH, WIDTH, SIZE, WEIGHT,
AMOUNT, QUANTITY, TEMPERATURE, PRICE, WORTH, VALUE, DURATION, EXTENSION, etc. or not,
as with the attributes NATIONALITY, SEX, COLOUR, SHAPE, CHARACTER, NATURE, MEANING,
FORM, etc.
Event attributes. This is a somewhat imprecise class of attributes that link the
possessor to events and activities. In the passport example, we encountered the
implicit attribute BIRTH of the bearer; there is also the implicit attribute ISSUE of the
passport which is involved in the attributes DATE OF ISSUE and also in AUTHORITY, as this
is meant to denote the issuing authority; finally there is the implicit attribute EXPIRY.
The values of these attributes are events uniquely connected to their ‘possessors’:
there is only one birth per person and one issue and expiry per passport. The
attributes of the values of event attributes (e.g. of the birth of the bearer and the
issuing of the passport) are thematic roles of the events: LOCATION (in PLACE OF BIRTH),
TIME (in DATE OF BIRTH/ISSUE/EXPIRY) and AGENT (in [ISSUING] AUTHORITY). Later we will
see that there is another group of event-related attributes: those specifying manners
of interaction with the referent of the frame or parts of it. As we saw with basic
categories in 11.3, manners of interaction are an important aspect of categorization.
12.1.6 Constraints
Each attribute in a frame takes values from a certain set of alternatives. These
alternatives are constrained. First, an attribute returns values of a certain type, e.g.
a date, a name of a person, a measurement in centimetres, etc. Then, among the
logically possible values, it may be the case that only some are eligible. In the case of
the passport frame attributes, the attributes underlie constraints such as these:
∑
HEIGHT: possible heights of
∑
DATE OF BIRTH: dates in the past that result in a possible age of
∑
GIVEN NAMES: names admissible as given names for persons
∑
NAME: a registered family name
∑
COLOUR OF EYES: a colour possible for human eyes
∑
FACE: a photographic picture that looks like a human face
persons
the bearer
In some cases, the choice is constrained to a fixed and closed set of alternative values,
e.g. for the attributes SEX and NATIONALITY of the bearer and TYPE of the passport.
Further constraints on the values of arguments arise from the fact that the values
of attributes may depend on, or correlate with, the values of others. For example,
sex and nationality of the bearer constrain the possible values of the attribute GIVEN
NAMES, at least in some countries (a male German bearer cannot have ‘Angelika’ as
an entry for his given name). Sex and age will constrain to a certain extent what the
Frames
311
face of a person may look like; the name and given names correspond to a certain,
if often modest, extent with the shape of the signature, and so on. These constraints
apply to the passport frame in general; their violation would result in an inconsistent
or at least implausible description. In the rest of this chapter we will leave constraints
more or less aside. This part of frame theory is not as well developed as the theory
of attributes and values.
We will now turn to various applications of Barsalou frames in semantics, first
looking at verbs (12.2) and then at nouns (12.3).
12.2 VERBS AND FRAMES
12.2.1 Verbal argument frames
The argument structure of a verb is a simple straightforward frame; the referent
is the referential argument of the verb; the attributes are the thematic roles of
the arguments. In fact, such argument frames were the first examples of frames
discussed in linguistics in Fillmore (1968) and are an important precursor to the
more general present notion of the frame. Fillmore observed that each argument of a
verb is of a different general role (he called them ‘cases’) and that each of the general
roles can occur only once in a sentence. Given that any argument is only specified
once in a sentence, argument frames satisfy the three uniqueness conditions in
Definition 1: there is a central element (UR), each role can occur only once (UA) and
the roles correspond to exactly one argument (UV). Figure 12.8 shows the argument
frame for a ditransitive verb like give or send. In the figure, the argument nodes are
indicated by rectangles rather than circles. Rectangles are used for open argument
elements in a frame (except for referential arguments).
Figure 12.8
Frame graph of the argument structure of ditransitive verbs
AGENT
THEME
RECIPIENT
From the perspective of frame analysis it is not only natural, but in fact necessary
to assume that verbs have a referential event argument (recall the discussion in
5.3.2): it is the possessor of the argument attributes.
The condition UA is responsible for a constraint on the argument structures that
is otherwise unexplained: a verb cannot have two equal semantic roles – even if the
event expressed involves participants that play equal roles in this kind of situation.
312
Understanding semantics
For most two-place or three-place verbs, the roles of the arguments are clearly
different; for example with ‘hit’; the AGENT and the PATIENT are clearly involved in
the event in very different ways. However, there are reciprocal (8.3.6) verbs such as
struggle, agree or correspond where the interacting participants factually play the
same role. Yet they have to be assigned different semantic roles and linked differently,
e.g. as subject and with-PP; alternatively they can be both packed into a conjunction
and linked as subject (recall Levin’s marry class of verbs discussed in 6.1.3). In this
conceptual variant, the predication applies to only one, though complex, argument.
It must be cautioned that an argument frame is still a long way off representing
the full meaning of the verb. Hundreds if not thousands of verbs may have the same
argument frame. The argument frame contains little information about the event
itself, the kind of process or action it constitutes. In addition, these ‘flat’ argument
frames, where all arguments are directly assigned as attributes to the referent, fail
to reflect possible conceptual dependencies among the arguments. For example, a
verb like go in the construction ‘go from A to B’ would have a similar structure: three
argument attributes, in this case labelled AGENT, SOURCE and GOAL. While these three
attributes are represented on a par, they are in fact conceptually dependent on each
other: the SOURCE is the LOCATION OF AGENT at the beginning of the event and the GOAL
is the LOCATION OF AGENT at the end. How this can be modelled in frames is not yet
settled. More generally, the problem of modelling the situation structure of an event
(recall 6.2), e.g. initial and resultant condition, is not yet solved. Nevertheless, verb
argument frames are very useful, as we will see.
12.2.2 Deverbal nouns
The meaning of deverbal nouns, i.e. nouns derived from verbs, can to a great extent
be understood as the result of simple operations on the argument frame of the
underlying verb. Figure 12.9 shows three nouns derived from the verb walk. The verb
concept has an AGENT attribute and a PATH attribute; the latter is syntactically not
obligatory. It can be partly or completely specified by source, goal and/or path
specifications like from the station, to the post office, through the park, respectively. The
first variant, as in I had a nice walk this morning, refers to the event itself. The two
original verb arguments are two potential relational arguments of the noun walk1.
Therefore they are depicted as rectangles. The argument frame of the noun is
Figure 12.9
Frame graph for three nouns derived from the verb walk
(act of walking)
(route for walking)
walk1
walk2
walk
AGENT
PATH
walker
walk
walk
AGENT
PURPOSE
PATH
PATH
AGENT
AGENT
PATH
ACTIVITY
Frames
313
identical with the argument frame of the verb walk (in one of its alternates). The
principal difference between a verb concept and a noun concept – whatever it is – is
not captured by the mere case frames.
The second variant walk2 denotes a route for walking. Its referent is the path that
is walked along. Thus the referential node is shifted from the event to the path. The
frame contains not only a link from the ‘walk’ node to the referential node, but also
a reverse link PURPOSE. This is owed to the meaning of walk2: it is a path FOR walking.
The attributes AGENT and PATH of the event node are not deleted. They belong to the
concept of walk2: a walk is the path of walking and there is no walking without
walkers. However, neither the event node nor the agent node is a relational argument
of the noun walk2, since the reference of ‘path’ can be identified without determining
somebody engaged in an act of walking on the path.
The concept ›walker‹ is formed from the verb concept by shifting the referential
node to the AGENT of walking. Again, there is an inverse link from the ‘walker’ node
back to the event node. The corresponding attribute can be added and labelled
‘ACTIVITY’: a ‘walker’ is someone engaged in an activity, their activity (a functional
3
concept) is walking.
In this way, frames can be used to analyse the meanings of certain types of word
formation. We will see in the next section that frame theory can not only model the
meanings of certain types of deverbal noun, but also helps us to understand the
semantics of nominal compounds.
12.3 NOUNS AND FRAMES
Nouns have few if any relational arguments. Thus, the argument frames of nouns do
not offer much access to noun concepts. It may well be the case that the concepts for
complex things like humans and animals or cars and universities are, in themselves,
very complex. In general, it is very hard to get at the internal components of noun
meanings.
12.3.1 Metonymy
One way of getting at the components of noun meanings is to study metonymy.
In 3.4.3.1, metonymy was introduced as shifting the reference of an expression to
something that BELONGS to the original kind of referent. For example, metonymy
applied to the concept ›university‹ may shift reference to the university campus,
the university administration, the university staff, the university’s student body,
the university’s course system, and so on. From the point of view of frame theory,
these things belonging to the original referent are just the values of corresponding
attributes: the campus is the value of the attribute CAMPUS of the university, and so on.
Therefore we can infer from the possible metonymical shifts that the original concept
3 The walker might be one who is engaged in walking in a particular situation, or habitually. This
kind of difference cannot be modelled in the simple frames used here.
314
Understanding semantics
contains certain attributes of the referent. In the case of the concept ›university‹ these
include the attributes CAMPUS, ADMINISTRATION, STAFF, STUDENTS BODY and COURSE SYSTEM.
Using frames, we can characterize metonymy much more precisely: reference is
shifted to the value of one of the original referent’s attributes. This is, however, only
a necessary, not a sufficient condition. Metonymy is not possible for every attribute
of universities. For example, every university has a year of foundation, say 1869.
But one cannot use the term university for referring metonymically to the year of
its foundation, like in Boston University was a year of considerable unrest. In order to
understand the restriction, let us first have a look at the university example. Figure
12.10 illustrates the effect of applying the metonymical shift from the institution to
the campus to the concept ›university‹.
Figure 12.10
Metonymical shift ›university‹ to ›campus‹
INSTITUTION
CAMPUS
CAMPUS
LOCATION
LOCATION
The original frame on the left contains an attribute CAMPUS of the referent node,
and many other nodes, only vaguely indicated. The value of the CAMPUS attribute,
i.e. the campus of the university, has its own attributes, among them an attribute
LOCATION which is responsible for the possibility to form sentences like the university
is on a hill. With the metonymical shift, the value of CAMPUS becomes the referent
node. The new referent can be linked back to the original referent by an attribute
which takes the university itself as its value. Such a functional concept does exist,
although we do not have a specific functional noun for it; in the figure, I have
simply dubbed the attribute INSTITUTION. It is a proper functional concept because,
according to our notion of a campus, a campus will host exactly one university (or
similar institution), so there is a one-to-one correspondence between campuses and
institutions with a campus.
The example shows that there is another necessary condition for metonymic
shifts: the referent node can only be shifted to an element of the frame that is linked
back to the referent node by an appropriate attribute. There must be a one-to-one
correspondence between the original referents and referents resulting from the
metonymic shift. This is in accordance with a general condition for metonymy.
Frames
315
In metonymic use, the literal referent ‘stands for’ the metonymic referent: the
university stands for its campus – and vice versa: if the campus is a location
called, say, Ormond Hill, the name of the location can be metonymically used for
the university with its campus there. This is impossible if there is more than one
campus for a university, or if a certain area hosts more than one university. In fact,
if these conditions are not given, metonymy fails. For example, Tokyo University has
two separate campuses, one at Komaba and one at Hongo. Thus, one cannot say, for
example, (5a), because this would apply only to the Komaba campus. Conversely, (5a)
would not refer to Tokyo University as a whole.
(5)
a.
Tokyo University lies close to Shibuya.
b.
Hongo has entrance examinations tomorrow.
If you take a second look at the types of metonymy that were mentioned in 3.4.3 and
listed in (10) in 3.5.2, you will realize that this condition is fulfilled in all cases. For
example, the pars-pro-toto type metonymies (cf. redneck, asshole) shift reference to
a part of the original referent; parts belong to a unique whole. The analogue holds
for types like ‘carrier for content’ (write a paper), ‘container for content’ (drink a
bottle), ‘residence for government’ (Beijing reacted immediately), ‘clothing for wearer’
(blue helmets), etc. In the case of fluid or celebrity, an instance of something having
a property stands for the carrier: this instance has a unique carrier, and the carrier
displays this unique instance.
Cognitive linguists like Lakoff consider metonymy (along with metaphor) a
fundamental conceptual mechanism of cognition (cf. Lakoff 1987). Indeed, the
mechanism of shifting the referent node in a frame and linking it back to the original
referent node is not restricted to metonymical shifts. The examples of deverbal
nouns analysed in 12.2.2 are based on the same conceptual mechanism.
12.3.2 Affordances
We saw in 11.3 on levels of categorization that one crucial aspect of categorization is
the manner in which we interact with members of a category. This holds in particular
for categories of artefacts. Artefacts exist because people invented and produced
them, and – except in the case of art – they did so in order to use them for certain
practical purposes. The purpose for which an artefact is designed constrains its
other characteristics such as its shape and size and the material it is made of. But, at
the same time, a given purpose can be achieved in very different ways; just think of
the practically unlimited ways in which chairs may be designed. What unites these
exemplars of the general category of chairs is that they are made for sitting on them
and therefore more or less sit-on-able. (Note that not everything sit-on-able is a chair,
e.g. a sofa, a bike or a camel.) We may therefore assume that the types of activities
for which an artefact is designed are included in the conceptual representation
4
of the corresponding category: the frames for concepts like ›cup‹, ›bowl‹, ›vase‹,
4 Recall the discussion of the Labov experiments in 11.4.2.
316
Understanding semantics
›hammer‹, ›car‹, ›saxophone‹ or ›T-shirt‹ have an attribute that specifies certain
ways of interacting with this kind of object. Taking up a term from philosophy
and generalizing it for our purposes, we will call these interactional properties of
objects ‘affordances’: chairs have the affordance that they are for sitting on; a cup
has the affordance of being for drinking from; a saxophone is for playing music on;
a car for sitting in and driving from A to B. In the frame graphs, I will use the label
‘FOR’ for the affordance attribute. The affordance attribute links the referent to an
activity; the activity carries its argument frame. One of the arguments is identified
with the referent of the main frame; the corresponding role link is the inverse of the
affordance link. Thus, an affordance attribute not only provides a link to an activity
but also specifies the role that the frame referent plays in the activity. Figure 12.11
shows the frame for ›cup‹ with the drinking affordance. All other attributes of the
cup, indicated by the little arrows pointing off the referent nodes, are not elaborated.
The cup fills the container role in the drinking activity.
Figure 12.11
Frame for ›cup‹ with drinking affordance
drink
FOR
AGENT
THEME
CONT.
cup
The THEME and AGENT attributes provide conceptual links from the frame referent to a
beverage and a drinking person. These links can be employed for the interpretation of
possessive constructions such as my cup or cup of tea, which can be taken as ›cup speaker
drinks from‹ and ›cup containing tea to drink‹. (Recall the discussion of the interpretation
of possessive constructions with sortal nouns in 4.1.4: one needs some relation between
the possessor and the referent for interpreting possessive constructions.)
There are concepts and categories at the superordinate level of categorization
which are essentially reduced to a certain affordance specification: for example,
›food‹, ›garment‹, ›footwear‹, ›vehicle‹, ›vessel‹, ›aircraft‹, ›musical instrument‹, etc.
12.3.3 Nominal compounds
Affordance attributes are a key to certain types of nominal compounds. Consider
the case of coffee cup. In one of its meaning variants, coffee denotes a beverage. As
such it has the affordance of playing the THEME role in a drinking activity. Since both
Frames
317
parts of the compound coffee cup have a drinking-affordance, the two frames can be
connected by unifying the drinking nodes in the two frames. Figure 12.12 shows the
two frames of ›coffee‹ and ›cup‹ with their drinking affordances and indicates the
unification of the two ‘drink’ nodes.
Figure 12.12
Unification of frames for ›coffee‹ and ›cup‹
drink
drink
FOR
FOR
CONT.
THEME
coffee
cup
The unification has the effect of integrating the ‘coffee’ node and the ‘cup’ node
into the same ›drink‹ frame (see Fig. 12.13). The referent node of the ›cup‹ frame
becomes the referent node of the total frame, since cup is the head of the compound.
The ‘coffee’ node loses its referent status (otherwise the frame would violate UR).
This is in accordance with the fact that the compound coffee cup, unlike the syntactic
combination cup of coffee, does not refer to any coffee – a coffee cup is a coffee
cup even when it is empty, or filled with water or another substance. According to
condition UR, every node in a frame is to be connected to the referent node by a
chain of attributes. The ›coffee cup‹ frame complies with UR due to the FOR link from
the ‘cup’ node to the ‘drink’ node, and the THEME link from there to the ‘coffee’ node.
Figure 12.13
Frame for ›coffee cup‹
drink
coffee
FOR
FOR
THEME
CONT.
cup
318
Understanding semantics
The example is representative of a very frequent conceptual pattern of
compounding. It is not necessary that the linking activity frame is an affordance
for both parts of the compound. For book shop we need not assume that one of the
affordances of a book is to be sold in a shop. The ›shop‹ frame contains a selling
(or buying) affordance in which the shop takes the role of the location of the event.
For the regular interpretation of book shop we will unify the THEME argument of
the ›selling‹ frame with the referent node of the ›book‹ frame. This is admissible in
terms of the selectional restrictions of the THEME argument; but they require less than
the affordance of being for sale. I would like to call this type of compound a frame
compound. There are three more types of compound that can be straightforwardly
described within the frame approach.
Value compounds. This type is represented by plastic bag, company flat or park
bench. The modifier directly specifies the value of an attribute: the MATERIAL of the
bag is plastic; the OWNER of the flat is the company; the LOCATION of the bench is a park.
Thus the referent node of the modifier concept is unified with the value node of one
of the attributes in the head frame (cf. Fig. 12.14), thereby losing its referent status.
Figure 12.14
Unification of frames for ›plastic bag‹
plastic
MATERIAL
bag
Argument compounds. If the head noun is relational or functional, its frame
provides for a possessor argument or other relational arguments. There are
compounds which are interpreted as the modifier specifying an argument of the
head frame, e.g. air pressure ›pressure of (the) air‹, oil price ›price of oil‹ or chicken
leg ›leg of a chicken‹. Figure 12.15 shows the way in which the frame for air pressure
comes about. Again, the modifier node is stripped of its referent status.
Figure 12.15
Unification of frames for ›air pressure‹
air
PRESSURE
pressure
Synthetic compounds. So-called synthetic compounds are cases such as piano
player or bus driver. The head is a deverbal noun which comes with the verb frame
Frames
319
integrated into its frame (cf. the frame for walker in Fig. 12.9). The modifier specifies
another argument of the event; the unification mechanism is the same as with frame
compounds (cf. Fig. 12.12).
Reflection on the analysis of regular compounds. The examples illustrate that
the frame approach to word meaning is able to account for central aspects of the
semantics of regular compounds.
∑ The basic mechanism is unification. By means of unification the conceptual
content of the modifier is added to the frame of the head and connected to the
referent.
∑ The referent of the head becomes the referent of the compound.
∑ The meaning of the head is contained in the meaning of the compound.
∑ Therefore, the compound is a hyponym of the head (cf. 8.2.1).
∑ Unification leads to a specific relation between the meaning of the modifier and
the meaning of the head (cf. 8.2.2).
12.4 FRAMES AND COMPOSITION
12.4.1 Composing a predication
Frame theory also allows us to model elementary steps of composition. The basic
mechanism is unification. Predicate expressions carry open argument nodes as the
values of the role attributes. These open nodes are unified with the referential nodes
of the frames for the arguments. This will be illustrated with the first example from
chapter 5:
(6)
Johnny sent money to a dubious company.
The frame for the verb send is an argument frame with the three attributes AGENT,
and RECIPIENT, like the frame in Fig. 12.8. For ›company‹ we will just use
an unanalysed frame, i.e. a referent node with unspecified attributes. The frame
for the proper name Johnny will be just the referent node with one attribute NAME
that assigns the value Johnny. The adjective dubious requires a frame without a
referential node (because adjectives do not have a referential argument). We will
represent it as an open argument node with one attribute, in this case RELIABILITY,
assigning the value ‘low’. Figure 12.16 shows the ingredient frames for the sentence
and the unifications that bring everything together into one frame for the complex
sentence predication.
THEME
320
Understanding semantics
Figure 12.16
Unification of frames for ›Johnny sent money to a dubious company‹
send
AGENT
THEME
RECIPIENT
money
Johnny
company
NAME
RELIABILITY
low
Johnny
The grammatical structure of the sentence indicates which nodes are to be unified
in order to link all five frames to each other. The frame for the adjective is linked
by unification of its argument node with the referent node of the ›company‹ frame
because the adjective dubious is syntactically combined with the noun company.
According to the linking rules of English, the subject of the sentence specifies the
agent of the sending; the direct object specifies its theme and the to-PP the recipient.
This leads to the unification of the referent nodes of the frames for ›Johnny‹, ›money‹
and ›company‹ with the respective argument nodes of the ›send‹ frame. Figure 12.17
shows the result.
Figure 12.17
Frame for ›Johnny sent money to a dubious company‹
send
AGENT
THEME
RECIPIENT
NAME
Johnny
Johnny
money
company
RELIABILITY
low
The frame for the whole sentence violates the UR condition on frames. UR only
holds for lexical frames. In the frame of a sentence, the referents of the referring
Frames
321
elements are all marked. Composition connects all conceptual information into
one coherent whole. Thus, every piece of information is in the context of the rest
of the sentence. It is this frame in which the meaning variants of the lexemes and
grammatical forms occurring in the sentence are selected, and eventually shifted, to
make everything fit (recall the discussion of the Principle of Consistent Interpretation
in 3.4). In the case of our example, the combination of dubious with company rules
out the meaning variant ›hesitating, doubting‹ because the company is very unlikely
to be in that state. Company is taken in the ›commercial business‹ sense, required by
its role as the recipient of sending; the meaning ›being with someone‹ (as in I enjoyed
5
his company) is ruled out in the setting of sending.
12.4.2 Open problems
The frames in Fig. 12.16 and Fig. 12.17 represent only the predicate-argument
structure of the proposition, and this is even without the tense argument (recall
6.3.5, 6.4). Tense and aspect are not represented; this is connected to the fact that the
situation structure is not represented in the verb frames. Similarly, the frames we
used do not capture the determination of noun phrases: in the example, we would
get the same frame if money or a dubious company were replaced by the money or the
dubious company. The frames capture only the core of the verbal and the nominal
onions (6.6.1, 4.5.5). All of the functional elements of the sentence are missing. This
does not mean that representation of these elements is principally impossible in
the frame approach; it just represents the present stage of applying frame theory in
semantics.
12.5 FRAMES AND COGNITION
12.5.1 Frames vs features
As a theory of concept and category representation, the frame approach is superior to
the feature approach in three important respects. First, the frame approach employs
explicit attributes while these are left implicit in the feature approach. For example,
the three features that make up the description of ›girl‹ in Table 9.3, [+HUMAN], [–
ADULT], [–MALE], implicitly specify the values of three attributes:
(7)
[+HUMAN]
[–ADULT]
[–male]
____.SPECIES: human
____.AGE: young
____.SEX: female
Thus, the features used in BFA are in fact fixed combinations of an attribute and one
out of two possible values. It goes without saying that semantic analysis should be
explicit about the notions actually employed. Barsalou cites empirical evidence that
5 The meaning descriptions relate to the Oxford English Dictionary, http://oxforddictionaries.com/,
accessed 18 September 2012.
322
Understanding semantics
shows that humans actually use attributes for categorization rather than just feature
lists (cf. Barsalou 1992b: 25–29).
Second, BFA restricts the description to binary features. This is psychologically
unmotivated. In fact there are very few attributes that admit only two values;
SEX would be one. Given that visual perception plays a very important role in
categorization, there is no reason whatsoever to exclude attributes such as COLOUR or
SHAPE, which can take an unlimited number of values. Due to its restriction to binary
features, BFA is unable to deal with these aspects of objects. There is no feasible
way of describing the shape of a pig or the colour of its skin by a combination of
binary features. At best, values of non-binary attributes are just turned into so many
different features, like [HUMAN], [ROUND], [GREEN], etc.
Third, BFA descriptions lack recursion; no information can be included about
the values of the attributes of the referent. Recall the problem we encountered with
a feature description of bikini in 9.3.2. If one wanted to include into the meaning
description of bikini that the wearer is female, there is no way to do so by using the
feature [–MALE] (one could only use an ad hoc feature [–WORN BY MALE PERSONS]).
With the frame approach, this property of bikinis could be easily represented as:
bikini.FOR:wear.WEARER:[wearer].SEX:female. The lack of recursion is an immediate
consequence of the restriction to binary features: there is no information which one
could reasonably add about the + or – value of a binary feature.
12.5.2 Frames vs prototypes
Barsalou emphasizes that prototypes can be integrated into frame theory (Barsalou
1992a: 161ff.). Prototypical properties can be integrated into frames as default values
for certain attributes; for example, there might be a default entry for the SIZE value in
the ›bird‹ frame. Also, one might add, certain prototype effects can be captured by
constraints. Again relating to birds, the fact that most, but not all kinds of bird are
able to fly can be captured by a constraint that links the ability to fly to the relative
size of the wings of the bird and to its weight. Using a frame model for concept
representation would also offer the possibility of coping with another question left
unanswered by prototype theory: PT bases category membership on the degree
of similarity to the prototype, but it does not say which aspects of the objects
considered matter for the similarity judgement: shape? size? colour? affordance?
(recall 11.2.9). Frame theory would be able to define the criterion of similarity in
terms of the attributes involved.
12.6 CONCLUSION
The application of Barsalou’s frame concept to semantics is a recent development. It
offers possibilities for analysing word meanings in terms of attributes and the values
they take. Superior to both the binary feature approach and prototype theory, it is the
first general model of concepts that promises applicability to the decomposition of all
parts of speech as well as to the questions of semantic composition. It is also the first
Frames
323
theory of representation which is, at the same time, formally more precise and explicit
than its competitors and based on experimental cognitive research. As a framework
for doing conceptual semantics, the theory is still in its early stages, but it is very
promising, both as a perspective for the classical problems of semantics – lexical
semantics (including the semantics of word formation), grammatical meaning and
compositional semantics – and as a perspective for cognitive linguistics in general.
As far as compositional semantics is concerned, there is another framework
incomparably further developed than what we have sketched above: formal
semantics, to which an in-a-nutshell introduction will be given in the next chapter.
Formal semantics, however, is an approach which does not aim at the cognitive level
of meaning; in fact, the cognitive level is not included at all in that framework.
EXERCISES
1. Characterize the crucial properties of attributes in Barsalou frames.
2. Identify attributes of all the four types distinguished in 12.1.5 that might figure
in the frame for a shoe.
3. Translate the frames in Figs 12.9 (all three of them), 12.10 (both), and 12.13 into
frame matrices. Proceed as follows:
a. Disregard the attributes rendered in grey.
b. Disregard the attributes indicated by unlabelled small arrows without value
nodes.
c. Insert variables x, y, z for the values of argument nodes (rectangles)
d. Use entries of the form ‘(attribute)’ for the specification of unspecified
attribute values, for example: … .AGENT: (agent).
4. Determine the readings of the following deverbal nouns in terms of the verb
argument they refer to, and draw the respective frame graphs like those in
Fig. 12.9:
a. drink
b. building (two readings)
c. baker
5. Characterize metonymy in terms of frames. What are the crucial conditions for
this operation?
6. Try to construct a frame for the metonymical expression blue helmet.
7. Describe the notion of ‘affordance’.
8. Categorize the following compounds according to the four types mentioned
in 12.3.3.
a. bus ticket
b. bus driver c. school bus
d. bus stop e. Greyhound bus
Describe their meanings in terms of a frame analysis. Feel free to choose
appropriate names for the attributes involved.
324
Understanding semantics
9. Explain why frames are superior to binary features as a format for representing
meanings.
FURTHER READING
Ungerer and Schmid (2006) contains a chapter about ‘frames’ and ‘scripts’ and their
applications in cognitive semantics; the notion of a frame used there is similar, but
not the same as the one used here. Kövecses (2006, chs 5–9) applies a very broad
notion of frames to phenomena including metonymy and metaphor. Barsalou
(1992b) gives an introduction to the notion of frames (as used here) and their
applications; Barsalou (1992a, ch. 7) discusses frames in the general context of an
introduction to cognitive psychology. On frames vs feature lists and the relation of
frames and prototype theory see Barsalou (1992a, 7.2).
There is a huge online resource of frames related to English words: the FrameNet
site at www.icsi.berkeley.edu/framenet, initiated and supervised by Charles Fillmore.
It contains a network of thousands of frames, in terms of core and non-core attributes
and their relations to other frames. The network is permanently being extended and
refined; it is a very valuable resource of lexical semantics – browse it!
13
Formal semantics
This chapter deals with composition. More precisely, it offers an elementary
introduction into formal semantics (alternatively: model-theoretic, truth-conditional,
referential, logical or possible-world semantics). In formal semantics, meaning is a
matter of reference and truth conditions. It is the most technical and difficult variety
of semantics, very mathematical, but it is the main framework in which sentence
semantics has been developed.
13.1 JAPANESE NUMERALS: A SIMPLE EXAMPLE OF A COMPOSITIONAL
ANALYSIS
13.1.1 The system of numerals
Japanese has two sets of numerals, a native set of expressions for the numbers 1 to
10, 20 and a couple of isolated greater numbers such as 100 and 1000. The numeral
system proper is of Chinese origin (as is the case in many languages neighbouring
China). It is chosen here because it is perfectly regular, unlike the system of number
names in English or those of any other European language. The basis of the system
for the first ninety-nine numerals consists of just ten lexical units, the expressions
for the first ten numbers:
(1)
ichi
1
ni
2
san
3
yon
4
go
5
roku
6
nana
7
hachi kyū
8
9
jū
10
When we attach the word for 10 jū to the words for the digits 2 to 9 we obtain the
terms for the tens higher than 10:
(2)
jū
10 nijū
20 sanjū 30 yonjū 40
…
kyūjū 90
Thus ‘thirty’ is expressed as ‘three ten’. Terms for the numbers 11, …, 19, 21, … 29
and so on are formed by attaching the words for the digits 1, …, 9 to the terms for
the tens, including jū. The number 11 is expressed as ‘ten one’ and 21 as ‘two ten one’.
The complete system for 1 to 99 is given in (3). The overall system continues in the
same regular and economic fashion, but we will confine the analysis to the numerals
for 1 to 99:
326
Understanding semantics
jū
(3)
10
ni
jū
20
…
kyū jū
90
ichi
1 jū ichi
11
ni
jū
ichi
21
…
kyū jū ichi
91
ni
2 jū ni
12
ni
jū
ni
22
…
kyū jū ni
92
san
3 jū san
13
ni
jū
san
23
…
kyū jū san
93
…
…
kyū
9 jū kyū
…
19
ni
jū
kyū
…
29
…
kyū jū kyū
99
13.1.2 Formal description of the system
The system in (3) is extremely economical: with as few as ten basic elements, the
numerals for 1 to 10, a total of ninety-nine numerals is given or can be formed.
The meanings of the complex numerals derive unambiguously from the meanings
of their components and their order. For example, the meaning of nijūsan can be
calculated from the components ni 2, jū 10 and san 3 as 2 × 10 + 3, i.e. 23. Therefore,
the complex numerals constitute a good example of expressions with compositional
meaning. Recall that, according to the Principle of Compositionality, the meaning of
a complex expression is determined by the lexical meanings of its components, their
grammatical meaning (irrelevant here) and the way in which they are combined.
There are two types of composite expressions: the terms for the tens in (2) and the
terms for the numbers between the tens in (3). Both types are formed by the same
morphological operation, called concatenation. Concatenation (from Latin ‘chain
together’) forms one expression out of two by attaching the second to the end of the
first. While both types of composites are formed by concatenation, the interpretation
of the combinations differs. The word sanjū ‘three ten’ for 30 means ›three TIMES ten‹,
but jūsan ‘ten three’ for 13 means ›ten PLUS three‹. If the first part denotes a smaller
number than the second, concatenation is interpreted as multiplication. Otherwise
concatenation stands for addition. Now, if a numeral is composed of three parts,
it must have been built up in two steps, since concatenation only allows for the
combining of two expressions. Therefore, in a first step two basic elements have to be
combined to form a two-part numeral and this in turn combined with a third part.
These two steps can be taken in either order. For example, sanjūroku could, for its
mere form, be derived in two ways:
(4)
a.
(3 10) 6:
First concatenate san 3 with jū 10 to form sanjū, then
concatenate sanjū with roku 6.
b.
3 (10 6)
First concatenate jū 10 with roku 6 to form jūroku, then
concatenate san 3 with jūroku.
When we follow (4a), we obtain the interpretation 30 for sanjū (smaller number 3
times greater number 10) and 36 for sanjūroku (greater number 30 plus smaller
number 6). But with (4b) we obtain the interpretation 16 for jūroku and would then
Formal semantics
327
have to multiply this with 3, obtaining 48. Only the first derivation is correct: first
form the ten word and then attach the digit word.
We are now in a position to give a compositional description of the system. A
1
compositional description always consists of four components.
FB
Formation Base
A list of the basic expressions.
IB
Interpretation Base
A specification of the interpretations of the basic expressions.
FR
Formation Rules
A specification of the formation rules for complex expressions.
IR
Interpretation Rules
A specification of the corresponding interpretation rules for complex
expressions.
2
FB was defined in (1).
FB
Basic expressions are ichi, ni, san, yon, go, roku, nana, hachi, kyū, jū
FR consists of two rules:
F1
form ‘UV’
where
and
U = ni, san, yon, go, roku, nana, hachi or kyū
V = jū
F2
form ‘ZX’
where
and
Z = jū, nijū, sanjū, …, kyūjū
X = ichi, ni, san, yon, go, roku, nana, hachi or kyū
We will use square brackets for denoting interpretations: [A] is ‘the interpretation of
A’. IB assigns interpretations to the ten basic expressions:
IB
[ichi] = 1, [ni] = 2, [san] = 3, … , [kyū] = 9, [jū] = 10
1 Compared to the general scheme of semantic composition given in Fig. 1.1, the intermediate level
of grammatical meaning is missing here. It will be neglected throughout this chapter.
2 We use the more general term interpretation instead of meaning. Interpretations are whatever the
expressions are assigned in such formal systems. How far these interpretations can be considered
meanings in the proper sense will be discussed later.
328
Understanding semantics
IR consists of two interpretation rules, I1 and I2, that correspond to F1 and F2,
respectively. They determine the way in which the interpretation of a complex
expression derives if it is built according to these rules.
I1
If UV is formed according to rule F1,
then [UV] = [U] × [V]
I2
If ZX is formed according to rule F2,
then [ZX] = [Z] + [X]
We can now ‘compose’ the meanings of the complex numerals. In (5), this is done for
nanajūroku 76. On the basis of the given rules, the expression can only be derived
by first applying F1 and then F2: nana and jū combine to nanajū 70, then nanajū is
concatenated with roku 6. If we first derived jūroku 16 by F2, we could not concatenate
this with nana, as neither formation rule allows for jūroku as the second part.
(5)
f.
formation of nanajūroku
Step 1: F1: nana, jū → nanajū
Step 2: F2: nanajū, roku → nanajūroku
i.
interpretation of nanajūroku
Step 1: I1: [nanajū]
=
=
=
Step 2: I2: [nanajūroku] =
=
=
[nana] × [jū]
7 × 10
70
[nanajū] + [roku]
70 + 6
76
(IB)
(above, IB)
Figure 13.1 shows the parallel steps in deriving the form and the interpretation of
nanajūroku.
Figure 13.1
Formation and compositional interpretation of nanajūroku
冦
nana, jū
Interpretations
7 , 10
冦
Expressions
×
nanajū, roku
70 , 6
冦
I1
冦
F1
+
F2
nanajūroku
76
I2
Formal semantics
329
13.1.3 The general scheme of meaning composition
The formal treatment of the numeral system illustrates the general scheme of
compositional semantics as given in Fig. 13.2.
Figure 13.2
The general scheme of semantic composition
LEXICON
FB
basic expressions
F1,
IB
interpretations of
basic expressions
F2, …
complex expressions
GRAMMAR
I1,
I2, …
interpretations of
complex expressions
SEMANTICS
The essence of composition is the stepwise parallel formation of a complex expression
and its interpretation. The basis of the system consists of the components FB and IB.
These constitute the lexicon: a list of the lexical items with specified interpretations
(= lexical meanings). Unlike in our example, the lexical items are usually categorized
as, for instance, nouns, intransitive verbs, prepositions, etc. Such grammatical
categories are sets of expressions with the same combinatorial properties. They are
needed for the formulation of the formation rules, because these rules usually apply
only to certain categories of expression. For example, the rules F1 and F2 require four
different input categories for which we used the variables U, V, X and Z. U consists of
the unit words for 2 to 9 because these numerals share the combinatorial property
that they can form the first part of a complex ten word (for 20 to 90). The output of
a formation rule can be categorized, too: F1 generates expressions in the category of
tens, which in turn is an input category for F2. Both F1 and F2 generate expressions
that can be attached to the hundreds, not covered by our little system.
The components FB and FR form the grammar of the system, i.e. the part that
determines which expressions, basic or complex, are regular expressions and to
which categories the regular expressions belong.
The interpretation of the basic expressions, component IB, and the set IR of
interpretation rules I1, I2, … constitute the semantics of the system, i.e. the
apparatus that provides each regular expression with an interpretation. The basic
expressions receive interpretations directly by IB. The interpretations of complex
regular expressions are derived step by step from the interpretations of the basic
expressions they contain. Each time a formation rule applies to derive a new
expression, the corresponding interpretation rule applies to derive its interpretation.
The broken-line arrow in the bottom part of the scheme represents the resulting
assignment of interpretations to complex expressions.
330
Understanding semantics
13.2 A SMALL FRAGMENT OF ENGLISH
In formal semantics, the description of composition is usually broken down into
two steps. The first consists of a ‘translation’ of natural language sentences into an
appropriate formal language, usually some variant of predicate logic. This step itself
is compositional in the sense that the translations are systematically derived step by
step in the same sort of parallelism as semantic composition. In a second step, these
logical formulae receive a standard compositional interpretation. While the two-step
procedure may seem complicated at first sight, it is in fact of great advantage. The
translation into a logical language is the essential step of the analysis. Since logical
formulae are, to the trained eye, semantically completely transparent, the translation
of a sentence reveals its semantic structure as far as it can be expressed in the logical
language. The second step, a compositional interpretation of the logical language, is
standard and trivial, once the general rules are defined; actually it is usually omitted.
The logical formulae are therefore practically considered as meaning representations.
It is technically possible to skip the translation into logic and supply the natural
language sentences directly with the interpretations the translation formulae would
receive. In fact some variants of formal semantics proceed in this way.
Semantic theory is still far from being able to provide this kind of account for the
major part of any language. The analysis is therefore always restricted to a ‘fragment’,
a limited set of representative lexemes and a limited set of formation rules, for
which the corresponding interpretation rules are set up. Compositional analysis of
natural language is difficult because syntactic combinations tend to allow for several
semantic interpretations. In what follows we will treat a small fragment of English
containing just a handful of basic expressions. For the sentences of the fragment, we
will derive translations into predicate logic formulae similar to those used in 5.5.
13.2.1 The grammar of the fragment
The lexicon of the fragment is minimal, just designed to contain representatives
of the major word classes and to allow for minimal variation. In order to keep
the intermediate level of free grammatical forms out of the system, the forms are
simply fixed: all nouns are singular, all adjectives in their positive form and all verbs
in simple present tense, 3rd person singular, indicative, active. NPs are restricted
3
to proper names and singular indefinite count NPs. The basic expressions are
categorized in the traditional way.
FB Basic expressions are the members of the following categories:
NP
noun phrases
Mary, John
3 It must also be mentioned that we disregarded quantification (recall 4.5). Quantification is a
central issue of formal semantics; many introductions into formal semantics start from it. For
the purpose of this chapter, quantification can be left aside; its treatment is not necessary for
explaining the basic notions and methods of formal semantics.
Formal semantics
331
FB contd
N
nouns
person, fox
A
adjectives
young, busy, wicked
D
articles
a
VP
verb phrases
squints, smokes
TV
transitive verbs
knows, dislikes
CV
copula verb
is
4
Formation rules. In a system with categories, a formation rule specifies three things:
(i) the categories of the input expressions; (ii) the form of the output expression; and
(iii) its category. According to the first rule, a sentence may consist of an NP followed
by a VP. This is the only rule for forming a sentence. It supplies the VP with a subject
NP, i.e. a specification of its first argument:
F1
NP VP → S
The notation of the rule represents the general format used here. On the left side of the
arrow, the input categories are specified. The order in which they are arranged indicates
the form in which they must be combined. Rule F1 takes as input an NP and a VP and
forms a string consisting of the NP followed by the VP. On the right side of the arrow, the
output category is specified. F1 reads as follows: take an NP and let it be followed by a
VP; the result is of category S (sentence). Members of S are always complex expressions.
If we had only this formation rule, it would allow us to form four sentences:
(6)
NP
VP
result: S
Mary
squints
Mary squints
Mary
smokes
Mary smokes
John
squints
John squints
John
smokes
John smokes
The system contains two basic NPs, Mary and John. In addition to these, complex NPs
can be formed by combining nouns (N) with the indefinite article (D).
F2
D N → NP
4 All VPs of the fragment are to be taken in their habitual reading. Smokes means as much as is a
smoker, squints is to be taken as has eyes that look in different directions. The restriction enables us
to use the verbs in the simple present tense and assume present time reference.
332
Understanding semantics
F2, so far, allows the formation of the complex NPs a fox and a person. F2 enlarges
NP, an input category of F1. So we are able now to derive four more sentences: a fox
smokes, a person squints, etc.
With F3, we can combine adjectives with nouns. The output category is N, i.e. the
category of expressions that yield an NP when combined with the indefinite article.
AN→N
F3
When we apply this rule once, we can form six AN combinations, among them young
fox and wicked person. But since the output of rule F3 is again of category N, the
rule can be applied repeatedly. It allows for the formation of young wicked person
out of the derived N element wicked person and the adjective young and so on.
Consequently, an N can be preceded by any number of adjectives, for example young
wicked young young young wicked fox. This consequence of the rule, though it looks
strange, is harmless. There is nothing ungrammatical to such formations. F3 also
indirectly enlarges the number of NPs and sentences, since all N elements provide
input to rule F2 and the output of F2 is input to F1. With F1, F2, F3 we can form
sentences such as a young busy person smokes.
Let us now turn to the rules for forming complex VPs. The two transitive verbs
and the copula is can be combined with an NP to form a VP. There are two different
formation rules F4 and F5 because, as we will see, the interpretation of the two types
of VPs differs. The last rule allows the formation of a VP by combining the copula
with an adjective.
F4
TV
NP
→
VP
F5
CV
NP
→
VP
F6
CV
A
→
VP
Now the grammar of the system is complete. Using F4, F5 and F6, we can form
sentences like these:
(7)
a.
John is busy
F6: CV is, A busy → VP is busy
F1: NP John, VP is busy → S John is busy
b.
Mary is a busy person
F3: A busy, N person → N busy person
F2: D a, N busy person → NP a busy person
F5: CV is, NP a busy person → VP is a busy person
F1: NP Mary, VP is a busy person → S Mary is a busy person
Formal semantics
c.
333
a young fox dislikes John
F3: A young, N fox → N young fox
F2: D a, N young fox → NP a young fox
F4: TV dislikes, NP John → VP dislikes John
F1: NP a young fox, VP dislikes John → S a young fox dislikes John
13.2.2 The predicate logic language PL-F: its grammar
For the translation of the fragment, we define an appropriate predicate-logic
language, PL-F. All PL languages make use of the same repertoire of formation
rules, but they differ in the choice of the basic expressions, i.e. predicate constants,
5
individual constants and individual variables.
PL-F contains just the basic expressions we need: individual constants for Mary
and John, one-place predicate constants for the adjectives, nouns and intransitive
verbs and two-place predicate constants for the transitive verbs. In addition we
will need two variables for individuals. Both, individual constants and individual
variables, belong to the category T of [individual] ‘terms’. In addition, individual
variables form a category of their own, V (variables). When specifying FB for PL-F,
bold face is used for the constants that correspond to lexical items of the natural
language fragment. As PL-F does not obey the rules of English syntax, the ending -s
is omitted for the predicate constants that correspond to verbs.
FB
Basic expressions of PL-F are the members of the following categories:
T
individual constants, terms
m
j
V, T
individual variables, terms
x y
P1
one-place predicate constants
fox person
young busy wicked
squint smoke
P2
two-place predicate constants
know dislike
The formation rules for PL-F are only a subset of the usual PL system. F1 and F2
allow the formation of so-called prime formulae by attaching the required number of
argument terms to predicate constants (recall the description of PL formulae in 5.5):
F1
P1(T) → S
F2
P2(T, T) → S
5 Expressions in PL are either basic or complex; basic expressions are either variables or constants.
In first-order logic as used here there are individual constants and individual variables; basic
predicate expressions are all constants.
334
Understanding semantics
S, again, is the category of sentences (or formulae). The two rules allow the formation
of a limited number of formulae (60 in total), among them:
(8)
a.
person(m)
young(x)
b.
dislike(j,y)
know(x,m)
squint(j)
In addition to this type of prime formulae we can form identity statements by
connecting two terms with the ‘equals’ sign, yielding formulae such as ‘x = m’:
T=T→S
F3
Semantically, the equals sign is just another two-place predicate constant. It could be
introduced as such and treated syntactically as the others. We would then write ‘=
(x,m)’ instead of ‘x = m’. The usual way of writing identity statements is commonly
preferred in PL. But, as you see, the price of doing so is an extra syntactic rule.
Next we conjoin two formulae by logical conjunction. More connectives are used in
richer PL languages, such as ∨ for ‘and/or’ and → for ‘if … then …’. But we can do
without them here. We also do not include negation.
S∧ S→S
F4
F4 can be applied repeatedly. It allows the formation of complex formulae:
(9)
a.
young(j) ∧ wicked(j)
b.
fox(x) ∧ fox(y)
c.
fox(x) ∧ young(x) ∧ dislike(x,y) ∧ y = m
(9a) says that John is young and wicked, (9b) is a way of expressing that both x and
y are foxes, while the longish formula (9c) tells us the following: x is a fox and young
and dislikes y and y is Mary. In other words: x, a young fox, dislikes Mary.
The last formation rule introduces the existential quantifier ∃. This operator,
combined with a variable, is used as a prefix to a formula. The formula is enclosed in
parentheses in order to mark the part of the total expression to which the quantifier
applies, its so-called scope.
F5
∃V(S)→S
The existential quantifier is to be read and interpreted as ‘there is at least one V
such that S’. Usually, the ‘at least’ part is omitted in reading (but nevertheless to be
understood). The rule gives us formulae like these:
Formal semantics
335
(10) a. ∃x (young(x))
there is an x that is young
b. ∃x (young(x) ∧ person(x)) there is an x that is young and a person
= there is an x that is a young person
c. ∃y (know(x,y) ∧ x = m)
there is a y such that x knows y and x is Mary
= there is a y that Mary knows
Both F4 and F5 can be applied repeatedly and in combination with each other. The
formation potential of the system is considerable, but we will confine our examples
to such cases as will actually occur as translations of English fragment sentences, for
example:
(11) ∃x (x = m ∧ ∃y (dislike(x,y) ∧ fox(y) ) )
The formula reads as follows: ‘there is an x such that x is Mary and there is a y
such that x dislikes y and y is a fox’. (11) is somewhat complex, so we will give
its derivation in full detail. First the prime formulae are generated (steps 1 to 3).
Then they are packed according to the structure of the total formula. Following the
bracketing, we first conjoin the second and the third subformula (step 4) and then
apply ∃y in step 5. Next we conjoin ‘x = m’ to the ∃y-subformula (step 6). ∃x is
applied in the last step.
(12) step 1
F3
x=m
step 2
F2
dislike(x,y)
step 3
F1
fox(y)
step 4
F4
dislike(x,y) ∧ fox(y)
step 5
F5
∃y(dislike(x,y) ∧ fox(y))
step 6
F4
x = m ∧ ∃y(dislike(x,y) ∧ fox(y))
step 7
F5
∃x(x = m ∧ ∃y(dislike(x,y) ∧ fox(y)))
The formation rules of PL allow for two different uses of variables. First, by F5, they can
occur in the scope of an existential quantifier with that variable. Such occurrences are
called bound by the quantifier. Alternatively, a variable may occur in a formula without
being bound by a quantifier. It is then called free. To see the difference, consider the
stepwise derivation of the formula in (12). In step 1, x is free and it remains free until
it is bound by ∃x in step 7. The variable y is introduced in step 2 as a free variable. It is
eventually bound in step 5 by ∃y. Note that in step 5 ∃y only binds y.
It is important to note that free variables are interpreted on a par with individual
constants. They stand for a particular individual, functioning roughly like ‘the x’.
Existentially bound variables, however, achieve the reading of ‘some x’. Therefore,
steps 4 and 5 are interpreted as follows:
(13) a.
b.
dislike(x,y) ∧ fox(y)
the x dislikes the y and the y is a fox
∃y(dislike(x,y) ∧ fox(y)) there is some y that the x dislikes and y is a fox
i.e. there is some fox that the x dislikes
336
Understanding semantics
Note that the bound variable can be eliminated when the formula is expressed in
ordinary language (there is some fox …). The free one cannot, because it essentially
works like a pronoun.
13.2.3 Translating the fragment into predicate logic
13.2.3.1 Challenges for a compositional account
The system for the English fragment is quite small, yet it already poses a reasonable
set of compositional challenges. To begin with, the translation of the basic expressions
must be carefully arranged in order for the rest of the system to work out properly.
Then there are a number of problems concerning the system of composition rules.
Referential and predicative NPs. The system allows for both referential and
predicative NPs. If an NP is used according to F1 or F4, it is a verb complement with
a referent, i.e. referential. If F5 is applied, the NP is part of a copula VP, e.g. is a young
fox, is Mary, etc., it is then predicative and has no referent of its own (5.4.3). The rules
of the system have to do justice to both kinds of NPs.
Attributive and predicative adjectives. A parallel problem arises with adjectives.
How can they be analysed in a uniform way that covers their attributive use (F3) as
well as their predicative use (F6)?
Proper names and indefinite NPs. There are NPs of two different kinds, proper
names and indefinite NPs. In this chapter, we follow the usual approach taken in
formal semantics of treating proper names like individual terms and indefinite NPs
as predicators. For example, a fox dislikes John would be rendered as (14):
(14)
fox(x) ∧ dislike(x,j)
The proper name NP John directly provides the second argument of the verb
dislike. By contrast, the indefinite NP a fox contributes a further predication, fox(x),
about the first argument of the verb. The translation system has to take care of the
difference. At the same time, however, it should manage to treat both types of NPs in
a uniform way: for each of the three formation rules F1, F4 and F5, there should be
only one translation rule.
Argument sharing. A more general problem is a proper account of argument
sharing. The translation rules have to produce an appropriate account of which
predicates share which arguments. We have to aim at solutions such as fox(x) ∧
dislike(x,j) for a fox dislikes John. Here, the use of the same variable x as an argument
term of the subject noun fox and the verb dislike shows that they share one argument.
Thus, the subject specifies the first argument of the verb.
13.2.3.2 The translation system
The translation of the fragment into PL-F follows the scheme of semantic composition.
The notion of interpretation is to be replaced by translation. The general schema is
given in Fig. 13.3.
Formal semantics
337
Figure 13.3
The general scheme of the translation procedure
FB
basic expressions
F1,
translations of
basic expressions
TB
F2, …
T1,
T2, …
translations of
complex expressions
complex expressions
GRAMMAR
TRANSLATION
The translation base TB fixes the translations of the basic expressions. Those given
here will not be regular expressions of PL-F inasmuch as they contain empty slots
for argument terms. In the course of applying the translation rules, these empty slots
are properly filled with individual terms. In the end, each fragment sentence receives
a regular PL-F formula as its translation. The use of empty slots helps us to keep
track of the translation steps. There are two basic expressions of the fragment that
do not receive a translation: the indefinite article and the copula. These will be taken
care of in the translation rules concerning the categories D and CV. An underlined
expression stands for the translation of the expression. (John is the translation of
John, Mary is a young person is the translation of Mary is a young person.)
13.2.3.3 The translation base
Except for the copula and the indefinite article, we treat all basic expressions
including proper names as predicate expressions.
TB
NP
N
A
VP
TV
John:
_=j
Mary:
_=m
fox:
fox( _ )
person:
person( _ )
young:
young( _ )
busy:
busy( _ )
wicked:
wicked( _ )
squints:
squint( _ )
smokes:
smoke( _ )
dislikes:
dislike( _ , _ )
knows:
know( _ , _ )
338
Understanding semantics
You may wonder why the proper names John and Mary are not translated into oneplace predicate constants, rendering the translations John( _ ) and Mary( _ ). The
reason is that proper names correspond to particular individuals. Translating Mary
as ‘_ = m’ takes account of both aspects of proper names. On the one hand, the
proper name is treated as a predicate constant, since the empty slot in ‘_ = m’ has to
be filled with an argument term. On the other, the resulting predication only holds
for one single individual, namely Mary. Other one-place predicate constants such as
the translations of ordinary nouns, verbs and adjectives, may be true of any number
of individuals, zero, one or more. For a predicate constant such as person there is no
particular individual p such that ‘x is a person’ would be tantamount to ‘x = p’.
13.2.3.4 The translation rules
The simplest rules are those handling the indefinite article and the copula. As a
matter of fact, both expressions are treated as semantically zero:
T2
aN
=N
T5
is NP = NP
T6
is A
=A
A noun, e.g. fox, constitutes a one-place predicate constant that is true of all foxes.
An indefinite NP such as a fox provides the same predication. If an indefinite NP is
combined with the copula, the result is a one-place verb phrase, another predicate
6
constant with the same meaning. The NP a fox and the VP is a fox differ only in
one respect: the NP, if used referentially, predicates about its own referent, while
the VP predicates about the referent of the subject of the sentence. Such differences
are, however, eliminated in predicate logic as all one-place predicate constants are
grammatically treated in a uniform way. In predicate logic, differences such as those
between verbs, adjectives and nouns do not exist.
Rule TR6 is motivated similarly: an adjective like busy provides the same predicate
as the copula expression is busy. The three rules lead to the results in (15):
(15) a. a fox
= fox
b. is a fox
= a fox
c. is Mary
= Mary
d. is wicked = wicked
= fox( _ )
= fox = fox( _ )
= _ =m
(same for a person)
(same for is a person)
(same for is John)
= wicked( _ ) (same for is young, is busy)
6 Strictly speaking, the meanings of a fox and is a fox are not the same, as only the latter carries
present tense. The difference is not relevant in the fragment, because tense is neglected in our
treatment.
Formal semantics
339
We interpret the combination of A and N as logical conjunction, which is adequate
for the adjectives in the fragment, because these adjectives are predicating adjectives
(see 5.4.2.3 for cases of non-predicating adjectives).
T3
AN=A∧ N
Together with T2 and T5 we get for example:
busy person
= busy ∧ person = busy( _ ) ∧ person( _ ) T3
b.
a busy person
= busy person
c.
is a busy person = a busy person = busy( _ ) ∧ person( _ ) T5, b
(16) a.
= busy( _ ) ∧ person( _ ) T2, a
The rules of combining a verb with an argument, T1 and T4, introduce an
existential quantifier for the argument.
T1
NP VP = ∃x( |NP ∧ VP| _ = x )
The notation ‘|…| _ = x’ stands for the expression that we obtain when we insert
the variable x into the last empty slot of each predicate constant within |…|. Thus,
for example, |young( _ )| _ = x yields young(x) and |know( _ , _ )| _ = y results
in know( _ ,y). The procedure will be referred to as variable insertion. All of the
predicate constants treated so far are one-place. Thus, the last argument slot is the
only one. Let us apply T1 to the sentence Mary squints.
(17) Mary squints
= ∃x( |Mary ∧ squints| _ = x )
T1
= ∃x( | _ = m ∧ squint( _ )| _ = x )
TB
= ∃x( x = m ∧ squint(x) )
x insertion
Note that x insertion can only be performed when the resolution of the translation is
down to the basic translations with the empty slots. Next, we combine the NP Mary
with the VP is a busy person from (16):
(18) Mary is a busy person =
= ∃x( |Mary ∧ is a busy person| _ = x )
T1
= ∃x( | _ = m ∧ busy( _ ) ∧ person( _ )| _ = x )
TB, (16)
= ∃x( x = m ∧ busy(x) ∧ person(x) )
x insertion
The translation rule for the combination of a transitive verb with its object NP is
similar to T1. Instead of x, the variable y is used for the object referent. Rule T4 is
340
Understanding semantics
applied before T1 (because the VP must be formed before it can be combined with
the subject NP). Therefore the provision that the variable be inserted into the last
empty slot of each argument term ensures that the object argument slot is filled first.
TV NP = ∃y( |TV ∧ NP| _ = y )
T4
One example, a fox dislikes John, may suffice to show how this rule works in
combination with T1. The formation of the sentence is given in (19f), its translation
in (19t).
(19) f.
(19) t.
F4:
TV dislikes, NP John → VP dislikes John
F2:
D a, N fox → NP a fox
F1:
NP a fox, VP dislikes John → S a fox dislikes John
a fox dislikes John =
=
∃x( |a fox ∧ dislikes John| _ = x )
T1
=
∃x( |fox ∧ dislikes John| _ = x )
T2
=
∃x( |fox ∧ ∃y(|dislikes ∧ John| _ = y) | _ = x )
T4
=
∃x( |fox(_) ∧ ∃y(|dislikes(_,_) ∧ _ = j| _ = y) | _ = x )
TB
Now that we are down to the translations of the basic expressions with empty slots
for arguments, we perform the variable insertions, first the inner insertions of y and
then the outer insertions of x.
(19) (contd.) = ∃x( |fox(_) ∧ ∃y(dislikes(_,y) ∧ y = j) | _ = x )
= ∃x( fox(x) ∧ ∃y(dislikes(x,y) ∧ y = j) )
y insertion
x insertion
13.2.3.5 Discussion of the analysis
How does the system cope with the problems addressed at the end of 13.2.1? To
begin with, TB assigns the proper types of translations, i.e. one-place or two-place
predicate constants, to the basic expressions of the fragment. The four problems are
then taken care of as follows:
Referential and predicative NPs. T1 and T4 introduce referents only for referential
NPs, i.e. NPs in subject or object position. For each referent there is a variable bound
by an existential quantifier. This is adequate since reference necessarily involves
existence of the referent. Rule T5 for predicative NPs does not introduce an existential
quantifier but leaves the argument slot(s) of the NP unoccupied.
Attributive and predicative adjectives. The system treats both uses of adjectives
alike. T3 for attributive and T6 for predicative adjectives both add the adjective
meaning as a predicate conjunct to the meaning of N or to the zero meaning of is.
Formal semantics
341
Proper names and indefinite NPs are treated in the same way. They uniformly add
a predicate conjunct to the verb predication or to the zero meaning of the copula.
Argument sharing. The system properly accounts for argument sharing among
predicate constants. Two predicate terms that share an argument (A and N, NP and
VP, TV and NP) translate into predicate expressions with an empty slot and are
connected by conjunction; then the same variable is inserted into the empty slots of
conjoined predicate expressions. For the sentences in (18) and (19), Fig. 13.4 displays
argument assignments for the predicate constants and referent assignments for the
NPs (recall the analyses in Fig. 5.2 and Fig. 5.7).
Figure 13.4
Referents and arguments of two sentences
sentence
NPs
[Mary]NP is [a busyA personN]NP
(no referent)
referents
x
translation
∃x(x=m ∧ busy(x) ∧ person(x))
sentence
NPs
[a fox N]NP dislikesTV [John]NP
referents
translation
x
y
∃x(fox(x) ∧ ∃y(dislike(x,y) ∧ y=j ))
The translation system yields PL representations that essentially correspond
to the intuitive representations we used in 5.5. Superficially, the translations are
different. For example, a fox dislikes John would have been analysed there as (20a)
instead of our actual result (20b):
(20) a.
b.
fox(x) ∧ dislike(x,y) ∧ John(y)
∃x( fox(x) ∧ ∃y(dislike(x,y) ∧ y = j ) )
The difference between the two formulae is due to two properties of the translation
system. First, proper names result in identity statements like ‘y = j’ rather than
predications like ‘John(y)’. The two treatments are logically equivalent (if we assume
342
Understanding semantics
that the individual term ‘j’ denotes the only individual for which the predicate
constant John is true, which we did indeed do here). We can thus replace ‘John(y)’ by
‘y = j’. This yields (21) out of (20a):
(21)
fox(x) ∧ dislike(x,y) ∧ y = j
The result lacks the existential quantifiers binding x and y. We will see in the next
section that we actually need them if we want to achieve the intended reading.
(20b) can be replaced by a simpler logical equivalent: observing the conjunct
‘y=j’, we can replace ‘y’ by ‘j’ and eliminate the quantifier ‘∃y’ and the identity
conjunct:
(20) c.
∃x(fox(x) ∧ dislike (x,j)
Let me add a few critical remarks about this system. First, while the system
illustrates the general way in which formal semantics operates, it involves the
unusual employment of empty slots and argument insertion. This allowed me to
avoid additional formalism (such as λ operators, for those who know). But due to
this, only sentences receive regular PL-F expressions as translations: translations
of everything below sentence level contain empty argument slots, which are not
provided for by the syntax of PL-F. Second, the reader must be cautioned that the
system takes advantage of considerable restrictions of the lexical input. For example,
not all types of one-place adjectives can be treated in that way, as was pointed out
in 5.4.2.3. None of the rules applies to all adjectives, all one-place nouns, all NPs, all
intransitive or transitive verbs.
13.3 MODEL-THEORETIC SEMANTICS
We will now undertake the next major step, the introduction of a formal semantics for
the PL language and thereby indirectly for the fragment. A formal semantics for a PL
language serves one purpose: the assignment of a truth value to every formula. The
interpretation base of such a system is a so-called model. It fixes the interpretations
of all individual terms and predicate constants. The truth values of formulae derive
from these fixings by means of general rules valid for all PL languages. Thereby,
a general definition of the truth conditions of PL formulae is achieved. For each
language, an infinity of different models can be defined. We will first fix a model for
PL-F and then introduce the general interpretation rules.
13.3.1 A model for PL-F
A model for a PL language consists of two components, a so-called universe
(also ‘universe of discourse’) and an interpretation base. The universe is needed
for the interpretation of quantifiers and predicate constants. The basic idea of
interpreting an existential quantification is this: there is a given domain of cases;
Formal semantics
343
these cases are checked for the condition formulated in the scope of the quantifier;
if there is at least one case in the domain for which the scope formula is true, then
the existential statement is true, otherwise it is false. The universe of the model
provides this domain of cases. It is the ‘domain of quantification’ and plays the
same role as the domain of quantification of referential quantificational NPs in
natural language (recall 4.5). If no such domain were fixed in advance, it would
be impossible to assess that an existential quantification is false. Even if all cases
checked so far were false, there would be others yet to be checked and therefore
possibly true.
The universe indiscriminately provides the domain of cases for all quantifying
formulae. What are the ‘cases’? They are the potential arguments of the predicates
employed in the language. For example, if we are to determine the truth value of the
formula ‘∃x(fox(x) & dislike(x,y) ∧ y = j)’, the cases to be checked are the entities
the variables x and y can stand for, and these are the possible arguments of the
predicates denoted by fox and dislike. It is common practice to use the general term
individuals for all those abstract or concrete things that might be arguments of the
predicates of the PL language. Depending on the model, i.e. on the application of the
PL language, the individuals may be persons, or numbers, or points in time or sets
or whatever.
IB1 The universe U of our model of PL-F is the set consisting of the
following four individuals:
– Ken: a squinting dog
– Fritz: a young, wicked fox
– Mary: a person, neither young nor busy nor wicked nor squinting
nor a smoker
– John: a young, busy, wicked smoker
Having fixed the universe, we can assign interpretations to the basic expressions. Each
individual term receives one of the individuals in the universe as its interpretation.
The model must fix interpretations for the variables too because they can occur free
in formulae such as ‘fox(x)’ or ‘x = y’. If no interpretations for variables were fixed, it
would not be possible to determine the truth values of such formulae.
IB2 [j] = John
[m] = Mary
[x] = Ken
[y] = Fritz
The interpretations of predicate constants are entities that assign a truth value to
7
every individual in the universe of the model. As a convenient notational convention,
7 Note the different use of small capitals here; the interpretations of PL predicate constants are
neither categories nor frame attributes.
344
Understanding semantics
from here on small capitals are used for the interpretations of predicate constants,
FOX for [fox], KNOW for [know] and so on. FOX assigns 1 to Fritz and 0 to Ken, Mary
and John, because in our model Fritz, unlike Ken, Mary and John, is a fox. The table
in IB3 displays the assignments of truth values to individuals for PERSON, FOX, YOUNG,
BUSY, WICKED, SQUINT and SMOKE.
IB3
Ken
Fritz
Mary
John
[person] =
PERSON
0
0
1
1
[fox]
=
FOX
0
1
0
0
[young]
=
YOUNG
0
1
0
1
[busy]
=
BUSY
0
0
0
1
[wicked] =
WICKED
0
1
0
1
[squint]
=
SQUINT
1
0
0
0
[smoke]
=
SMOKE
0
0
0
1
Two individuals, Mary and John, are persons, one is a fox. Two individuals each are
young and wicked (Fritz and John). YOUNG and WICKED assign the same truth value in
every case, as do BUSY and SMOKE. Apart from specifying Ken as neither being a person
nor a fox, the model leaves open what species Ken is (he must have two eyes, though,
because he squints).
FOX, YOUNG, WICKED, etc. assign truth values to individuals; they are thus functions
in the mathematical sense. The particular type of function that assigns truth
values to its arguments is called a ‘predicate’ in mathematics. Since we have been
using the term predicate for a certain type of concept, we will call predicates in the
mathematical sense ‘m-predicates’.
DEFINITION 1 M-predicate
An m-predicate over a set A is a function that assigns a truth value to every
element of A.
M-predicates resemble predicates-as-concepts – let us call them ‘c-predicates’
for short – in that both yield truth values when applied to their arguments. But
while c-predicates are part of our cognitive systems, m-predicates are abstract
set-theoretical constructs, namely arbitrary pairings of arguments and truth values.
For example, the function YOUNG is the set of the four pairs Ken→0, Fritz→1,
Mary→0 and John→1. That is why m-predicates can be defined in the form of
tables that list the truth values for all arguments as we did in IB3. If two m-predicates
(or functions in general) assign equal values to all arguments, they are identical.
Formal semantics
345
Hence, in our model, YOUNG = WICKED and BUSY = SMOKE. The corresponding mental
predicates ›young‹ and ›wicked‹, or ›busy‹ and ›smoke(r)‹, are of course by no means
identical. They just happen, in this model, to be true of the same individuals. In Fig.
13.5, the seven m-predicates of our model are depicted as what they just are: sets of
argument-value pairings.
Figure 13.5
The five one-place m-predicates of the model
1
[person]
2
[fox]
3
[fat]
[wicked]
4
[busy]
[smoke]
5
[squint]
KenÆ 0
KenÆ 0
KenÆ 0
KenÆ 0
KenÆ 1
FritzÆ 0
FritzÆ 1
FritzÆ 1
FritzÆ 0
FritzÆ 0
MaryÆ 1
MaryÆ 0
MaryÆ 0
MaryÆ 0
MaryÆ 0
JohnÆ 1
JohnÆ 0
JohnÆ 1
JohnÆ 1
JohnÆ 0
There are one-place and multi-place m-predicates. The interpretations of the twoplace predicate constants know and dislike are two-place m-predicates; they assign
a truth value to pairs of individuals. Their definition requires a different form of
table. In IB4, a1 stands for the subject argument of the corresponding transitive verb,
a2 for the direct-object argument.
Fritz
Mary
John
Ken
Fritz
Mary
John
0
1
1
0
Ken
1
1
1
1
Fritz
1
0
0
1
Fritz
1
1
0
1
Mary
1
0
0
1
Mary
1
0
1
1
John
1
1
1
0
John
1
1
1
1
a2
a1
a2
Ken
Ken
IB4
a1
DISLIKE
KNOW
According to IB4, Ken dislikes Fritz and Mary, Fritz dislikes Ken and John, and so on.
In connection with m-predicates we use the common mathematical notation
‘f(x) = y’:
(22)
FOX(Fritz) = 1
for:
FOX assigns 1 to Fritz
DISLIKE(Ken, Fritz) = 1
for:
DISLIKE assigns 1 to Ken and Fritz
(in this order, i.e. Ken dislikes Fritz)
346
Understanding semantics
With the definition of the universe and interpretations for all basic expressions of
PL-F, the model – one possible model – is complete. Since the same interpretation
rules apply to all PL languages and all models, these rules are not part of a particular
model.
13.3.2 Interpretation rules for PL-F
Since all complex expressions in PL-F are formulae, we can give the interpretation
rules in a uniform format:
[S] = 1
iff
(conditions)
The conditions on the right side of iff are conditions for the sentence S being true,
i.e. its truth conditions. The first three interpretation rules take care of the prime
formulae. T and T’ are used when two terms must be distinguished:
I1
[P1(T)] = 1
iff
[P1]([T])=1
I2
[P2(T,T’)] = 1
iff
[P2]([T],[T’]) = 1
I3
[T = T’] = 1
iff
[T] = [T’]
In (23), rule I1 is applied to ‘fox(j)’:
(23) a.
[fox(j)] = 1
iff [fox]([j]) = 1
by I1
This step takes us down from the interpretation of the formula to the interpretations
of the predicate constant and its argument term. [fox] is the m-predicate FOX, [j] is
the individual John. In the next step, these are inserted:
(23) b.
[fox(j)] = 1
iff
FOX(John) = 1
by IB
We now have to check the truth value of the m-predicate FOX for the individual John
in our model. It is 0 (see IB3). Thus, the statement on the right side is false, and
consequently the one on the left side is too. In other words: [fox(j)] = 0.
An application of I2 is given in (24). (25) illustrates the interpretation of identity
statements.
(24)
[dislike(y,m)] = 1 iff [dislike]([y],[m]) = 1
iff
DISLIKE(Fritz,Mary) = 1
Since DISLIKE(Fritz, Mary) = 1, [dislike(y,m)] = 1.
by I2
by IB2, IB3
Formal semantics
[m = x] = 1
(25)
iff [m] = [x]
by I3
iff Mary = Ken
by IB2
347
Since Mary is not Ken, [m = x] = 0.
Two rules are left. The interpretation of conjunction is straightforward. It states that
a conjunction is true iff both conjuncts are true (cf. the definitions in 7.4).
I4
[S ∧ S’] = 1
iff
[S] = 1 and [S’] = 1
(26) [wicked(x) ∧ know(x,y)] = 1
iff [wicked(x)] = 1
and
[know(x,y)] = 1
by I4
iff [wicked]([x]) = 1 and
[know(x,y)] = 1
by I1
iff [wicked]([x]) = 1 and
[know]([x],[y]) = 1 by I2
iff
KNOW(Ken,Fritz) = 1
WICKED(Ken) = 1
and
by IB2,3,4
Since WICKED(Ken) = 0 and KNOW(Ken,Fritz) = 1, the conjunction is false:
[wicked(x) ∧ know(x,y)] = 0
Rule I5 for existential quantification is in need of some explanation. Grasping the
sense of such formulae is one thing, giving a precise interpretation rule, another. Let
us consider an example first:
(27) a.
∃x(fox(x) ∧ wicked(x))
In a given model, the formula is true if there is at least one individual in the universe
that is both a fox and wicked. How can this be put on terms with the general design
of the interpretation rules? What we have to do in order to check whether (27a) is
true is check the truth value of the formula fox(x) ∧ wicked(x), for different possible
interpretations of x. Is it true if we assume that [x] = John? Is it true for [x] = Mary?
And so on for each individual in our universe. If the formula is true for at least
one individual, then the existential quantification is true. This yields the following
formulation of I5, where V represents the bound variable:
I5
[∃ V(S)] = 1
iff
there is at least one individual u in U such that,
for [V] = u, [S] = 1
Applied to (27), the new rule yields the following derivation:
348
Understanding semantics
(28) [∃x(fox(x) ∧ wicked(x))] = 1
iff
iff
iff
iff
there is at least one u in U such that, for [x] = u,
[fox(x) ∧ wicked(x)] = 1
by I5
there is at least one u in U such that, for [x] = u,
[fox(x)] = 1 and [wicked(x)] = 1
by I4
there is at least one u in U such that, for [x] = u,
[fox]([x]) = 1 and [wicked]([x]) = 1
by I1
there is at least one u in U such that, for [x] = u,
FOX([x]) = 1 and WICKED([x]) = 1
iff
by IB3
there is at least one u in U such that
FOX(u) = 1 and WICKED(u) = 1
Since Fritz is such an individual, the formula is true.
We can now see why employment of an existential quantifier makes a big difference.
Compare (27) with the ‘naked’ (27b):
(27) a.
b.
∃x(fox(x) ∧ wicked(x))
fox(x) ∧ wicked(x)
The quantifier-free formula in (27b) is just about a single case, namely about the
individual we happened to fix as the interpretation of the variable x in case it is
used as a free variable. In our model, (27b) thus happens to be false because [x]
is defined as Ken, and Ken is neither a fox nor wicked. If (27b) were true, it would
be a coincidence. The use of a quantifier with a certain variable cancels the value
assignment for the bound variable. This is why the distinction between free and
bound variables is so important: free variables are simply terms denoting a certain
individual specified in the model as the value of the variable; bound variables,
however, are ‘real’ variables that stand for all individuals in the universe.
Now that the interpretation system for PL-F is complete, let us have a look at
how it works and what it buys us. It yields the assignment of an interpretation (in
the technical sense) to every basic or complex expression of the language. Thus, in
mathematical terms, an interpretation system is a complex definition of the function
[ ] which assigns each expression its interpretation in the model. Different categories
of expressions receive different types of interpretations: we assign individuals to
individual terms, m-predicates to predicate constants and truth values to sentences.
The interpretation base and the rules are designed in a way that the value of a formula
is always a truth value. For example, the interpretation rule I1 for the formula type
P1(T) produces a truth value by the following mechanism: it takes the individual [T]
and the m-predicate [P1] and lets the m-predicate work on the individual to produce
a truth value. This, then, is the resulting ‘interpretation’ of the formula, its truth value
Formal semantics
349
in the model. Table 13.1 displays the general picture for PL-F and how it works out
for such a formula.
Table 13.1
Categories of expressions and types of values
Category
individual term
predicate constant
formula/sentence
Expression
m
busy
busy(m)
Ken
→ 0
Fritz → 0
Interpretation
Mary
Mary → 0
0
John → 1
Type of interpretation
individual
m-predicate
truth value
13.3.3 Application to the translations of fragment sentences
With the interpretation system for PL-F, a first formal semantics for our fragment
of English is complete. The semantics consists of two steps. First, sentences from
the fragment are translated into PL-F. In a second step, their truth values in the
given model are determined in the interpretation system for PL-F. We will illustrate
the process for one sentence of the fragment, a fox dislikes John. The example is
somewhat complex because we have to deal with double quantification.
(29) a.
a fox dislikes John
b.
translation (cf. (19)):
c.
truth value:
∃x(fox(x) ∧ ∃y(dislike(x,y) ∧ y = j))
[∃x(fox(x) ∧ ∃y(dislike(x,y) ∧ y = j))] = 1
iff
iff
iff
there is at least one u in U such that, for [x] = u,
[fox(x) ∧ ∃y(dislike(x,y) ∧ y = j)] = 1
by I5
there is at least one u in U such that, for [x] = u,
[fox(x)] = 1 and [∃y(dislike(x,y) ∧ y = j)] = 1
by I4
there is at least one u in U such that, for [x] = u,
8
[fox(x)] = 1 and there is at least one v in U such that,
for [y] = v, [dislike(x,y) ∧ y = j] = 1
by I5
8 Each time when I4 is applied, a new variable must be chosen, hence v instead of u.
350
Understanding semantics
iff
there is at least one u in U such that, for [x] = u,
[fox]([x]) = 1 and there is at least one v in U such that, for
[y] = v, [dislike]([x],[y]) = 1 and [y] = [j]
by I1,2,3,4
iff
there is at least one u in U such that
FOX(u) = 1 and there is at least one v in U such that
DISLIKE(u,v) = 1 and v = John
iff
by IB2,4
there is at least one u in U such that
FOX(u) = 1 and DISLIKE(u,John) = 1
Since the condition is fulfilled – Fritz dislikes John, see IB4 – the sentence a
fox dislikes John is true in the model.
What is the benefit of all this formal apparatus? After all, the result looks utterly
trivial. Is not the result of the formal analysis of the sentence a fox dislikes John just a
somewhat more cumbersome way of saying the same thing? Yes, it is. But the crucial
point is: it must be that way. If the resulting truth conditions were not recognizably
those of the original sentence, the analysis would be plainly wrong. After all, semantic
analysis is to reproduce our semantic intuitions, and these are with respect to the
sentence a fox dislikes John just that the sentence says ‘a fox dislikes John’.
The main achievement of the analysis is a description of HOW THE TRUTH CONDITIONS
COME ABOUT. In the course of deriving the truth conditions for the sentences of the
fragment, we had to settle a number of semantic questions that are far from trivial.
What are the meanings of the basic expressions and what are their respective
contributions to the truth conditions of a sentence? How do their meanings combine
with those of the other parts of a sentence? How can expressions of different
meanings and forms but of the same syntactic category, e.g. proper names and
indefinite NPs, be analysed in a uniform way? The answers to these questions are
nothing less than a description of central composition rules – the main objective of
sentence semantics.
13.3.4 Model-theoretic semantics
The interpretation system for PL-F is an example of model-theoretic semantics.
Model theory is a branch of mathematics concerned with the interpretation of
formal languages such as PL. It provides general definitions of possible models and
the general rules of deriving truth conditions for the formulae of such languages on
the basis of a given model. A model-theoretic semantics is a semantics in a technical
sense in that it assigns non-linguistic entities to linguistic expressions, entities such
as referents, m-predicates and truth values which are at least somehow related to
meaning in the usual sense. When natural language sentences are translated into a
formal language and then given a model-theoretic interpretation (see Fig. 13.6 for
the general scheme), this kind of ‘semantics’ is applied to, or supplied for, natural
language. ‘Meanings’ in model-theoretic semantics are whatever interpretations the
function [ ] assigns to natural language expressions. Let us take a look at what kind
of semantics we have achieved so far.
Formal semantics
351
Figure 13.6
Two-step interpretation assignment in formal semantics
NATURAL LANGUAGE
FRAGMENT
sentences
translation system
FORMAL LANGUAGE
formulae
model-theoretic
semantics
interpretations […]
The first thing to observe is that the expressions receive interpretations only for
a given model. What is such a model? A model fixes the reference of all referring
expressions and the facts relevant for determining the truth value of every sentence.
For a natural language, a model would just be a context of utterance. There are as
many possible models as there are assignments of referents and constellations of
facts in CoUs. Hence the model-theoretic interpretation (i.e. ‘meaning’) of a proper
name is its referent in the CoU given by the model and the interpretation of a
sentence is its truth value in the model.
Clearly, the truth value of a sentence in one particular CoU is not its meaning. If
it were, all sentences true in that CoU would have the same meaning, and all false
sentences alike. Likewise, the referent of an NP cannot be equated with its meaning.
Nor does an m-predicate that assigns a truth value to every individual in a given
CoU constitute the meaning of a predicate constant. Recall how the interpretations
of young and wicked coincided in our model. Therefore, these interpretations cannot
be considered the meanings proper.
What we need in order to account for the meanings of natural language expressions
is more: a description of the potential of the expressions with respect to ALL possible
CoUs. This would, after all, gain a description of GENERAL truth conditions of predicate
expressions and sentences. In the next section of this chapter, this is accomplished by
extending model-theoretic semantics to ‘possible-world semantics’.
352
Understanding semantics
13.4 POSSIBLE-WORLD SEMANTICS
13.4.1 Possible worlds
Let us assume we are dealing not with a small fragment of English but have managed
to set up a full-fledged formation system of English, a complete lexicon and grammar.
Let us further assume that we have a translation system that yields translations of
arbitrary English sentences into an appropriate logical language LE (for Logical
English). LE has constants corresponding to all lexical items of English, in particular
predicate constants for all verbs, adjectives and nouns. What would it mean to give
a model for LE? Recall that a model is the basis for assigning truth values to all
formulae of the logical language. Since these comprise the translations of all English
sentences, the model must provide the relevant information for assigning TRUE or
FALSE to every possible English sentence. Now, sentences are not simply true or false
(except for logically true or false sentences, 7.2). Rather their truth values depend on,
and change with, the CoU. For example, the sentences in (30) keep changing their
truth values through time; at some times one is true and the other false, at others
both are true or both false.
(30) a.
b.
Mary is tired.
Mary was tired.
The time dependence is due to the fact that the VPs carry tense. That is why the time
of utterance is one component of the CoU (1.1.2). In addition, the truth values of
the sentences also depend on who is referred to by the name Mary, again a feature
of the CoU. And even if time and person reference is fixed, the truth values depend
on the particular facts given in the CoU, on what Mary has been doing and how
much and well she has slept. That is why the given facts must also be considered a
component of the CoU. When we defined the model for PL-F, we implicitly fixed all
these components of a possible CoU.
We assigned referents to the proper names; we assumed certain facts as given,
such as Fritz being young and John being a young, wicked, busy smoker; we also
implicitly fixed a reference time and interpreted all predications as relating to this
time of utterance.
What we have been calling a CoU is called a possible world in formal semantics.
There are possible worlds that correspond to reality in sharing all their facts with
what we consider our real world, while other worlds represent alternative ways of
how things might be. There is not only one ‘real’ world but many, since even if the
9
facts are real, different CoUs could be fixed within reality. Alternative, non-real
9 In some variants of formal semantics, what we call a possible world is divided into two
components. One called context comprises the parameters of the situation of utterance, i.e. it takes
care of fixing reference, while a world consists of those facts that are independent of the CoU.
In such frameworks, there is only one ‘real’ world, which, however, can be paired with different
‘contexts’.
Formal semantics
353
worlds will not be completely different, but differ only in some details. For example,
Mary may be tired in a real world w1, but might just have had a nap in some other
world w2 which is otherwise not so different from w1. This is the kind of world one
would refer to in a statement like (31):
(31) If Mary had had a nap, she would not be so tired.
The verb forms in the if-clause and the main clause indicate a shift of reference from
the real world into an alternative world. In this world Mary is said to be not as tired
as in the real one.
In actual applications of possible-world semantics (PWS), i.e. the branch of formal
semantics that employs this notion, concrete possible worlds are never actually
spelled out. This would, of course, be impossible. We are not omniscient. Possible
worlds are simply assumed as given, and even the ingredients that make up a possible
world such as time of utterance, speaker, addressee, location and circumstances are
only made explicit to the extent that is required by the semantic data to be analysed.
Possible worlds are theoretical constructs for what the truth value of a sentence may
depend on. This construction enables us to describe how reference and truth values
can be derived IF we are given the necessary information in a particular CoU.
13.4.2 Intensions
With the notion of possible worlds, we are able to assign general interpretations to
linguistic expressions. These interpretations are functions that assign the expression
not just its referent, truth value, etc., but a referent, truth value, etc., FOR EVERY POSSIBLE
WORLD.
Consider an arbitrary sentence, such as the above Mary is tired. In every possible
10
world, the sentence is either true or false. Let W be the set of all possible worlds.
Then a function can be defined that assigns to every possible world w in W the truth
value of Mary is tired in that world. In PWS the usual notation for the function is
double square brackets 冀 冁. 冀Mary is tired冁 is the function that assigns the sentence a
truth value for every possible world; the truth value in a particular world w is 冀Mary
is tired冁w .
DEFINITION 2
For arbitrary sentences S, 冀S冁 is a function that assigns to each possible world
w in W the truth value of S in that world, i.e. 冀S冁w.
10 Strictly speaking, the sentence is only true or false in those possible worlds where its presuppositions
are fulfilled. We will ignore the issue of presuppositions here.
354
Understanding semantics
This function is considered the meaning, and called the proposition, of the sentence
in PWS. You may think of the proposition of a sentence S as an infinite table that lists
a truth value for every possible world, as illustrated in (32):
(32)
Proposition of S, 冀S冁
world w1
truth value of S
1
w2
w3
w4
w5
w6
w7
w8
…
1
0
1
0
0
0
1
…
Formally, for every sentence S its proposition is an m-predicate over the set of
possible worlds. It is true of those worlds in which S is true and false of all other
11
worlds.
Let us next consider the PWS interpretations of natural language individual
terms, for example the name Mary, taken as a proper name, or the pronoun I. For
these, interpretations are defined that yield a (potential) referent for every possible
world. We will assume that part of what makes up a particular possible world w is
a ‘universe’ Uw of its own. Thus, an individual term in each world w is assigned an
individual in that universe.
DEFINITION 3
For arbitrary individual terms T, 冀T冁 is a function that assigns to each possible
world w in W 冀T冁w, the individual in Uw that is the referent of T if T is used
referentially.
For the pronoun I, 冀I冁w is the speaker who utters the sentence in world w. For Mary,
the interpretation assigns to each possible world the person referred to as ‘Mary’ in
that world. If a possible world were to have more than one such person Mary, we
would have to distinguish different names Mary1, Mary2, etc. in the lexicon and assign
them different interpretations. The interpretations for individual terms in PWS are
called individual concepts (although, again, these entities are not really concepts,
but rather merely mathematical functions). Again, you may think of an individual
concept as an infinite table like that in (32) that lists an individual for every possible
world.
11 It must be noted that a proposition in the sense of formal semantics is not a proposition in the
sense introduced in chapter 2. In perfect analogy to the distinction between m-predicates and
c-predicates, one might talk of m-propositions and c-propositions. C-propositions are mental
descriptions, i.e. concepts, of situations, while m-propositions are set-theoretical constructs,
namely assignments of truth values to possible worlds. As we saw in 5.8, c-propositions can
be considered c-predicates about situations potentially referred to. An m-proposition is the
m-predicate determined by this c-predicate about situations. Having stated this point, I will
continue to talk of ‘propositions’ instead of ‘m-propositions’ in the given context of possible-world
semantics.
Formal semantics
355
In PWS, functions of the type 冀…冁, e.g. propositions and individual concepts, are
generally called intensions. The intension of an expression assigns it its extension
for, or in, every possible world. Thus the intension of a sentence S is a proposition
and the extension of S in a particular world is its truth value there; the intension of
an individual term is an individual concept, its extension in w is an individual in U w.
DEFINITION 4 Intension, extension
For arbitrary expressions E, the intension of E, 冀E冁, is a function that assigns
to every possible world w in W an entity in w of the appropriate type. 冀E冁w is
the extension of E in the world w.
For a predicate constant, its extension in a world w is an m-predicate over the
respective universe Uw. We will therefore define the intensions and extensions of
predicate constants as follows.
DEFINITION 5
For arbitrary predicate constants P, 冀P冁 is a function that assigns to each
possible world w in W an m-predicate over Uw as its extension 冀P冁w.
Intensions of one-place predicate constants are called properties, those of multiplace predicate constants, relations. Table 13.2 gives a survey of the three types of
intensions and extensions introduced.
Table 13.2
Intensions and extensions
Expression
E
Its intension
冀E冁
Its extension in world w
冀E冁w
sentence
proposition
truth value
individual term
individual concept
individual
predicate term
(1-place)
property
function that assigns truth values
to individuals
predicate term
(2-place)
relation
function that assigns truth values
to pairs of individuals
13.4.3 Intensional models
In application to formal languages, a possible world corresponds to a model: it fixes
reference and truth values. Hence, the interpretation of an expression E in a particular
356
Understanding semantics
model corresponds to the extension of E in the corresponding possible world w: [E]
(in that model) is 冀E冁w. We will therefore call the type of models introduced in 13.3.1
extensional models.
In an intensional model for a formal language, the basic expressions receive
intensions as interpretations. Individual terms and predicate constants receive
interpretations of the following forms (for individual terms T and one-place
predicate constants P1):
(33) a.
b.
冀T冁
=
a function that assigns to each possible world w in W an
individual 冀T冁w in Uw.
冀P1冁
=
a function that assigns to each possible world w in W a oneplace m-predicate 冀P1冁w.
What we then need are intensional interpretation rules for deriving the intensions
of complex expressions. This too is straightforward. Recall the extensional rule for
prime formulae of the form ‘P1(T)’ (we attach the subscript e for extensional):
I1e
[P1(T)] = 1 iff [P1]([T]) = 1
The INtensional interpretation rule must ensure that I1e remains valid at the
extensional level. We can ensure this if we simply define I1i (i for intensional) as
follows.
I1i
for each possible world w in W 冀P1(T)冁w = 1 iff 冀P1冁w (冀T冁w) = 1
Since 冀A冁w is just [A] in the model corresponding to the world w, I1i means that the
former extensional rule I1e is to be applied for every possible world w.
The use of intensions as interpretations brings us an important step further in
coping with the problems pointed out at the end of 13.3.4. First, the interpretations
are no longer restricted to one particular CoU (possible world). Rather they cover the
range of all possible CoUs. Second, the interpretations no longer massively coincide
for expressions with intuitively different meanings. For example, the adjectives
young and wicked may happen to be true for the same set of individuals in some
possible worlds. But there will always be other worlds where they differ, because it
is POSSIBLE that someone is young and not wicked or wicked and not young. Thus,
there are always possible worlds w and individuals i in Uw such that 冀young冁w(i) ⫽
冀wicked冁w(i). Therefore 冀young冁w and 冀wicked冁w are different and hence also the
intensions 冀young冁 and 冀wicked冁. As we will see, however, intensionalization still does
not guarantee that expressions with different meanings in fact are assigned different
interpretations in PWS.
Formal semantics
357
13.4.4 Logical properties and relations
When an intensional model is supplied for the formal language into which the
sentences of natural language are translated, this yields an intensional interpretation
system for the latter. Such a system provides general truth conditions for all sentences.
For example, for the truth conditions of a fox dislikes John we obtain formally (34)
(see (28) for the derivation):
(34) for every possible world w in W, a fox dislikes John is true iff
let 冀j冁w = John; then there is at least one individual u in Uw such that
冀fox冁w(u) = 1 and 冀dislike冁w(u,John) = 1.
The availability of general truth conditions allows a very important step: the formal
definition of the logical notions introduced in Chapter 7:
DEFINITION 6
a. A sentence is logically true iff it is true in every possible world.
b. Sentence S1 logically entails sentence S2 iff for no possible world S1 is true
and S2 is false.
c. Two expressions E1 and E2 are logically equivalent, iff they have equal
extensions in all possible worlds.
We can therefore check the results of a PWS analysis by testing if the formal
interpretations produce intuitively correct logical properties and relations. For
example, the A + N combination young fox intuitively entails both young and fox:
young foxes are foxes and young foxes are young. This observation is correctly carried
out by our treatment, because it yields the following translations for the respective
test sentences with a free individual variable x:
x is a young fox
fox(x) ∧ young(x)
b.
x is a fox
fox(x)
c.
x is young
young(x)
(35) a.
Given the intensional model sketched in 13.4.3, we can easily verify the entailments.
According to what would be I4i, in every possible world w the formula ‘fox(x) ∧
young(x)’ is true iff ‘fox(x)’ and ‘young(x)’ are both true in w. Thus, whenever (35a)
is true, (35b) and (35c) are true too.
However, the intensional interpretation given so far only captures those logical
relations that originate from the application of the composition rules. The system is
not able to derive entailments based on meaning relations between lexical items, for
example that x is a dog entails x is an animal. There are two ways to handle this kind
of phenomenon.
358
Understanding semantics
First, one can set up a collection of meaning postulates such as dogs are animals,
children of the same parents are siblings, pork is meat from pigs, etc. (cf. 9.3.4). These
are declared axioms of the interpretation system, i.e. sentences that must come
out true in every possible world. The possibilities of defining the overall intension
function 冀 冁 would thereby be considerably constrained. Note that, up to now, we have
only given a general characterization of intensions. In principle, 冀 冁 can be defined
in a way such that, for example, for some world w there are individuals for which
冀duck冁w and 冀human冁w both assign 1; these individuals would then be both ducks
and human beings. If we want to exclude this kind of contradiction, we can take care
of the matter by appropriate meaning postulates.
Alternatively, we can incorporate lexical decomposition into this kind of approach.
In fact, Dowty’s theory of decomposition (9.4.1) was developed within the framework
of PWS. Lexical items to be decomposed are then translated not directly into
corresponding predicate constants (fox → fox) but into decompositional formulae.
For example, the nouns woman and man could be translated as follows:
(36) a.
b.
woman
= human(_) ∧ ¬ male(_)
man
= human(_) ∧ male(_)
We can then immediately derive that the nouns woman and man are logically
contradictory (7.5). This procedure does not constrain the intension function 冀 冁.
Rather it reduces the number of predicate constants in LE for which intensions have
to be defined. But ultimately we would of course want to constrain the interpretations
of the remaining semantic primitives, i.e. the irreducible basic expressions of LE.
13.5 THE SCOPE AND LIMITS OF POSSIBLE-WORLD SEMANTICS
13.5.1 Scope and potential
Above all, PWS is a theory that provides a precise theoretical notion of meaning:
in PWS, the meaning of a lexical or non-lexical expression is its intension in an
intensional model. For sentences, intensions are functions that assign the sentence a
truth value for every possible world, where possible world is a theoretical construct
for the CoU. The semantic description of a sentence in PWS is a definition of its truth
conditions. For all kinds of expressions, the approach provides not only a notion of
meaning (namely, intension) but also a notion of reference: for a given possible world
w, the extension of an expression E can be considered its actual or potential referent
in that world. Thus, PWS can claim to provide theoretical definitions for the following
central semantic notions:
∑ the meaning of a lexical or syntactically complex expression: its intension
∑ the context of utterance: a possible world
∑ the (potential) referent of an expression: its extension in a given world
∑ the truth conditions of a sentence: a description of its intension
Formal semantics
359
As a semantic framework, PWS is primarily aimed at describing semantic composition.
As we saw in the fragment, lexical meanings are essentially treated as simply being
given; by translating them into corresponding constants of the formal language, e.g.
know = know(_,_), they are themselves not analysed. Semantic analysis in these
cases is restricted to the assignment of the appropriate logical type (individual term
or x-place predicate constant). This corresponds to the aim of sentential semantics to
address questions such as: GIVEN the meanings of the adjective young and the noun
fox, what is the meaning of the A + N combination young fox? Neither structuralist
semantics nor cognitive semantics have so far managed to come up with a theory of
composition (but recall the application of frame theory to composition, 12.4). This is
the main asset of PWS. PWS describes composition by translating natural language
expressions into a formal logical language. The two-step procedure yields explicit
meaning representations in the form of semantically transparent and explicit logical
translations. It further allows the description of logical properties and relations.
Thus, as a descriptive apparatus, PWS provides the following:
∑ meaning representations in a logical language
∑ a description of composition rules for sentences and complex expressions in
general
∑ the derivation of logical properties and logical relations such as entailment
In addition, as we saw in the last section, PWS provides a framework for decomposition
and thereby for describing, to a certain extent, word meanings and meaning relations
between lexical items.
13.5.2 Limits
While the basic notion of intension opens up all the possibilities PWS offers, it also
demarcates its limits. First of all, meaning in PWS is defined in terms of reference and
truth conditions. Consequently all the restrictions apply to PWS that were lined out
in 7.6 for the logical approach to meaning in general:
∑ PWS cannot account for non-descriptive meaning, in particular social and
expressive meaning and the meaning of sentence type (cf. 2.2.3).
∑ Even descriptive meaning is only captured indirectly.
In terms of the semiotic triangle, PWS exclusively describes the relation between an
expression and its extensions, i.e. the right side of the triangle (Fig. 13.7).
In fact, it is not that PWS just focuses on the right side of the triangle; rather it
disregards the conceptual level completely. The two-step link from an expression via
its meaning to its extensions or denotation is bypassed, as it were, by the intension,
i.e. by the direct assignment of extensions to the expression. Since to PWS the
intension IS the meaning, there is nothing in the PWS model that corresponds to the
component ‘meaning’ in the semiotic triangle. To PWS it does not matter how the
truth conditions of a sentence come about. What counts is what the truth conditions
objectively are. But can we do without meaning? Can we account for the semantic
360
Understanding semantics
Figure 13.7
Semiotic ‘triangle’ for possible world semantics
expression
intension
descriptive
meaning
extension
data on the basis of direct expression → intension → extension assignments? The
answer is: partly, but not totally.
In PWS, two expressions with the same intension must be considered synonymous,
because synonymy is defined as meaning identity and the intensions are the
meanings. However, as we saw in 7.6.1 and 8.1, logical equivalence does not even
amount to descriptive synonymy (let alone total synonymy, which would also include
non-descriptive meaning). In particular, sentences with identical truth conditions
need not have the same descriptive meaning. One extreme consequence of the PWS
approach is that all logically true sentences receive the same meaning, because they
all have the same intension, namely the one and only function that assigns the value
TRUE to every possible world. Thus, contrary to our semantic intuitions, the following
sentences all have the same meaning in PWS:
(37) a.
Donald is a duck or Donald is not a duck.
b.
All dogs are dogs.
c.
Two plus two equals four.
etc.
The list grows considerably, if meaning postulates are added as axioms. All meaning
postulates would mean the same because they would all come out as logically true
in the resulting interpretation system. Intuitively, however, all these sentences, for
instance the examples in 9.3.5, have quite different meanings.
It is not only logically true sentences whose meanings coincide, contrary to
intuition, in a PWS approach. Likewise, pairs of sentences such as the following
turn out semantically indistinguishable (recall the discussion in 7.6), if the system
contains the relevant meaning postulates, i.e. that the day following Sunday is a
Monday:
Formal semantics
(38) a.
b.
361
Today is Sunday.
Tomorrow is Monday.
What holds for natural language expressions also applies to the meaning
representations in the formal language: they may look like having (and thereby
representing) different meanings, but actually fail to do so. For example, one might
prefer a translation system that produces (39a) rather than (39b) as the translation
of Mary squints, i.e. an analysis that treats proper name NPs as directly inserted into
argument slots:
(39) a.
b.
squint(m)
∃x(x = m ∧ squint(x))
When we interpret the formulae, as we inevitably do, in the same way as we interpret
natural language sentences, the two formulae are semantically different. One is a
simple predication with a one-place predicate applied to one definite argument. The
other is much more complex: an existential quantification applied to a conjunction of
two different predications. But as long as the formulae receive the PWS interpretations
they do, the difference is an illusion: the formulae have the same meaning because
they are logically equivalent. At best, they can be seen as representing the meaning
of Mary squints in different ways. But they do not represent two different meanings.
13.5.3 Possible-world semantics vs mentalist semantics
The shortcomings of PWS with respect to the semantic distinction of logically
equivalent expressions are immediately due to eliminating meanings proper from
the semiotic scheme by equating sentence meaning with truth conditions. To PWS,
meanings are but assignments of extensions to CoUs (possible worlds). Thus, PWS
provides a model for capturing the context dependence of truth and reference. But
it does not provide an answer to the question of how all this works at the cognitive
level. Clearly, we do not have ready-made intensions in our minds that produce the
relevant extensions (truth values, referents, m-predicates) for any context we might
encounter. Intensions are infinite, because there are an infinite number of logically
possible CoUs. Of course, PWS does not claim that we have intensions in our minds.
Rather, the approach does not aim at all at the cognitive level. But the cognitive level
exists and we would like to understand how it works.
From the mentalist perspective on meaning which has been advocated throughout
this book, truth conditions and reference are but an EFFECT of meaning, not meaning
itself. For example, the meanings of (38a) and (38b) are two situation concepts
that hold true of the same situations in the same CoUs, but they do so for different
reasons. They are different concepts. Ultimately, the task of semantics is a description
of the concepts that constitute the meanings of the expressions we use. PWS
contributes to this task in describing what these concepts must yield in terms of
truth and reference.
362
Understanding semantics
Let us compare the situation of a semanticist with that of someone who is to
describe a computer program. Essentially, computer programs are entities that
produce certain outputs when fed with certain inputs. With a PWS-like approach one
would give a complete account of the input-output characteristics of the program,
a table which for each input gives the resulting output, similar to the table given
in (32) for propositions in a PWS approach. An approach in the vein of mentalist
semantics would aim at a description of the program itself, its structure and the way
it works. Given such a description, the input-output characteristics of the program
can be predicted and explained. Clearly, differently written programs may have the
same input-output characteristics. And since this is so, we would not agree that we
know what is essential about a program as long as we just know its input-output
characteristics. Analogously, we can consider sentence meanings as mental software
that when fed with the description of a situation produces the output TRUE or FALSE.
PWS tries to capture the input-output characteristics of the software, but mentalist
semantics would try to describe the software itself, how it works and why it produces
the truth values it does. We do not know what is essential about the meaning of
a sentence as long as we only know its truth conditions. The crucial thing about
meaning is the kind of software it constitutes.
EXERCISES
1. Describe the structure of a compositional interpretation system in your own
words.
2. Introduce categories in the original system for Japanese numerals. (You can
choose arbitrary names for the categories, just be consistent and introduce
different names for different categories.) Assign categories to the basic expressions
noting that an expression may belong to more than one category. Reformulate the
formation and interpretation rules in the style of those for the English fragment.
3. The Japanese word for 100 is hyaku. The word for 200 is nihyaku (lit. ‘two
hundred’), and so on, up to kyūhyaku for 900. (Actually, some terms for the
hundreds are slightly irregular due to phonological effects of concatenation, e.g.
sanhyaku → sanbyaku. These irregularities are to be ignored in the exercise.)
To the terms for the hundreds as first parts the terms for the numbers from 1 to
99 are directly concatenated. For example, the term for 539 is gohyaku sanjūkyū.
Modify the system of rules and categories from exercise 2 to cover all numerals
up to the word for 999.
4. Describe the formation and compositional interpretation of the numeral
nanahyaku nijūichi in the style of (5).
5. a. Give a full account of the formation of the fragment sentence
A wicked fox squints in the way illustrated in (7).
b. Derive, step by step, the translation of the sentence as is done in (17), (18).
Formal semantics
363
c. Derive, step-by-step, the truth value of the translation of the sentence in the
given extensional model of PL-F (cf. (28)).
6. a. Derive the translation of A young person is Mary (the sentence is odd, but do
it for the sake of the exercise).
b. Reduce the formula by quantifier elimination.
7. Consider the PL-F formula wicked(m).
a. Derive the truth conditions of the formula in the given model for PL-F.
b. Find a fragment sentence that has the same truth conditions.
c. Translate the sentence into PL-F.
d. Derive its truth conditions.
8. a. Define an extensional model for PL-F with only three individuals, where Mary
is a young squinting fox, John a wicked dog, x is Mary and y is Sheila, a busy
smoking person. Every individual knows every other individual (including
themselves) and dislikes Mary (including Mary herself). Give complete
definitions of the interpretation of all basic terms of PL-F in that model.
Which m-predicates are equal in this model?
b. Determine the truth values in this model for the formulae in (23), (24), (25),
(27), (28), making use of the derivations given there.
9. Explain the notions possible world, intension and extension and how they are
related.
10. Describe the overall scheme of possible-world semantics. How is meaning
defined in PWS? What is the method of semantic description in PWS?
11. Explain why extensional models are insufficient for a semantic description of
natural language sentences.
12. Explain why meaning cannot be fully described in terms of truth conditions and
reference.
13. What is the difference between m-predicates and predicates?
14. Explain the difference between a proposition in the conceptual sense introduced
in 2.2 and a proposition in the sense of PWS.
15. In which respect is PWS superior to cognitive semantics? In which respect is
cognitive semantics superior to PWS?
FURTHER READING
There are several introductions to formal semantics. Some of the more recent ones
are, in alphabetic order: Cann (1993), Chierchia and McConnell-Ginet (1990), de
Swart (1998), Heim and Kratzer (1998). Bach (1989) offers an informal introduction
to central topics of formal semantics.
References
Aitchison, Jean. 2012. Words in the Mind. Oxford: Blackwell.
Anderson, Stephen R. and Edward L. Keenan. 1985. Deixis. In Shopen, Timothy (ed.),
Language Typology and Syntactic Description. Volume III: Grammatical Categories and the
Lexicon. Cambridge: Cambridge University Press, pp. 259–308.
Andersson, Lars-Gunnar and Peter Trudgill. 1990. Bad Language. Oxford: Blackwell.
Aristotle. 1958. Metaphysics. A revised text with introduction and commentary by W.D. Ross.
Oxford: Clarendon Press.
Bach, Emmon. 1989. Informal Lectures on Formal Semantics. Albany, NY: State University of
New York Press.
Barsalou, Lawrence W. 1992a. Cognitive Psychology. An Overview for Cognitive Sciences.
Hillsdale, NJ: Erlbaum.
Barsalou, Lawrence W. 1992b. Frames, Concepts, and Conceptual Fields. In A. Lehrer and
E.F. Kittay (eds), Frames, Fields, and Contrasts: New Essays in Semantic and Lexical
Organization. Hillsdale, NJ: Lawrence Erlbaum Associates, pp. 21–74.
Berlin, Brent and Paul Kay. 1969. Basic Color Terms. Their Universality and Evolution. Berkeley,
CA and Los Angeles, CA: University of Los Angeles Press.
Bickerton, Derek 1992. Language and Species. Chicago: Chicago University Press.
Bierwisch, Manfred. 1982. Formal and lexical semantics. Linguistische Berichte 80: 3–17. Also
published in Hattori, Shiro and Kazuki Inoue (eds), Proceedings of the XIIIth International
Congress of Linguistics, August 29–September 4, 1982, Tokyo. Tokyo, pp. 122–36.
Bierwisch, Manfred and Ewald Lang (eds). 1983. Dimensional Adjectives: Grammatical
Structure and Conceptual Interpretation. Berlin, New York, Tokyo: Springer.
Brown, Penelope and Stephen Levinson. 1978. Universals in language usage: politeness
phenomena. In E. Goody (ed.), Questions and Politeness: Strategies in Social Interaction.
Cambridge: Cambridge University Press, pp. 56–311.
Brown, R. and A. Gilman. 1960. The pronouns of power and solidarity. In T. Sebeok (ed.),
Style in Language. Cambridge, MA: MIT Press.
Bryson, Bill. 1990. The Mother Tongue. New York. Harper.
Cann, Ronnie. 1993. Formal Semantics: An Introduction. Cambridge: Cambridge University
Press.
Cann, Ronnie. 2011. Sense relations. In Maienborn, von Heusinger and Portner (eds, 2011),
pp. 456–80.
Carlson, Gregory. 2012. Genericity. In Maienborn, von Heusinger and Portner (eds, 2012),
Volume 2, pp. 1153–85.
Chierchia, Gennaro and Sally McConnell-Ginet. 1990. Meaning and Grammar. An Introduction
to Semantics. Cambridge, MA: MIT Press.
Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press.
Comrie, Bernard. 1985. Tense. Cambridge: Cambridge University Press.
References
365
Cooke, Joseph R. 1968. Pronominal Reference in Thai, Burmese, and Vietnamese. Berkeley and
Los Angeles: University of California Press.
Corbett, Greville G. 2000. Number. Cambridge: Cambridge University Press.
Croft, William and D. Alan Cruse. 2004. Cognitive Linguistics. Cambridge: Cambridge
University Press.
Cruse, D. Alan. 1986. Lexical Semantics. Cambridge: Cambridge University Press.
Davis, Anthony R. 2011. Thematic roles. In Maienborn, von Heusinger and Portner (eds,
2011), pp. 399–419.
Dedrick, Don. 1998. Naming the Rainbow. Colour Language, Colour Science, and Culture.
Dordrecht: Kluwer.
Deutscher, Guy. 2010. Through the Language Glass. Why the World Looks Different in Other
Languages. London: Cornerstone.
de Swart, Henriëtte. 1998. Introduction to Natural Language Semantics. Stanford: Centre for
the Study of Language and Information.
Diessel, Holger. 1999. Demonstratives. Form, Function and Grammaticalization. Amsterdam/
Philadelphia: Benjamins.
Diessel, Holger. 2012. Deixis and Demonstratives. In Maienborn, von Heusinger and Portner
(eds, 2012b), Volume 3, pp. 2407–2432.
Dillon, George L. 1977. Introduction to Linguistic Semantics. Englewood Cliffs, NJ: PrenticeHall.
Dowty, David R. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel.
Drivonikou, G.V., P. Kay, T. Regier, R.B. Ivry, A.L. Gilbert, A. Franklin and I.R.L. Davies. 2007.
Further evidence that Whorfian effects are stronger in the right visual field than in the
left, Proceedings of the National Academy of Sciences 104: 1097–102.
Dryer, Matthew S. and Haspelmath, Martin (eds). 2011. The World Atlas of Language
Structures Online. Munich: Max Planck Digital Library. Available online at http://wals.
info
Engelberg, Stefan. 2011a. Lexical decomposition: foundational issues. In Maienborn, von
Heusinger and Portner (eds, 2011), pp. 124–45.
Engelberg, Stefan. 2011b. Frameworks of lexical decomposition of verbs. In Maienborn, von
Heusinger and Portner (eds, 2011), pp. 358–89.
Fillmore, Charles J. 1968. The Case for Case. In E. Bach, R. T. Harms (eds). Universals in
Linguistic Theory. New York. Holt, Rinehart & Winston, pp. 1–90.
Foley, William A. 1997. Anthropological Linguistics: An Introduction. Oxford: Blackwell.
Galton, Antony. 1984. The Logic of Aspect. Oxford: Clarendon Press.
Givón, Talmy. 1993. English Grammar. A Function-based Introduction. Amsterdam and
Philadelphia, PA: John Benjamins.
Goddard, Cliff. 1998. Semantic Analysis. A Practical Introduction. Oxford and New York:
Oxford University Press.
Gussenhoven, Carlos and Haike Jacobs. 2011. Understanding Phonology. Third edition. New
York, London: Routledge.
Harris, Roy. 1983. F. de Saussure: Course in General Linguistics. Trans. Roy Harris. Oxford:
Gerald Duckworth.
Haspelmath, Martin. 2010. Understanding Morphology. Second edition. New York, London:
Routledge.
Heim, Irene and Angelika Kratzer. 1998. Semantics in Generative Grammar. Oxford: Blackwell.
Huang, Yan. 2007. Pragmatics. Oxford, New York: Oxford University Press
Huddleston, Rodney and Geoffrey K. Pullum 2002. The Cambridge Grammar of the English
Language. Cambridge: Cambridge University Press.
366
Understanding semantics
Hughes, Geoffrey. 1992. Swearing. A Social History of Foul Language, Oaths and Profanity in
English. Oxford: Blackwell.
Isac, Daniela and Charles Reiss. 2008. I-language. An Introduction to Linguistics as a Cognitive
Science. Oxford: Oxford University Press.
Jackendoff, Ray. 1990. Semantic Structures. Cambridge, MA: MIT Press.
Jackendoff, Ray. 2009. Compounding in the parallel architecture and conceptual semantics.
In Rochelle Lieber and Karol Štekauer (eds), The Oxford Handbook of Compounds. Oxford,
New York: Oxford University Press, pp. 105–28.
Jackendoff, Ray. 2011. Conceptual semantics. In Maienborn, von Heusinger and Portner (eds,
2011), pp. 688–709.
Kay, Paul and Chad K. McDaniel. 1978. The linguistic significance of the meaning of basic
color terms. Language 54: 610–46.
Kay, Paul and T. Regier. 2003. Resolving the question of color naming universals. Proceedings
of the National Academy of Sciences of the United States of America 100: 9085-9089.
Published online 10 July 2003, http://www.pubmedcentral.nih.gov/articlerender.fcgi?tool
=pmcentrezandartid=166442#ref1.
Kearns, Kate. 2011. Semantics. Second edition. Basingstoke, New York: Palgrave Macmillan.
Kövecses, Zoltán. 2006. Language, Mind and Culture. Oxford: Oxford University Press.
Kummer, Manfred. 2005. Politeness in Thai. In Richard J. Watts, Sachiko Ide, Konrad Ehlich
(eds), Politeness in Language: Studies in its History, Theory, and Practice. Berlin, New York:
De Gruyter, pp. 325–36.
Kuroda, S.-Y. 1973. Where epistemology, style and grammar meet: a case study from
Japanese. In Kiparsky, Paul and Stephen Anderson (eds), A Festschrift for Morris Halle.
New York: Holt, Rinehart and Winston, pp. 377–91.
Lakoff, George. 1987. Women, Fire, and Dangerous Things. What Categories Reveal about the
Mind. Chicago, IL: The University of Chicago Press.
Lee, Penny. 1996. The Whorf Theory Complex. A Critical Reconstruction. Amsterdam,
Philadelphia, PA: Benjamin.
Levin, Beth. 1993. English Verb Classes and Alternations. A Preliminary Investigation. Chicago,
IL: The University of Chigaco Press.
Levinson, Stephen C. 1983. Pragmatics. Cambridge: Cambridge University Press.
Löbner, Sebastian. 1985. Definites. Journal of Semantics 4: 279–326.
Lucy, John A. 1992. Language Diversity and Thought. Cambridge: Cambridge University Press.
Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press.
Lyons, John. 1995: Linguistic Semantics. An Introduction. Cambridge: Cambridge University
Press.
Maienborn, Claudia, Klaus von Heusinger and Paul Portner. 2011. Meaning in linguistics.
In Maienborn, von Heusinger and Portner (eds, 2011), pp. 1–10.
Maienborn, Claudia, Klaus von Heusinger and Paul Portner (eds). 2011. Semantics: An
International Handbook of Natural Language Meaning. Volume 1. Berlin, New York: De
Gruyter.
Maienborn, Claudia, Klaus von Heusinger and Paul Portner (eds). 2012a. Semantics: An
International Handbook of Natural Language Meaning. Volume 2. Berlin, New York: De
Gruyter.
Maienborn, Claudia, Klaus von Heusinger and Paul Portner (eds). 2012b. Semantics: An
International Handbook of Natural Language Meaning . Volume 3. Berlin, New York: De
Gruyter.
Martin, Samuel E. 1975. A Reference Grammar of Japanese. New Haven, CT, and London: Yale
University Press.
References
367
Matthews, Peter. 2001. A Short History of Structural Linguistics. Cambridge: Cambridge
University Press.
Matthewson, Lisa. 2011. Methods in cross-linguistic semantics. In Maienborn, von Heusinger
and Portner (eds, 2011), pp. 268–85.
The New Oxford Dictionary of English. 1998. Oxford: Clarendon Press.
Olsen, Susan. 2012b. Semantics of compounds. In Maienborn, von Heusinger and Portner
(eds, 2012b), Volume 3, pp. 2120–2150.
Palmer, Frank R. 1994. Grammatical Roles and Relations. Cambridge: Cambridge University
Press.
Palmer, Frank R. 2001. Mood and Modality. Cambridge: Cambridge University Press.
Palmer, Gary B. 1996. Toward a Theory of Cultural Linguistics. Austin, TX: University of Texas
Press.
Partee, Barbara H., Alice ter Meulen and Robert E. Wall. 1993. Mathematical Methods in
Linguistics. Dordrecht: Kluwer Academic Publishers.
Plag, Ingo. 2003. Word-formation in English. Cambridge: Cambridge University Press.
Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, MA: MIT Press.
Radford, Andrew. 1988. Transformational Grammar. Cambridge: Cambridge University Press.
Riemer, Nick. 2010. Introducing Semantics. Cambridge: Cambridge University Press.
Salzmann, Zdenek. 1993. Language, Culture and Society. An Introduction to Linguistic
Anthropology. Boulder, San Francisco and Oxford: Westview Press.
Sapir, Edward. 1921. Language. An Introduction to the Study of Speech. New York: Harcourt,
Brace and World.
Seuren, Pieter A.M. 1991. Präsuppositionen. In: A. von Stechow and D. Wunderlich (eds),
Semantik. Ein internationales Handbuch der zeitgenössischen Forschung/Semantics. An
International Handbook of Contemporary Research. Berlin, New York: De Gruyter, pp.
286–318.
Siewierska, Anna. 2004. Person. Cambridge: Cambridge University Press.
Shibatani, Masayoshi. 1990. The Languages of Japan. Cambridge: Cambridge University Press.
Sohn, Ho-Min. 1999. The Korean Language. Cambridge: Cambridge University Press.
Suzuki, Takao, 1978. Japanese and the Japanese. Words in Culture. Tokyo, New York, San
Francisco: Kodansha.
Tallerman, Maggie. 2011. Understanding Syntax. Third edition. New York, London: Routledge.
Tyler, Andrea and Hiroshi Takahashi. 2011. Metaphors and metonymies. In Maienborn, von
Heusinger and Portner (eds, 2011), pp. 597–621.
Ungerer, Friedrich and Hans-Jörg Schmid. 2006. An Introduction to Cognitive Linguistics.
Second edition. Harlow: Pearson Education.
Van Valin, Robert D. 2001. An Introduction to Syntax. Cambridge: Cambridge University Press.
Van Valin, Robert D. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge
University Press.
Vendler, Zeno. 1957/67. Verbs and times. Philosophical Review 56: 143–160. Reprinted 1967 in
Vendler, Z., Linguistics in Philosophy. Ithaca: Cornell University Press, pp. 97–121.
Verschueren, Jef. 1999. Understanding Pragmatics. New York, London: Routledge.
Wales, Katie. 1996. Personal Pronouns in Present-day English. Cambridge: Cambridge
University Press.
Watts, Richard J. 2003. Politeness. Key-topics in Sociolinguistics. Cambridge: Cambridge
University Press.
Wierzbicka, Anna. 1996. Semantics. Primes and Universals. Oxford, New York: Oxford
University Press.
Whorf, Benjamin Lee. 1940. Science and linguistics. In Whorf (1956), pp. 207–19.
368
Understanding semantics
Whorf, Benjamin Lee. 1956. Language, Thought and Reality. Selected Writings by Benjamin
Lee Whorf, ed. John B. Carroll. Cambridge, MA: The MIT Press.
Wittgenstein, Ludwig. 1958. Philosophische Untersuchungen/Philosophical Investigations.
Trans. G.E.M. Anscombe. Oxford: Blackwell.
Index
In the index, ‘f ’ refers to the following page, ‘ff ’ to the following two pages; wider ranges of
pages are given explicitly. Consult the bold type page numbers for the pages where the term is
introduced and defined. If you look up a combination of adjective plus noun, e.g. ‘intransitive
verb’, look for a sub-entry of the noun, except for fixed terms like ‘personal pronoun’, ‘social
meaning’ or ‘possible world’.
accommodation 50, 100f
accomplishment term 141–4, 146, 155, 211
achievement term 144ff, see also simple
change term
active voice 124, 136f
activity term 142ff, see also process term
addressee 5, 28ff, 32, 62ff, 67, 70, 94, 113,
192, 253, 353
adjective: as predicate term 117ff;
attributive use 117, 336, 340;
comparative 118; non-predicational
119, 207; predicative use 118, 336, 340;
scalar 210; two-place 118
adjunct 108, 135
adverbial quantification 92f
adverbs/adverbial 33f, 210, 214, 291; deictic
71f; expressive 34f; of frequency 92;
quantifying 93, temporal 72f, 159
affix 9, 226f
affordance 315f
agent (role) 122–6, 128f, 135–44, 165, 214,
227, 230, 238, 310, 312, 316, 319f
aggregate mass nouns 81
agreement 124f
aktionsart 141, see aspectual class
alienability 104, 117, 218
all 74, 85, 89, 90, 171, 210, 291
all-or-nothing contrast 88f, 91f, 98, 290
alternation 136–40; causative/inchoative 139,
237; conative 139; dative 139f; simple
reciprocal 140; understood reciprocal
object 140; understood reflexive object
140; unspecified object 139
ambiguity 41–59, 148, 221; compositional 48f;
contextual 49–54; contextual vs lexical
56ff; of grammatical forms 49; lexical 44ff
anaphor, anaphora 64, 72; definite
descriptions 78
anchoring 15, 62, 74, 162
antecedent 64f, 78f
anticausative 138f
antipassive 137, 139f, 206
antonymy/antonym 209f, 213, 215f, 236, 244
appositions 78
arbitrariness 223
argument 108–11; core ~ 135; implicit
232f, 241, 253; oblique 135; parasitic
110; referential 110f, 113, 115, 116,
151, 241; relational 116f, 119, 312f, see
also possessum; situational 162
argument compound 318
argument frame 311f
argument sharing 120f, 130f, 336, 341
argument structure 123, 135–40, 162, 311f
argument term 108–11
Aristotle 168, 170, 267
arity 108
article 10; articles across languages 83;
definite 5, 19f, 72, 75–9, 258; indefinite
81ff
aspect 150–7, 162; habitual 149f, 154, 160;
imperfective 150, 153f; perfect 154f;
perfective 150–3; prospective 155f; in
English 163; in German 163; in Russian
163; in Japanese 164; interaction with
tense 157ff
370
Understanding semantics
aspectual class 140–50
associative plural 66, 104
at-adverbial 146, 147, 149
attribute: (in a frame) 302–11; (in
grammar) 48; (in logic) 168f; ~ terms
77, 307; ~ types 309f
attributive use 117ff, 131, 336, 340
Austin 6
auxiliary 172
bare mass noun 81, 83, 86
bare plural 81, 83, 86
Barsalou 266, 301, 302, 307, 321f
Barsalou frames 301–23
basic category 276–9, 286, 298, 310
basic expression 327
basic level 276–9
BCT basic colour term see colour term
be about to 155
be going to 155
BECOME 238, 244f
Berlin and Kay 258–61, 268
BFA see Binary Feature Approach
Bickerton 13
Bierwisch 51, 54
big 209f, 236, 244, 287f
Binary Feature Approach 228–36, 245,
268, 298, 301, 321f
binary features 228–36, 321f
body part term 35, 218, 233, 235
both 74, 85, 89, 90
bottom-up process 11, 41, 56
bound morpheme 227
bound variable 335, 348
bra 193, 257f, 278
break 138f, 243f
brother 82, 190, 217, 22, 257
Burmese: personal pronouns 67; social
meaning 31
c-predicate 354
case 124
categorization 266–84
category 19, 24; basic 276–9; ~ boundary
268, 280f, 288, 292; cognitive 266f;
cultural 292–8; grammatical 42, 226,
237, 329, 331f; semantic 292–8
causative voice 137f, 162
CAUSE 238, 240f, 244f
change verb 145ff, 148, 149, 155, 164, 179,
206, 211
Chinese (Mandarin): agreement 124;
articles 83; personal pronouns 67;
social meaning 31; tense 72
citation form 43
CKA see Cultural Knowledge Approach
classeme 234
coda 225
coercion 58, 144, 147, 149, 212
cognate 44
cognitive semantics 269, 284f, 294, 299,
324, 359
coherence 95
collective predication 87
colour term 47, 57, 212, 224, 244, 258–61,
268, 287f, 289
communicative meaning 6f, 12, 15, 16, 28,
30, 38, 41
comparative 8, 27, 42, 117f, 210, 212, 214,
227
complement 96, 108ff, 112, 118ff, 124f,
127f, 130f, 230; clause ~ 98, 112; core
~ 135; oblique ~ 135
complementarity: logical 190f, 194, 196,
212, 219; opposition 212, 216, 230,
232, 236, 245
complex words 8, 12, 226f
composition 7–12, 56, 131, 319ff, 326–9
composition rule 14, 15, 327, 328, 346ff,
356
compositional ambiguity 48f
compositional meaning 7
compositionality: Principle of 10ff, 14
compound 12f; and cultural knowledge
294ff; argument ~ 318; frame ~
317f; meaning 13, 207, 294ff, 316–19;
regular 207, 294ff, 316–19; synthetic
318f; subtypes 207, 316f; value ~ 318
conative 139, 143, 144, 147
concatenation 326
concept 19f, 22, 24f, 36, 47, 52ff, 58, 80f,
108, 140–9, 192f, 219, 239ff, 266f,
296f, 302
Conceptual Semantics 239ff
conclusion (logical) 175
condition (of categorization): necessary
~ 267f, 270, 272f, 279f, 283f, 288; see
also features
Index
conjunction (‘and’) 121, 127, 184, 235, 334;
in logic 184; logical vs natural 185f;
presuppositions 200
connotation 36–9
constituent 10, 225
constraint (in frames) 310f
content word 2, 16, 20–5, 38, 130, 191, 203
context 3ff, 45, 80, 114,125, 128, 155, 169,
249, 287f; sentential 56, 128, 321; social
78; see also context of utterance
context dependence 4–7, 49–56, 62–74, 78ff,
85, 143, 276, 283, 292
context of utterance 5, 15, 19–26, 28, 30, 33f,
47, 49, 55f, 64f, 69, 72, 73–9, 91f, 94ff,
98–102, 110, 111, 170–4, 197f, 200, 282,
291, 351ff, 356, 358, 361
contextual ambiguity 49–59
contextual information 49
contextual knowledge 16
contextual meaning shifts 51–4, 55–9
contextual polarity 198
contingent sentence 173f, 177f, 181ff, 193,
196
contradiction, logical/logically
contradictory 179f, 182f, 190, 196, 209,
230, 358
contrariety, logical/logically contrary 179f,
182f, 194, 212f, 285
contradictory see logically false
contrastive property 226
converses 213f
conversion (in word formation) 12
copula 118, 119
core argument 135
coreferent, coreference 79
CoU see context of utterance
count noun 80f, 83
cue validity 274
culmination 141, 145
cultural see category; concept; knowledge
Cultural Knowledge Approach 294ff
culture 263
Danish: formal personal pronouns 67
de Saussure 223f, 246
declarative see sentence
decomposition 221–45, 322, 357f
definite article 15, 19f, 75–9, 258; evolution
72; function 79; indexicality 78
371
definite descriptions 75, 82; anaphoric use
78; deictic use 78
definite NPs 82; ~ and quantification
85f; mass ~ 87; plural ~ 86f;
presuppositions 96
definiteness 74–80, 115; of demonstratives
72; of personal pronouns 64
deictic 62
deictic centre 62
deictics: adnominal 74; adverbial 74;
pronominal 74
deixis 62–74, 101; person ~ 63–70; place
~ 70, 119; social ~ 40, 104; time ~ 72f,
119, 158; see also tense
demonstratives 70ff; adjectival 71, 72;
adnominal 71, 79, 82; anaphoric use 72;
definiteness 72; distal 71, 72; English
71; German 71; Japanese 70f; meaning
71f; medial 71, 72; pronominal 71, 82;
proximal 71; Spanish 71; systems 71
denotation 23–6, 188–96, 265f, 288, 359f;
actual 24; of colour terms 258–62; of a
content word 24; of mass nouns 80; of
plural nouns 80; of a sentence 25f
derivation 12, 226f, 312
descriptive meaning 20–30, 32, 34–8, 71,
73, 192–6, 204, 222, 266, 359; of content
words 22; of expressions with social
meaning 28f; of expressives 35; of
grammatical forms 23; of sentences 23
determination 74–83; and predication 115
determiner 71, 74, 84f, 89f, 93
deverbal noun 312f
diathesis 136–9, 162
dictionary form 43
differentiation: of categorization 289ff;
conceptual 53f, 58f, 101, 174, 206;
lexical 289
dimensions of meaning 18–39, 192, 203f,
359
directional opposites 210f
disambiguation 45, 49, 50f, 129
discourse roles 63, 67
discrimination 37
disjunction: in logic 184; logical vs natural
186
distal 71
distinguisher 234
distributive predication 87f
372
Understanding semantics
domain of quantification 83f, 96, 101, 343
DoQ see domain of quantification
Dowty 237ff
drink 240; English, German, Japanese 249
dual (number) 66
each 74, 84, 85, 89, 90, 96
eat 114, 139, 141, 143; English, German,
Japanese 249
economy of language 46, 47, 50, 290, 297
egressive 145
English: 2nd person pronouns 67; causative
138; demonstratives 71; perfect 155;
possessive constructions 116f; sibling
terms 190; tense and aspect system 163
entailment, logical 99, 175f, 180–3, 188,
195f, 359; and meaning 195f; mutual
175; ~ reversal 176; transitivity 178;
unilateral 176; vs if – then 177, 181; of
words 188
episodic 91, 149, 153, 156; quantification
92f, 96, 101
equivalence, logical 176, 178f, 183, 357; and
meaning 192–6, 203f, 360f; of words
187f, 203f, 213
euphemism 37, 204
event 113, 153
every 74, 85, 89, 90, 93, 171f
evidential 254ff
exclamation 34
existential quantifier 334, 347
experiencer (role) 123, 251
expression meaning 1–4, 14ff
expressive 33ff; ~ adverbs 34; vs
propositional meaning 34
expressive meaning 33ff, 192, 196, 203f,
359; and truth conditions 192; of an
utterance 38
extension 354f
face-to-face 62
facets of meaning 58, see dimensions of
meaning
factive verb: presuppositions 98
family resemblance 270f, 272, 273
feature 228–36, 268, 270f, 273f, 283, 288,
294ff, 322f; ~ semantics 228–36, 322f
felicity conditions 30
Fillmore 311, 324
focal colour 259f, 262, 268, 287
for-adverbial 144, 146, 147, 149
formal semantics see semantics
formality 16, 28f, 31f, 38, 66f, 69, 192, 222,
251
formation base 327, 330, 332
formation rule 327, 331f, 333f
fountain pen 193
fragment (of English) 330–41, 349f
frame 245, 301–23; basic structure 306f
frame compound 317f
frame referent 304, 307
FrameNet 324
free morpheme 227
free variable 335, 348
Frege 11, 170
French: definite articles 72;demonstratives
71; formal personal pronouns 67, 69;
gender 65, 258; headache construction
251ff; indefinite articles 81; [r] sound
223
function (of objects) 78, 217, 278, 283
function word 2f, 13, 20, 23
functional concept 307ff
functional noun 77, 115f, 308f, 318
fusion 127ff, 131, 241
future tense 161f, 163f
fuzziness 269f, 280–3
gender 42, 64f, 104f, 128, 258; of pronouns
63f, 66, 115, 128
gender languages 64f, 258
gender systems 104
Generalized Quantifier Theory 105
genericity/generic: habitual 149f; NPs 90–3
German: anticausative 139; bitte, danke
29; demonstratives 71; future tense
161; gender 65; headache construction
251f; non-past 160; passive 137;
personal pronouns 63f; progressive 163;
pronouns of address 28f, 31, 63f, 66f,
192, 250; reflexive verbs 140; tense and
aspect system 163; word order 125
give 139f, 240
GO 240f
goal (role) 123
graded membership 271ff, 279ff, 285f, 291,
298f
graded structure 269, 271ff, 279ff, 284, 292
Index
grammatical category 42
grammatical form 8, 42; ambiguity 49
grammatical meaning 8f, 12, 14, 15, 20, 37f,
48f, 135, 323; see also aspect; number;
sentence type; tense
grammatical person see person
(grammatical)
grammatical property 42
Grice 99
group 80
habitual aspect 149f, 154, 160
he, she 64f, 94ff, 234
head: of a compound 204ff, 236, 295, 298,
317ff; of an NP 115, 117
hedge 289
heteronymy 212f
hierarchy: of categories 276–9; lexical 189f,
213, 216f; mereologial 218f; of thematic
roles 135; of verb functions 162
historic present 149, 160, 164
holonym 219
homography 45
homonymy 43ff; and polysemy 46f
homophony 45
Hungarian: formal personal pronouns 67;
gender 66; headache construction 252;
interjections 33; possessive affixes 68,
252; sibling terms 190; verb agreement
124
hyperonym 205
hyponymy 191, 194, 195, 204–7, 213, 216f,
218, 219, 222, 227, 230, 232, 236, 245,
319
I (pronoun) 2, 4f, 15, 32, 34, 62, 64, 68, 115,
242f, 250, 354
idiom 41, 59
ignorance 282
I-language 14
illocutionary act 6f
imperative 15, 27, 29, 112, 285
imperfective aspect 150–4, 156–61, 163f; see
also progressive aspect; habitual aspect
imperfective verb 42, 163
impersonal constructions 137
implicature, conversational 6, 99
in-adverbial 143, 144, 146, 147
inalienability 104, 117, 218
373
inclusive/exclusive we 104
incompatibility, logical 179, 182, 230; of
words 189, 196, 209, 230, 231, 245
indefinite article 19, 74, 75, 81ff, 90, 331,
338
indefinite NP 81ff, 93, 336, 338, 341; and
quantification 85f, 88, 89; generic 90ff
indefinite pronoun 83
indexicality 15, 62, 67, 75, 78, 80, 94, 101
individual (logic) 343f
individual concept (in possible world
semantics) 354f
individual constant 120, 333, 335, 342
individual noun 77f, 115f
individual term 120f
individual variable 120f, 333, 335, 342, 348
indivisibility, presupposition of ~ 98, 151,
198, 290
inference 6, 99f, 101, 177
inflection 68, 112
information: contextual 20, 49, 352f;
semantic 2, 14, 27f, 71, 75, 86, 93, 94f,
107f, 127f, 134, 152f, 174, 177f, 321f; ~
transfer 18ff
informativeness 174, 178, 279
ingressive 145
initial condition 97, 144ff, 312
instrument (role) 123, 124, 135, 139
intension 353ff
interaction, social 6, 7, 18, 28ff, 36, 39, 66,
251, 302; with objects 278, 316
interface 15; of expression meaning and
utterance meaning 15, 56, 57f, 62,
101; of expression meaning with
communicative meaning 15, 30; of
expression meaning with utterance
meaning 15, 56, 101; of language and
thinking 38
interjection 33, 244
interpretation base 327, 329, 343ff, 348
interpretation rule 12f, 327f, 329, 330, 346f,
356; see also composition rule
interrogative sentence 4, 15, 27, 99f, 285f;
meaning 27; presuppositions 99;
proposition 27
intransitive (verb) 42f, 111, 114, 333
Irish: articles 83
irregularity (semantic) 129f
Italian: formal personal pronouns 67
374
Understanding semantics
Jackendoff 13, 127, 239
Japanese: addressing 251, 257; agreement
124; articles 83; brother terms 222;
causative 137f; demonstratives
70f; evidentials 254ff; headache
construction 252–6; homophones
45; numerals 325ff; omission of verb
complements 114; order of verbal
affixes 162; personal pronouns 67;
possessive pronouns 68; progressive
164; self-reference 250; seichō 45;
sibling terms 190; simple change terms
146; social meaning 31, 113; tense and
aspect system 164
Kay 258
knowledge: cultural 292–9; expert ~ 16,
293; personal 292ff; semantic 16, 232,
292–9; semantic vs cultural 296ff; world
~ 16, 138, 177, 232f, 302
Korean: agreement 124; personal pronouns
67; plural 9; social meaning 31, 113
Labov 269, 281, 283
Lakhota: (in)alienable possession 117
Lakoff 315
Law of Contradiction 168f, 170, 185
Law of the Excluded Middle see Principle of
Polarity
levels of meaning 1–7, 15f, 37f, 41
Levin verb classes 139f
lexeme 41–4, 46f, 224, 226f, 306; composite
43
lexical entry 43
lexical field 214–19
lexical meaning 7, 41ff, 49, 50, 221, 231, 245,
267, 302, 326, 329, 359
lexicon 7, 8, 15, 41, 47, 50, 57, 236, 256, 289,
297, 309, 329
light 44
linking 123ff, 139, 227, 241, 320
localism 240
locative 123
locutionary act 6
logic 167–200; and meaning 191–200; and
semantics 196–200; see also predicate
logic; sentential logic
logical exclusion 179
logical independence 183
logical relation 175–83, 187–96, 203,
230, 245, 357ff; between contingent
sentences 182f; between sentences
175–83; between words 187–91;
and meaning 182, 183, 191–200;
pathological cases 180ff; survey 180,
190
logically false sentence 25, 49, 173f, 180ff,
185, 193, 352
logically true sentence 25, 173f, 180ff, 185,
193, 215, 234, 352, 357, 360
many 83, 90, 92, 290f
marker 234
marry 116, 128, 140, 186, 214, 312
mass noun 80ff, 86, 106, 249; bare 81
mass/count distinction 80f
material implication see subjunction
meaning 1–16, 18–39, 191–6, 223, 226ff,
358–62; communicative 6f, 12, 15,
16, 28, 30, 38, 41; compositional 7; of
compounds 13, 207, 294ff, 316–19; of
a content word 20; descriptive 20–30,
32, 34–38, 71, 73, 192–6, 204, 222,
266, 359; grammatical 8f, 14, 15, 49;
lexical 7, 41ff, 49, 50, 221, 231, 245, 267,
302, 326, 329, 359; and logic 191–200,
358–62; of personal pronouns 64ff; of
possessive pronouns 68f; of a sentence
20; unpredictable 13
meaning postulate 234, 358, 360
meaning relation 58, 123, 182, 183, 191–6,
203–19, 221f, 227, 230, 232, 236f, 239,
244f, 265, 297ff, 357ff
meaning shift 5, 14, 15, 49–54, 56, 57ff,
101, 110, 128ff, 144, 147, 149, 160, 174,
299, 314, 321; differentiation 53f, 58,
59, 101, 174, 206; metaphor 52f, 56–59,
101, 129, 194, 206, 208, 299, 315, 324;
metonymy 51f, 53, 56, 57ff, 101, 129,
153, 233f, 313ff; and polysemy 58f; and
presuppositions 101; and selectional
restrictions 129
meaning variant 44–50, 53, 57ff, 127, 203,
204, 207, 216, 249, 299, 316, 321
meaning variation 281
medial (demonstratives) 71
mentalist semantics 14, 38, 285, 361f
mereology 217ff, 278
Index
meronymy 217ff
metaphor 52f, 56–9, 101, 129, 194, 206,
208, 299, 315, 324
metonymical relation 233f
metonymy 51f, 53, 56, 57ff, 101, 129, 153,
233f, 313ff
microsenses 58
modal adjectives 119
modal verb 135, 291
modality 135, 161, 162
modalization 291
model 342–5; extensional 356; intensional
355–8
model-theoretic semantics 342–58
modification 206
monosemy 46
Montague Grammar 237
mood 8, 135, 162
morpheme 226f
motivation (linguistic); of meaning
components 232f
m-predicate 344ff, 348, 350, 351–6, 361
m-proposition 354n
Natural Semantic Metalanguage 242ff, 245
negation 3, 87, 90, 98ff, 134, 145, 171f,
180, 285f; external vs internal 89;
of generic sentences 91f; in logic
184; in logic vs language 185; and
quantification 88ff; radical 199f
negation test for presuppositions 100
no 73, 83, 84, 85, 86, 89, 172, 210
none 86
non-past tense 159ff, 163f
noun 9f, 13, 42f; 115ff, 130, 134f; count ~
80f, 83; deverbal 113, 312f, 315, 318;
functional 77f, 115, 308f; gender 42f,
65, 104f, 258; grammatical meaning
9; individual 77f, 115; inherent
uniqueness 76ff, 116; logical relations
187–91; mass 80ff, 86, 106, 249;
meaning relations 207, 210, 212, 214,
217, 218; one-place 115; partitive
84ff; predicative use 119f, 338;
relational 69, 77, 115ff, 235; sortal 69,
78, 115ff
noun phrase see NP
NP 10; definite 74–80, 82, 86f, 96, 197;
generic 90–3, 96; indefinite 81ff, 85f,
375
90ff, 330, 336, 341; predicative 336,
340; quantifying 83–89, 92f, 197;
referential 336, 340; simple indefinite
83, 86, 90ff
NSC (necessary and sufficient conditions)
see condition (of categorization)
nucleus (syllable) 225
number (grammatical): 9; dual 66; paucal
66; pronouns 63, 65, 66; singular and
plural 80; trial 66; in verb inflection
68, 112, 124
numeral 78, 80, 81, 83, 85, 90, 215, 236,
249; composition 325ff; Japanese
325ff; ordinal 78
object (grammatical): direct 10, 106, 112,
122, 124f, 135ff, 140, 320; indirect 106,
122, 135, 137
oblique see complement
occurrence terms, simple 146ff
one (numeral) 85; (pronoun) 115, 126, 137
one-place, two-place see arity
onion: nominal 90; verbal 162
onset (syllable) 225
open 95, 97, 112, 114, 122f, 126f, 135, 138,
139, 206, 211, 237f, 243
opposition 207–14, 227
orthograpy 43
paradigm 63; in structuralism 225f
pars pro toto 58
part, constitutive 219, 309
participant 122, 135, 162
participants see thematic roles
particular sentence see episodic
partitive 84ff
partitive NPs 84ff
partonomy 219
parts of speech 42
passive voice 8, 123f, 136f, 139, 165, 214;
of causative 138
past tense 8f, 31, 42, 49, 72f, 113, 131, 150,
151, 154, 156–9, 163f, 211
path (role) 123, 135, 239ff, 312f
patient (role) 123, 125, 135ff, 140, 312
paucal (number) 66
perfect aspect 20, 154–9, 163f;
experiential 155; of persistent
situation 155, resultative 154f, 164
376
Understanding semantics
perfective aspect 8, 134, 150–3, 155–61,
163f; vs imperfective 150–3; with
present tense 159f
perfective verb 42, 163
performative verb 149, 160
person (grammatical) 63, 64, 68
person deixis 63–70
person terms 65, 216, 219
personal pronoun 2, 4, 63–8, 82, 95f, 115,
250; definiteness 64; gender 64f, 104f;
languages without 67; meaning 63–6;
number 104f; paradigms 66, person
meaning 64; plural 65f; strategies of
formal address 66f
phoneme 223, 225, 228
phonology 228
place deixis 70, see demonstratives
plural 8f, 42f, 33, 65ff, 74, 80–3, 124, 204,
227; associative 66, 104; collective 87;
definite 88f, distributive 87
polarization 171, 198, 285f, 288f, 291f, 298f
politeness see formality
political correctness 37, 204
polysemy 44, 45ff, 124, 127, 161, 203, 208,
216, 249, 299; and homonymy 46f;
and meaning shifts 58f; principled 58;
systematic 58
possession 69, 116f; (in)alienable 117, 218;
verbs of change of ~ 139, 208f, 235, 240
possessive chain 309
possessive construction 4, 116f, 120, 316
possessive pronoun 2, 4, 68–71, 74, 82, 90,
95f, 116; meaning 4, 67–70; reference
21, 116; social meaning 28, 69
possessive suffix (Hungarian) 68, 252
possessor 68, 76, 77, 116, 303f, 306–11, 318
possessum 69f
possible world 352–8
possible-world semantics 352–62: limits
359ff; vs mentalist semantics 361f;
scope 358f
potential referent 19, 21–4
power button (terms for) 194
PP 48, 55, 108, 112, 118, 120, 134, 262; by-PP
124, 136; of-PP 84, 116; to-PP 106, 320;
with-PP 124, 312
pragmatics 6, 16, 62
pragmatism 282
predicate 108–11; c-predicate 344;
m-predicate 344; of the sentence 10, 62,
169
predicate constant 120, 333
predicate expression 108–11, 115–20, 123,
125–131, 187f, 190, 213, 228, 235f, 241,
286, 298, 319, 337, 341, 351
predicate logic 120ff; semantics 342–9;
syntax 333–6
predicate term 108–11; see also predicate
expression
predication 6, 16f, 62, 106–30, 134, 197ff,
284, 286, 298, 319f; central 134;
collective 87; and determination 115;
distributive 87f, 98; quantified 83–6, 89,
93, 98
predicative: adjective 117ff, 125, 131, 336,
340; noun, NP 82, 119f, 125, 131, 336,
340
prefix 9
premise 175
preposition 2, 42, 124, 214f
prepositional object, adjunct, complement
112, 122, 135
prepositional phrase see PP
present tense 3f, 8, 72f, 124, 149f, 157–61,
163f, 198, 227; of dynamic predications
149; with perfective aspect 159f; with
timeless states 160
presupposition 62, 94–101, 151f, 170f, 185,
196–200, 285; accommodation 101; of
achievement terms 146; ~ carriers 96ff;
of definite NPs 96; existential 96; of
factive verbs 98; of indivisibility 98, 151,
197f, 290; ~ and meaning shifts 101;
and the Principle of Polarity 196–200;
projection problem 200; of quantifers
96; selectional restrictions 97; tests 100;
and truth conditions 98ff; of verbs of
change 97, 145
prime formula 333f
Principle of Compositionality 10ff, 14, 56,
326
Principle of Consistent Interpretation 49,
55ff, 100f, 110, 125, 128, 169, 174, 321
Principle of Polarity 170f, 174, 184f, 197–
200, 285f
principled polysemy 58
Index
process term 142ff, 147, 149, 150, 153, 158,
164
productive 12
progressive aspect 8, 56, 143, 144, 146, 147,
149, 150, 153, 156, 157, 163f; of state
predications 149
projection problem of presuppositions 200
pronominal deictic 74
pronoun 2, 15, 19; impersonal 137;
indefinite 83; of address 28ff, 40, 63,
66f, 250; see also personal pronouns;
possessive pronouns
proper names 51, 82, 83, 95f, 106, 108, 115,
319, 330, 336ff, 341, 151, 354
property 58; ~ attribute 309f; combinatorial
225, 226, 228, 230, 234, 241, 245, 329;
compositional 222, 245; contrastive
226, 234, 235; distinctive 44, 123;
grammatical 42ff, 124; interactional
316; logical 172–5,192, 194, 196f,
202, 357, 359; paradigmatic 226, see
contrastive property; in possible world
semantics 355; of prototypes 271, 322;
syntagmatic 226, 228
proposition 22–8, 29, 34, 38, 112, 116, 130f,
193, 284, 286, 321; in possible-world
semantics 354f
propositional meaning see descriptive
meaning
prospective aspect 155ff, 158, 163f
protolanguage 13
prototype semantics 284–92
prototype theory 267f, 279–84, 291f
prototypes 267–76, 268f, 322; features 274;
similarity 275f
proximal (demonstrative) 71
punctuality 146, 148, 151, 156, 158
Pustejovsky 241, 245
quantification 83–90, 289ff, 330; adverbial
92f; and negation 88ff; domain of ~ 84,
343; episodic 93; generic 92f; universal
see all; each; every
quantified predication 84
quantifier 83ff, 89f, 92f, 171f;
presuppositions 96
quantifier, existential 334, 347ff
quantity specification 81ff, 86, 90, 92
quantized object 143
377
question test 100
questions 285; see also integrative sentence
radical negation 199f
reading 49
reciprocal 186, 214, 312; ~ alternation 139,
140
reference 2, 62, 196, 325, 351, 355, 358, 361;
of adjectives 21; of content words 21f;
and descriptive meaning 21f; of nouns
21; of sentences 22; of tense 21; of verbs
21
referent 2; of a frame 304, 307
referential argument 110; of nouns 110,
117; of verbs 113, 151, 241, 311
relation: in possible world semantics 355;
mereological 233; metonymical 233;
paradigmatic 224f, 230; syntagmatic
224f, 227f; transitive 178, 219; see also
logical relation; meaning relation
relational noun 69, 77f, 115ff, 120, 135, 218,
235
relativism 256ff, 262
Relevance Theory 6
repetition, repetitive 147, 150, 153, 154, 156
representation: meaning ~ 238, 240, 242,
245, 330, 359, 361; mental 266
resultant condition 96, 144–8, 155, 211, 312
rhyme (syllable) 225
roles see thematic role
Rosch 269, 273, 277
running commentary 149, 160
Russian: articles 83; formal personal
pronouns 67; gender 42, 65; headache
construction 252f; no copula 134;
perfective and imperfective verbs 42;
tense and aspect system 163f
Sapir 256f, 262, 292f
Saxon genitive 82, 309
scalar adjectives 210
scenario 301
schema 301
scope: of possible world semantics, 358f; of
a quantifier 334; of semantics 14ff
script 301
Searle 6
selectional restrictions 56, 97, 101, 125–30,
221f, 228, 230, 232, 235, 240, 308
378
Understanding semantics
self-contradiction 49ff, 55, 168, 173n; see
also logically false
self-reference 31, 40, 250f
sell (opposites) 208, 209, 211, 213, 214
semantic irregularity 129f
semantic knowledge 16, 232, 292, 299
semantic prime 242ff
semantic roles see thematic role
semantic uniqueness sees uniqueness
semantics 13–16; cognitive 269, 284f, 294,
299, 324, 359; formal ix, 80, 299, 323,
ch. 13; and logic 196–200; mentalist
14, 38, 285, 361f; model-theoretic 342;
possible-world ~ 351–62; prototype
~ 284–92; structuralist 226ff; of word
formation 15
seme 234
semelfactive 147; see also simple occurrence
terms
semiotic triangle 24, 25, 26, 38, 111, 131,
194, 203, 205, 224, 265, 284, 296, 359
sentence: contingent 174; declarative
4, 15, 26, 27, 30, 99, 100, 170ff, 197,
285f; imperative 15, 27, 38, 112, 285;
interrogative see interrogative sentence;
logically false 25, 49, 173f, 180ff, 185,
193, 352; logically true 25, 173f, 180ff,
185, 193, 215, 234, 352, 357, 360
sentence meaning 7–12, 14ff, 22–8, 47f, 55f,
130f, 196, 298f, 319ff, 336–42, 359–62
sentence type 15, 27, 162; declarative
27; imperative 27; interrogative 27;
meaning 27f, 37, 192, 359
sentential logic 183–7
shared knowledge 177
she, he 64f, 94ff, 234
shift see meaning shift
sign 223f
similarity to ptototype 170ff, 279f, 283f, 322
simple change term 145ff, 148, 149, 153,
164
simple occurrence term 146ff, 149, 150, 153,
155, 164
singular 8f, 20, 74, 80
situation 20; ~ described, expressed 22, 27,
34, 141, 157, 178, 193; ~ referred to 22,
130f, 135, 174, 178
situation structure 140–50, 312
sm 82f, 86
social deixis see social meaning: personal
pronouns
social interaction 6, 7, 18, 28ff, 36, 39, 66,
251, 302
social meaning 16, 28–31, 113, 192, 196,
203f, 250, 359; vs expressive meaning
35f; personal pronouns 66f; of a
sentence 28; and truth conditions 192;
of an utterance 38
some 84, 89, 171f, 290, 335f
sortal noun 69, 76, 78, 115f, 117
sound form 43
source (role) 135, 312
source domain 53
Spanish: anticausative 138; definite article
72; demonstratives 71; formal personal
pronouns 67; gender 65, 258; person
inflection of verbs 68, 112f; terms for
‘bra’ 193
speaker 5, 29, 32, 62ff, 67, 70f, 94, 113, 192,
253, 353
speech act 6, 15, 27, 30, 149
speech act theory 6
speech community 14, 16, 256f, 281, 289,
293
spelling 43
Sperber and Wilson 6
state term 56, 148ff, 151, 153, 156, 163,
237ff
stem 9
structuralism 222–36
subcategory 188, 266, 273f, 296
subject 10, 32, 83, 111f, 122–5, 135f
subjunction 184ff; ~ and entailment 186;
vs if 186
subordination/subordinate 188, 189, 205; vs
hyponymy 195
suffix 9, 226f
sum 80
superlatives 8, 42, 47, 78, 210, 212
superordinate 188
swear word 34f
synchronization 14
syncretism 63
synonymy 191, 203f, 360; partial 204
syntactic structure 12
syntagm 225
syntagmatic property see property:
combinatorial
Index
taboos 257f
tautology see logically true
taxonomy 216f, 276–9
tense 3, 72f, 113, 124, 131, 156–62; absolute
157; in English 163; in German 163;
intraction with aspect 157ff; in Japanese
164; relative 157; in Russian 163f
tense argument 156, 157ff
tertium non datur see Principle of Polarity
Thai: personal pronouns 67; social meaning
31
thanking 29, 36
thematic role 122–5, 135, 241, 310f
theme (role) 123, 125, 135ff, 140
theory of common usage 282f
theta role see thematic role
they 2n, 64, 66, 96; impersonal 137
thou 67
time of utterance 72, 158
Tok Pisin, term for bra 193, 252
top-down process 11, 56
transitive see relation; verb
translation 248–51; into predicate logic
336–40
tree 189
trial (number) 66
truth 4f, 25f, 29, 98ff, 110, 169–73, 192–6,
197f, 285f, 325, 351, 361
truth conditions 25f, 30, 38, 99, 170, 192f,
197f, 325, 342, 346, 351, 357–60; and
presuppositions 98ff, 197f
truth table 184
truth value 25, 170
truth-value gap 197, 285n
Turkish: articles 83; sibling terms 190
Ungerer and Schmid 294ff
uniqueness 75–9; inherent 76ff; pragmatic
75; semantic 76f, 79
uniqueness conditions (frames) 307
uniqueness presupposition 95
universal quantifier see all; each; every
universalism 256ff, 262
universe (of discourse) 342
used to 154
utterance meaning 4ff, 12, 16, 37f, 41, 49f,
55f, 91, 110
vagueness 47f, 282, 286ff, 291f, 299
379
value compounds 318
value of a frame attribute 303, 307
variable 120f; bound 335; free 335
Vendler 150
verb 111–14, 134f; accomplishment 141f,
143f; achievement 144ff; activity 142ff;
causative 206f, 237ff, 243f; of change
211; of change of possession 139;
of change of state 139; ditransitive
112; factive 98; finite 21f, 27, 31f,
63, 113, 124f, 171, 185; of grooming
and bodily care 140; inchoative 139,
145, 206, 237f; of ingesting 139;
intransitive 111; number of arguments
113f; performative 149, 160; process
142ff; punctual 146, 156; reciprocal
312; referential argument 113, 151,
311; simple change 144ff; simple
occurrence 146ff; of social interaction
214; state 148ff, 153; transitive 112;
zero-place 135
verb of change: presuppositions 97
verb phrase see VP
Vietnamese: personal pronouns 67; social
meaning 3
voice (of sounds) 225
voice (of verbs) 122, 136–9; see also active;
passive; conative
VP 10, 42, 55, 331f
wash 143
we 63n, 64ff; associative plural 65; inclusive
and exclusive 66
Whorf 256f, 262
Wierzbicka 242ff, 245
Wittgenstein 271
word class 12, 42, 111, 114, 130, 134, 228,
330
word formation 7f, 12f, 226f, 312f;
semantics 12f, 15, 207, 226f, 294ff,
316–19
word order 27, 112, 124f
work: English, German, Japanese 249
world knowledge 16, 138, 177, 232f, 293, 302
world view 256, 258, 264
you 2, 28f, 35, 62, 65ff, 96, 115, 137, 192,
242f, 250; impersonal 137
you all, you guys 67
Praise for the first edition
‘Understanding Semantics is an original and innovative resource for introductory
courses on linguistic semantics, excelling in particular in conciseness and accessibility
of presentation.’
Linguistics 43-2
‘It is a careful and thorough volume which should be the set text of choice for every
undergraduate course in semantics. The lecturer is presented with a ready-made
course; the student with a key tool for the study of semantics. Both will benefit from
Professor Löbner’s clear presentation of the breadth of ideas and methods of a broad
subject.’
Graeme Davis, University of Northumbria, UK
Praise for the second edition
‘Sebastian Löbner’s Understanding Semantics is a rare treat for both seasoned
researchers in semantics and novices to the field who are just embarking on the
journey towards the ultimate goal in the study of language: the understanding of
what meaning is. Löbner’s direct and engaging style makes even the most difficult
concepts easy to understand and I believe this book should be on every semantics
curriculum as it has been on mine since it first appeared.’
Luna Filipovic, University of East Anglia, UK
‘This excellent and well-written introductory textbook provides students with easy
access to the fascinating world of semantics. It is unique in successfully combining
lexical, cognitive and formal approaches to semantics, and in offering a coherent
picture of elementary concepts as well as of advanced theories.’
Klaus von Heusinger, University of Cologne, Germany
‘Sebastian Löbner deserves great credit for producing an introduction to linguistic
meaning that is both accessible and thorough. The approach, which integrates
the key concepts of formal semantics with a cognitive perspective on meaning, is
engaging and refreshing: readers will come away with an understanding of how
truth-conditional semantics is done, but, more importantly, they will have a clear
sense of how meaning works in natural language – and why this is such a rich and
rewarding field of study.’
Chris Lucas, School of Oriental and African Studies, University of London, UK
‘Without stinting on presenting core areas of sentential logic, predication, and lexical
relations, the new edition of this comprehensive survey of semantics is enhanced
by three new chapters. Students will benefit from the exercises that cover the
intersection of semantics and pragmatics at the lexical, sentential and cognitive
levels and require research within both their own, as well as typologically diverse,
languages.’
Laurel Smith Stvan, University of Texas at Arlington, USA