Understanding Semantics [second edition] 9780415826730, 9781444122435, 9780203528334

Citation preview

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, so¯ and a¯ 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)

Imperfective verb

Perfective verb

ita-la

pro ita-la

Present tense (1st sing.)

ita-ju

---

Future tense (1st sing.)

budu itat’

pro ita-ju

Past tense (fem. sing.)

6

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 kyu¯ 8 9

ju¯ 10

When we attach the word for 10 ju¯ to the words for the digits 2 to 9 we obtain the terms for the tens higher than 10: (2)

ju¯

10 niju¯

20 sanju¯ 30 yonju¯ 40

…

kyu¯ju¯ 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 ju¯. 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 ju¯

(3)

10

ni

ju¯

20

…

kyu¯ ju¯

90

ichi

1 ju¯ ichi

11

ni

ju¯

ichi

21

…

kyu¯ ju¯ ichi

91

ni

2 ju¯ ni

12

ni

ju¯

ni

22

…

kyu¯ ju¯ ni

92

san

3 ju¯ san

13

ni

ju¯

san

23

…

kyu¯ ju¯ san

93

…

…

kyu¯

9 ju¯ kyu¯

… 19

ni

ju¯

kyu¯

… 29

…

kyu¯ ju¯ kyu¯

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 niju¯san can be calculated from the components ni 2, ju¯ 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 sanju¯ ‘three ten’ for 30 means ›three TIMES ten‹, but ju¯san ‘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, sanju¯roku could, for its mere form, be derived in two ways: (4)

a.

(3 10) 6:

First concatenate san 3 with ju¯ 10 to form sanju¯, then concatenate sanju¯ with roku 6.

b.

3 (10 6)

First concatenate ju¯ 10 with roku 6 to form ju¯roku, then concatenate san 3 with ju¯roku.

When we follow (4a), we obtain the interpretation 30 for sanju¯ (smaller number 3 times greater number 10) and 36 for sanju¯roku (greater number 30 plus smaller number 6). But with (4b) we obtain the interpretation 16 for ju¯roku 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, kyu¯, ju¯

FR consists of two rules: F1

form ‘UV’

where and

U = ni, san, yon, go, roku, nana, hachi or kyu¯ V = ju¯

F2

form ‘ZX’

where and

Z = ju¯, niju¯, sanju¯, …, kyu¯ju¯ X = ichi, ni, san, yon, go, roku, nana, hachi or kyu¯

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, … , [kyu¯] = 9, [ju¯] = 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 nanaju¯roku 76. On the basis of the given rules, the expression can only be derived by first applying F1 and then F2: nana and ju¯ combine to nanaju¯ 70, then nanaju¯ is concatenated with roku 6. If we first derived ju¯roku 16 by F2, we could not concatenate this with nana, as neither formation rule allows for ju¯roku as the second part. (5)

f.

formation of nanaju¯roku Step 1: F1: nana, ju¯ → nanaju¯ Step 2: F2: nanaju¯, roku → nanaju¯roku

i.

interpretation of nanaju¯roku Step 1: I1: [nanaju¯] = = = Step 2: I2: [nanaju¯roku] = = =

[nana] × [ju¯] 7 × 10 70 [nanaju¯] + [roku] 70 + 6 76

(IB)

(above, IB)

Figure 13.1 shows the parallel steps in deriving the form and the interpretation of nanaju¯roku. Figure 13.1 Formation and compositional interpretation of nanaju¯roku

冦

nana, ju¯

Interpretations 7 , 10

冦

Expressions

× nanaju¯, roku

70 , 6

冦

I1

冦

F1

+

F2 nanaju¯roku

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 kyu¯hyaku 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 sanju¯kyu¯. 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 niju¯ichi 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