Category Mistakes (Oxford Philosophical Monographs) [1 ed.] 9780199572977, 0199572976

Citation preview

Category Mistakes

OXFORD PHILOSOPHICAL MONOGRAPHS Editorial Committee anita avramides r. s. crisp william child antony eagle stephen mulhall other titles in this series include Quantum Information Theory and the Foundations of Quantum Mechanics Christopher G. Timpson The Critical Imagination James Grant Nietzsche and Metaphysics Peter Poellner Understanding Pictures Dominic Lopes Things That Happen Because They Should A Teleological Approach to Action Rowland Stout The Ontology of Mind Events, Processes, and States Helen Steward Wittgenstein, Finitism, and the Foundations of Mathematics Mathieu Marion Semantic Powers Meaning and the Means of Knowing in Classical Indian Philosophy Jonardon Ganeri Hegel’s Idea of Freedom Alan Patten Metaphor and Moral Experience A. E. Denham Kant’s Empirical Realism Paul Abela Against Equality of Opportunity Matt Cavanagh The Grounds of Ethical Judgement New Transcendental Arguments in Moral Philosophy Christian Illies Of Liberty and Necessity The Free Will Debate in Eighteenth-Century British Philosophy James A. Harris Plato and Aristotle in Agreement? Platonists on Aristotle from Antiochus to Porphyry George E. Karamanolis Aquinas on Friendship Daniel Schwartz The Brute Within Appetitive Desire in Plato and Aristotle Hendrik Lorenz

OXFORD PHILOSOPHICAL MONOGRAPHS

Category Mistakes OFRA MAGIDOR

1

3

Great Clarendon Street, Oxford, OX2 6DP, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © Ofra Magidor 2013 The moral rights of the author have been asserted First Edition published in 2013 Impression: 1 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer British Library Cataloguing in Publication Data Data available ISBN 978–0–19–957297–7 Printed in Great Britain by CPI Group (UK) Ltd, Croydon, CR0 4YY Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work

To my parents, Sarah and Menachem Magidor

This page intentionally left blank

CON TEN TS

Acknowledgements

ix

1 Introduction

1

§1 §2 §3 §4 §5

The Phenomenon of Category Mistakes Category Mistakes in the Philosophical Literature Category Mistakes in the Linguistics Literature Category Mistakes in Computer Science Conclusion

2 The Syntactic Approach

25

§1 The Syntactic Approach to Category Mistakes §2 Some Unsatisfactory Arguments Against the Syntactic Approach

§3 §4 §5 §6 §7

1 7 15 20 23

25 30

§2.1 The simplicity argument §2.2 The meaningfulness argument §2.3 The argument from universality

30 32 32

The Argument from Particularity The Argument from Embedding Interactions with Meanings Interactions with Extra-Linguistic Facts Interactions with Context

35 38 39 40 42

3 The Meaninglessness View

43

§1 The Meaninglessness View of Category Mistakes §2 The Argument(s) from Compositionality §2.1 Atomic category mistakes §2.2 Type-theoretic semantics to the rescue? §2.3 Conjunctions and quantifier phrases

§3 The Argument from Synonymy §4 The Argument from Propositional Attitude Ascriptions

vii

43 46 46 48 56

58 59

CONTENTS §5 The Argument from Metaphor §6 Arguments in Favour of the Meaninglessness View? §6.1 The imagination motivation §6.2 The motivation from alternative theories of meaning §6.3 The nonsense motivation

4 The MBT View

66 74 75 76 79

80

§1 The MBT View §2 A General Argument Against the MBT View §3 Arguments in Favour of the MBT View §3.1 The infelicity argument §3.2 Routley’s transfer argument §3.3 The arbitrariness argument

§4 The Supervaluationist Treatment of Category Mistakes §4.1 The formal details §4.2 Validity and implication §4.3 The problem of complex category mistakes

5 The Pragmatic Approach

80 83 91 91 94 95

99 99 101 106

110

§1 The Pragmatic Approach §2 The Naïve Pragmatic Approach §3 The Background Framework: Pragmatic Presuppositions §3.1 Tests for presupposition §3.2 Foundational issues

§4 A Presuppositional Account of Category Mistakes

110 111 116 117 124

131

§4.1 A basic example §4.2 Other cases §4.3 Characterizing category mistakes?

131 140 146

§5 Merits of the Account §6 Postscript: Some Final Reflections on the Implications of the Account

148 154 159 167

References Index

viii

ACKNOWLEDGEMEN TS

The material in this book has evolved in various stages: starting with a BPhil thesis and a DPhil thesis submitted to the University of Oxford in 2004 and 2007 respectively, and then rewritten more recently in the form of the current manuscript. As such, it is hard to track all the individuals who have provided input to this material over the years. I would like to thank all of those who did have such input, and apologize in advance to anyone who may have been accidentally omitted from these acknowledgements. For helpful comments and discussions of various parts and versions of this material, I would like to thank audiences in Durham, Jerusalem, Lisbon, MIT, NYU, Oxford, and Rutgers; I would also like to thank Brian Ball, Emmanuel Chemla, Antony Eagle, Danny Fox, Anandi Hittiagandi, John Hyman, Gail Leckie, Stephen Kearns, Hemdat Lerman, Julia Markovits, Menachem Magidor, Adrian Moore, Jessica Moss, Sarah Moss, Anders Nes, Daphna Oren-Magidor, Oiwi Parker Jones, Carl Posy, Richard Price, Katrina Przyjemski, Daniel Rothschild, Nathan Salmon, Philippe Schlenker, Jason Stanley, Daniel Star, Judith Jarvis Thomson, Emanuel Viebahn, Sebastian Watzl, and Bruno Whittle. Four individuals played a special role in the previous incarnations of this manuscript: Dorothy Edgington co-supervised my BPhil thesis; Timothy Williamson supervised both my BPhil and DPhil theses; John Hawthorne and Robert Stalnaker acted as examiners for my DPhil thesis. I am fortunate to have benefited from the advice of these great philosophical minds: they each provided comments which were at the same time incisively critical as well as generous and helpful. I am especially grateful for the abundant advice I received from Tim Williamson: the importance of his philosophical work is clearly no secret to the profession, but I also had the opportunity to discover what a dedicated, kind, and supportive mentor he is. I am extremely indebted to him for all that he has taught me throughout the years. ix

ACK NOW LEDGEMENTS For their help in the writing of the current version of the manuscript, I would like to extend special thanks to Márta Abrusán, Cian Dorr, Aynat Rubinstein, and Giora Sternberg, as well as to two anonymous readers for Oxford University Press who provided highly detailed and helpful comments on the manuscript. The material in Chapter 3 is a revised version of an article published in Linguistics and Philosophy: I would like to thank the journal for their permission to reprint the material, as well as the anonymous readers of that manuscript. Various institutions were helpful in facilitating the writing of this material, most notably Queen’s College, Oxford, for supporting me via a Junior Research Fellowship in 2005–2007, and Balliol College, Oxford, the Faculty of Philosophy at the University of Oxford, and MIT for supporting my research leave in 2010. Finally, I would like to thank my family, and most of all my parents Sarah and Menachem, and my partner Giora. They have given me more love and support than I could have wished for, and I can only hope they know how important they are to me.

x

Imagine a piano keyboard, eighty-eight keys, only eighty-eight and yet, and yet, hundreds of new melodies, new tunes, new harmonies are being composed upon hundreds of different keyboards every day in Dorset alone. Our language, Tiger, our language, hundreds of thousands of available words, frillions of legitimate new ideas, so that I can say the following sentence and be utterly sure that nobody has ever said it before in the history of human communication: “Hold the newsreader’s nose squarely, waiter, or friendly milk will countermand my trousers.” Perfectly ordinary words, but never before put in that precise order. A unique child delivered of a unique mother. And yet, oh and yet, we all of us spend all our days saying to each other the same things time after weary time: “I love you”, “Don’t go in there”, “Get out”, “You have no right to say that”, “Stop it”, “Why should I”, “That hurt”, “Help”, “Marjorie is dead”. That surely is a thought to take out for a cream tea on a rainy Sunday afternoon. (Stephen Fry and Hugh Laurie)

xi

This page intentionally left blank

1

2

Introduction

§1

The Phenomenon of Category Mistakes

Category mistakes are sentences such as ‘The number two is blue’, ‘The theory of relativity is eating breakfast’, or ‘Green ideas sleep furiously’. Such sentences strike most English speakers as highly infelicitous, and infelicitous in a similar way.1 Like most terms in language, one can come to understand what is meant by ‘category mistake’ without an explicit definition of the term: its meaning can be learnt by ostension, via an ample range of relevant examples. Consider the following sentences:2 (1) (2) (3) (4) (5) (6)

John is drinking water. *John is shminging water. *John is drinking water and John is drinking water. *The king of the United States is drinking water. *John is drinking the theory of relativity. *The theory of relativity is drinking beer.

Sentence (1) is completely felicitous, and is clearly not a category mistake. But even among a range of infelicitous sentences, we can distinguish those which constitute category mistakes from those which do not: sentences (2)–(6) are all infelicitous, but (5) and (6) are category mistakes, while (2), (3), and (4) are not. The observation that there is a distinctive 1 Here and throughout, I use the term ‘infelicitous’ to mean something like ‘seems odd or inappropriate’, without making any theoretical commitment as to the source of oddness or inappropriateness. 2 I use the symbol ‘*’ throughout to mark infelicity. Note that use of this symbol is intended to be completely neutral with respect to the source of the infelicity.

1

INTRODUCTION class of infelicitous sentences, ones that seem infelicitous in a similar manner to sentences (5) and (6), points to a linguistic phenomenon: the phenomenon of category mistakes. Of course, as with many phenomena, there are borderline cases (Is ‘The stone is thinking about the theory of relativity’ a category mistake? How about ‘The squirrel is thinking about the theory of relativity’?). But the claim that the boundaries of a phenomenon are vague does not entail that the phenomenon is any less real. The project I am concerned with in this monograph is to account for the phenomenon of category mistakes, in the following sense: I would like to explain what makes category mistakes infelicitous. In seeking such an explanation, I adopt as a working hypothesis the assumption that it is possible to give a uniform account of category mistakes, or in other words that the infelicity of different category mistakes arises for similar reasons. This uniformity hypothesis is motivated by the observation that different category mistakes at least resemble each other by exhibiting a very similar phenomenology of infelicity (after all, it is precisely this distinctive phenomenology that has been used to characterize the relevant class of sentences), and more straightforwardly by a preference for theoretical simplicity. Unless it turns out that no uniform account is ultimately successful, this hypothesis should be maintained. As we shall see, one issue that complicates the task of explaining the infelicity of category mistakes is that the phenomenon is at the same time highly widespread and extremely diverse. Nearly any predicate can be used to form a category mistake (e.g. ‘green’, as in ‘The number two is green’; ‘hungry’ as in ‘My chair is hungry’; ‘prime’ as in ‘My mother is prime’; and so forth).3 Category mistakes can involve expressions of a wide variety of syntactic types (e.g. verbs as in ‘The theory of relativity is sleeping’; adverbs, as in ‘sleeps furiously’; or prepositions as in ‘underneath the number two’). 3 Possible exceptions include predicates such as ‘interesting’ or relations such as ‘identical’, though even this is controversial. For example, various authors have argued that identity statements arising in the context of the ‘Julius Caesar’ problem in philosophy of mathematics constitute category mistakes (see e.g. Benacerraf (1965), pp. 63–7, Routley (1966), p. 204, and Shapiro (1997), p. 79).

2

THE PHENOMENON OF CATEGORY MISTAK ES Nor are category mistakes restricted to atomic sentences. For example, ‘The number two isn’t blue’ or ‘Either the number two is blue or it is prime’ seem to be infelicitous in a very similar manner to the atomic sentence ‘The number two is blue’. Finally, category mistakes seem to occur across a very wide range of languages (possibly all of them).4 Another issue that complicates the task of accounting for the phenomenon of category mistakes is that there are many subtleties to the phenomenon, some of which are too often overlooked. For example, some sentences that exhibit the relevant infelicity in some contexts, do not exhibit it in other contexts. Consider the sentence ‘That is green’. Relative to a context where it is clear that the demonstrative refers to the number two, the sentence is infelicitous in the relevant manner, whereas relative to a context where a pen is referred to, the sentence does not exhibit the relevant phenomenology. A fully satisfactory account of category mistakes must accommodate not only simple cases such as ‘The number two is green’, but also the full range of complex and subtle data pertaining to category mistakes. Having stated the aim of my project, it is also important to point out one thing that is not an aim of this project. I do not aim to offer a precise analysis of the concept of ‘category mistake’, provide informative and necessary and sufficient conditions for being a category mistake, or explain what makes category mistakes into a distinctive class of sentences. The following two questions ought to be separated: first, what is required of a certain sentence to belong to the class of category mistakes; second, what makes those sentences that do belong to this class infelicitous. My focus in this monograph will be on the second rather than the first (though I do offer some brief remarks on the first question in Chapter 5, §4.3). Why this choice of focus? Recent years have seen a certain general trend away from traditional analysis projects. Philosophers working in a 4 I do not know of any systematic research on this point but an informal survey I conducted among a range of language users suggests the phenomenon occurs at least in Arabic, Catalan, Chinese, Danish, Dutch, English, Finnish, French, German, Hawaiian, Hebrew, Hungarian, Irish, Italian, Japanese, Norwegian, Portuguese, Sanskrit, Spanish, Swedish, Swiss German, and Turkish.

3

INTRODUCTION range of domains have, on the one hand, become increasingly sceptical about the feasibility of many such projects, especially given repeated epicycles of counter-examples and amendments. On the other hand, it has transpired that in many cases one can offer interesting and illuminating discussions of a philosophical concept without attempting to provide a reductive analysis or delineating it precisely.5 Similar considerations apply, in my view, to the concept of category mistakes: I am sceptical regarding the prospects of offering a full analysis of the concept, and at any rate, I think addressing the second question—namely that of accounting for the infelicity of category mistakes—is an interesting and illuminating project in its own right. Two additional points ought to be emphasized in this context. First, the various theories of category mistakes I discuss (including those I reject) are also primarily concerned with the second project: they offer little if any by way of addressing the question of analysis6 and my objections to the alternative theories focus on their failure to correctly account for the phenomenon rather than on a failure to delineate it accurately. Second, one might worry that without a precise definition of the concept of category mistakes much of the debate will turn on the question of which examples are subsumed under the phenomenon. As it turns out, however, this worry does not bear out. While (as noted above) there certainly are borderline cases of category mistakes, many of the examples discussed involve uncontroversial paradigmatic cases of the phenomenon. In other cases, where I discuss more controversial examples, I often point out that these are examples my opponents also acknowledge to be category mistakes (or at least to exhibit

5 For positions along these lines see for example Stalnaker on possible worlds (Stalnaker (1976)); Campbell on colour (Campbell (1993)); Carroll on laws of nature and on causation (Carroll (1994)); and Williamson on knowledge (Williamson (2000)). 6 The only potential exception is the meaninglessness view, which could attempt to characterize category mistakes as the only sort of sentence that is syntactically well-formed but meaningless. But even given that one adopts the meaninglessness view this characterization would be highly controversial (for example, a defender of the view might well accept that the Liar sentence is syntactically well-formed and meaningless, without classifying it as a category mistake), and at any rate defenders of the view have not highlighted this potential benefit of their view.

4

THE PHENOMENON OF CATEGORY MISTAK ES the relevant kind of phenomenology), so the disagreements do not ultimately turn on how to classify such examples.7 Finally, it is worth noting that while accounting for the infelicity of category mistakes forms a distinct project from delineating the concept, such an account may nevertheless be helpful in adjudicating at least some borderline cases.8 Why study the phenomenon of category mistakes at all? For a start, explaining systematic infelicities is a paradigmatic example of the type of question that linguists and philosophers of language are interested in.9 Especially so when, as is the case here, the type of infelicity in question is very widespread, both within and across languages. But what makes category mistakes particularly interesting is that a plausible case can be (and indeed has been) made for explaining the phenomenon in terms of each of syntax, semantics, and pragmatics. This means that the topic of category mistakes offers a particularly fruitful ground for exploring a range of foundational questions concerning each of these three realms of language, and the boundaries and interactions between them. Some such foundational issues depend directly on one’s ultimate theory of category mistakes. Consider for example the question of whether or not category mistakes are meaningful. If they are, one might be able to accept a strong form of the principle of compositionality, according to which any meaningful expressions combined in a syntactically wellformed manner, compose a meaningful expression.10 On the other hand, if category mistakes are not meaningful then, assuming they are syntactically well-formed, one can at best accept only a weaker principle of compositionality (one which states that if a syntactically well-formed sentence is meaningful, then its meaning is a function of the meaning of its parts). 7

See for example, f.n. 25 in Chapter 3, f.n. 22 in Chapter 4, and f.n. 43 in Chapter 4. See f.n. 26 in Chapter 2, and f.n. 48 in Chapter 5. 9 Indeed, in a recent retrospective interview, Barbara Partee describes the explanation of ambiguities and infelicities as the kind of issue that “linguists were always thinking about”, and mentions Chomsky’s famous category mistake (‘Colourless green ideas sleep furiously’) in this context. (Partee (interview)). 10 Note that something like this strong form of compositionality is assumed by Montague (1970), in his treatment of compositionality as a homomorphism between the syntax and semantics of a language. 8

5

INTRODUCTION Relatedly, since category mistakes are arguably the only example of sentences that might be considered syntactically well-formed but meaningless, the issue of whether they are in fact meaningless may decide a crucial question concerning the relationship between syntax and semantics: whether being syntactically well-formed is a sufficient condition for meaningfulness.11 Conversely, other foundational questions have a direct impact on one’s theory of category mistakes. For example, as we shall see, the issue of whether syntactic features may supervene on semantic features; of whether there are such things as partial propositions (propositions that are truth-valueless relative to some possible worlds); and of whether presupposition failures entail truth-value gaps, play central roles in motivating different theories of category mistakes.12 Finally, exploring the issue of category mistakes can shed light on the fundamental methodological question of how one might decide whether to treat a phenomenon as syntactic, semantic, or pragmatic. One challenge that I have faced in writing this book is that the existing literature on category mistakes is rather heterogeneous and dispersed: it includes authors from different academic fields, writing in different periods of time, and against the background of different traditions. In discussing the various views on the topic, I have attempted to present reasonably clean and systematic versions of the arguments. This often entailed weaving together proposals from various authors, and making explicit what is sometimes merely implicit in the literature. Given that much of the interest in the phenomenon lies in its interaction with the fields of syntax, semantics, and pragmatics, I found it best to divide the accounts of category mistakes into four general approaches. The first (‘the syntactic approach’) maintains that category mistakes are infelicitous because they are syntactically ill-formed. The second and third approaches are both semantic, but each accounts for category mistakes in terms of a different facet of semantics. The second approach (‘the 11

For discussion of whether syntactic well-formedness is a necessary condition for meaningfulness, see Magidor (2009a). 12 For discussion of these issues see, respectively, Chapter 2, §2.3; Chapter 4, §2; and Chapter 5, §3.2.

6

CATEGORY MISTAK ES IN THE PHILOSOPHICAL LITER ATUR E meaninglessness view’), places the phenomenon at the level of meaning, and maintains that category mistakes are meaningless. The third approach (‘the MBT view’) places the phenomenon at the level of content or reference, and maintains that category mistakes are meaningful but truth-valueless (either because they express no content, or because they express a truth-valueless proposition). The final approach (‘the pragmatic approach’) maintains that category mistakes are syntactically wellformed, meaningful, and truth-valued, but that they are nevertheless pragmatically inappropriate. The structure of the book follows this division. I discuss the syntactic, meaninglessness, and MBT views in Chapters 2, 3, and 4 respectively, and I argue that each of these three approaches ought to be rejected. The fifth and final chapter is devoted to the pragmatic approach. I begin by presenting and rejecting one version: ‘the naïve pragmatic approach’. I then proceed to propose and defend an alternative version: ‘the presuppositional account of category mistakes’. The remainder of the current chapter is devoted to a brief survey of the (modern) history of the topic of category mistakes in philosophy, linguistics, and computer science.

§2

Category Mistakes in the Philosophical Literature

The philosophical interest in category mistakes dates at least as far back as Aristotle,13 but one can trace the origins of the modern debate to Russell’s theory of types. In his 1903 book Principles of Mathematics (Russell (1903)), Russell published his now-famous paradox (§78), and remarked that “It is the distinction of logical types that is the key to the whole mystery” (p. 105). The general idea for how types might help in resolving the paradox is laid out in Appendix B of the book, where Russell suggests:

13

See especially Aristotle’s remarks on negation, in Categories 10. Aristotle claims that when Socrates is too young for it to be natural for him to possess sight then both ‘Socrates has sight’ and ‘Socrates is blind’ are false (13b22–24). This seems to suggest that Aristotle took atomic category mistakes to be false. Other relevant discussions of Aristotle are his discussion of ‘snub-nose’ in Metaphysics Zeta 5, and his discussion of per se predications in Posterior Analytics I.4.

7

INTRODUCTION Every propositional function φ(x)—so it is contended—has, in addition to its range of truth, a range of significance, i.e. a range within which x must lie if φ(x) is to be a proposition at all, whether true or false. This is the first point of the theory of types; the second point is that ranges of significance form types, i.e. if x belongs to the range of significance of φ(x), there is a class of objects, the type of x, all of which must also belong to the range of significance of x (p. 523).

Russell goes on to argue that in a statement of the form ‘x is a u’ (and correspondingly, ‘x is a not-u’), “x and u must be of different types”, and hence that “ ‘x is an x’ must in general be meaningless” (p. 524). This presumably resolves the paradox because sentences containing descriptions of the form ‘P is not predicable of P’ or ‘x is not a member of x’ are meaningless, and the paradox cannot even get off the ground. In his 1908 paper ‘Mathematical logic as based on theory of types’, Russell expanded the theory into what is now known as ‘the ramified theory of types’. The ramified theory was intended to address not only the original paradox, but a wide range of semantic paradoxes such as the Liar, Richard’s, and Grelling’s paradoxes.14 Leaving the details aside, the underlying idea involved a complex hierarchy of types, and a ban on entities of certain types from appearing in the relevant positions in propositions, thus ensuring the relevant paradoxes become effectively inexpressible.15 In the 1930s Russell’s treatment of paradoxes through the theory of types inspired two related ideas. The first was that some seemingly grammatical sentences are meaningless because they involve some kind of ‘type confusion’ or ‘type mismatch’.16 The second is that apparent philosophical puzzles can be resolved by declaring the statements of the puzzle meaningless. (The ideas are clearly connected, in that the relevant puzzles can be stated using seemingly grammatical sentences, and thus the first point is needed to ensure that the relevant statements can 14

See Russell (1908), and the further development of the theory in Russell and Whitehead’s Principia Mathematica. 15 For a helpful introduction to Russell’s theory of types and some of the problem it faces see Copi (1971). 16 I say ‘seemingly grammatical’ because under one interpretation of Russell’s theory of types, it places syntactic rather than semantic constraints.

8

CATEGORY MISTAK ES IN THE PHILOSOPHICAL LITER ATUR E nevertheless be classified as meaningless). These ideas found further support in the then prominent logical positivist movement and the work of early Wittgenstein. One philosopher with which these ideas resonated deeply was Gilbert Ryle. In his 1938 paper ‘Categories’, Ryle stated that “we are in the dark about the nature of philosophical problems and methods if we are in the dark about types and categories”.17 Ryle went on to claim that whenever a grammatical sentence constructed out of meaningful words is “(not true or false but) nonsensical or absurd”, this is so because “at least one ingredient expression in it is not of the right type to be coupled or to be coupled in that way with the other ingredient expression or expressions in it”, and he labelled this kind of mismatch as a ‘type trespass’.18 Ryle notes the interest of logicians (primarily Russell) in type trespasses such as those occurring in the Liar Paradox, but also complains that logicians’ interests in types is too narrow, and that we should be equally concerned with more obvious type trespasses, i.e. those occurring in sentences such as ‘Saturday is in bed’—namely category mistakes.19 Ryle also proposed to use category mistakes in order to define the notion of an ontological type. He claimed that two things a and b are of different types just in case there are two sentences which differ only in that one has an expression denoting b where the other has an expression denoting a, such that one sentence is ‘absurd’ (i.e. a category mistake) and the other is not. For example, on Ryle’s view, the fact that the sentence ‘Saturday is asleep’ is a category mistake, while the sentence ‘Tony Blair is asleep’ is not, suffices to show that Saturday and Tony Blair belong to different ontological types. According to Ryle, then, the notion of a category mistake plays a crucial role in philosophical methodology. While in ‘Categories’ Ryle put forth a philosophical methodology, The Concept of Mind constituted his main attempt to apply this methodology. In the book’s introduction, he reiterates his commitment to the importance

17

18 Ryle (1938), p. 189. Ryle (1938), p. 200. Ibid, pp. 200–1. Note though that the term ‘category mistakes’ never appears in ‘Categories’ and is only introduced by Ryle later, in The Concept of Mind (Ryle (1949), p. 16). 19

9

INTRODUCTION of type distinctions in philosophy: “Philosophy is the replacement of category-habits by category-disciplines”.20 The rest of the book is an attempt to argue against the dualist position that contrasts mind and body, by claiming that it is “one big mistake and a mistake of a special kind. It is, namely, a category mistake”.21 However, it is far from clear what Ryle took the central category mistake in the dualistic position to be. Sometimes he talks as if it is speaking of the mind in mechanistic-like terms (as in talk of mental causation).22 In other places, he seems to be worried by a kind of ‘double counting’ problem. He notes that “a purchaser may say that he bought a left-hand glove and a right-hand glove, but not that he bought a left-hand glove, a right-hand glove, and a pair of gloves” (p. 22). The proposal is that what is wrong with this description is that it involves an illegitimate mix of types—talking of the types ‘glove’ and ‘pair of gloves’ in the same occasion. Equally, he argues, mind and body are of different types, so one cannot legitimately discuss both at the same time.23 At any rate, his general position seems to be that most classic problems in the philosophy of mind (e.g. the question of how the mental and the physical interact or the problem of free will) arise merely because of some sort of a category mistake. But other than a few rather obscure examples, Ryle does not say much about what category mistakes are or how one should account for them.24 Both Russell’s theory of types and Ryle’s generalization of it to a wider philosophical context generated a range of debates concerning category mistakes. Some of the debates were centred on the attempt to define 20

Ryle (1949), p. 8. Ryle (1949), 16. As far as I can tell, this is the first time the concept of a category mistake is referred to using this label. 22 Ryle (1949), pp. 19–20. 23 Ryle (1949), pp. 22–3. Note that Ryle’s analysis of the double counting problem is clearly wrong: it would be equally inappropriate to describe buying one left-hand glove by saying ‘I bought a left-hand glove and a left-hand glove’, and it would be perfectly appropriate to describe buying three gloves two of which were left-handed by saying ‘I bought a pair of gloves and a left-hand glove’. The double counting problem thus has nothing to do with ‘mixing types’. 24 He also seems to shift between the view that category mistakes are meaningless or ‘absurd’ and the view that they are false (see for example his claim in p.16 that the dualistic doctrine is “entirely false”). 21

10

CATEGORY MISTAK ES IN THE PHILOSOPHICAL LITER ATUR E ontological types via category mistakes.25 Others dealt with the semantic status of category mistakes. Strawson, for example, endorsed the Ryle/ Russell position that category mistakes are grammatical but meaningless, and in particular endorsed Russell’s idea that predicates have ‘ranges of significance’.26 Pap was generally sympathetic to this position, but (with some hesitation) claimed that there is one set of sentences which seem to him like category mistakes but are nonetheless simply false rather than meaningless. These were sentences that made explicit class attributions, such as ‘Socrates is a number’ or ‘The theory of relativity is a concrete object’.27 By the 1950s the claim that category mistakes are meaningless (or at least truth-valueless) seemed to be the prominent view. So much so, that in his 1954 article ‘Entities’, Arthur Prior complained that “the person who maintains that virtue is not a square must nowadays count himself among the heretics”.28 However, this orthodoxy was not uniformly accepted. Even before the 1960s the view was contested by Ewing, Prior and Quine, who maintained that category mistakes are perfectly meaningful: a sentence such as ‘The number two is green’ is simply false and its negation ‘The number two is not green’ is true.29 Quine, for example, complained that the view that took category mistakes to be meaningless was “just a spontaneous revulsion against silly sentences”,30 and argued that the meaninglessness view is ill-motivated, while its alternative is justified by “the considerable theoretical simplifications that are gained by lifting such [category] bans”.31 However, since all three of these writers were primarily interested in 25 See for example Smart (1953), Baker (1956), and Sommers (1963). For a more recent contribution to this debate see Westerhoff (2005). 26 Strawson (1952), p. 112 and pp. 226–7. However, Strawson was later drawn to the idea that category mistakes are simply ungrammatical (see Strawson (1970)). 27 Pap (1960). Pap’s motivation for this concession was that he wanted to explain why a sentence such as ‘Socrates is prime’ is meaningless, by relaying on the claim that Socrates is not a number. But this required him to accept that the latter sentence is true rather than meaningless. 28 Prior (1954), p. 160. 29 See Ewing (1937), Prior (1954) and Quine (1960). As I note below, in the ensuing two decades this view received further support (see e.g. Lambert (1968), Haack (1971), and Goldstick (1974)). 30 31 Quine (1960), p. 229. Quine (1953), p. 449.

11

INTRODUCTION refuting the then widespread view that category mistakes are meaningless and in particular the applications of this view to problems in metaphysics, they devoted little discussion to the issue of how one should nevertheless account for the infelicity of category mistakes.32 The 1960s saw a more direct interest in category mistakes as an independently significant linguistic phenomenon, rather than merely as a tool for metaphysical theorizing. In 1966, Drange published an entire monograph (entitled ‘Type Crossings’) devoted to the topic.33 In the same year, two papers—one by Routley and one by Goddard—appeared in the same issue of the Australasian Journal of Philosophy, both discussing how one should address category mistakes within formal logic.34 The publication of these papers spurred a lively debate in the Australasian Journal, which focused primarily on the question of whether or not category mistakes are truth-valued.35 While Lambert and Haack supported the view that category mistakes are true or false (this was labelled as the ‘no-type view’, or on some versions as the ‘falsidal’ view), Routley and Brady rejected this claim. One of the central issues in this debate was a classic argument in favour of the no-type view which relied on a simple syllogism of the following form: ‘The theory of relativity is an abstract object. No abstract object is blue. Therefore, the theory of relativity is not blue’.36 No-typers argued that this establishes that ‘The theory of relativity is not blue’ is true, and hence that the atomic category mistake ‘The theory of relativity is blue’ is false. Their opponents, however, complained

32 Quine does, however, note that category distinctions may be of interest to linguists (Quine (1953), p. 449), and Ewing briefly suggests that the infelicity of category mistakes could be explained via pragmatic considerations (Ewing (1937), pp. 60–1), though he never spells out what these considerations are. 33 Drange (1966). Drange’s book contains some interesting material, but overall supports some very odd views such as the view that atomic category mistakes are both meaningless and false. 34 Goddard (1966) and Routley (1966). 35 In addition to the above mentioned papers this included Lambert (1968), Routley (1969), Haack (1971), Brady & Routley (1973), and Haack (1975). 36 It seems that a version of this argument is first presented in Prior (1954), p. 159, but raised by various others (See e.g. Drange (1966), p. 24, Lambert (1968), p. 83, and Goldstick (1974), p. 342).

12

CATEGORY MISTAK ES IN THE PHILOSOPHICAL LITER ATUR E (not unreasonably) that the argument is question begging, because it assumes that the apparent category mistake ‘No abstract object is blue’ is true rather than meaningless. For a period of over a decade between the late 1960s to the early 1980s, there appeared a series of works attempting to address the topic of category mistakes using a range of different formal tools.37 This period culminated with the publication of another book devoted entirely to the topic category mistakes: Shalom Lappin’s Sorts, Ontology and Metaphor.38 However, by the mid 1980s the topic of category mistakes seemed to have fallen out of fashion: the concept certainly continued to be discussed in the context of other debates,39 but for over twenty years it was, by and large, neglected in the philosophical literature as a topic in its own right.40 37 See for example Goddard (1968) and Lappin (1981) for three-valued approaches, Martin (1975) for a four-valued approach, Thomason (1972) for a supervaluationist approach, and Bergmann (1977) for an approach that combines the latter two but nonetheless assigns to category mistakes bivalent truth-values. 38 Lappin (1981), which is a revised version of Lappin’s 1975 PhD dissertation. 39 One context in which the notion still comes up fairly often, is in debates in metaphysics concerning the question of the putative identity of objects. Thus for example, in arguing for the non-identity of a statue and a coincident lump of clay, Kit Fine maintains that the statue cannot be identical to the lump, because while ‘The statue is Romanesque’ might be true, ‘The lump of clay is Romanesque’ is meaningless, presumably because it is a category mistake (Fine (2003), pp. 207–8). For a general discussion of the role of category mistakes in such non-identity arguments see Schnieder (2006), §2, and Magidor (2011), §4. Other contemporary discussions in which the topic of category mistakes is brought up include Structuralism in philosophy of mathematics (see Shapiro (1997), ch. 3, §2); the semantics of metaphor (see Stern (2006),§3); and the issue of whether there can be vagueness in the world (see Williamson (1994), ch. 9). 40 Very recently, however, the topic seems to be enjoying a small revival with the publication of Camp (2004), Asher (2011), in addition to my own work on the topic (Magidor (2007), Magidor (2009a), and Magidor (2009b)). Camp (2004) argues that category mistakes are meaningful (Cf. Chapter 3 below). Since Asher’s book appeared very close to the completion of the current manuscript, I was not able to incorporate a detailed discussion of it, but let me make a few brief remarks. The focus of the book is not category mistakes, but rather a range of other phenomena other than category mistakes which concern predication (such as restricted predication, co-predication, and resultative constructions). However, Asher motivates his general framework, a system of very fine-grained types, by its treatment of category mistakes (ibid p.4). Like the proposal defended in Magidor (2007) and in the current work, Asher suggests that category mistakes are a presuppositional phenomenon (Asher (2011), pp.6–9), but as opposed to my own view he opts for a semantic treatment of presuppositions, one which entails that category

13

INTRODUCTION Since category mistakes are no longer widely discussed in the philosophical literature, it is not entirely straightforward to determine what the prominent views on the topic currently are. Looking at the historical debate described above, we can crudely characterize it as consisting of two kinds of views. According to one, category mistakes are defective sentences: they are either meaningless or at the very least truthvalueless. This view is often accompanied with the thought that the concept of a category mistake has some significant role to play beyond the philosophy of language, e.g. in the philosophy of mind or in metaphysics. According to the second view, category mistakes are perfectly acceptable (even if somewhat unusual) meaningful and truth-valued sentences. This second view is typically accompanied with a dismissive attitude towards the phenomenon of category mistakes or at least towards its philosophical interest. It seems that similar remarks are apt with respect to current attitudes towards the topic. The view that category mistakes are meaningless certainly continues to receive endorsement in contemporary philosophical literature—especially by those who wish to apply the concept in the context of other debates.41 On the other hand, anecdotal evidence suggests that many (if not most) contemporary philosophers take the second, more dismissive attitude. The position to be defended in this book offers a third, surprisingly underexplored alternative: one that accepts (as proponents of the second view do) that category mistakes are meaningful and truth-valued, at the same time as taking category mistakes to be a substantial phenomenon in need of an account. mistakes are at the very least truth-valueless (ibid, p.7), if not entirely meaningless (see e.g. ibid p. 5). Cf. my criticisms of semantic theories of category mistakes more generally (Chapter 3 and 4), and of type-theoretic treatments of category mistakes in particular (Chapter 3, §2.2 and Chapter 4, §2). Asher also maintains that the presuppositions associated with category mistakes cannot be accommodated, at least in the standard sense (see e.g. Asher (2011), pp. 8–9). Contrast this with my discussion of accommodation in Chapter 5, §4 below. 41 See for example, Shapiro (1997) p. 79, Steward (1997) p. 96, Fine (2003) pp. 207–8, Diamond (2001), Sorensen (2001) p. 89, Beall & van Fraassen (2003) p. 125, Stern (2006), p. 252, and Ludlow (2011), p. 65.

14

CATEGORY MISTAK ES IN THE LINGUISTICS LITER ATUR E

§3

Category Mistakes in the Linguistics Literature

Almost entirely disjoint from the philosophical debate, category mistakes were also extensively discussed in the linguistics literature.42 The starting point of the debate in linguistics is Chomsky’s revolutionary monograph Syntactic Structures.43 Category mistakes do not play a very central role in the book, but do come up in several arguments. Chomsky argues that syntax is an independent field from semantics. To support this claim he argues that the now-famous category mistake ‘Colourless green ideas sleep furiously’ is grammatical but meaningless.44 The thought is that the fact that the rules of grammar do not determine the semantic status of the sentence, lends support to syntax being an autonomous field. Chomsky also brings up category mistakes in motivating some of the details of his syntactic system. The system is divided into two components: first there is a set of rules (phrase structure rules coupled with obligatory transformations) which generate a set of the basic sentences called ‘kernel sentences’ (the notion of a kernel sentence can be thought of as the early ancestor of the notion of deep structure). Second there are (optional) transformations which allow one to derive various additional sentences from a given kernel sentence. In particular, Chomsky argues that a passive sentence such as ‘Sincerity is admired by John’ is derived 42 Note that the term ‘category mistakes’ is not very common in linguistics. Relevant terms in linguistics include ‘selectional violations’, ‘selectional restrictions’, and more rarely ‘sortal presuppositions’. Category mistakes have also been occasionally discussed in the psycholinguistics literature (see for example Johnson-Laird (1983), ch. 10). 43 Chomsky (1957). 44 Ibid, p. 15. Later in the book (p.42) Chomsky suggests that category mistakes are grammatical to a lesser degree than ordinary sentences. It should also be noted that the famous ‘colourless green ideas’ sentence was actually first introduced by Chomsky in his 1955 work The Logical Structure of Linguistic Theory (not published until twenty years later, as Chomsky (1975)) to make a rather different point. It was introduced there in order to argue against a statistical account of syntax. Chomsky argued that ‘Colourless green ideas sleep furiously’ appears in corpuses just as rarely as ‘Furiously sleep ideas green colourless’, but while the former is grammatical the latter is not, and thus grammaticality cannot be defined in terms of statistical distribution (Chomsky (1975), p. 145. This point is also reiterated in Chomsky (1957), p. 16).

15

INTRODUCTION from the active kernel sentence ‘John admires sincerity’. In support of this claim he remarks that ‘John admires sincerity’ and ‘Sincerity is admired by John’ are both perfectly legitimate sentences, while ‘Sincerity admires John’ and ‘John is admired by sincerity’ are both, as he sees it, nonsensical. The fact that active and passive forms behave similarly with respect to whether or not they constitute category mistakes is taken to support the claim that they originate from the same kernel sentence. (In particular, note that Chomsky also argues that the kernel from which a certain sentence is derived is crucial to its semantic interpretation, so a meaningless sentence should presumably be derived from a meaningless kernel).45 The next influential work in linguistics in which category mistakes played a central role was Fodor and Katz’s ‘The structure of a semantic theory’ (Fodor & Katz (1963)). Taking on board Chomsky’s syntactic framework, Fodor and Katz set out to incorporate a semantic theory into the framework. They maintained that semantics should explain all aspects of linguistic competence that are not explained via syntax (or as they put it “Linguistic description minus grammar equals semantics”, p. 173). One aspect of linguistic competence that Fodor and Katz were particularly interested in, was speakers’ ability to recognize which sentences are ‘semantically anomalous’ (a label they applied to category mistakes). They maintained, for example, that it is part of being a competent English speaker that one can recognize that ‘He painted the wall with red paint’ is an acceptable sentence, while the category mistake ‘He painted the wall with silent paint’ is not.46 Following Chomsky’s analysis in Syntactic Structures they argued that the distinction between these two sentences is not syntactic, and concluded from their general principle that it must therefore be semantic. Another reason that Fodor and Katz were interested in category mistakes is because of their role in the disambiguation of sentences. The 45 See Chomsky (1957), ch. 5 for the introduction of kernel sentences; ch. 9 for their importance in semantic interpretation; and p. 42, 78 for the importance of category mistakes in justifying the passive transformations. 46 Fodor & Katz (1963), p. 485.

16

CATEGORY MISTAK ES IN THE LINGUISTICS LITER ATUR E English word ‘ball’, for example, is ambiguous between its ‘party’ meaning, and its ‘play-object’ meaning, and consequently, the sentence ‘The ball is nice’ is ambiguous. However, according to Fodor and Katz, the sentence ‘I hit the ball’ has just one reading (the one that takes ‘ball’ in the play-object sense). The reason is that if one were to try to interpret ‘ball’ in the party sense, the sentence ‘I hit the ball’ would, they maintained, be meaningless. They thus conclude that there is only one possible reading of the sentence ‘I hit the ball’, and the sentence is ultimately unambiguous.47 With these motivations in hand, Fodor and Katz proceeded to suggest a semantic theory that renders category mistakes grammatically wellformed but meaningless. But in his 1965 book Aspects of the Theory of Syntax, Chomsky revised his earlier position.48 He now claimed that category mistakes are ungrammatical or syntactically ill-formed. The issue at stake was not simply whether one ought to count category mistakes as grammatical or not.49 Rather, the issue was the more general one of how much information should be encoded into deep structures: the view that renders category mistakes syntactically ill-formed requires syntactic structures which are rich enough to account for this. And indeed, in the Aspects, Chomsky proposed a syntactic theory which was able to rule category mistakes as syntactically ill-formed. The theory followed many of the technical details of Fodor and Katz’s treatment of category mistakes, but this time treating as syntactic, features which were treated by Fodor and Katz as semantic.50 By the late 1960s, the debate regarding whether various linguistic phenomena, among them category mistakes, should be treated via syntax or semantics played a key role in the divide between Chomskians and a 47 It is worth noting in this context that on the view which takes category mistakes to be meaningful, the sentence is ambiguous (both readings are possible). But opponents of the meaninglessness view can still maintain that one of the two reading (the one that does not result in a category mistake) is strongly preferred. 48 Chomsky (1965). 49 Even in Aspects he claims that category mistakes are ungrammatical only to a certain degree, a position he briefly mentions already in Syntactic Structures. 50 For further discussion of Chomsky’s syntactic treatment of category mistakes, see Chapter 2 below.

17

INTRODUCTION new movement in linguistics: generative semantics. Chomsky and his followers (whose position was labelled ‘interpretative semantics’) claimed that a sentence is first assigned a syntactic structure by an autonomous syntactic module, and this structure is then provided as input for semantic interpretation. As opposed to this, generative semanticists (the chief figures of the movement being George Lakoff, Haj Ross, James McCawley, and Paul Postal) claimed that the structure of a sentence is determined by both ‘syntactic’ and ‘semantic’ considerations which interact with each other in complex ways, and to a large extent simply rejected the syntax/semantics distinction. Consequently, they took infelicitous sentences such as category mistakes to be simply ‘ill-formed’, without attempting to pin-down whether this ill-formedness was syntactic or semantic.51 It is not entirely clear how the generative semanticists’ treatment of category mistake relates to the question of whether they are meaningful or truth-valued. On the one hand, generative semanticists claimed that the rules which determine which sentences are ‘well-formed’ are exactly the same rules that determine their meaning. This suggests that they would have taken ill-formed sentences in general, and category mistakes in particular, to be meaningless.52 On the other hand, generative semanticists often applied the label ‘ill-formed’ to any sentence which seems infelicitous or odd to speakers, and admitted that the occurrence of such infelicities often depend on speakers’ presuppositions and context.53 This suggests that they would have allowed at least some ‘ill-formed’ sentences to be treated as meaningful but pragmatically inappropriate. But pragmatics was then only in its very early days and this line of thought was not fully developed. 51 For a classic exchange between interpretative semanticists and generative semanticists, one in which category mistakes play a central role, see Katz (1970) and the response of McCawley (1971). For an excellent extensive history of the debate between these two camps see Harris (1993). 52 This position is sometimes explicitly endorsed. See for example McCawley’s claim that what makes ‘Robert own’s John’s after image’ odd is that “an afterimage is something that it makes no sense to speak of someone’s owning”. (McCawley (1971), p. 294). 53 See especially Lakoff (1971).

18

CATEGORY MISTAK ES IN THE LINGUISTICS LITER ATUR E In the 1960s and 70s the question of how to analyse category mistakes thus played an important role in the foundation of linguistics: because it was not clear whether the phenomenon should be treated as syntactic or semantic, the question of how to handle it formed a fruitful ground for exploring what the border between the two fields is, if there is one at all. But by the mid-1980s, probably due to a general move away from foundational issues, the debate concerning category mistakes seemed to have petered off.54 Although in more recent years category mistakes received fairly little discussion in the linguistics literature as a topic in its own right, they continue to be routinely mentioned (albeit briefly)—especially in textbooks. If one is to try and extract a common view from these brief remarks, then it seems that the Chomsky’s position of Syntactic Structures—namely that category mistakes are syntactically well-formed but meaningless—is a very standard one.55 In other places, category mistakes are classified more loosely as ‘semantically ill-formed’, ‘semantically deviant’, or ‘semantically anomalous’.56 It is far from clear, however, how to interpret such labels, and whether they commit one to the view that category mistakes are meaningless, meaningful but truthvalueless, or suffer from some third kind of semantic defect.57 The position that category mistakes are ‘semantically anomalous’ is often cashed out further by the claim that they violate ‘selectional restrictions’ (which are taken to be a species of the more general category of ‘s-selection’).58

54 Though one prominent exception is Horn (2001) (first published in 1988) which dedicates a section to category mistakes, and in linguistics too the topic seems to be regaining some recent momentum, e.g. with the discussion of category mistakes in the context of the triggering problem (see Abrusán (2011a)). 55 See e.g. Lyons (1995), p. 136, Sauerland & von Stechow (2001), p. 15413, Givón (2001), p. 10, and Carnie (2011), p. 16. 56 See e.g. Bouchard (1995), p. 44, Carnie (2002), p. 11, Fromkin (2000), §3.2.7, and Denham & Lobeck (2010) p. 287. 57 Indeed, it is not even obvious that ‘semantic anomaly’ is not sometimes taken to be a species of syntactic ill-formedness. 58 See e.g. van Valin (2001), p. 87, Carnie (2002) p. 167, or Fromkin (2000), §3.2.7. See also Chapter 2 below, for a brief discussion of the potential relevance of more contemporary syntactic theories of argument realization to the topic.

19

INTRODUCTION The idea is that predicative lexical items impose certain restrictions (‘selectional restrictions’), which restrict the kinds of arguments they accept in certain positions. For example, the verb ‘murder’ accepts in its agent positions only arguments which are human, and thus sentences such as ‘The rock murdered John’ are taken to suffer from selectional violations, and deemed to be ‘semantically anomalous’. But the treatment of category mistakes via s-selection does not seem particularly well-developed (one does not find much more beyond a few cursory remarks), and as noted above, there is little discussion devoted to what precisely ‘semantic anomaly’ amounts to. The alternative proposal that category mistakes be treated as a pragmatic phenomenon, and in particular as an instance of presupposition failure, is also occasionally briefly mentioned in the literature, but it is not fully developed or defended.59

§4

Category Mistakes in Computer Science

Although somewhat more tangential to the main thread of the current work, the discussion would not be complete without noting that issues very much related to category mistakes play a significant role in computer science. One way in which category mistakes are relevant to computer science is in the field of computational linguistics. For example, Resnik’s 1993 PhD thesis in the field is devoted entirely to the topic of category mistakes.60 Resnik employs tools from information theory in order to suggest an analysis of category mistakes and also applies this analysis to some practical problems in automated natural language processing. He proposes, for instance, that an automated detection of category mistakes can aid an automatic parser in disambiguating sentences. The idea, already familiar from Fodor and Katz’s work, is that when a supposedly ambiguous sentence has two readings, one which is a category mistake 59 60

See Seuren (1988), Beaver (1997), p. 994, and Abbott (2006), p. 4. Resnik (1993).

20

CATEGORY MISTAK ES IN COMPU TER SCIENCE and one that is not, we should opt for the latter reading.61 A computational approach to category mistakes can therefore improve the automatic processor’s ability to replicate the behaviour of a natural-language speaker. But the relevance of the topic to computer science goes much further than the field of computational linguistics. Since computer languages typically involve variables and values which are classified into different types, they also crucially face the issue of what type restrictions are placed, and how to handle various kinds of type violations—issues with are naturally connected to the topic of category mistakes. One aspect of this issue, which comes up both in the theory of particular programming languages as well as in the related field of lambda calculus, is that of static versus dynamic typing. It is easiest to explain the issue by way of illustration. Consider a programming language which falls under the Object Oriented Programming (OOP) paradigm. Such a language typically allows one to define a range of ‘objects’ which fall under different ‘classes’ or types. Each class consists of an abstract definition which specifies a range of attributes that objects belonging to the class can have, and a range of functions they can perform. For example, in a program designed to run a bank, we can define the class ‘Customer’— which specifies that objects of this class will have an attribute describing the current balance of the particular customer-object’s account, and a function ‘add-balance(x)’ which adds a specified amount to the costumer’s balance. The language may also allow objects to belong to multiple classes: for example, all objects belonging to the class ‘Business Customer’ may, by default, also belong to the super-class ‘Customer’, and all objects of any type belong by default to the ultimate super-class ‘Object’. But allowing objects to belong to multiple classes raises difficult issues about how to handle variable typing. First, suppose a particular program contains a variable ‘Jones’, of the class Customer. Suppose the command

61 Though pace Fodor and Katz, Resnik accepts that the categorically-mistaken reading is theoretically available, but rightly notes that the preferred reading should be the noncategorically-mistaken one.

21

INTRODUCTION ‘Jones.add-balance(100)’ requires the object which is the value of ‘Jones’ to add 100 pounds to its balance attribute. Next, consider a variable ‘Obj’, of the most general class Object. Now suppose a certain program contains the command ‘Obj.add-balance(100)’. Note the value of ‘Obj’, might be an object which also belongs to the class Customer (for example, the above command might be preceded by the instruction to assign to ‘Obj’ the value of ‘Jones’), in which case, the command can—at least in principle—be carried out smoothly. On the other hand, the value of Obj might be an object which does not belong to the sub-class Customer, in which case our command might well be undefined, and lead to a kind of type error. Finally, note that the question of whether or not Obj belongs to the class Costumer, may not be computable simply by reviewing the program’s code (this follows both from general theorems concerning the limitations of computability, and more simply, because the value of ‘Obj’ may depend on some choice the user inputs during runtime). Should the programming language allow for such a command? As it turns out, different languages take different approaches to this issue. Languages which employ static typing attempt to detect and prevent all type errors at compilation time. Since at compilation time we cannot determine what the value of ‘Obj’ is and what classes it belongs to, we cannot be certain that the command ‘Obj.add-balance(100)’ is well-defined, and hence in a language employing static typing, the complier would deem this statement to involve a type error. On the other hand, languages which employ dynamic typing perform type-checks only at runtime. Typically, in a dynamically typed language, the compiler would accept the above command, but if it turns out at runtime that the value of ‘Obj’ is not an object of the right class, a runtime type-error will occur. The upshot is that statically-typed languages are in a sense ‘too safe’—they ensure no type violations occur at runtime by refusing to compile some commands that might be well-defined. On the other hand, dynamically-typed languages are ‘not-safe enough’, because they allow the possibility of some runtime type-errors. And of course, various hybrids between static and dynamic typing are also possible. 22

CONCLUSION The question of static versus dynamic typing is a central issue in computer science. Although the connection of this issue to the topic of category mistakes in natural languages is somewhat tangential, there are interesting analogies between the debates. At least to a first approximation, the static typing approach, which detects type violations at the level of compilation, seems analogous to the view which locates the problem with category mistakes at the level of meaning (cf. the meaninglessness view discussed in Chapter 3); While the dynamic-typing approach, which detects type violations only at the runtime stage, can be compared to the view that locates the problem with category mistakes at the level of content or reference (cf. the MBT view, discussed in Chapter 4). As we shall see, one crucial issue that divides the meaninglessness view from the MBT view, is how to treat a sentence such as ‘That is green’, where the demonstrative ‘that’ can refer to a standard concrete object in some contexts, but to an abstract object in other contexts. The meaninglessness view must either deem the sentence to be in general meaningless (though few proponents of the meaninglessness view would accept such an extreme conclusion), or else deny the sentence is a category mistake— even in those contexts where the demonstrative refers to an abstract object. The MBT view, on the other hand, has the flexibility of maintaining that in such contexts, the sentence is a meaningful but truth-valueless category mistake. Adherents of this view must concede, however, that category mistakes are in general meaningful. It is not hard to see that this dilemma is very much related to the question of how programming languages should handle commands of the sort discussed above. For the remainder of the book, however, I leave aside such issues in computer science, and return to the question of how to account for category mistakes in natural language.

§5

Conclusion

Category mistakes form a substantial linguistic phenomenon which interacts with central foundational questions in the philosophy of language. The phenomenon was once widely discussed in both philosophy 23

INTRODUCTION and linguistics, but while it continues to play a role in both fields, it has not received sufficient attention in recent years. The past literature on the topic contains much interesting material, but an adequate account of category mistakes is still lacking, and it is time to revisit the debate from a contemporary perspective. In the remainder of the monograph, I present and discuss what I take to be the four main approaches of category mistakes (the syntactic approach, the meaningfulness view, the MBT view, and the pragmatic approach). The discussion will culminate with a defence of a presuppositional account of category mistakes: one according to which category mistakes are meaningful, truth-valued, but suffer from (pragmatic) presupposition failures. This discussion will hopefully serve to reconstitute this important debate, at the same time as bringing us closer to its successful resolution.

24

2

2

The Syntactic Approach

§1

The Syntactic Approach to Category Mistakes

The first approach for explaining the infelicity of category mistakes (‘the syntactic approach’) states that category mistakes are infelicitous because they are syntactically ill-formed or ungrammatical.1 The syntactic approach received its most significant support in the 1960s, when it was defended in Chomsky’s Aspects of the theory of Syntax.2 Although to some extent outdated, it is well worth considering Chomsky’s discussion in some detail. Chomsky’s discussion provides the single most detailed and explicit attempt to defend the syntactic approach. Moreover, the problems with Chomsky’s theory are ones that afflict the syntactic approach more generally. Chomsky’s treatment of category mistakes in the Aspects came against the background of his pioneering development of the theory of syntax. The role of syntax, on this approach, is to devise for each language a ‘generative grammar’ for that language. Roughly, a generative grammar is a finite set of rules that generate all and only the grammatical sentences of the language. Moreover, the generation process should produce for each sentence what Chomsky calls a ‘phrase marker’: a generation tree that reveals the deep structure of the sentence. The representation of the deep structures via phrase markers enables one to account for important 1 I use these two terms interchangeably, but note that the two notions are sometimes distinguished—especially in the earlier literature. (See e.g. Ziff (1964), p. 392, and Jackendoff (1972) p. 17). 2 Chomsky (1965). The syntactic approach is also supported in Ziff (1964), Katz (1964), and Strawson (1970).

25

THE SYNTACTIC APPROACH syntactic distinctions. For example, one can use it to explain how sentences such as ‘John is easy to please’ and ‘John is eager to please’ can have very different deep structures (ones that reveal that the latter sentence concerns John as a pleaser, while the former concerns John as the one being pleased), despite the fact that they have very similar surface structures (noun-copula-adjective-infinitive verb). As opposed to traditional grammar that limited itself largely to the classification of words into parts of speech, Chomsky’s notion of deep structure enabled him to treat a much wider set of linguistic phenomena as syntactic. But this opened up the question: which linguistic phenomena should be treated as syntactic? In his Aspects, Chomsky notes that we must somehow account for the deviance of sentences such as ‘John solved the pipe’ or ‘The boy may frighten sincerity’. Although he admits that such sentences are not “clear-cut cases of violation of purely syntactic rules” (clear-cut violations are sentences such as ‘sincerity frighten may boy the’), he claims that they “deviate in some manner from the rules of English” (p. 76). He grants that this still leaves room for treating category mistakes as a semantic phenomenon (as he himself suggested in his earlier monograph Syntactic Structures), but nonetheless settles on treating category mistakes as a syntactic phenomenon.3 Chomsky’s treatment of category mistakes in the Aspects proceeds roughly as follows. Each lexical item is assigned not only a general grammatical category such as ‘noun’ or ‘verb’, but also what Chomsky calls ‘selectional features’. For example, the word ‘sincerity’ receives selectional features such as ‘+abstract’, ‘−animate’, and ‘−human’, while the word ‘boy’ is assigned features such as ‘−abstract’, ‘+animate’, and ‘+human’.4 Verbs and adjectives also receive selectional features, which mark which kinds of arguments they are able to accept. For example 3

See Chomsky (1965), §2.3.1. It may be that some features can be systematically predicted on the basis of others, and thus do not need to be explicitly specified in the lexicon. For example, we can add to the system rules that allow us to infer from the fact that a lexical item has the feature ‘+abstract’ that it also the feature ‘−human’. See Chomsky (1965), pp. 164–70. 4

26

THE SYNTACTIC APPROACH TO CATEGORY MISTAK ES ‘frighten’ will receive a special selectional feature indicating that it cannot take a lexical item with a selectional feature of ‘-animate’ as a direct object.5 Finally, certain syntactic rules ensure that the selectional features of different lexical items match up, so that no category mistakes are generated by the syntax. So, for example, ‘The dog frightens the cat’ and ‘The dog frightens the elephant’ come out as syntactically permissible, but not ‘The dog frightens sincerity’. In a fascinating section of his book entitled ‘The boundaries between syntax and semantics’ (§4.1 of the Aspects), Chomsky reopens the question of whether category mistakes should be treated as semantic or syntactic. His discussion is somewhat tentative, but Chomsky suggests at least one compelling argument in favour of the syntactic approach. The core of the argument is this. Chomsky acknowledges that category mistakes are at best a borderline case of syntactically ill-formed sentences, but he argues that the syntactic machinery that he uses in order to address the phenomena (namely, selectional features) is independently motivated. Consider for example the following two sentences: (1) The boy who is next to the table is large. (2) *The book who is next to the table is large.

According to Chomsky, (2), as opposed to (1), is ungrammatical: a relative clause which modifies the noun phrase ‘the book’ should be formed using the relative pronoun ‘which’ rather than ‘who’. Furthermore, he suggests that in order to account for this difference between (1) and (2) we must appeal to the fact that ‘boy’ has the selectional feature ‘+human’ while ‘book’ has the selectional feature ‘−human’. But once it is conceded that selectional features must anyhow be taken as an integral part of syntax, it is only natural to extend this treatment to sentences such as ‘The book is studying linguistics’. Later in this chapter, I will turn to assess Chomsky’s who/which argument (see §3 below), and the syntactic approach more generally. But it is

5 The selectional features of verbs (or other multi-place predicates) thus have a more complex form than those of nouns because they mark different restrictions relative to different argument positions.

27

THE SYNTACTIC APPROACH worth preceding the discussion with a few remarks on how it might be affected by more recent advances in linguistics. Since the publication of Chomsky’s Aspects, much has developed in the study of syntax in general, and the study of the relationship between verbs and predicates and their arguments in particular. Do these developments bear on the syntactic approach to category mistakes? There seem to be no contemporary attempts to defend the syntactic approach. Indeed, where category mistakes are mentioned in the contemporary syntax literature, it is merely in order to set aside the phenomenon and claim that it is not ultimately syntactic.6 Moreover, it is not obvious how a contemporary version of the syntactic approach might look like. Nevertheless, let me offer a few brief (and non-exhaustive) remarks on the apparent relevance of some more recent developments to the syntactic approach. The most important developments for current purposes occur in the field of lexical semantics and in the related research into the syntaxsemantics interface. A central issue for this field (often described under the heading of ‘argument realization’), is to account for the way in which the different argument positions of various verbs are represented in syntax. Consider for example the sentence ‘John ate the apple’. According to one standard account, this sentence describes an eating event, in which John is the agent (in this case—the eater), and the apple is the theme (in this case—the object being eaten). Leaving aside the tricky question of precisely what the labels ‘agent’ or ‘theme’ mean,7 it is important to note that these are semantic labels: they tell us something about the role that John and the apple play in the eating event described. Now the fact that the sentence in question describes an event in which John and the apple play these particular semantic roles is reflected in the syntactic structure of the sentence: that ‘John’ is the grammatical subject of the sentence reflects, in this case at least, the fact that John plays the role of the agent in

6 See for example Carnie (2011), p. 16, Givón (2001), p. 10, Grimshaw (2005), p. 80, and Sauerland & von Stechow (2001), p. 15413. 7 For a helpful discussion of this issue see Parsons (1995).

28

THE SYNTACTIC APPROACH TO CATEGORY MISTAK ES the event.8 To put things otherwise, we cannot describe an eating event in which John is the agent and the apple is the theme using the sentence ‘The apple ate John’. While this observation might seem at a first glance to be relevant to a syntactic analysis of category mistakes, it is far from clear that it is. Contemporary syntactic theory may well play a crucial role in ruling out a potential reading of ‘The apple ate John’ in which the sentence means that John ate the apple, but, crucially, this is not a reading where the sentence is interpreted as a category mistake. On the other hand, one could try to rule out the potential reading of ‘The apple ate John’ where it is interpreted as category mistake by insisting that an apple cannot play the role of an agent in an eating event.9 But note that such an account would basically be a semantic account of category mistakes rather than a syntactic one: according to the suggestion, it is not the rules of syntax that are responsible for the infelicity of the sentence, but rather the semantic claim that there is no coherent property of events which classifies them as eating events having an apple as an agent. Since our current focus is on the syntactic approach, we can thus leave such proposals aside for now. The interaction between syntax and semantic plays another role in contemporary theories of argument realization. Given the wide range of data concerning the way in which different verbs syntactically represent their arguments, one can ask if these data are to be explained via arbitrary and idiosyncratic syntactic features which are stipulated for each verb, or are the different syntactic representations associated with each verb something that is determined (or at least highly predictable) from its semantic interpretation. Recent research into this question provides

8 Note though that the precise relationship between the event described and its morphosyntactic realisation is a highly complex matter. For example, subjects (at least surface subjects) of an active sentence don’t always pick out the agent of the event described: a sentence containing an unaccusative verb such as ‘The door opened’, describes an event in which the door is the theme and not the agent. (See Levin & Rappaport Hovav (1995)). 9 Note, though that an apple can at least play the agent role in other events, as in ‘The apple pleased John’ (see Baker (1997), §2.1 for a defence of this claim.)

29

THE SYNTACTIC APPROACH increasing support for the latter hypothesis.10 One might hope that this kind of interaction between syntax and semantics would allow for a much larger range of properties to act as input for syntactic rules and thus provide us with more flexibility in dealing with the phenomenon of category mistakes (cf. the discussion of the argument from particularity in §3 below). But this would be a misunderstanding of the nature of the interaction. Even those who maintain that the syntactic argument realization properties supervene on the semantic interpretations of verbs, hold that it is only a very small number of some highly restricted semantic features that are relevant to determining a verb’s syntactic properties.11 Thus even if syntactic argument realization properties are ultimately determined by semantic features, this does not entail that they are sufficiently rich to allow for a syntactic treatment of category mistakes. With these initial remarks in place, I turn to an assessment of the syntactic approach. The structure of this chapter is somewhat atypical: while I ultimately reject the syntactic approach, I think the approach has more appeal than it is usually taken to have, and that it ought not to be dismissed too quickly. I begin in §2 by presenting several arguments that have been raised against the syntactic approach, but which I find to be unsatisfactory. In §3–7, however, I turn to present other, more compelling, arguments against the syntactic approach, and conclude that the approach should, after all, be rejected.

§2

Some Unsatisfactory Arguments Against the Syntactic Approach

§2.1 The simplicity argument One argument that has been brought forth against the syntactic approach is that it makes syntax too complex. Thomason, for example, argues that 10 See Levin & Rappaport Hovav (2005) for a comprehensive survey of the support for this hypothesis. I return to the issue of supervenience of syntactic features on semantic ones in §2.3 below. 11 See Levin & Rappaport Hovav (2005), ch. 4.

30

SOME UNSATISFACTORY ARGUMENTS the role of syntax is “generating the grammatical sentences by means of effective rules operating on a finite vocabulary classified into a finite number of syntactic categories. These rules, of course, must be as elegant and as simple as possible”.12 Thomason proceeds to argue that treating category mistakes as a syntactic phenomenon would require an extremely complicated syntactic theory, violating the requirement that the rules of syntax ought to be as simple as possible. Thomason is right to think both that the rules of syntax must be as simple as possible and that treating category mistakes as syntactic would add serious complications to one’s syntactic theory. Nevertheless, his argument is not convincing. The requirement for simplicity of the rules of syntax is more carefully formulated as follows: given the set of grammatical and ungrammatical sentences, we should produce a set of rules generating all and only the grammatical sentences. Of all the possible sets of rules which generate all and only grammatical sentences, we should choose the simplest.13 But the requirement for simplicity should not determine which sentences are taken to be grammatical. Even if syntax should be as simple as possible it may be impossible to devise an adequate syntactic theory unless it deems category mistakes to be ungrammatical. Moreover, if the infelicity of category mistakes is not explained via syntactic means, it must be explained by some other means, i.e. by a semantic or a pragmatic theory. So while removing the burden of accounting for category mistakes from syntax would certainly simplify syntax, it is

12

Thomason (1972), p. 211. Even this formulation is too crude, because it requires only what Chomsky has called ‘a descriptively adequate grammar’. But Chomsky places a further desideratum on grammar: that it be ‘explanatorily adequate’ (See Chomsky (1965), pp. 24–7). Chomsky’s original gloss on this further requirement is that the grammar would be psychologically and biologically real, in particular in a way that would explain how language acquisition is possible. But the further requirement can also be cashed in other ways: e.g. that the syntax would generate phrase markers that reveal certain aspects of the sentences’ structure or that the syntax would help explain various cross-linguistic generalisations. The fact there may be further requirements on syntax other than providing a descriptively adequate grammar only serves to weaken Thomason’s objection. 13

31

THE SYNTACTIC APPROACH likely to complicate other aspects of linguistic theorizing. But equally, semantics and pragmatics should be as simple as possible. As with any phenomenon, one would like an account of category mistakes which is not unnecessarily complicated but this does not entail that the burden of accounting for the phenomenon should not be placed in particular on syntax.

§2.2 The meaningfulness argument Another argument that has been raised against the syntactic approach runs as follows.14 Although category mistakes are infelicitous, they are nevertheless interpretable or meaningful. But only syntactically wellformed sentences can be meaningful. Thus category mistakes must be syntactically well-formed. One potential problem with this argument is that it relies on the highly controversial premise that category mistakes are meaningful. However, in Chapter 3 I will argue in detail that this premise is in fact correct and thus this point is not in itself a reason to reject the argument. But a second problem with this argument is that it implicitly relies on yet another assumption: that syntactically ill-formed sentences are meaningless. The issue is a complex one, but in Magidor (2009a) I argue that this assumption is false: syntactically ill-formed sentences can be meaningful, and thus the fact that category mistakes are meaningful is not sufficient to rule out the syntactic approach.15

§2.3 The argument from universality A third argument that has been raised against the syntactic approach has to do with the universal nature of category mistakes.16 14

See Arad (1996), p. 222. Interestingly, Chomsky’s Aspects contains a rare endorsement of the position I argue for. Chomsky argues that sentences such as those containing ‘who/which’ confusions are “uniquely, uniformly, and immediately interpretable . . . although they are paradigm examples of departure from well-formedness” (Chomsky (1965), p. 151). 16 The claim that the universal nature of category mistakes is an argument against the syntactic approach is made in Fodor (1977), p. 98, and also mentioned sympathetically (though not endorsed) in Chomsky (1965), p. 77. 15

32

SOME UNSATISFACTORY ARGUMENTS Syntax is language relative.17 Different languages have different grammars and consequently different kinds of strings are deemed to be syntactically ill-formed. For example, while (3) is grammatical in French, the corresponding string (4) in Hebrew is ungrammatical, because the noun ‘mita’ (bed) in Hebrew has the grammatical gender ‘feminine’. (3) Le lit est grand The bed is bigmasc (4) *Ha-mita gadol The-bed bigmasc The bed is big

On the other hand, the kind of infelicity which category mistakes exhibit does not seem language relative in this manner: given a particular category mistake s in a language L1 and another language L2 it is usually easy to find a category mistake in L2 that would correspond to s, i.e. would be infelicitous in a very similar way. (Think for example of the translation of ‘Two is green’ to a range of languages . . .)18 The argument from universality relies on this observation, and runs as follows. Syntactic deviancies are often idiosyncratic to specific languages. This might be thought to be especially so when the deviance in question is one generated due to lexically determined features, such as grammatical gender of common nouns. But the infelicity of category mistakes seems to be universal across languages, and this is so despite the fact that plausibly any syntactic account of category mistakes would have to depend on lexically determined features (such as the selectional features of Chomsky’s account). The thought, then, 17 Note that Chomskians do not dispute this claim. Although Chomsky thinks that all languages share a core universal structure (sometimes called ‘universal grammar’) he also maintains that the particulars of grammar differ from language to language. This is achieved both via the differences between the lexicons of different languages, as well as by the fact that (at least according to the ‘principles and parameters’ model) the universal grammar contains certain parameters which receive different values in the grammars of different languages. 18 It is worth noting, though, that this may not hold of all category mistakes. Cf. the discussion of highly particularised category mistakes in §3 below.

33

THE SYNTACTIC APPROACH is that the cross-linguistic uniformity of the phenomenon of category mistakes is an argument against treating it as syntactic. The problem with the argument from universality is that not all lexically determined syntactic features are as arbitrary as the gender of common nouns. Various syntactic features of lexical items are highly predictable by (even if they do not fully supervene on) the semantic values of these items. Consider for example the classification of lexical items into lexical categories such as ‘verb’ or ‘noun’. The lexical category of a word is widely agreed to be a syntactic feature of it. Nevertheless, which lexical category the item receives seems to be, at least to a large extent, predictable from its semantic values (roughly speaking, verbs tend to denote actions or properties of events, while nouns tend to denote properties of standard objects). Furthermore, it is often the case that synonymous words in different languages receive the same classification. Although it should be noted that the classification of words into lexical categories is not perfectly uniform across languages,19 this classification is nevertheless relatively stable. The crucial point is that this relative stability is not sufficient to undermine the claim that lexical categorization is syntactic. The classification of such categorization as syntactic seems primarily driven by the role such features play in explanations or generalizations that relate to what are unambiguously syntactic phenomena, rather than on their stability across languages.20 19 To point to two exceptions: the English word ‘must’ is an auxiliary verb, but its translation into Hebrew is an adjective and English adjectival phrases such as ‘is tall’ are translated into Hawaiian as stative verbs. 20 As I have noted in §1, another area in syntax where many linguists have posited a supervenience of syntactic features on semantic ones is in the study of argument realization (see Levin & Rappaport Hovav (2005)). For an illuminating discussion of why this supervenience does not undermine the fact that such properties in question are syntactically represented (specifically with respect to unaccusativity) see Levin & Rappaport Hovav (1995), especially §1.2.2. In the context of the syntactic approach to category mistakes, Chomsky himself observes that the question of to what extent semantic features determine syntactic ones is independent of the question of which features should be treated as syntactic, when he says that category mistakes raise “several difficult and rather vexing questions. First, it is not obvious to what extent this information should be provided by the syntactic component at all. Second, it is an interesting question whether or to what extent semantic considerations are relevant in determining such subcategorizations . . . These are distinct questions, though they are often confused. They are connected only in that if the basis for making the distinctions is purely syntactic, then surely the information must be presented by the syntactic component of grammar. We might call these questions the questions of presentations and justification, respectively”. (Chomsky (1965), p. 75).

34

THE ARGUMENT FROM PARTICUL AR IT Y Thus even if the treatment of category mistakes required syntactic lexical features (such as Chomsky’s selectional features), it is not inconceivable that these features might themselves be determined by semantic features, and hence uniform across languages. Cross linguistic variation might be a good sufficient condition (at least as a rule of thumb) for classifying a phenomenon as syntactic (cf. the case of gendered nouns), but it is much less compelling as a necessary condition. In particular, the crosslinguistic uniformity of category mistakes is not a strong argument against the syntactic approach. So far, I have discussed three arguments against the syntactic approach which I think are ultimately unsuccessful. In what follows I turn to present other arguments against the syntactic approach which are, I think, much more compelling.

§3

The Argument from Particularity

An important observation is that in order to offer a syntactic account for the infelicity of category mistakes in general, one needs to appeal not only to very general selectional features such as ‘+animate’ or ‘+human’ but also to extremely particular features.21 For example, Hebrew has special verbs for describing the picking of different types of fruit: ‘livtzor’ is used to describe picking grapes, ‘ligdod’ is used to describe picking dates, and ‘le-erot’ to describe picking figs. Thus while sentence (5) is completely acceptable, sentence (6) is a category mistake:22 (5) Ani botzer et ha-anavim. I am (grape)picking the grapes. (6) *Ani goded at ha-anavim. I am (date)picking the grapes.

21

See McCawley (1968), p. 134, who suggests the examples from English below. In case the English notation below is misleading, let me clarify that the Hebrew verbs are not morphologically complex. 22

35

THE SYNTACTIC APPROACH Examples in English include the verb ‘diagonalize’ which arguably can only be properly applied to matrices (note the infelicity of ‘I diagonalized the number three’), or the adjective ‘benign’ which, at least in one sense of the word, seems to only be properly applied to tumours. This means that a syntactic theory of category mistakes would require not only general syntactic features such as ‘+human’, but also highly particular ones such as ‘+grape’, ‘+matrix’, or ‘+tumor’. A syntactic treatment of category mistakes which relies on a very large number of highly particularized syntactic features certainly seems highly unappealing. But some care is needed in pinning down precisely what the problem with such a treatment is. One worry might be that adding so many features will make for a highly complicated theory of syntax. But this is simply a version of the simplicity argument, which I have already dismissed in §2.1. Another, somewhat more subtle concern, is suggested by McCawley.23 The worry is that the proliferation of highly particularized syntactic features would entail that syntax essentially encodes the same lexical information as semantics, in effect collapsing the distinction between syntax and semantics, or the need for both facets of linguistic theorizing. Take for example the syntactic feature ‘+matrix’. Plausibly, only nouns that denote the property of being a matrix would be marked with this highly idiosyncratic feature. Thus it seems that syntax and semantics are both representing the same information. McCawley’s worry is, however, too quick. For a start, note that syntactic features such as ‘+animate’ or ‘−matrix’ are merely labels: the syntactic theory would have achieved exactly the same results if one replaced the labels ‘+matrix’ or ‘−matrix’ with more arbitrary labels such as ‘+1’ and ‘−1’. On the other hand, the semantic theory assigns the noun ‘matrix’ a semantic content (e.g. a property or a function), and not merely a label. Thus even given highly particularized syntactic features, the semantic component encodes more information than is encoded by the syntactic 23

See McCawley (1968), p. 135. The duplication complaint is also reiterated in a slightly different context in Grimshaw (1979), p. 317.

36

THE ARGUMENT FROM PARTICUL AR IT Y features. Perhaps then, the worry is better understood as the complaint that given the semantic-values, the syntactic features are redundant? This, however, brings us back the discussion in §2.3: there are cases where we would want to classify a certain class of features as syntactic, even though these features are fully determined by semantic properties. Syntactic features are those which play a distinctive role in our theory of syntax, independently of what our meta-syntactic theory claims about their relationship to semantic features. What then, is the problem, with highly particularized syntactic features? In whichever way one ultimately explains the infelicity of category mistakes, category mistakes are clearly not a paradigmatic or uncontroversial case of syntactically ill-formed sentences. (Note that even Chomsky, when defending the syntactic approach in the Aspects, accepts this). A plausible methodological principle for classifying such borderline cases is the following. One starts with the paradigmatic cases of syntactic phenomena; One devises a syntactic theory that best accounts for these cases; Other things being equal, one only treats the remaining borderline cases via one’s syntactic theory if one can do so using independently motivated syntactic machinery.24 Note that this methodological principle is precisely what underlies Chomsky’s who/which argument. According to the argument, category mistakes should be classified as syntactic, precisely because they can be explained via features such as ‘+/−animate’ that are independently motivated by their role in accounting for another phenomenon (the ‘who’/’which’ distinction)—one which, according to Chomsky, is a clearly syntactic one.25 The problem, however, is that once one realizes that a full treatment of the phenomenon of category mistakes requires a much wider range of features the argument collapses. 24 The qualification ‘other things being equal’ is important here: there may be other overriding reasons to treat the phenomenon as syntactic. 25 Interestingly, though, the claim that the ‘who/which’ distinction is syntactic is highly controversial. It was contested already in the late 1960s and early 1970s (see McCawley (1968), p. 139; Lakoff (1971), pp. 330–2; and Jackendoff (1972), ch. 1). Moreover, at least one standard contemporary view treats the distinction between ‘who’ and ‘which’ (and similarly that between ‘he’/‘she’/‘it’) as presuppositional rather than syntactic. (See for example Cooper (1983), ch. 7, and Heim & Kratzer (1998), pp. 123–8 and pp. 244–5).

37

THE SYNTACTIC APPROACH Even if having a syntactic feature such as ‘+animate’ is independently motivated, it is highly doubtful that any paradigmatically syntactic phenomenon would require features such as ‘+grape’ and ‘+tumor’. Thus given the methodological principle proposed above, the phenomenon of category mistakes should not be treated as syntactic.26

§4

The Argument from Embedding

Another argument that tells against the syntactic approach is the argument from embedding.27 When one embeds an ungrammatical sentence, in particular inside a ‘that-clause’, the resulting embedding usually remains ungrammatical and thus infelicitous. For example, embedding the mildly ungrammatical ‘Me likes apples’ in the followings contexts results in sentences that are still ungrammatical: (7) * John said that me likes apples. (8) * John dreamt that me likes apples. (9) * It is nonsense to say that me likes apples.28

Thus if category mistakes are syntactically ill-formed one would predict that they would remain syntactically ill-formed and hence infelicitous 26 This methodological principle might also help tease apart similar looking cases. Consider for example the distinction between ‘John ate that the world is round’ and ‘John ate the proposition that the world is round’. According to one plausible hypothesis the former is syntactically ill-formed, because ‘eat’ cannot take a complement phrase as an argument (note that both the claim that verbs syntactically select the lexical category of their arguments, and the claim that CP is a lexical category have independent syntactic motivations), while the latter is a syntactically well-formed category mistake. 27 See Grimshaw (2005), p. 80, Jackendoff (1972), p. 18, and McCawley (1968), p. 128. 28 Note that the claim that (9) is ungrammatical is slightly more controversial than the other cases. Chomsky discusses embeddings under operators such as ‘it is nonsense to say that’ in Chomsky (1965), pp. 157–8. He allows that such embeddings are acceptable when the embedded sentence is a category mistake (as in example (12)), but suggests they may also be acceptable in other cases where the embedded sentences is syntactically ill-formed (as in 9). After offering some brief notes on the technical problem of how to admit such embeddings while ruling out the generation of the embedded sentences, he concedes that this issue “provides a slight consideration in favor of the decision to eliminate selectional rules from the syntactic component, and to modify the theory of the semantic component in some ways so as to allow it to accommodate these phenomena”.

38

INTER ACTIONS W ITH ME ANINGS under these embeddings. This prediction turns out, however, to be incorrect. The following embeddings for example, seem unproblematic: (10) John said that the number two is green. (11) John dreamt that the number two is green. (12) It is nonsense to say that the number two is green.

Unless one maintains that category mistakes behave differently under such embeddings than other syntactically ill-formed sentences and provides some non-ad-hoc explanation for this difference, the embedding data strongly suggests that category mistakes are not syntactically ill-formed.

§5

Interactions with Meanings

Another problem for the syntactic approach is that some category mistakes seem to arise due to complex interactions between the meanings of various words in a sentence. Janet Fodor, for example, observes that in so far as we take (13) to be a category mistake, (14) should receive a similar verdict:29 (13) This corpse admires sincerity (14) This man that I proved that John was mistaken in believing to be alive admires sincerity.

The example is perhaps not perfect (many metaphysicians maintain that a man—even a dead man—is not a corpse). But other examples in the same vein are readily available. If (15) and (17) are category mistakes then so are (16) and (18): (15) 2.2 is prime. (16) This number that I proved that John was mistaken in believing was a natural number is prime. (17) This man is pregnant. (18) This person that I proved that John was mistaken in believing was a woman is pregnant.

29

Fodor (1977), p. 98. See Jackendoff (1972), pp. 18–19 for a similar argument.

39

THE SYNTACTIC APPROACH The problem with such cases is the following. One might envisage a syntactic theory where, for example, ‘is pregnant’ requires an argument with the syntactic feature ‘+female’, and where ‘man’ is marked with the feature ‘−female’. But it is hard to see how any such theory could possibly account for the infelicity of (18). The fact that the noun phrase in (18) represents a person which is not female depends, among other things, on the very specific meanings of ‘prove’ (in particular, that it is factive), and ‘mistaken’ (in particular, that it is anti-factive) and on highly complex interactions between these meanings. It is highly unlikely that the syntactic component can systematically predict the relevant features for such complex noun phrases.30

§6

Interactions with Extra-Linguistic Facts

A related problem for the syntactic approach is this. Whether or not a sentence is classified as a category mistake depends not only on purely linguistic knowledge, but also on knowledge (or at least beliefs) of extralinguistic empirical facts.31 Consider for example the following category mistake: (19) This rock is thinking about the theory of relativity.

It is doubtful that our knowledge that rocks are inanimate, non-sentient, or incapable of thought is really part of our linguistic knowledge concerning the word ‘rock’. Rather, it seems that we have acquired the knowledge that, for example, monkeys can think but rocks cannot think by 30 One anonymous reader raised the worry that this argument will over-generalise to show that the gender agreement of pronouns is also not a syntactic feature (cf. ‘This person that I proved that John was mistaken in believing was a woman said that she is coming to dinner’). However, this consequence is not necessarily unpalatable: as noted in f.n. 25 above, a standard contemporary theory holds that such gender agreement features are indeed not syntactic. 31 This observation is noted in McCawley (1968), p. 129, Johnson-Laird (1983), p. 236, Jackendoff (1972), pp. 19–20, and something like this argument is also suggested by Grimshaw (2005), pp. 79–80, and Arad (1996), p. 222. Those with Quinean qualms about whether there is a real distinction between facts about meaning and extra-linguistic facts, can see the argument in this section as another version of the one in §5 above.

40

INTER ACTIONS W ITH E XTR A-LINGUISTIC FACTS empirical investigation (possibly, empirical investigation that was carried out after one already had the relevant words in one’s lexicon). But if so, then the infelicity of (19) cannot arise due to a purely linguistic, and in particular syntactic, features of the word ‘rock’. The point is perhaps strengthened if we consider sentences that in certain contexts would be deemed to be category mistakes, but which given additional background factual information may receive a different verdict. Consider for example: (20) This priest is pregnant. (21) This woman fathered my children. (22) This machine is thinking about the theory of relativity.

One familiar only with the Catholic Church may well deem (20) to be a category mistake, but not so once one learns that some Christian denominations allow women to be ordained. (21) may sound highly infelicitous, until one realizes that this woman may have previously been a man who fathered my children, and then went on to have a sex-change operation. And (22) would probably be judged to be a category mistake by one who has not learnt of computers with sophisticated artificial intelligence capabilities.32 One might argue that (20)–(22) are not really category mistakes: rather, some erroneously judge them to be category mistakes because they lack the relevant empirical information. This proposal has some serious disadvantages (for one thing, it divorces the notion of a category mistake from the phenomenological quality that was used to characterise the phenomenon in the first place). But even if the proposal were adopted, it would entail that in order to discover whether or not a sentence is a category mistake, one must appeal to many extra-linguistic empirical facts. This, however, would be a significant problem for the syntactic approach: it is highly implausible that the medical discovery of sex-change operations or the invention of sophisticated computers brought with it the

32

Of course, the infelicity judgments may be harder to get for readers who already possess the relevant background knowledge.

41

THE SYNTACTIC APPROACH syntactic discovery that the words ‘woman’ or ‘machine’ had different syntactic features than we have previously assumed.33

§7

Interactions with Context

A final problem for the syntactic approach concerns the fact that whether a certain sentence is taken to be a category mistake often depends on the context in which it is uttered.34 Consider the following: (23) The thing Jane is thinking about is green. (24) John’s best friend is pregnant.

When (23) is uttered in a context where it is clear that Jane is thinking about a book, the utterance is perfectly acceptable. But relative to a context in which it is known that Jane is thinking about a number, (23) would be deemed to be a category mistake. Similarly, (24) would be perfectly acceptable when uttered in a context where it is clear that John’s best friend is a woman, but not so in one where it is well-known that John’s best friend is a man. Thus at least in some contexts, (23) and (24) constitute category mistakes. But both sentences are clearly syntactically wellformed, and thus the syntactic approach cannot account for the infelicity of these sentences in the relevant contexts. The syntactic approach to category mistakes is far from being the nonstarter that it is often taken to be. Nevertheless, the approach should ultimately be rejected.

33 Nor is it particularly plausible that the empirical discovery brought with it an immediate change in the syntax of one’s language. 34 See e.g. Thomason (1972), pp. 212–15.

42

3

2

The Meaninglessness View1

§1

The Meaninglessness View of Category Mistakes

Having rejected the syntactic approach to category mistakes, I turn to discuss semantic approaches to the phenomenon. The first semantic approach to category mistakes is one that maintains that category mistakes are meaningless (‘the meaninglessness view’). The meaninglessness view has been endorsed throughout the years by a large number of philosophers and linguists, and is certainly the most popular approach to category mistakes, at least amongst those who have written on the subject. Indeed, although the view was no doubt more common a few decades ago than it is these days, contemporary endorsements of the view continue to appear. Thus for example, in a recent article arguing for the distinctness of a material thing and its matter, Kit Fine says: “It is worth emphasizing, in this connection, that these differences lie not merely in the correct but also in the meaningful application of the predicates. A chair can meaningfully be said to be comfortable or uncomfortable, though not the wood from which it is made . . . and one can meaningfully be said to spend a penny or a dollar, though not some metal or paper”.2 In a similar style of argument, Helen Steward argues that processes are distinct from events, because “processes have properties which it would be inappropriate to ascribe to

1 This chapter is a revised version of an article which appeared in Linguistics and Philosophy (Magidor (2009b)). I am grateful the editors of the journal for permission to reprint this material. 2 Fine (2003), pp. 207–8. (His emphases).

43

THE ME ANINGLESSNESS V IE W events, and vice versa. For example . . . the humming of my computer in the process sense can be persistent; but it does not really make sense to think of an event as persistent”.3 Timothy Williamson notes that “To apply the concept of vagueness to anything other than a representation may be treated as a category mistake . . . [This claim] makes ‘The facts are precise’ and ‘The facts are vague’ . . . equally meaningless”.4 More directly, Sauerland and von Stechow describe sentences such as Chomsky’s famous ‘Colourless green ideas sleep furiously’ as ones which “are syntactically well-formed but do not make any sense”,5 and Beall and van Fraassen explain that “[T]here are different ways in which a (declarative) sentence might properly be called ‘meaningless’. Perhaps the best example involves so-called category mistakes”.6 It is not difficult to see why the meaninglessness view has been taken to be particularly appealing. The view provides a simple explanation for why category mistakes are highly infelicitous. Moreover, in so far as one is concerned with the project of explaining what is distinctive about category mistakes (a project which, as noted in the introduction, I will mostly leave aside), the meaninglessness view has a potential account to offer: if category mistakes are meaningless, then (assuming the syntactic approach is rejected) they are arguably distinctive in being the only kind of sentence that is at the same time syntactically well-formed and meaningless. Despite its initial appeal, I argue in this chapter that category mistakes are meaningful (call this ‘the meaningfulness view’), and thus that the 3

Steward (1997), p. 96. (My emphasis). Williamson (1994), pp. 249–50. It is worth noting, though, that Williamson is only committing here to the claim that if a sentence is classified as a category mistake it is meaningless. He is not committed to the claim that the example in question (or indeed any other example) is in fact a category mistake. 5 Sauerland & von Stechow (2001), p. 15413. 6 Beall & van Fraassen (2003), p. 125. For other endorsements of the meaninglessness view see Russell (1908); Ryle (1938); Strawson (1952); Chomsky (1957); Smiley (1960); Benacerraf (1965), pp. 63–7; Drange (1966); Routley (1966); McCawley (1971), p. 294; Searle (1979), p. 93; Lappin (1981); Lyons (1995), p. 136; Shapiro (1997), p. 79; Diamond (2001); Givón (2001), p. 10; Hodges (2001) pp. 7–8; Sorensen (2001) p. 89; Stern (2006), p. 252; Asher (2011), p. 5; Carnie (2011), p. 16; and Ludlow (2011), p. 65. 4

44

THE ME ANINGLESSNESS V IE W OF CATEGORY MISTAK ES meaninglessness view ought to be rejected.7 A couple of preliminary notes are in order. First, when I say that category mistakes are meaningful, I mean that they are literally meaningful. Consider for example the category mistake ‘The theory of relativity eats Newtonian mechanics’. It is clear that speakers can use this sentence pragmatically or figuratively, in order to communicate the content that the theory of relativity is the better theory. Nevertheless, it is clear that the aforementioned content is not the sentence’s literal meaning (if it has one). The availability of such figurative interpretations is thus not in itself sufficient to prove that category mistakes are meaningful in the sense that I am concerned with here.8 Second, my arguments do not rely on any particular theory of meaning. I have intentionally tried to remain as neutral as possible with respect to the preferred theory of meaning, so as to make my conclusion as general as possible.9 Of course, this does not imply that I do not help myself to any assumptions about meaning. My arguments rely both on some intuitive claims concerning meaning (for example that certain pairs of sentences have the same meaning) and on some more theoretical assumptions (for example, certain formulations of the principle of compositionality). But the crucial point is that the claims on which I rely are, by and large, widely accepted, and in particular should seem equally plausible (at least prima facie) to proponents of the meaninglessness view. Of course, if my arguments are successful, proponents of the view might ultimately choose to reject some of these assumptions rather than

7 Note that in addition to the view that category mistakes are in fact meaningless (in English, or other familiar natural languages), there is a slightly weaker version of the meaninglessness view, one which maintains that there are possible languages in which category mistakes are meaningless (see McDaniel (MS) for an endorsement of this weaker claim). For current purposes my interest is in the stronger view, but much of my argument can be used to undermine the weaker view as well. 8 I leave aside the notoriously difficult question of how precisely we ought to draw the distinction between literal and non-literal meanings, and simply follow the widely accepted assumption that there is such a distinction to be drawn. 9 Note that the argument in §2.2 constitutes no departure from this methodology. I do not assume there that type-theoretic semantics is the correct semantic theory. Rather, I argue that type-theoretic semantics are the best resort for a proponent of the meaninglessness view, but that the view fails even if one takes on board this framework.

45

THE ME ANINGLESSNESS V IE W abandoning their view. But this kind of reaction is available in response to any valid argument (‘one person’s modus ponens is another’s modus tollens’), and its availability does not entail that my arguments are in any way question begging. Relatedly, it is worth noting that when I say that my arguments will not assume any specific theory of meaning, this does not commit me to the claim that my conclusion is compatible with any theory of meaning.10 If some theory of meaning appears to accept the background assumptions I rely on but nevertheless maintains that category mistakes are meaningless, then that theory suffers from an internal inconsistency that ought to be resolved. With these comments in place, I turn to defend the claim that category mistakes are meaningful. In §2–5 I provide a series of arguments against the meaninglessness view, while in §6, I briefly discuss some of the positive motivations that count in favour of the view and argue that they are unconvincing.11

§2

The Argument(s) from Compositionality

§2.1 Atomic category mistakes Perhaps the first argument that springs to mind in favour of the meaningfulness view is the argument from compositionality. Speakers of natural languages have the capacity to understand indefinitely many new sentences. This suggests that meaning must be compositional: the meaning of a sentence is composed from the meanings of its parts, and grasping the meaning of the parts enables speakers to grasp the meaning of a sentence. Moreover, some particular principles which govern compositionality of meaning and understanding suggest themselves. Take for example simple subject-predicate sentences of the form ‘Fa’ (where ‘a’ is a singular term, ‘F’ is a predicate, and ‘Fa’ is the grammatically correct sentence that has ‘a’ as its subject and ‘F’ as its predicate). Putting

10

See §6.2 below for more on this issue. For another recent defence of the meaningfulness view see Camp (2004). (The considerations in Camp’s paper are closest to those raised in §5 and §6 below). 11

46

THE ARGUMENT(S) FROM COMPOSITIONALIT Y category mistakes to the side for a moment, the following principle seems appealing: (Principle 1) If S is a generally competent speaker of a language L and S understands the terms ‘a’ and ‘F’ of L, then S understands the sentence ‘Fa’.

Principle 1 can be taken either as a quasi-empirical claim about which subjects understand sentences such as ‘Fa’, or as a constitutive condition on what it is to understand terms such as ‘F’ and ‘a’.12 Either way, the principle seems correct: assuming one is a generally competent speaker of English and understands the phrases ‘is red’ and ‘the parrot’, one would thereby understand the sentence ‘The parrot is red’. But if Principle 1 is correct, category mistakes must be meaningful: competent English speakers understand the phrases ‘the number two’ and ‘is green’, so according to Principle 1, they thereby understand the category mistake ‘The number two is green’. Since understanding a sentence requires grasping its meaning, this entails that the sentence must have a meaning to be grasped, i.e. it must be meaningful. Principle 1, then, entails very straightforwardly that category mistakes are meaningful. One might worry, though, that the principle begs the question against the meaninglessness view. Proponents of the meaninglessness view might insist on adopting instead the following, more qualified, principle: (Principle 2) If S is a generally competent speaker of a language L and S understands the terms ‘a’ and ‘F’ of L, then S understands the sentence ‘Fa’, if this sentence is a meaningful sentence of L.

According to Principle 2, understanding both ‘F’ and ‘a’ is sufficient for understanding ‘Fa’, providing the sentence can be understood at all. This still leaves open the possibility that sentences such as ‘The number two is green’ cannot be understood, because they are meaningless. Underlying Principles 1 and 2 are two somewhat different conceptions of 12 The latter approach is taken by Evans when he introduces the generality constraint (see Evans (1982), pp. 100–5). Interestingly, Evans qualifies the generality constraint so that it does not apply to category mistakes (see footnote 17, p. 101).

47

THE ME ANINGLESSNESS V IE W compositionality: according to the stronger conception underlying Principle 1, meaningful expressions combined in a syntactically appropriate way always compose a meaning.13 According to the weaker conception underlying Principle 2, meaningful complex expressions are composed out of the meanings of their parts, but no commitment is made into which simpler expressions compose meaningful complex expressions. It is not unreasonable for proponents of the meaninglessness view to shift to such weaker principles of compositionality, but it is a reaction that carries with it a non-trivial challenge: proponents of the meaninglessness view need to offer a semantic framework which on the one hand shows how the meaning of ‘acceptable’ sentences can be successfully derived in a compositional manner, but on the other hand explains what blocks such a derivation in the case of category mistakes. In the next sub-section, I describe what I take to be the most promising way for proponents of the view to address this challenge, and argue that this attempt ultimately fails.

§2.2 Type-theoretic semantics to the rescue? Proponents of the meaninglessness view are presented with the challenge of offering an adequate semantic framework that can support their view, and there appears to be one popular semantic framework that is best suited for addressing this challenge: functional type-theoretic semantics such as Montague Grammar and its variants. On this framework the semantic-values of predicates are taken to be functions from some type of entity to truth-values. Working within this framework, a proponent of the meaninglessness view could put forward the following proposal. (Proposal 1): Take the semantic-value of a predicate ‘F’ to be function the domain of which includes an object c, only if ‘Fc’ is not a category mistake. For example, take the semantic-value ‘green’ to be a function from concrete

13 As noted in Chapter 1, something like this strong form of compositionality is assumed by Montague (1970), in his treatment of compositionality as a homomorphism between the syntax and semantics of a language.

48

THE ARGUMENT(S) FROM COMPOSITIONALIT Y objects to truth-values, of ‘is eating’ to be a function from animate objects to truth-values, and so forth.

Given this proposal, it seems that on the one hand one can explain the mechanism that enables compositionality in the case of benign sentences such as ‘My chair is green’: the mechanism is simply that of functional application. On the other hand, one can also explain what “goes wrong” in the case of category mistakes such as ‘Two is green’: since the number two is an abstract object, it is not in the domain of the function denoted by ‘green’ and functional application fails in this case. The challenge I presented above thus seems to be met. The problem with this suggestion is that Proposal 1 forces us into an implausible semantic theory. Proposal 1 might seem plausible when we restrict our view to a very limited fragment of language, but once we try to generalize the proposal for larger fragments of language, serious problems arise. To see why, it will be helpful to fix on one type-theoretic semantic theory. Let us consider the most prominent theory of this sort, namely Montague Grammar.14 In Montague Grammar proper names do not directly denote individuals, but rather they denote the functional analogue of second-order properties, namely functions from first order-properties to truth-values.15 The semantic-value of ‘Two’ is thus taken to be λX.X(two) and the semantic-value of ‘prime’ is taken to be λx.prime(x), where ‘x’ ranges over individuals (i.e. it is a variable of type e) and ‘X’ ranges over functions from individuals to truth-values (i.e. it is a variable of type ). The meaning of sentences such as ‘Two is prime’ is then yielded by applying the function that ‘Two’ denotes to the function that ‘prime’ denotes so that we get λX.X(two)(λx.prime(x)), which reduces to the desired prime(two).

14

Montague (1974). For simplicity, I will ignore whenever possible the intensional aspects of Montague Grammar. I return to the question of whether moving to other versions of type-theoretic semantics might help the proponent of the meaninglessness view below. 15 This treatment of proper names is motivated by the syntactic similarities between proper names and quantified terms such as ‘Every man’ or ‘A woman’. Since the latter are given semantic-values of type , so are the former.

49

THE ME ANINGLESSNESS V IE W Now consider Proposal 1. Since ‘x is prime’ is a category mistake if and only if x is not a number (or some such similar restriction), then according to Proposal 1 the function that ‘prime’ denotes is ‘λxn.prime(xn)’, where xn is a variable of a type (call it ‘n’) which contains all and only numbers. Thus ‘prime’ denotes a function of type , i.e. a function from numbers to truth-values. But since the semantic-value of ‘Two’ is a function that only takes as arguments functions of type (i.e. functions whose domain includes all individuals), we now face the absurd situation that functional application breaks down in the case of ‘Two is prime’. One might propose that in the light of this problem we ought to revise the semantic-value of ‘Two’ so as to belong to the type , i.e. so that its domain consists of functions from numbers to truth-values. This, however, will not do. The reason is that some predicates which are appropriately applicable to ‘Two’ will have to receive a semantic-value whose domain includes both numbers and other kinds of objects (examples of such predicates are ‘is interesting’, ‘is an object’, and ‘is thought about’). But on the revised proposal, we cannot combine such functions with the semantic-value of ‘Two’ and thus we reach another absurd conclusion, namely that ‘Two is interesting’ or ‘Two is thought about’ are meaningless. Several responses to this argument could be suggested by a proponent of the meaninglessness view. First, she might point out that the Montague Grammar view of singular terms as denoting second-order properties is not universally accepted—other versions of type-theoretic semantics treat singular terms as denoting entities of type e, i.e. individuals.16 This point will not help, though, because the above argument can be generalized to other expressions, ones that receive less controversial semantic treatments. Consider for example the word ‘very’. It is natural to think that ‘very’ receives an interpretation of type , i.e. it denotes a function from first-order properties to first-order properties. For example, the expression ‘very green’ involves applying the function that ‘very’ denotes to the property that ‘green’ denotes, which yields the property of 16

The latter approach is taken, for example, by Heim & Kratzer (1998).

50

THE ARGUMENT(S) FROM COMPOSITIONALIT Y being very green. But according to Proposal 1, ‘green’ does not denote a function from individuals to truth-values (i.e. an entity of type ) but rather a function from (say) concrete objects to truth-values (call this an entity of type ‘’). It follows that the function ‘green’ denotes is not in the domain of the function ‘very’ denotes which, by the lights of the proposed view, entails that ‘very green’ is meaningless. One could claim that ‘very’ is ambiguous, receiving one meaning of type and another of type . But as soon as we consider the fact that Proposal 1 is formulated so as to block all subject-predicate category mistakes, it becomes apparent that this ambiguity suggestion is unfeasible. For according to the intended generalization, there will be some firstorder predicates that denote functions from individuals to truth-values (e.g. ‘is interesting’), some from concrete objects to individuals (e.g. ‘is green’), some from humans to individuals (e.g. ‘likes dancing’), some from numbers to individuals (e.g. ‘is larger than 5’), and so forth. This would entail that ‘very’ is a massively ambiguous word, which does not seem plausible. (And of course parallel arguments would show that many other words would also have to be massively ambiguous). A second response might be to claim that words such as ‘two’ have semantic-values which belong to a special kind of type, which contains functions with a gerrymandered domain, one that includes functions from individuals to truth-values (such as the function denoted by ‘is interesting’), functions from numbers to truth-values (such as the function denoted by ‘is prime’), and so forth. On this view, the semantic-value of ‘two’ is the function λX.X(two), where X is a variable which belongs to the special gerrymandered type described above. Similarly, one might argue that the semantic-value of ‘very’ belongs to a special kind of type containing functions with a gerrymandered domain which includes, among other things, entities of type and .17 The problem with 17 Note, though, that ‘very’ cannot receive any argument which is of type , with x being some sub-type of e—as can be attested by the infelicity of ‘very prime’. And things might be even more complex than this. Suppose, for example, that one maintains that there is a special sense of ‘large’ which applies only to numbers. (This might be motivated by the need to explain the infelicity of ‘The number fifteen is larger than the table’ or ‘Two

51

THE ME ANINGLESSNESS V IE W this proposal is that it constitutes a radical divergence from the typetheoretic semantics under consideration. A large part of the attraction of type-theoretic semantics is its simplicity. One starts with a very small number of basic types (often only two or three), and derives all other types by the single rule that if a and b are types then is a type (containing functions from entities of type a to entities of type b). Moreover, there is a tight link between the syntactic category of an expression and the type of its semantic-value. This simplicity is completely lost on the suggested response. In particular, note that this suggestion will more or less require that every single word is associated with a unique type. Consider for example the type of the semantic-value of ‘two’. The domain of the functions it contains should clearly include functions from prime numbers to truth-values, but also functions from even numbers to truthvalues. But then ‘two’ will be the only (or almost only) word which receives a semantic-value of this type, because the number two is the only object which is both prime and even. It therefore turns out that the variable X above will have to belong to a type which only plays a role in the interpretation of the word ‘two’. But this would be a serious blow to the universal character and simplicity of type-theoretic semantics. A final response, which strikes me as much more promising than the previous one, is to retain the original type-theoretic idea that expressions of the same syntactic category receive semantic-values of the same type, but slightly modify the definition of types. On the revised definition, if a and b are types, then [a,b] is a type denoting the collection of all partial functions from entities of type a to entities of type b. According to the revised proposal, the semantic values of predicates are just as in Proposal 1, but their semantic type is [e,t]. Similarly, the semantic-value of ‘two’ is a function of type [[e,t],t], the semantic-value of ‘very’ is a function of type [[e,t], [e,t]], and so forth. On this suggestion, merely knowing the types of the semantic-values of ‘two’ and of ‘green’ is not sufficient for determining

is large, and so is the table’.) In that case, ‘very’ would be applicable to some expressions of type ‘’ (‘large’, in the relevant sense), but not to all expressions of this type (e.g. not to ‘prime’).

52

THE ARGUMENT(S) FROM COMPOSITIONALIT Y whether or not an attempted interpretation of ‘Two is green’ will involve some breakdown in functional application. However, knowing the particular semantic-values that ‘two’ and ‘green’ end up receiving is sufficient in order to determine that the partial function that ‘green’ denotes is not in the domain of the partial function that ‘two’ denotes. For the moment, I will not argue that this proposal is incorrect.18 Rather, I would like to argue that even if it were correct, the best way to interpret it is as saying that category mistakes such as ‘Two is green’ are truth-valueless rather than meaningless. Compare this with Heim and Kratzer’s discussion of empty definite descriptions.19 According to Heim and Kratzer, the semantic-value of the determiner ‘the’ is a function of type , although strictly speaking the semantic-value of ‘the’ is only a partial function from entities of type to truth-values.20 A function f of type will only be in the domain of the semantic-value of ‘the’ if it yields the value true for one and only one individual. Thus for example, the function denoted by the predicate ‘queen of France’ is not in the domain of the function denoted by ‘the’, and thus the expression ‘the queen of France’ involves a failure of functional application. But clearly, the expression ‘the queen of France’ is meaningful and suffers at worst from denotation failure (not from meaningfulness failure). Consequently the sentence ‘The queen of France is rich’ suffers at worst from failure to express a proposition or to possess a truth-value, but is clearly nevertheless meaningful.21 One might argue that this case is different than that of ‘The number two is green’ because while in the case of the category mistake one only needs to know the meanings of the words in order to find out that there is a failure of functional application, in the case of ‘the queen of France’ one 18

Although I think it is incorrect, and argue for that in Chapter 4. (See especially Chapter 4, §2). 19 Heim & Kratzer (1998), pp. 81–2. 20 Thus using my above notation the semantic-value of ‘the’ is of type [[e,t],e] or at least [,e]. 21 The claim that (at least some) sentences containing empty definite descriptions are meaningful is rarely contested. For example, it is the one clear point of agreement between Russell and Strawson on this issue (see Russell (1905), p. 484 and Strawson (1950), p. 321).

53

THE ME ANINGLESSNESS V IE W needs to know further contingent or empirical facts about the world. But as Heim and Kratzer rightly argue, this is a bad criterion for meaningfulness. It is implausible to think that the sentence ‘The greatest prime number is odd’ is any less meaningful than ‘The queen of France is rich’, even though one does not need to know any empirical or contingent facts to discover that the function which (according to the Heim and Kratzer’s theory) ‘greatest prime number’ denotes is not in the domain of the function that ‘the’ denotes. As Heim and Kratzer put it: “In the case of an uninterpretable structure, information about the type of each subtree is sufficient to decide that the structure receives no denotation. To detect presupposition failure, by contrast, we must know more about the denotations of certain subtrees than the mere semantic types”.22 As further support for this claim, consider the sentence ‘That is green’. It is plausible to think that as uttered in a context where ‘that’ refers to the number two, ‘that’ will receive the same semantic-value that ‘two’ receives, and hence an attempted interpretation of ‘That is green’ will involve the same breakdown in functional application that ‘Two is green’ is said to involve. But it would be absurd to suppose that the possibility of such breakdowns entails that ‘That is green’ is a meaningless sentence. At worst, in the relevant contexts, ‘That is green’ fails to express a proposition or to possess a truth-value. A similar point holds when we consider sentences such as ‘The thing I am thinking of is green’, as evaluated relative to a possible world in which the thing I am thinking of is the number two. At least according to some theories of definite descriptions (roughly, ones of a Fregean orientation) the definite article ‘the’ as evaluated in a possible world will denote a function from properties to individuals, or rather on its Montague grammar version—a function from properties to second-order properties.23 Thus as evaluated relative to a possible world w* in which

22

Heim & Kratzer (1998), p. 82. In general, ‘The’ will denote the function λw.λY.λX.Xw(ιx.Yw(x)), with Y and X being variable of type , and thus as evaluated in a world w*, ‘The’ will denote the function λY.λX.Xw*(ιx.Yw*(x)). 23

54

THE ARGUMENT(S) FROM COMPOSITIONALIT Y the thing I am thinking of is the number two, ‘The thing I am thinking of’ will denote the function λX.Xw*(two), and the interpretation of ‘The thing I am thinking of is green’ will involve exactly the same breakdown in functional application as is involved with ‘Two is green’. But it would be absurd to conclude from this that ‘The thing I am thinking of is green’ is meaningless. Again, it does not help to point out that, as opposed to the case of ‘Two is green’, the semantic-value of ‘The thing I am thinking of is green’ is such that as evaluated relative to some worlds, no breakdown of functional application occurs. For suppose I am actually thinking of the number two, then ‘The thing I am actually thinking of is green’ will involve (on the proposed view) the relevant kind of breakdown in functional application as evaluated relative to every possible world, but is nevertheless meaningful.24 Nor will it help to point out that one needs to know some empirical fact to infer the alleged failures in functional application. For example, assuming that ‘x is prime’ is a category mistake when ‘x’ denotes a non-natural number, the sentence ‘The ratio between the circumference of a circle and its diameter is prime’ will involve (on the proposed view) the relevant kind of breakdown in functional application as evaluated relative to any world. But although one would not need to know any empirical facts to discover this, the sentence is nonetheless meaningful. I conclude that if the final proposal is accepted, category mistakes are at worst truth-valueless, but not meaningless. In Chapter 4, I argue that category mistakes are not truth-valueless, but my aim at the moment is the more modest one of arguing that they are meaningful. In particular, I have argued that the attempt to use type-theoretic semantics to explain how category mistakes can be meaningless fails. That is, on most versions the attempt leads to absurd conclusions and on its most plausible version it at most entails that category mistakes are truth-valueless, not 24 It also does not help to note that ‘The thing I am thinking of is actually green’ will involve no breakdown as uttered in some different contexts of utterance. If ‘w0’ denotes the actual world then ‘The thing I am thinking of in w0 is green’ will involve the relevant kind of breakdown relative to any context of utterance and context of evaluation pair (fixing the time and the speaker).

55

THE ME ANINGLESSNESS V IE W that they are meaningless. The challenge to the meaninglessness view thus remains unanswered.

§2.3 Conjunctions and quantifier phrases The argument from compositionality is not restricted to the case of predicative expressions and their arguments. Take for example conjunctions. The following principle seems highly plausible: (Principle 3) If ‘p’ and ‘q’ are meaningful declarative sentences, then ‘p and q’ is a meaningful sentence.

But now consider the following example: (1) That is a number and that is green.

Assuming the two occurrences of ‘that’ are co-referential, (1) seems to be a category mistake. After all, an utterance of (1) where it is clear that the two occurrences of ‘that’ are co-referential, is infelicitous in just the same way as an utterance of ‘That is a green number’. But it is obvious that each of the conjuncts ‘That is a number’ and ‘That is green’ is meaningful. It follows by Principle 3 that the conjunction is meaningful, and hence that category mistakes are meaningful. Similar problems arise when we consider category mistakes containing quantifier phrases. Consider the following example: (2) Some number is green.

This sentence seems to be a category mistake.25 The following considerations suggest that it is also meaningful. Assume for simplicity, that the logical form of the sentence is ‘∃x(number(x)∧green(x))’.26 First, note that ‘green(x)’ and

25 Note that this judgment is explicitly endorsed by Lappin (1981), pp. 69–71, who is a defender of the meaninglessness view. 26 The argument can be easily generalized for other proposed logical forms. For example, if the logical form of the sentence is ∃x(λx.(number(x)∧green(x)), all we need is to replace Principle 4 with the corresponding principle for lambda terms; If the logical form of the sentence is that of a binary second-order quantifier, then the simplest way to generalize the argument is to consider the category mistake ‘Something is a green number’.

56

THE ARGUMENT(S) FROM COMPOSITIONALIT Y ‘number(x)’ are both meaningful. After all, they both occur as constituents in some meaningful sentences (e.g. ‘Something is green’ and ‘Something is a number’), and since the meaning of a meaningful sentence is composed out of the meaning of its constituents, these constituents must be meaningful as well. Next, it follows from Principle 3, that ‘number(x)∧green(x)’ is meaningful.27 Finally, we can conclude that ‘∃x(number(x)∧green(x))’ is meaningful, using another principle of compositionality: (Principle 4) If ϕ(x) is meaningful, then ∃xϕ(x) is meaningful.

This principle seems highly plausible. After all, we know precisely how to derive the truth-condition for ‘∃xϕ(x)’ from those of ‘φ(x)’: ‘∃xϕ(x)’ is true just in case there is an assignment function g, relative to which ‘φ(x)’ is true.28 Simple and plausible principles of compositionality thus entail that category mistakes such as (1) and (2) must be meaningful. 29 27 One might worry that Principle 3 requires the two conjuncts to be declarative sentences, and open formulas are not declarative sentences. However, it should be noted that the only reason I restricted Principle 3 to declarative sentences was to ensure that ‘p and q’ was well-formed. One can therefore replace Principle 3 with the principle that whenever ‘p’ and ‘q’ are meaningful, and ‘p and q’ is well-formed, ‘p and q’ is meaningful. Since ‘green(x)∧prime(x)’ is well-formed, this suffices to show that it is meaningful. 28 Indeed, considering these truth-conditions suggest that ‘∃x(number(x)∧green(x))’ is not only meaningful, but false. Even if one conceded that relative to every assignment ‘number(x)∧green(x)’ is (meaningful but) truth-valueless, the above truth-condition for the existential statement are most naturally matched with a falsity-condition according to which ‘∃xφ(x)’ is false just in case there is no assignment g, relative to which ‘φ(x)’ is true. But then if ‘φ(x)’ is truth-valueless relative to every assignment, then there is no assignment relative to which it is true, and the existential statement is thus false. One could, of course, suggest alternative falsity conditions such as those proposed by trivalent logics. But note that on a Weak Kleene interpretation of the quantifier, one would get the unappealing result that benign sentences such as ‘Something is green’ must be truth-valueless (because the quantifier ranges over some values x, for which ‘x is green’ is a category mistake), and a Strong Kleene interpretation is harder to motivate philosophically in this context. Thus the above argument can also be used to support the claim that category mistakes are truthvalued, not merely meaningful. (For brevity, I do not repeat this point in Chapter 4). 29 The arguments for compositionality of conjunction and quantifier phrases are particularly important if one adopts a neo-Davidsonian view of the semantics of verbs. On this view, even atomic category mistakes such as ‘The rock is thinking’ will receive an analysis along the following lines: ‘∃e(thinking(e)∧agent(e) = the rock). But each of the conjuncts here is clearly meaningful which, following the above argument, suggests that the whole condition must be meaningful as well.

57

THE ME ANINGLESSNESS V IE W

§3

The Argument from Synonymy

Two sentences are said to be synonymous if and only if they have the same meaning. This suggests that if two sentences are synonymous then they must be meaningful: if the two sentences have the same meaning, then each of them has a meaning. Now at least on the face of it, the English sentence ‘Two is green’ is synonymous with the French sentence ‘Deux est vert’. This suggests that the two sentences—both category mistakes—are meaningful. Of course, a proponent of the meaninglessness view can insist that this intuition is misleading, and the two sentences are not really synonymous. But denying a plausible intuition is a price for the view to pay. Moreover, it is worth noting that it is not entirely straightforward to explain away this intuition. For example, one might try to argue that two sentences merely seem synonymous because they are word-to-word synonymous (that is, there is a one-to-one correspondence which preserves synonymy between the words of the English sentence and those of the French sentence). This, however, will not do: the Hebrew translation of ‘Two is green’ is ‘Shtaim yarok’ which contains only two words (Hebrew does not require a copula in such constructions), and is thus not word-to-word synonymous with the English sentence. Still, the Hebrew sentence seems entirely synonymous with the English sentence (no less so than the above French sentence). It should be acknowledged that in literary contexts one does encounter a notion of ‘adequate translation’ which does not assume preservation of meaning: there can be better and worse translations of Jabberwocky despite the fact that it contains meaningless words. Still the case of category mistakes seems very different: it takes no great literary ingenuity to translate ‘Two is green’ to other languages and it is seems perfectly clear which translations are correct and which are incorrect. A proponent of the meaninglessness view can try to come up with a sophisticated explanation of this phenomenon which does not entail that category mistakes are meaningful, but the simplest explanation for the apparent synonymy

58

THE ARGUMENT FROM PROPOSITIONAL AT TITUDE ASCR IPTIONS of certain category mistakes in different languages is that they are in fact synonymous.30

§4

The Argument from Propositional Attitude Ascriptions

Category mistakes can be embedded in propositional attitude ascriptions.31 Consider the following: (3) John said that the theory of relativity is eating breakfast. (4) Jane believes that the number two is green. (5) George dreamt that his toothbrush was pregnant.

How do such embeddings support the meaningfulness view? Consider the following four claims: (M): (3)–(5) are meaningful sentences. (T): For each of (3)–(5), there is some possible circumstance in which it is true. (M-entailment): If M is true, then the category mistakes embedded in (3)–(5) are meaningful. (T-entailment): If T is true, then the category mistakes embedded in (3)–(5) are meaningful.

30 Another possibility is that the proponent of the meaninglessness view acknowledge that ‘Two is green’ is synonymous with ‘Deux est vert’, but claim that this is so simply because both sentences are meaningless, and thus they both have the same “null” meaning. It is not clear that a proponent of the view would want to opt for this proposal, because plausibly they would maintain that ascribing synonymy to meaningless sentences is itself a category mistake, and hence meaningless. And at any rate, the proposal does not work because given the meaninglessness view, it would entail that ‘Deux est vert’ is synonymous with every English category mistake. But this is the wrong result: ‘Deux est vert’ is not synonymous with ‘The theory of relativity is eating breakfast’. 31 Cf. McCawley (1970), p. 168 who suggests such embedding data supports the thought that “it appears incorrect to regard many so-called “selectional violations” as not corresponding to possible messages”. Note, though, that elsewhere McCawley expresses sympathy to the meaninglessness view (e.g. McCawley (1971), p. 294) and that most discussions of similar embedding data are raised in the literature in the context of attacking the syntactic approach rather than the meaninglessness view (see Chapter 2, §4).

59

THE ME ANINGLESSNESS V IE W Clearly, either M and M-entailment or T and T-entailment are sufficient to establish the meaningfulness view. Since a sentence can only be true if it is meaningful, T entails M and M-entailment entails T-entailment. One can therefore either choose to defend M, which is the weaker of the first two claims, and then establish the stronger of the latter two claims, namely M-entailment. Alternatively, one can attempt to defend the stronger claim T, leaving us with the need to establish only the weaker T-entailment. In what follows I will argue that all four claims are highly plausible, leaving either course of argument open. I shall not say much to defend M. It seems to me clear that we can understand (3)–(5), and so they are meaningful sentences. A die-hard proponent of the meaninglessness view might insist on denying this claim, but I am guessing that few others would.32 How about T? It is easy to imagine a situation which would make (3) true: John simply utters the sentence ‘The theory of relativity is eating breakfast’. Now, perhaps that is not enough. After all, Jill can utter the sounds ‘blablabla’, and yet it is meaningless (and therefore it cannot be true) to say: ‘Jill said that blablabla’. But there are several reasons to think that the case of John’s utterance is quite different than that of Jill’s. We may adequately report John’s utterance. Moreover, we may do so using different words from the ones John originally used. For example, even if John actually uttered the French sentence ‘La théorie de la relativité est en train de prendre le petit déjeuner’, we can adequately report him using (3). We can also claim that John said something very odd (note the implication that he said something), ask John to explain why he believes what he said, and so forth. What about (4)? One might think that this sentence cannot be true for the following reason: if one does not possess the concepts of ‘two’ or of ‘green’ then one cannot believe that the number two is green. On the other hand, the reasoning goes, anyone who does possess the concepts of

32 Note that proponents of the meaninglessness view are at least likely to accept the meaningfulness of sentences such as ‘It is nonsense to say that the number two is green’, and this concession is sufficient for my purposes in this section.

60

THE ARGUMENT FROM PROPOSITIONAL AT TITUDE ASCR IPTIONS ‘two’ and of ‘green’ cannot believe that the number two is green, for it is a necessary condition on possessing these concepts that one would not believe the obviously absurd claim that the number two is green. So either way, the argument goes, it is impossible for anyone to believe that the number two is green.33 The problem with this argument is with the claim that it is a necessary condition on possessing the concepts of ‘two’ and ‘green’ that one does not believe that two is green. I think this claim is false. Consider the following scenario: Jane is a philosopher. She recently developed a new theory in the philosophy of mathematics according to which numbers are coloured, and the colour of the number two is green. To flesh out the example a bit more, suppose Jane holds some naturalist position according to which the number two is the set of all pairs of physical objects in the world. In addition, Jane holds that if most such pairs have a certain colour, then the set—and therefore the corresponding number—have this colour. (Compare this to an actual philosophical position, one held by Penelope Maddy: a set of physical object has a spatial location and it is located wherever its members are located).34 Thus, if it happens to be the case that most pairs of physical objects are green then, according to Jane’s theory, this makes the number two green. Suppose that following some empirical investigation Jane concludes that it is in fact the case that most pairs of objects are green and so, following her theory, she forms the belief that the number two is green (or, to not yet beg the question, at least comes to sincerely assent to the sentence ‘The number two is green’). But Jane, we may suppose, also knows a lot of mathematics, and clearly possesses the concept ‘two’. Also, we may suppose, Jane generally does a perfectly good job of telling which things are green and which are not. It thus seems wrong to say that Jane does not possess the concept ‘green’.

33

A thought along these lines is suggested by Asher (2011), when he argues that “according to most people’s intuitions, a competent speaker could entertain or even believe that tigers are robots; . . . it is much harder to accept the possibility, or even make sense of, a competent speaker’s believing or even entertaining that tigers are literally financial institutions” (p. 5). See also Asher (2011), p. 49 and p. 118. 34 See Maddy (1980).

61

THE ME ANINGLESSNESS V IE W I am not denying that usually, if someone were to seriously utter the sentence ‘The number two is green’, it would be sensible to conclude that they do not understand the meaning of either ‘two’ or ‘green’. (That is, either they do not know the meaning of the English words, or they do not even possess the relevant concepts). But although this conclusion is usually correct, it is not always correct. In particular, we should not conclude that in the above scenario, Jane does not possess the concepts of ‘two’ and ‘green’, and hence we have no reason to reject the claim that Jane believes that the number two is green, and that (4) is sometimes true.35,36 As for (5), it is an empirical fact that people often sincerely report having had dreams which involve category mistakes. The question of whether this constitutes conclusive evidence that these reports are in fact correct is a complex issue in the philosophy of mind which I cannot address here.

35 One might try to argue that although Jane possesses ordinary concepts of ‘green’ and ‘two’, when philosophizing she uses ‘green’ and ‘two’ with a slightly different meaning from the ordinary. However, we can further stipulate that Jane explicitly insists that ‘two’ and ‘green’ are used in her theory in the ordinary sense, and moreover that she is also happy to endorse sentences such as ‘My chair is green and so is the number two’ (note that the anaphoric reference ensures that the same property is ascribed to the number two and to the chair). 36 One could object that it is easier to construct such an argument for the particular category mistake I have chosen (namely, ‘the number two is green’) than for others (e.g. ‘the theory of relativity is eating breakfast’). I believe that with enough ingenuity similar situations can be constructed for almost any category mistake. To briefly point out a few further examples of cases where someone might sincerely believe a category mistake: First, some people who suffer from synaesthesia perceive certain sounds as always accompanied by a perception of a certain colour. It would probably be quite natural for such persons to believe that ‘This sound is green’ is literally true. Second, Lakoff (1971), p. 332, reports that a common belief among Papagos is that events have minds, so it might well be natural for such persons to believe that ‘My birthday is angry’ is literally true. Third, such cases are not even restricted to simple subject-predicate category mistakes. Armstrong et al. (1983) report an experiment in which subjects were given six even numbers and were asked to judge, on a scale of one through seven, how good an example each of them was of an even number. Surprisingly, not a single one of the subjects complained that the task was unintelligible, and the results show that on average, most subjects tend to think that 6 is much better example of an even number than 42, and that 42 is a better example of an even number than 34. There is of course some room for debate on how to interpret these results, but it does not seem out of the question to describe the participants in this experiment as sincerely believing that 6 is more even than 42, or that 6 is a very even number.

62

THE ARGUMENT FROM PROPOSITIONAL AT TITUDE ASCR IPTIONS I leave matters by noting that there are at least strong prima facie reasons to accept there are situations where ascriptions such as (5) are true. So far, I have defended M and T. I next wish to argue that under any standard theory of propositional attitude ascriptions T-entailment holds, and with some additional assumptions so does M-entailment. There are generally three types of views regarding the semantics of propositional attitudes: the propositional view according to which propositional attitudes are relations between agents and propositions, the Fregean view, according to which propositional attitudes are relations between individuals and meanings, and the sentential view according to which propositional attitudes are relations between individuals and sentences or utterances.37 How does each of these views fare with respect to T-entailment and M-entailment? First, consider the propositional view. Whatever we take propositions to be, the view entails that if ‘ϕ’ denotes a propositional attitude and if ‘S ϕs that p’ is true then the phrase ‘that p’ successfully denotes a proposition. Now assume that T holds, i.e. assume that there are possible situations in which the propositional attitude ascriptions are true. Relative to such situations, the category mistakes embedded in the ‘that’-clauses must successfully express a proposition. But since meaningless sentences cannot express propositions, it follows that the embedded category mistakes are meaningful. So T-entailment holds. What if we reject T and adopt only the weaker claim M, i.e. we claim that while ascriptions such as (3)–(5) are never true they are nonetheless meaningful? Of course on some versions of this proposal, the embedded category mistakes are trivially meaningful (e.g. if the embedded category mistakes express propositions albeit ones that the agent fails to stand in the relevant attitude towards, or if the embedded category mistakes are taken to be meaningful though fail to express a proposition at all). The only potential problem arises if it is possible for the embedded category 37 For typical defences of these three views see, respectively, Barwise & Perry (1981), Frege (1952), and Davidson (2001a). I am assuming here that we construe Fregean senses to be meanings. Note also that the propositional view includes positions according to which the ascriptions describe relations between an agent, a proposition, and some third factor (e.g. a mode of presentation).

63

THE ME ANINGLESSNESS V IE W mistakes to be meaningless, while the ascription sentences are meaningful. But this possibility can be ruled out if we accept the following two highly plausible assumptions: that the meaning of a sentence is composed of the meanings of its constituents, and that (given the propositional view) sentences (3)–(5) have the embedded category mistakes as constituents. If this is granted, then the embedded category mistakes cannot be meaningless while (3)–(5) are meaningful. So the stronger claim M-entailment also holds. Next, consider the Fregean view. According to this view, if ‘S ϕs that p’ is true then S stands in the relation ϕ to the meaning of ‘p’. But for this to be the case ‘p’ must have a meaning, i.e. it must be meaningful. So, if it is possible for (3)–(5) to be true, the embedded category mistakes must be meaningful, and so T-entailment holds. What if we assume only the weaker claim M? Again, our conclusion follows trivially if we accept a version of this scenario where the embedded category mistakes are meaningful (though the agent fails to stand in the relevant attitude towards the relevant meaning). So the issue comes down to whether there is a feasible version of the Fregean view where the embedded category mistakes are meaningless, but the propositional attitudes containing them are nevertheless meaningful. Ideally, we would like to rule out this possibility by appealing to the compositionality of meaning (as we did for the analogous version of the propositional view above). However, this case is a little trickier because according to the Fregean view we are currently considering, the embedded category mistakes do not contribute to the meaning of the propositional attitude ascriptions their regular meaning, but rather a second-order meaning. So in order to argue from the claim that the propositional attitude ascriptions are meaningful to the desired conclusion that the embedded category-mistakes have a first-order meaning, we must also adopt the following principle: an expression can only have a second-order meaning if it has a first-order meaning. It does seem, though, that from the perspective of the Fregean framework we are considering this principle should indeed be accepted. Plausibly, the theory should take second-order senses as computable from the first-order senses: otherwise speakers would need 64

THE ARGUMENT FROM PROPOSITIONAL AT TITUDE ASCR IPTIONS to learn two primitive senses for each embedded sentence, and this point is strengthened if one accepts that to accommodate sentences with multiple embeddings, one will need a distinct nth-order sense for any natural number n.38 So it seems that ultimately we can rule out the possibility that the embedded category mistakes are meaningless while the (3)–(5) are meaningful, and M-entailment holds. Finally, consider the sentential view. Admittedly, this view is the best avenue for the proponent of the meaninglessness view: after all, on the sentential view the semantic function of the embedded sentences is to refer to themselves and the claim that the embedded category mistakes successfully denote sentences (or utterances) does not seem in any way at odds with the claim that these sentence are meaningless. Nevertheless, I think that at least T-entailment is plausible even on the sentential view. In whichever way the details of the view are filled-out, the view must somehow account for the fact that we can truly ascribe a propositional attitude to an individual even if that individual uses different words to express their own attitude. For example, we can correctly claim that Galileo said that the earth moves, even if Galileo never uttered the English sentence ‘The earth moves’. Consequently, such views of propositional attitudes must assume some kind of relation between the utterances of the ascriber and the utterances or thoughts of the ascribee: a relation such as synonymy or having the same content or, following Davidson, of ‘being such as to make the ascriber and ascribee same-sayers’. But suppose that my utterance of ‘The theory of relativity is eating breakfast’ in (3) merely has the same meaning as or the same content as (rather than being an exact repetition of ) John’s original utterance. If my utterance and John’s have the same meaning or the same content, it suggests that both utterances are meaningful.39As with the argument from synonymy, 38

This claim is defended Dummett (1981) and contested in Parsons (1981). For a more recent defence of the claim which addresses Parson’s attack, see Boisvert & Lubbers (2003). 39 Cf. §3 of this chapter. Note that this argument is not in tension with Davidson’s defence of the sentential view, which wishes to avoid reference to intentional entities such as propositions or meanings. Although Davidson does not accept that there is strictly speaking a meaning or content that John and my utterances share, he nevertheless insists

65

THE ME ANINGLESSNESS V IE W the proponent of the meaninglessness view can insist that the relation that needs to hold between John’s original utterance and my report is not one of having the same meaning or content, but rather some other relation that does not require our utterances to be meaningful. But here as above, this places an explanatory burden on the proponent of the meaninglessness view, one which is far from trivial to meet. Finally, it is worth noting that the argument just presented does depend on the stronger claim T, rather than on M: if one can never truly report the contents of what John said, then we are under no obligation to claim that his utterance was meaningful. The sentential view, then, supports T-entailment, though probably not M-entailment.

§5

The Argument from Metaphor

Many, if not most, metaphors involve category mistakes.40 Consider for example ‘The silence was liquid’, ‘A sentence wears its meaning on its sleeve’ and ‘This poem is pregnant’.41 The fact that most metaphors involve category mistakes is not a coincidence. It seems that a big part of what makes metaphors the poetic or figurative devices that they are, has to do with connecting objects and properties that normally seem to belong to completely disjoint domains. Metaphors clearly have some communicative purpose: they are intended to communicate some content, or at least produce some effect in the hearer. What can we infer from this regarding the question of whether category mistakes have literal meanings? In this section, I argue as follows.

that the relation of same-saying requires synonymy between the utterances of the reporter and the reportee (see e.g. Davidson (2001a), p. 104 or p. 107) and he also claims that the utterance of the reporter must “serve at least the purpose of conveying the content of what someone said” (p. 107)—again suggesting that the utterance of the reporter must be contentful. 40 This observation also suggests another fairly straightforward argument against the syntactic approach, since metaphors are generally required to be syntactically well-formed. 41 A more complicated issue, which I shall not address here, is whether sentences involving explicit type attributions such as ‘John is an ice-cube’ should ultimately count as category mistakes.

66

THE ARGUMENT FROM METAPHOR I defend the claim that for metaphors in general to achieve their metaphorical communicative purpose, they must have literal meanings. This in turn entails that, since many metaphors involving category mistakes manage to achieve their metaphorical purpose, they must also have literal meanings, so category mistakes must be (literally) meaningful. How does one defend the claim that for metaphors to achieve their metaphorical purpose they must have literal meanings? I will discuss the most prominent linguistic theories of metaphor, and classify these theories into two categories: those that require metaphors to have literal meanings in order to achieve their metaphorical purpose and those that do not. I argue that theories which fall under the second category ought to be rejected, and hence that one ought to accept the conclusion that metaphors have literal meanings. To clarify, I will only be concerned with linguistic theories of metaphor rather than cognitive or psychological theories of metaphor. That is to say, I will be concerned with theories about the nature of metaphorical truth and metaphorical meaning, rather than theories concerning how subjects process those metaphorical meanings. I maintain that the linguistic and philosophical questions concerning metaphors are by and large independent of the cognitive issues: it may well be right that (at least some) metaphorical meanings are processed directly and not via their literal meanings. This is nevertheless compatible with the claim that metaphors can only have a metaphorical meaning or effect if they have literal meanings. (By analogy, suppose it turns out that we can process the meanings of some familiar sentences directly, without processing the meaning of each of the words in the sentence first. This is still compatible with the claim that the meaning of the sentence is linguistically composed out of the meaning of its parts, and thus that the sentence could not be meaningful unless the words appearing in it were meaningful). For this reason, I leave aside the rich literature on metaphor from psychology and cognitive science. 42

42 See Camp (2006a) for further defence of the claim that the question of which philosophical linguistic theories of metaphor ought to be adopted is in principle independent of the question of cognitive processing.

67

THE ME ANINGLESSNESS V IE W Reimer and Camp’s recent survey of the field presents the following four classes of theories as representative of the current state of the debate:43 interaction theories; simile theories (and related to them the substitution and expansion theories); Gricean theories; and non-cognitivist theories. I will not say much about the interaction theory: I find the view too vague to allow for a detailed discussion,44 and moreover it seems that any way of making the view more precise has it collapse either into a version of non-cognitivism (if the interaction effects are not taken to produce a specific metaphorical ‘content’, in the standard sense of the term) or into a version of the substitution view (if the interaction effect is taken to produce a specific content, albeit one that cannot be paraphrased in literal terms). My discussion of these other two kinds of theories should thus apply to the interaction theory as well. Turning to simile theories, on standard versions of the view, a metaphor means the same as its corresponding simile.45 For example, the metaphorical meaning of ‘Juliet is the sun’ is the same as the literal meaning of the simile ‘Juliet is like the sun’. The simile theory thus understood falls under the category of theories which allow for metaphors to have metaphorical meanings without the metaphorical sentence being meaningful: there is no obvious contradiction in claiming that ‘The poem is pregnant’ is (literally) meaningless, while the simile ‘The poem is like someone who is pregnant’ is meaningful. However, the standard simile view of metaphor ought to be rejected. First, as Davidson notes, it makes metaphors ‘too easy’ to figure out: It does not explain the general feeling that metaphors cannot be paraphrased in literal terms.46 Relatedly, the theory does not explain why we use metaphors as well as similes. But most importantly, even if the theory deals adequately with simple metaphors such as ‘Juliet is the sun’, it is not

43

Reimer & Camp (2006). As Reimer and Camp put it, their survey “is intended to be representative rather than exhaustive” (p. 851). 44 At least when it comes to Black’s defence of the theory, Reimer and Camp concur with this judgment (ibid., p. 855). 45 See discussion in Black (1962), Davidson (2001b), and Fogelin (1988). 46 Davidson (2001b), p. 254.

68

THE ARGUMENT FROM METAPHOR clear how it should handle metaphors such as ‘John rides his mind at a gallop in search of an idea’.47 If this metaphor is simply elliptical for ‘John is like one who rides his mind at a gallop in search of an idea’ the corresponding simile still leaves us with a metaphor. In response one might opt for a much more sophisticated simile theory—one that does not directly translate a metaphor of the form ‘A is B’ into the simile ‘A is like B’, but into some other, more complex, simile. For example, the above sentence might correspond to a less obvious simile such as ‘John is like someone who searches very hard for an idea’. But it is hard to see how such a proposal will get off the ground without taking into account the meaning not only of words, but also of complete phrases and ultimately of the metaphorical sentence as a whole (more on this below). So it seems that if the view of metaphor as simile has any plausibility, one will have to resort to a more sophisticated formulation of the theory, one which falls under the first category of theories (namely those that maintain that metaphorical meanings require the metaphorical sentence to be literally meaningful). Other (somewhat related) kinds of theories that arguably fall under the second category are the ‘expansion of meaning’ and ‘substitution’ views of metaphor.48 According to the substitution view, a word used metaphorically is merely a substitute for another word or phrase that expresses the same meaning literally. Thus for example when ‘ice-cube’ is used metaphorically (e.g. in the sentence ‘John is an ice-cube’), it acts as a substitute for the literal phrase ‘person who is cruel and unemotional’. However, we need not restrict the discussion to cases where there actually exists a linguistic substitute which literally expresses the metaphorical meaning. Let us thus understand the substitution view as making the more general claim that a particular word used metaphorically has a second, metaphorical meaning in addition to its literal meaning (whether or not that

47

Taken from Davidson (2001b) p. 253, who cites Virginia Woolf. See Black (1962), Moran (1997), p. 253, and Davidson (2001b), p. 248 for discussion of these suggestions. 48

69

THE ME ANINGLESSNESS V IE W second meaning can be expressed literally). The ‘expansion of meaning’ view is simply a version of the substitution view (understood in the wider sense), according to which the metaphorical meaning is an expansion of the literal meaning. Thus on the expansion view, when used metaphorically, the word ‘ice-cube’ will apply both to cubically shaped frozen pieces of water as well as to people who are cruel and unemotional. At least at a first pass, both these theories fall under the second category: there is no obvious contradiction in the claim that the literal meaning of ‘pregnant’ is not meaningfully applicable to poems (and hence that ‘The poem is pregnant’ is meaningless), while the metaphorical meaning of ‘pregnant’ is meaningfully applicable in this case. However, thus understood these theories ought to be rejected as well. First, it is not clear why, under these views, the metaphorical meaning is not simply a second literal meaning:49 If the sentence ‘John is an ice-cube’ is analysed so that ‘John’ refers to John, and ‘ice-cube’ refers to some property that literally applies to John, then at best ‘ice-cube’ can be seen as ambiguous, and ‘John is an ice-cube’ as literally true, on one disambiguation. Second, the view faces a similar problem to the one mentioned above concerning more complex metaphors. The metaphor ‘John rides his mind at a gallop’ does not seem to involve any words being individually used in a special metaphorical sense. Rather, it seems to be the entire verb-phrase which receives a metaphorical interpretation here. As above, this suggests that one ought to consider more sophisticated versions of the views, versions that allow the metaphorical meaning to be initially assigned to complex phrases rather than to individual words. But again, this pushes the views towards the first category, namely the claim that the metaphorical meaning is determined by the meaning of the sentence as a whole. Note in particular that although in the example we are considering it is the verb-phrase (rather than the entire sentence) that is the minimal unit requiring a metaphorical interpretation, it is precisely the verb-phrase which is responsible for making the sentence into a category mistake in this case, and hence it is the verb-phrase which, according to 49

This point is made by Davidson (2001b). pp. 248–9 and Searle (1979), p. 100.

70

THE ARGUMENT FROM METAPHOR the proponents of the meaninglessness view, ought to already be meaningless. If one is willing to assign a literal meaning to the categorically mistaken phrase ‘rides his mind at a gallop’, there should be no further obstacle to accept that the sentence as a whole is meaningful. Another problem with the substitution and expansion theories is that the same predicate can have very different metaphorical contributions in different contexts. Consider the sentence ‘Juliet is the sun’ as uttered by Romeo and the sentence ‘Stalin is the sun’ as uttered by a devoted communist. It is not one and the same property that we are ascribing to Juliet and Stalin: in the sense in which Juliet is the sun Stalin is not, and vice versa. Of course, one could claim that ‘sun’ has many metaphorical meanings, and that we are appealing to different meanings in different contexts. But this would have the unappealing consequence that every predicate is massively ambiguous. Perhaps a more attractive way to handle this problem would be to argue that expressions have only one metaphorical meaning, but that this meaning is context sensitive, i.e. it can receive different contents relative to different contexts. The most sophisticated defence of this idea appears in Stern (2000). In a brief, on Stern’s view there is an operator ‘Mthat’ which can take any simple or complex expression ‘φ’. ‘Mthat(φ)’ is a context sensitive expression, such that given a context c, its content is fixed by the semantic-values that are presupposed in c to be “m-associated” with ‘φ’. Finally, a sentence is interpreted metaphorically whenever it is interpreted as having at least one Mthatoperator in its scope. I cannot discuss Stern’s view here in the detail that it deserves, but let me make a few brief remarks. At a first pass, Stern’s view seems to have the odd consequence that metaphorical meanings need not depend even on the literal meanings of the individual words in a metaphorical expression. After all, m-association is a relation between agents and expressions (rather than agents and semantic-values or meanings), and since Stern claims there can be many different grounds for m-association it is not clear that all these grounds have to involve the meaning of ‘φ’ rather than the expression itself. But this consequence is odd because it fails to explain why we expect metaphors to be composed of grammatical sentences 71

THE ME ANINGLESSNESS V IE W built out of meaningful words (why can’t one have m-associations to completely meaningless expressions?). Perhaps Stern can insist that as a matter of fact, the grounds for m-association turn out to always involve the meaning of ‘φ’ and that consequently, m-association only holds (and is presupposed to only hold) over meaningful expressions. But this brings us back to the point of complex metaphors such as ‘John rides his mind at a gallop’. Stern’s view is able to accommodate such metaphors because it allows that ‘Mthat’ can operate on complex phrases as well as on individual words. However, the same line of thought discussed above applies here: if ‘Mthat’ operates on the complete verb-phrase and m-association requires meaningfulness, then the complete verb-phrase ought to be meaningful. But once the problematic verb-phrase generating the category mistake is deemed meaningful, there is no further reason not to accept that the sentence is meaningful as a whole.50 Next, we have Gricean theories of metaphor, which take metaphorical meanings as a species of the general phenomenon of conversational implicature.51 Different Gricean theories differ in the details of which implicatures generate the relevant metaphorical meanings, but the details need not concern us. What is important is that the Gricean framework in general assumes that conversational implicatures are generated via literal contents, and hence that a sentence cannot generate an implicature without 50 It is also worth noting that Stern’s view faces some non-trivial problems. For a start, the view does not address the question of why metaphorical meaning isn’t just a case of a second literal meaning. Moreover, although Stern’s move from many metaphorical meanings to a single but highly context-sensitive meaning addresses the massive ambiguity objection, it comes at the price of creating new problems. To point out two, both raised in Camp (2005): because Stern allows for embedding of any expression in a sentence under an Mthat-operator, Stern’s view avoids massive lexical ambiguity at the cost of introducing massive structural ambiguity (see Camp (2005), p. 717). Also, the move to context sensitive terms creates complications for the semantics of propositional attitude ascriptions: if ‘p’ is a context sensitive term, we expect its content to be fixed by the context of utterance, even if it is embedded in a propositional attitude ascription such as ‘John said that p’. This suggests that on Stern’s view there is no reading on which the ascription ‘Romeo said that Juliet is the sun’ ascribes to Romeo whatever property he (rather than the reporter) presupposes to be m-associated with ‘the sun’, but this seems wrong (see Camp (2005), §3). There is some scope for addressing this problem via more complex theories on how context interacts with propositional attitude ascriptions, but I will not explore the topic any further here. 51 See e.g. Martinich (1984) and Searle (1979).

72

THE ARGUMENT FROM METAPHOR being literally meaningful. The upshot is that Gricean theories fall under the first category, namely theories that imply that if a sentence has a metaphorical meaning it must be literally meaningful. Finally, we have non-cognitivist theories of metaphor, most prominently defended by Davidson.52 According to non-cognitivists there is no such thing as metaphorical meaning. Of course metaphors can produce (and be intended to produce) various effects in the hearer, but those effects are induced directly via the literal meaning of the metaphor.53 The upshot is again that the view falls under the first category: it entails that metaphors, and in particular metaphors involving category mistakes, have a literal meaning.54 To summarize: I have discussed what are taken to be the most prominent linguistic theories of metaphor. I argued that many of these theories require a metaphorical sentence to be literally meaningful in order to achieve its metaphorical purpose. I have also presented several theories that do not require metaphors to be literally meaningful (for example some simplistic versions of the simile and substitution views), but I argued

52

Davidson (2001b). It is important to note that Davidson explicitly endorses the claim that metaphors have literal meanings, and that it is via the grasp of these meanings that one achieves the metaphorical effect. For example, he says that the metaphorical effect “is something brought off by the imaginative employment of words and sentences and depends entirely on the ordinary meanings of those words and hence on the ordinary meanings of the sentences they comprise” (ibid, p. 247). 54 Camp presents an argument that resembles my argument from metaphor (see Camp (2004), pp. 223–6). However, she chooses to attack Davidson’s view as providing an incorrect account of metaphor (ibid, pp. 225–6). Whether or not her attack is successful, this move strikes me as dialectically unnecessary in this context because Davidson’s view anyhow takes metaphors to be literally meaningful. One might try to develop an alternative non-cognitivist view according to which the metaphorical effects are achieved directly via the literal meanings of the individual words in the metaphorical sentence, rather than through the literal meaning of the sentence as a whole. This, however, is far from straightforward since the grammatical structure of the sentence clearly matters to the meaning of the metaphor: ‘This man is a stone’ is not the same metaphor as ‘This stone is a man’. Moreover, it is not merely superficial aspects of the grammatical structure that play a role here, as is apparent from translating metaphors to languages with quite different syntactic structures than English. 53

73

THE ME ANINGLESSNESS V IE W that such approaches should be rejected. I thus conclude that metaphorical sentences, and in particular metaphorical sentences involving category mistakes, are literally meaningful. Obviously, my argument did not take into account every potential theory of metaphor.55 But it is worth noting that the considerations which were raised above point to a general challenge that any view which allows for metaphorical meanings without literal meanings would have to face. Metaphorical meanings often depend not only on the literal meanings of the words in the metaphor, but also on the way these words are combined and on complex interactions between these meanings. Yet on the proposed views, these interactions are somehow supposed to fall short of assigning a meaning to the metaphorical sentence in question. This challenge leaves us, I think, with good reasons to be sceptical about the prospects of theories of metaphor that allow for metaphors literally meaningless.

§6 Arguments in Favour of the Meaninglessness View? So far I have presented some direct arguments against the meaninglessness view. But especially given the prominence of the view, it may be instructive to consider whether there are any compelling positive arguments in favour of the meaninglessness view that need to be addressed. Proponents of the meaninglessness view rarely provide such arguments explicitly. Rather they tend to take it for granted that, in the absence of 55 One class of theories I have not discussed (merely because introducing them will require too much background) are contextualist theories of metaphor (see e.g. Bezuidenhout (2001) and Recanati (2001)). But since one can think of such theories as pragmatic analogues of semantic theories such as the substitution and expansion views, and Stern’s view in particular (cf. Recanati (2001), §5, and Bezuidenhout (2001), §5), very similar considerations to those raised above apply here as well. For example, since contextualists admit that the expanded meanings of metaphorical phrases are constructed (at least in part) out of the meanings of the original phrases (see e.g. Bezuidenhout (2001), p. 168), they too would need to admit that phrases such as ‘rides his mind at a gallop’ are literally meaningful. That aside, it is also worth taking note of Camp (2006b), which argues that contextualist theories of metaphor ought to be rejected, because even by the contextualists’ own criteria, metaphors should not be treated as part of ‘what is said’.

74

ARGUMENTS IN FAVOUR OF THE ME ANINGLESSNESS V IE W? any compelling arguments to the contrary, the default or natural position is the meaninglessness view.56 However, there seem to four main ideas that are taken to support the meaninglessness view. No doubt the most prominent one concerns the central issue of this book, namely explaining the infelicity of category mistakes. As I mentioned at the outset of the chapter, the meaninglessness view does seem to offer a simple and compelling explanation for the fact that category mistakes are highly infelicitous. However, this motivation is only convincing in so far as there are no equally (or more) compelling explanations of the phenomenon available. And in Chapter 5, I suggest and defend an alternative explanation for the infelicity of category mistakes, one that is consistent with the claim that category mistakes are meaningful. In the remainder of this section I briefly discuss three other motivations for the meaninglessness view and explain why they are not ultimately compelling.

§6.1 The imagination motivation One motivation which is sometimes mentioned for taking category mistakes to be meaningless is that one cannot even imagine ‘what it would take for ‘Two is green’ to be true’. Underlying this complaint is sometimes the thought that the meaning of a sentence is its truth-conditions and that a sentence cannot have truth-conditions if one cannot imagine ‘what it would take for the sentence to be true’.57 But this somewhat obscure complaint seems to amount to no more than a convoluted way of saying that one cannot imagine a situation in which ‘Two is green’ is true.58 And if so, the complaint is a rather puzzling one: since it is necessarily false that two is green, it is hardly surprising that we cannot imagine a situation in which ‘Two is green’ is true. Conceivability may not be an infallible guide to possibility, but it seems odd to

56

See for example Diamond (2001), p. 96 or Pap (1960), p. 41. See Lappin (1981), p. 67, and Stern (2006), p. 252 for versions of this line of thought. 58 Situations in which ‘Two is green’ simply has a different meaning from the one it actually has are presumably ignored for this purpose. 57

75

THE ME ANINGLESSNESS V IE W conclude from the fact that in this case conceivability is a successful guide to possibility, that the sentence in question is meaningless. Nonetheless, the objector might insist that for all other examples of meaningful but necessarily false sentences one can imagine situations in which the sentence in question is true. For example, even though ‘There is a counterexample to Fermat’s last theorem’ is a necessarily false sentence, one can perhaps imagine a situation in which is it is true (for example, one can imagine the newspaper headlines announcing ‘Famous mathematician finds mistake in proof for Fermat’s last theorem and produces a counterexample!’). Similarly, even though the sentence ‘The number 5 is even’ is necessarily false, one can perhaps imagine a situation in which it is true (for example, one imagines dividing five by two and being left with no remainder). But, the argument goes, the same is not true of category mistakes: as hard as one tries one cannot imagine a situation in which ‘Two is green’ is true.59 I am not sure that if the suggested disanalogy in our imagination powers were accurate, that would have proved much regarding the meaningfulness of category mistakes. But more straightforwardly, I don’t think there is any such disanalogy. If ‘imagining a situation in which s is true’ is interpreted in a sense which is permissive enough to include imagining dividing five by two and being left with no remainder, then it seems to me obvious that we can also imagine ‘Two is green’ being true (imagine that you check what colour the number two has, and it turns out to be green). Moreover, I take it that cases such as that of the philosopher of mathematics described in §4 above, provide us with a way of imagining that ‘Two is green’ is true, which does not require very wild powers of imagination. The imagination motivation is thus unconvincing.

§6.2 The motivation from alternative theories of meaning The imagination motivation was most naturally linked to the idea of meaning as truth-conditions. A second motivation for the meaninglessness 59 Asher (2011), p. 5 and p. 49 suggests something along these lines as a crucial difference between category mistakes and other necessarily false sentences.

76

ARGUMENTS IN FAVOUR OF THE ME ANINGLESSNESS V IE W? view stems from a preference for some alternative theory of meaning, coupled with the thought that one’s favoured theory of meaning is incompatible with the meaningfulness view. Consider for example verificationism about meaning. According to verificationism, the meaning of a sentence is its verification conditions, and a sentence is meaningful if and only if it is either verifiable or falsifiable. According to at least some version of verificationism (roughly, the version endorsed by the logical positivists—call this ‘traditional verificationism’) only two kinds of verification methods count as legitimate: judging that a sentence is true or false on the basis of direct sense experience and judging that a sentence is true or false on the basis of its being an analytic or logical truth or falsehood. One might argue that traditional verificationism entails that category mistakes are meaningless. No sense experience, the argument goes, shows that ‘Two is green’ is either true or false. But neither is ‘Two is green’ analytically true or analytically false. So ‘Two is green’ fails to have legitimate verification conditions and hence, by the lights of traditional verificationism, it is meaningless.60 Or consider conceptual role semantics. According to conceptual role semantics “the meaning of a representation is the role of that representation in the cognitive life of the agent, e.g. in perception, thought and decision-making”.61 One might argue that conceptual role semantics entails that category mistakes are meaningless. Sentences such as ‘Two is green’, the argument goes, play no role in the cognitive life of any agent and hence by the lights of conceptual role semantics they are meaningless. It is hard to address such worries properly without entering into a lengthy discussion of the subtleties of various theories of meaning. But let me mention two brief responses. First, it remains to be shown that any suggested theories of meanings are in fact inconsistent with the

60 See Reimer & Camp (2006), p. 847 for the suggestion that historically, verificationism motivated taking categorically-mistaken metaphors to be meaningless. 61 Block (1998).

77

THE ME ANINGLESSNESS V IE W meaningfulness view. The two arguments I have sketched above are certainly questionable. It is not clear that general verificationism about meaning entails the meaningless view, for one might argue that we can easily verify that ‘Two is green’ is false.62 Nor is it even clear that traditional verificationism entails the meaninglessness view. If one concedes that sense experience of non-black non-ravens can verify the claim that ravens are black, then one might also argue that sense experience of green non-numbers can verify the claim that numbers, and in particular the number two, are not green. Alternatively, one might try to argue that ‘Two is green’ is analytically false in much the same way that ‘2 + 2 = 5’ or ‘Something is green all over and blue all over’ can be argued to be analytically false. And finally, it is far from clear that conceptual role semantics really entails the meaninglessness view. One might argue, for example, that the uses of category mistakes in metaphors entail that they do play a role in the cognitive lives of agents. It is perhaps on the basis of her belief that (metaphorically) a poem is pregnant, that a publisher decides to publish the poem. The second response to the ‘theories of meaning’ argument is this. If it nonetheless turns out that a certain theory of meaning X is inconsistent with the meaningfulness view, then all the worse for that theory of meaning. The arguments I presented in this chapter in favour of the meaningfulness view did not rely on a particular theory of meaning. Rather, they relied on recognizing certain linguistic phenomena such as metaphorical uses of category mistakes or embeddings of category mistakes in propositional attitude ascriptions, as well as on some general principles of language such as the principle of compositionality—principles which, I assume, any reasonable theory of meaning should accommodate. It is insufficient for proponents of X to object to the arguments on the grounds that their conclusion is inconsistent with their favoured theory. One must either show that the arguments are unsound, or else abandon the theory in the light of their conclusion. 62

Compare this with the example of the continuum hypothesis, where there is much stronger case to be made for the claim that it is neither verifiable nor falsifiable.

78

ARGUMENTS IN FAVOUR OF THE ME ANINGLESSNESS V IE W?

§6.3 The nonsense motivation Perhaps a final somewhat flatfooted motivation for the meaninglessness view is simply the intuition that sentences such as ‘The number two is green’ or ‘The theory of relativity is eating breakfast’ should be classified as ‘nonsense’. But even if this is so, I think one ought to be careful not to confuse the philosophical claim that something literally has no sense, with the everyday use of the English phrase ‘nonsense’. My friend can say to me ‘My work is worthless!’, and I can reply ‘That’s nonsense, and you know it!’. Of course, I do not mean to claim that the sentence he uttered was literally meaningless. What I mean to say is that what my friend said was obviously false, or a ridiculous claim to make. In this sense, it seems perfectly appropriate to categorize category mistakes as nonsense: at least in most contexts, they would be ridiculous claims to make. But this does not entail that they literally have no sense or are meaningless. On the contrary, as I have argued in this chapter, category mistakes are perfectly meaningful sentences, and the meaninglessness view cannot be the correct explanation for their infelicity.

79

4

2

The MBT View

§1

The MBT View

In the previous chapter I discussed a semantic approach to category mistakes, one according to which category mistakes are meaningless. But many contemporary semantic theories postulate other semantic facets besides meaning. One such facet is content: as uttered in context, a sentence expresses a proposition or content. It is commonly assumed that meaning and content can come apart: the sentences ‘I am writing’ and ‘Ofra is writing’ may have different meanings, even if, as currently uttered, they express the same content. Conversely, the sentence ‘I am writing’ as uttered by me and as uttered by Paul McCartney have the same meaning, but express different propositions. Another facet is reference, which in the case of sentences amounts to truth-value. Reference is standardly assumed to diverge from both meaning and content: the sentences ‘I am sitting’ and ‘I am writing’ can both receive the same truth-value (true), even though they differ in both meaning and content. In this chapter, I discuss theories that provide a semantic account of category mistakes, but one that focuses on content or reference, rather than on meaning. A content-based account of category mistake posits that they are infelicitous because they fail to express a proposition. A reference-based account posits that category mistakes are infelicitous because they lack truth-value.1 Since a sentence which lacks content also 1

When I say that a sentence lacks truth-value, I mean that it lacks one of the two standard truth-values. I will also briefly discuss accounts according to which category mistakes cannot be asserted to have a truth-value, but also cannot be asserted to lack a truth-value (see §2).

80

THE MBT V IE W lacks truth-value, both accounts maintain that category mistakes are truth-valueless. The claim that category mistakes are truth-valueless also follows from the meaninglessness view (a sentence can only have a truthvalue if it is meaningful). But since the meaninglessness view has already been discussed at length in the previous chapter, it would be helpful to restrict this chapter to approaches that are clearly distinguished from the meaninglessness view. I will thus focus on those approaches which claim that category mistakes are meaningful but truth-valueless (call this ‘the MBT view’). It is worth noting at the outset, that the central assumption behind the MBT view, namely that a sentence can be meaningful but truth-valueless, is not untenable. Consider the sentence ‘That is red’. As uttered in a context where the demonstrative fails to refer, the sentence fails to express a proposition and is thus truth-valueless, despite the fact that the sentence is clearly meaningful. Similarly, according to some views of definite descriptions, ‘The queen of France in 2010 is bald’ is meaningful but truth-valueless. A more controversial example involves borderline cases of vague predicates. Suppose that Jack is a borderline case of baldness. According to supervaluationism about vagueness, the sentence ‘Jack is bald’ is truth-valueless. But since supervaluationism aims to provide a semantics for precisely such borderline cases, it seems that such sentences are taken by the theory to be meaningful. Moreover, since in some situations Jack may cease to be a borderline case of baldness (imagine Jack at an older age, after he lost every single one of his hairs), there are some contexts of utterance where ‘Jack is bald’ is just straightforwardly true, and hence meaningful. Thus it seems that any (non-nihilist) account of vagueness which takes ‘Jack is bald’ to be truthvalueless in some contexts, should accept that there are meaningful but truth-valueless sentences. The claim that some sentences are meaningful but truth-valueless is, then, widely accepted. But the question is whether category mistakes are such sentences, and in this chapter I argue that they are not. It is not easy to locate the MBT view within the literature on category mistakes. A central complication is that most of this literature was written 81

THE MBT V IE W at a time when the distinction between the different semantic facets presented above was underdeveloped and ill-understood. For example, many authors simply used the terms ‘truth-valueless’ and ‘meaningless’ as interchangeable.2 Nevertheless, there is a distinct body of literature which focuses in particular on the issue of the truth-values of category mistakes, and includes a range of attempts to address the phenomenon using a variety of non-classical logics: several three-valued logics,3 a fourvalued logic,4 and a supervaluationist logic.5 While not all authors in this tradition can be clearly identified as supporters of the MBT view,6 many of the arguments they provide and the logics they propose can naturally figure in a defence of the MBT view. It is also worth keeping in mind that some potential accounts of category mistakes which might be classified as pragmatic are (at least in a sense) versions of the MBT view. The accounts I have in mind are ones which take category mistakes to suffer from presupposition failures (cf. Chapter 5), but maintain that presupposition failures make for truthvalue gaps. Although my focus in this chapter will be with purely semantic versions of the MBT view, it is important to note that some of the considerations I raise against the MBT view (especially the argument offered in §2 below), will apply to such presuppositional accounts as well.7 The structure of the chapter is as follows. In §2, I provide a general argument against the MBT view. In §3, I present the most prominent arguments in favour of the MBT view and respond to them. In §4, I discuss in some detail what is probably the most sophisticated version of the 2 See e.g. Hallden (1949), p. 9 and Pap (1960), p. 43. See also Martin (1975), pp. 70–1 for a discussion of this terminological confusion. 3 Hallden (1949), Smiley (1960), Routley (1966), Goddard (1968), and Lappin (1981). 4 Martin (1975). 5 Thomason (1972). 6 In fact, some are quite explicit that they take category mistakes to be meaningless. See for example, Smiley (1960), Martin (1975), Routley (1966), and Lappin (1981). 7 The argument in §3.1 also applies, assuming that the relevant theories rely on a semantic notion of context. Moreover, the tension discussed in §4.3 below concerning how to handle category mistakes which are also instances of classically valid (or invalid) schema, arises for these theories as well. See f.n. 29 in Chapter 5 for more on this issue.

82

A GENER AL ARGUMENT AGAINST THE MBT V IE W view: Thomason’s supervaluationist approach to category mistakes, and I point out some serious difficulties with the approach. Taking the general objection to the MBT view together with the failure of its proponents to make a convincing case in its favour, I conclude that the MBT view should be rejected.

§2

A General Argument Against the MBT View

As noted above, there are two ways in which a sentence might be thought to be truth-valueless: according to the first, the sentence fails to express a proposition; according to the second, it expresses a proposition but is nevertheless truth-valueless. The first conception is unproblematic. It is easy to point to reasonably uncontroversial cases where a (meaningful) sentence fails to express a proposition and is thereby truth-valueless (e.g. the case of ‘That is red’, as uttered in contexts where the demonstrative fails to pick out an object). Moreover, it is not hard to understand how such truth-value gaps could arise: if sentences or utterances are true (false) in virtue of expressing propositions which are true (false), then a sentence which fails to express a proposition will thereby be truth-valueless. No revision of classical logic or of our concept of truth is needed to accommodate such truth-value gaps. It is, however, far less clear that there can be truth-value gaps of the sort postulated by the second conception (more on this below). In this section, I present a general problem for the MBT view. The core of the problem is that, as I will argue, the MBT view cannot rely merely on the first, more innocent conception of truth-value gaps. More precisely, I claim that the MBT view must accept either that there are sentences which succeed in expressing a proposition but are nevertheless truthvalueless (the second conception), or at least that there are sentences which succeed in expressing a proposition, but such that the proposition is truth-valueless relative to some possible worlds (the proposition is then said to be ‘partial’). However, I claim that both these conceptions of truth-value gaps (propositions that are actually truth-valueless or 83

THE MBT V IE W propositions that are possibly truth-valueless) are, for similar reasons, highly problematic. This conclusion leaves the MBT view in a theoretically difficult position. First, could the MBT view rely only on the first, less problematic conception of truth-value gaps, according to which sentences are truthvalueless when they fail to express a proposition? One reason to think that category mistakes must express propositions has to do with the fact that, as we have seen in §4 of Chapter 3, category mistakes can be embedded in true propositional attitude ascriptions. This argument, however, is not in itself fully conclusive, because some views of propositional attitude ascriptions might allow for such true ascriptions, even if the embedded sentences fail to express a proposition. For example, a Fregean view might allow that ‘S said that p’ can be true where ‘p’ is meaningful yet fails to express a proposition. Another reason to think that category mistakes express propositions is that straightforward cases where a meaningful sentence is thought to fail to express a proposition are those where the sentence contains some sub-sentential expression which, although meaningful, fails to refer. But this kind of failure does not apply in the case of category mistakes. For example, neither the subject term nor the predicate of ‘The number two is green’ suffer from reference failure. Thus if category mistakes fail to express a proposition, then at the very least one cannot rely on this most mundane kind of expression failure. Still, a defender of the MBT view could maintain that category mistakes exhibit a different sort of expression failure: even if all the subsentential expressions successfully express contents, perhaps these contents fail to compose together into a proposition that the sentence as a whole expresses. So, for example, one might argue that although ‘Two’ refers to the number two, and ‘is green’ expresses the property of being green, the property of being green somehow fails to apply to the number two, and thus the sentence ‘Two is green’ fails to express a proposition. The most promising way to develop this idea is to return to the typetheoretic semantics discussed in §2.2 of Chapter 3, but this time interpreting the semantic theory as a theory of content rather than as a theory 84

A GENER AL ARGUMENT AGAINST THE MBT V IE W of meaning.8 The idea would be to treat the content of ‘is green’ as a partial function, such that the number two falls outside of the domain of this function, and explain the failure of expression via failure of functional application. In the previous chapter (§2.2), I argued against the typetheoretic proposal as interpreted at the level of meaning, but conceded that the proposal was more plausible as interpreted at the level of content. As things stand, then, the type-theoretic proposal seems like the best avenue for defending the view that category mistakes are meaningful but truth-valueless. However, I would now like to argue that if the type-theoretic proposal is adopted, one must at least be committed to the existence of partial propositions. Why is that? Consider the sentence ‘The thing I am thinking of is green’, as uttered in a context in which the (unique) thing I am thinking of is a table. In this context, the sentence clearly expresses a proposition—call it p. Let us also make the following two assumptions: that the definite description in question is interpreted non-rigidly (i.e., p is true relative to a possible world w if and only if the thing I am thinking of in w is green);9 and that definite descriptions are interpreted as singular terms (i.e. that one takes the approach to the definite article).10 8 Note that the proponents of type-theoretic semantics are often unclear which of these two interpretations they take. 9 Rothschild (2007) argues that not all definite descriptions have non-rigid readings. His claim is that only ‘role-type descriptions’ have such readings, where a role-type description is one which involves a predicate which is presupposed in the conversation to have a unique satisfier across a range of relevant possible worlds (and where the satisfier sometimes varies among such worlds). Note that one can construct contexts in which the above description (‘The thing I am thinking of’) will count as a role-type description, and moreover, the argument can be rephrased in terms of other predicates that make for even clearer cases of roletype descriptions. Consider for example: ‘The main thing which will be discussed in the annual keynote lecture is green’, where the thing is in fact a statue, but could have been a theory; ‘The gift the ambassador brings is green’, where the gift is in fact a piece of jewellery but could have been a poem; or ‘The president is pregnant’, where the president is in fact a woman, but could have been a man. 10 See Heim & Kratzer (1998), §4.4. This assumption is needed merely to simplify the argument. Suppose one instead adopts a Russellian or neo-Russellian view of definite descriptions. Then the sentence ‘The thing I am thinking of is green’ will be interpreted (possibly after reduction) as: ‘there exists an x, such that x is the unique thing I am thinking of and x is green’. Now, relative to a world in which the unique thing I am thinking of is the

85

THE MBT V IE W Now let w* be a possible world relative to which the thing I am thinking of is the number two. Assuming the type theoretic semantics suggested above is adopted, in order to evaluate the truth-value of p relative to w*, we must apply the function denoted by ‘is green’ to the number two and, according to the suggested semantics, this results in a failure of functional application which entails that p is truth-valueless relative to w*. The upshot is that even if the MBT view wishes to maintain that ‘Two is green’ is truth-valueless in the more benign way (i.e. because it fails to express a proposition altogether), it must at least be committed to partial propositions, due to sentences such as ‘The thing I am thinking of is green’. Let us take stock. So far all I have shown is that the MBT view is committed either to the existence of sentences that express propositions but are nevertheless truth-valueless (because the view maintains that ‘Two is green’ is such a sentence), or at least to the existence of partial propositions (because the view maintains that ‘The thing I am thinking of is green’ expresses such a proposition in the relevant contexts). Why not simply endorse these commitments? Admittedly, many philosophers and linguists are willing to accept truth-valueless propositions, or at least propositions that are possibly truth-valueless.11 However, these commitments face serious theoretical difficulties, ones that are not unfamiliar, but too often ignored. number two, relative to an assignment which assigns to x the unique thing I am thinking of (namely the number two), the open sentence ‘x is the unique thing I am thinking of and x is green’ will (according to the type-theoretic view under discussion) be truth-valueless. But on any logic the MBT theorist is likely to adopt, this is sufficient to render the existential statement truth-valueless as well. (The existential statement would be truth-valueless at least on Strong Kleene three-valued logic, Weak Kleene three-valued logic, a simple version of supervaluationism according to which ‘Two is green’ receives the value true on some valuations and false on others, and Thomason’s more complex supervulationist logic discussed below. Note moreover that any MBT logic should operate in this way, if one wishes to maintain that ‘Something is a green number’ or ‘Some number is green’ is a category mistake, and hence also truth-valueless.) At any rate, independently of how the English determiner ‘the’ is interpreted, we can consider a variant of the argument that is phrased in terms of an artificially stipulated determiner ‘dthe’, which is treated as in the view (perhaps with some special provision as to how to handle cases where there is no unique satisfier). 11 See for example von Fintel (2004), pp. 271–2 and Soames (1999), pp. 167–8.

86

A GENER AL ARGUMENT AGAINST THE MBT V IE W Start with the suggestion that sentences can express truth-valueless propositions. Already at an intuitive level, this suggestion should strike us as odd: if a sentence successfully expresses the content that things are thus and so, it seems that it ought to be true if things are thus and so, and false otherwise. More importantly, Williamson offers a simple yet compelling argument, which shows that the claim that an utterance expresses a proposition but is neither true nor false leads to a direct contradiction, and is thus unfeasible.12 While Williamson’s original argument attacks only the claim that there are propositions which are actually neither true nor false, the argument can be generalized to show that there is a similar problem with the claim that there are propositions which are possibly neither true nor false. The more general, modalized version of Williamson’s argument proceeds as follows: (NT) Necessarily, the proposition that p is true if and only if p. (NF) Necessarily, the proposition that p is false if and only if not p. (1) Possibly, the proposition that p is not true and the proposition that p is not false. Therefore, (2) Possibly, not p and not not p.

(NT) and (NF) are necessitated versions of the standard Tarskian truth and falsity schemas. (It seems that any motivation for accepting the original schemas is ipso facto also a motivation for accepting the necessitated versions.)13 Premise (1) is the reductio assumption (the claim that a certain proposition p is possibly truth-valueless). The conclusion, (2) follows from the premises by elementary modal logic.14 But (2) asserts that it is 12 Williamson (1994), §7.2. A different argument in favour of the view that sentences which express propositions must have truth-values, relying on the idea of truth as the point of assertion, is given in Glanzberg (2004), §2. 13 Cf. Horwich (1990), pp. 22–3. 14 Here is a proof using the weak modal logic K (as presented in Hughes & Cresswell (2003)). Let ‘pt’ stand for ‘the proposition that p is true’ and ‘pf’ for ‘the proposition that p is false’. (1) L((p→pt)∧(pt→p)) [NT] (2) L((¬p→pf )∧(pf→¬p)) [NF] (3) M(¬pt∧¬pf ) [Reductio assumption]

87

THE MBT V IE W possible for a straightforward contradiction to be true, which is no less absurd than claiming that a contradiction is actually true. Assuming one accepts the (modalized) Tarskian bi-conditionals, premise (1) ought to be rejected, and we can conclude that propositions cannot be partial: sentences either fail to express a proposition altogether, or else they express a proposition which is truth-valued relative to every possible world.15 The upshot to our own case is that since ‘The thing I am thinking of is green’ is actually truth-valued, it must express a proposition, one that is (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)

L(p→pt) L(¬p→pf ) L(¬pt→¬p) L(¬pf→¬¬p) L((¬pt→¬p)∧(¬pf →¬¬p)) L((¬pt∧¬pf )→(¬p∧¬¬p)) L(¬(¬p∧¬¬p)→(¬(¬pt∧¬pf))) L(¬(¬p∧¬¬p))→L(¬(¬pt∧¬pf)) ¬L(¬(¬pt∧¬pf ))→¬L(¬(¬p∧¬¬p)) M(¬pt∧¬pf )→M(¬p∧¬¬p) M(¬p∧¬¬p)

[from 1, by K3] [from 2, by K3] [from 4 by DR1] [from 5, by DR1] [from 6,7 by K3] [from 8, by DR1] [from 9, by DR1] [from 10, by K] [from 11, by PC] [from 12, by K5] [from 3, 13 by PC]

15 One potential challenge to this argument is the worry that it would generalize to the case of contingent semantic paradoxes. This is a very interesting and deep challenge that I cannot hope to address properly in this context. It is worth noting, however, that it is not entirely obvious that contingent semantic paradoxes force us to accept partial propositions. Consider the sentence s: ‘The first sentence in John’s first book is not true’. Suppose one utters s in a context where the first sentence in John’s first book is ‘Grass is green’ (call this utterance u1). Clearly, u1 expresses a proposition (call it p). One might think that (assuming the definite description in s is given a non-rigid reading), relative to a world w in which the first sentence in John’s book is s, the proposition p would have to be neither true nor false, or else we would run into paradox. This conclusion, however, is too quick. We can think of s appearing as the first sentence in John’s book in w as consisting of another utterance, call it u2. Suppose we accept that on pain of paradox, u2 fails to express a proposition and is thus truth-valueless. Now relative to w, the proposition p asserts that u2 is not true. And since u2 fails to express a proposition, it is indeed not true, and p is simply true rather than truthvalueless. (The crucial point here is that although u1 and u2 are utterances of the same sentence type, they need not express the same proposition, or indeed both express propositions.) Alternatively, we might consider the sentence s*: ‘The first proposition John uttered on Monday is not true’. Suppose that in the actual world the first proposition that John uttered on Monday is the proposition that grass is green. An utterance of s* thus clearly expresses a proposition, call it p*. Now if there is a possible world w in which the first proposition that John utters on Monday is p*, we would be led to contradiction. But all we can conclude is that there is no such possible world.

88

A GENER AL ARGUMENT AGAINST THE MBT V IE W truth-valued relative to every possible world—even ones where the thing I am thinking of is the number two. Moreover, once it is conceded that relative to some worlds, the function denoted by ‘green’ can be successfully applied to the number two and yield a truth-value, there remains little reason to deny that straightforward category mistakes such as ‘Two is green’ are truth-valued, and hence that the MBT view ought to altogether rejected. Defenders of partial propositions more generally, and the MBT view in particular, can respond to the above argument in one of two ways. The first is to reject (either or both of) the Tarskian bi-conditionals. The problem with this response is that the bi-conditionals seem entirely fundamental to our understanding of truth and falsity: to claim that, for example, it is raining but not true that it is raining seems like an outright contradiction. As Williamson puts it: “In formal semantics it is consistent to label two properties ‘t’ and ‘f’, and suppose that some sentences have neither (or both, for that matter). Evidently, such a manoeuvre shows nothing of philosophical interest. No connection has yet been made between the properties labelled ‘t’ and ‘f’ and the properties of truth and falsity. If one claims that the former properties are the latter, one must be prepared to claim that they are governed by the same principles. What principles govern truth and falsity, if not (T) and (F) [the Tarskian bi-conditionals]?”16 The second response available to the defenders of partial propositions is to accept the Tarskian bi-conditionals, but argue that there is a weak sense in which one can support partial propositions (or at least fail to reject them), without thereby being committed to (1).17 Consider a I certainly do not wish to deny that there is a genuinely difficult issue here (a trickier case is one that concerns the sentence s**: ‘This proposition has the property John first referred to on Monday’, where in the actual world, John first refers to the property of being interesting, while in another world he first refers to the property of being false; and no doubt other revenge paradoxes are lurking in the background). But the issue of semantic paradoxes—in particular contingent semantic paradoxes—is a highly complex issue and one would need to carefully consider a detailed solution to such paradoxes in order to determine whether or not it ultimately forces us to accept partial propositions. (See also Williamson (1994), p. 197 for a brief discussion of the implications of his original argument to semantic paradoxes.). 16 Williamson (1994), pp. 190–1. 17 See Soames (1999), ch. 6 and Field (2003) for views along these lines.

89

THE MBT V IE W proposition p, which one might initially think of as neither true nor false. The current response takes a more cautious approach: p ought not to be asserted or accepted, and correspondingly we ought not to assert or accept that p is true; Conversely, p ought not to be denied or rejected, and correspondingly, we ought not to assert or accept that p is false. The suggestion is that we must remain completely silent with respect to the question of the truth-value of p: we ought not to assert that p is truth-valueless (neither true nor false), but nor should we assert that p has a truth-value. Correspondingly, premise (1) in my argument above is neither to be accepted (as straightforward defenders of partial propositions would like) nor rejected (as I claim it should be). But this quietist approach faces some serious theoretical difficulties of its own. Consider a quietist version of the MBT view. It would have seemed that, whatever one’s view of category mistakes is, one thing we can all agree on is that ‘The number two is green’ is not true. The quietist defender of the MBT view, however, would accept no such thing!18 Moreover, it is not easy for a defender of the quietist approach to explain precisely what the disagreement between them and someone who takes category mistakes to be truth-valued is. Suppose, for example, that the opponent asserts that ‘The number two is green’ is false. The quietist cannot complain that the opponent asserted something that isn’t true. Of course, the quietist could nevertheless insist that the opponent asserted something they ought not to have asserted, but it is not clear that the quietist can offer any explanation or justification for this complaint. (They cannot say the opponent asserted something false or not true, and it is not even clear that they have any grounds for claiming that the opponent said something that the opponent did not know.) The quietist defender of the MBT view thus not only offers the peculiar recommendation that we ought to remain silent on the issue in question, but also has little to say by way of justifying their recommendation, which seems highly unsatisfying.

18 This may be at least one reason for thinking that a quietist approach to the liar sentence is more plausible than a quietist version of the MBT view: there is far less temptation to think that we should accept that the liar sentence is not true.

90

ARGUMENTS IN FAVOUR OF THE MBT V IE W No doubt these brief remarks will not settle the ongoing debate between those who accept that propositions are necessarily truth-valued and those who do not. But it is worth highlighting that a commitment to partial propositions is not merely (as many seem to assume) a matter of devising a clever logic that tells one how to compute the non-standard truth-values of complex sentences. The commitment to partial propositions also comes with some deep and difficult theoretical problems. Since, as I have argued in this section, the MBT view of category mistakes is committed to partial propositions, it inherits these theoretical problems, and these should at least give us serious cause to question whether the MBT view is the right path to take.

§3 Arguments in Favour of the MBT View §3.1 The infelicity argument The primary motivation for the MBT view, as with the alternative accounts, is to account for the infelicity of category mistakes. Having rejected the syntactic approach and the meaninglessness view, one might feel that the last resort for explaining the defectiveness of category mistakes is to claim that they are truth-valueless (even if meaningful). A more specific version of the infelicity motivation is presented by Martin, who claims that the view that category mistakes are truth-valueless “may be grounded in our intuitions about the concept of presupposition”.19 The thought is that category mistakes exhibit a similar kind of infelicity to standard cases of presupposition failure. Since Martin takes the latter to be truth-valueless, he concludes that category mistakes must be truth-valueless as well. Both versions of the motivation are, however, too quick. Accepting the MBT view is neither our last, nor our best resort: as I will argue in Chapter 5, there is an alternative and preferable explanation for the infelicity of category mistakes, one that is consistent with the claim that they are truth-valued. Moreover, although I accept Martin’s observation that category mistakes exhibit the same kind of infelicity as sentences 19

Martin (1975), p. 67.

91

THE MBT V IE W suffering from presupposition failure, this does not entail that category mistakes are truth-valueless. Where Martin is wrong, is in assuming that sentences suffering from presupposition failure must be truth-valueless. As I will discuss in Chapter 5, there are compelling theories of presupposition according to which presupposition failures need not be associated with truth-value gaps, and thus claiming that the phenomenon of category mistakes is related to, or even an instance of, presupposition failure, does not mean that category mistakes are truth-valueless. Not only is the MBT view unnecessary for explaining the infelicity of category mistakes, but it is also not particularly well suited for providing such an explanation. We have already seen that whether or not a sentence exhibits the infelicity associated with category mistakes depends on the context in which it is uttered. Thus for example, the sentence ‘The thing I am thinking of is green’ may be completely innocuous in some contexts, but infelicitous in others. Since the MBT view places the phenomenon of category mistakes at the level of content or reference, it is better suited than the syntactic or meaninglessness views to accommodate this observation, since the same sentence may express different contents or have different referents relative to different contexts. However, more careful reflection shows that whether or not a sentence is infelicitous in the relevant sort of way, does not quite depend on actual referents, but rather on (roughly), what speakers believe the referents to be. Return to the example of ‘The thing John is thinking of is green’. Consider a context in which John is in fact thinking of a green table, but where speakers mistakenly assume that he is thinking of the number two. The sentence will seem to participants in the conversation just as infelicitous as ‘Two is green’, but on a natural interpretation of the MBT view the sentence will be rendered true.20 Conversely, consider a context in which John is in fact thinking of the number two, but where speakers mistakenly assume he is thinking of 20 I say ‘a natural interpretation’ because one could propose some revisionary semantics according to which the referent of the definite description depends not on the actual satisfier of the predicate, but rather on what is believed to be the satisfier. A view in this spirit, at least where rigid descriptions or demonstratives are concerned, is proposed by Stalnaker (see Stalnaker (1978)), and criticized in Hawthorne & Magidor (2009).

92

ARGUMENTS IN FAVOUR OF THE MBT V IE W a table. The sentence would seem to speakers perfectly felicitous, despite the fact that on a natural interpretation of the MBT view, the sentence will be rendered true-valueless. The upshot is that according to the MBT view’s own lights, a sentence’s being truth-valueless (for the right kind of reasons) is neither necessary nor sufficient for it to exhibit the relevant kind of infelicity. To be clear: there is nothing incoherent in the suggestion that speakers can be mistaken about whether a certain sentence is truth-valued or not. But the worry is that, in so far as the MBT view is motivated by the attempt to explain a certain phenomenology (the infelicity associated with category mistakes), the view does not successfully accomplish this task.21, 22 21 See also Magidor (2010), pp. 170–1, for a similar criticism of an analogous attempt to explain the infelicity of presupposition failures using truth-value gaps. Of course, defenders of the MBT view might maintain that a sentence is infelicitous when speakers think (perhaps mistakenly) that it is truth-valueless. But this is a much less satisfactory way to account for such infelicities. Compare, for example, a view that accounts for the infelicity of ‘Boy the here sitting’ by devising a syntax which deems the sentence to be ungrammatical, versus an account according to which the sentence is after all grammatical, but for some reason speakers mistakenly think that it is not. In so far as one’s theory of syntax is motivated by the wish to account for grammaticality intuitions, the former kind of account should obviously be preferred. Similarly, in so far as one’s motivation for the MBT view is simply to account for the infelicity of category mistakes, a view according to which truth-value gaps do not match up with infelicities is at the very least deficient. 22 Another interesting observation in this context is the following contrast between the MBT view and the syntactic or meaninglessness views. The MBT view is committed to what Thomason calls ‘The principle of referentiality’: “The only feature of singular terms relevant to determining the sortal correctness in subject-predicate sentences is their reference” (Thomason (1972), p. 212). This in turn means that the MBT view is committed to treating sentences such as ‘The thing I am thinking of is green’ relative to contexts in which the thing I am thinking of is the number two, as category mistakes. By contrast, it is in principle open for the syntactic and meaninglessness views, to maintain that such utterances do not constitute genuine category mistakes: only sentences that exhibit the relevant kind of infelicity relative to every context are genuine category mistakes. (See Lappin (1981) for a view along these lines. Lappin does concede that sentences such as ‘The thing I’m thinking of is green’ exhibit, in the relevant contexts, ‘sortal deviance’ (ibid., p. 81) but does not classify them as full-blown category mistakes.) Note that despite the above noted problem in how the MBT view handles contextual variance, this contrast does not point to an advantage of the syntactic or meaninglessness views: for one thing, the sentence ‘The thing I am thinking of is green’, exhibits, in the relevant contexts, exactly the kind of infelicity we are trying to explain. For another, as I will go on to argue in Chapter 5, even paradigmatic category mistakes such as ‘Two is green’ can vary in their felicity relative to the context of utterance.

93

THE MBT V IE W

§3.2 Routley’s transfer argument The following argument suggested by Routley may provide another reason for thinking that category mistakes are truth-valueless.23 According to Routley, “Sentences like ‘book a is (is not) blue’ and ‘figure f is (is not) square’ can in suitable contexts be transferred, preserving truth and signification, into respectively, ‘the colour of book a is (is not) blue’ and ‘the shape of figure f is (is not) square”. Similarly, Routley claims, ‘The theory of relativity is (is not) blue’ can be ‘transferred’ into ‘The colour of the theory of relativity is (is not) blue’. Now, since the theory of relativity does not have a colour, the definite description ‘the colour of the theory of relativity’ is empty. But then assuming one adopts a Strawsonian theory of definite descriptions, it follows that ‘The theory of relativity is blue’ is truth-valueless. It is not entirely clear what Routley means by saying that transferring a certain sentence s1 into a sentence s2 ‘preserves truth and signification’, but at the very least, he seems to require that s1 entail s2. The problem is that while this entailment might hold for the positive sentences (a is blue entails that the colour of a is blue), there is no reason to accept that it also holds for the negative sentences. Consider, for example, a book that is striped with alternating green and red stripes. It is true that the book is not blue. But plausibly, the book does not have a unique colour. So on the Strawsonian position envisaged, ‘The colour of the book is not blue’ would be truth-valueless rather than true.24 So Routley is wrong that his proposed transfer preserves ‘truth and signification’. Once we realize this, there is no problem in supposing that ‘The theory of relativity is not blue’ is true (contra the MBT theorist), but that ‘The colour of the theory of relativity is not blue’ is either truth-valueless or false.

23 See Routley (1969), p. 371. Note though that Routley uses the argument to make a slightly different point than the one I present here. Routley accepts a Russellian rather than a Strawsonian view of definite descriptions, and his argument is intended to show that the sentence ‘The theory of relativity is not blue’ cannot be true, as his opponents (defenders of what he calls ‘the falsidal theory’) maintain. His argument depends on the additional assumption that the negation in the sentence takes narrow rather than wide scope. 24 It would also fail to be true on a Russellian view with the negation interpreted as taking narrow scope.

94

ARGUMENTS IN FAVOUR OF THE MBT V IE W

§3.3 The arbitrariness argument The most widely discussed argument in favour of the claim that category mistakes are truth-valueless, is based on the complaint that “any assignment of truth-values to category mistakes must be arbitrary”.25 On the face of it, the complaint seems entirely wrong. If one is forced to choose one of two truth-values for the sentence ‘The number two is green’, one would no doubt opt for ‘false’. (Compare this with choosing a truth-value for a sentence one does not understand or for the Truth Teller.) This observation gives rise to what is known as the ‘falsidal’ theory of category mistakes, which holds that standard atomic category mistakes such as ‘Two is green’ are false.26 It is slightly less obvious, however, what the falisdal theory should say about sentences containing negation. Consider the following: (3) (4) (5) (6)

The number two is breakable. It is not the case that the number two is breakable. The number two is not breakable. The number two is unbreakable.

Since (3) is deemed by the falsidal theory to be false, it seems uncontroversial that its straightforward negation (4) should be true. On the face of it, (5) should also be rendered true. Routley complains, however, that this verdict is problematic, because (5) is just as infelicitous as the original category mistake (3), and thus cannot be true.27 The response Routley envisages on behalf of falsidalist is to claim that (5) is after all false, and that despite appearances it is not really a negation of (3). However, a much more compelling response is to note that some sentences can be at the 25 Thomason (1972), p. 218. The same argument is made in Routley (1969), p. 368, and occupies much of the debate in Haack (1971), Brady & Routley (1973), and Haack (1975). 26 The origin of the term seems to be Lambert (1968), p. 80, though Lambert uses the term ‘falsidical’, which is then altered by Routley to ‘falsidal’. Routley defined falsidal theories to be “no-type theories which assign to non-significant simple subject-predicate sentences with suitable (positive) predicates the truth-value false”. (Routley (1969), p. 368). The term ‘no-type theories’ is used in this literature to describe views on which category mistakes are truth-valued (remaining neutral on which truth-value they receive). 27 Routley (1966), pp. 180–1.

95

THE MBT V IE W same time infelicitous and true. That is, one can agree with Routley that (5) is just as infelicitous as (3), yet suggest an explanation of this infelicity that is consistent with the claim that (3) is true. (Indeed, the explanation I will go on to develop in Chapter 5 has precisely these features.) By contrast, it is plausible that the falisidalist would accept that (6) is false, but is not really the negation of (3). On this view, the claim that a is breakable and that a is unbreakable are not contradictories, merely contraries.28 It is worth noting that independently of the issue of category mistakes, there is evidence that prefixes such as ‘in’, ‘dis’, or ‘un’ in English behave differently than a negation operator such as ‘not’. For example, Horn notes that (at least assuming a Russellian view of descriptions) the sentence ‘The king of France is not interested in music’ seems to have both a true and a false reading (depending on the scope of the negation). On the other hand, the sentence ‘The king of France is uninterested in music’ has only a false reading.29 The distinction between (4) and (5) on the one hand, and (6) on the other hand, raises the suspicion that the falsidal theory relies merely on the fact that syntactically, (6) is an atomic sentence while (4) and (5) are not. The proposal is, then, that the falsidal theory avoids assigning truthvalues arbitrarily by adhering to the following rule:30 Truth Assignment Rule (TAR): Atomic category mistakes are false, their negations are true, their double negation is false, and so forth.

Thomason, however, argues that by adopting such a rule, the falsidal theory faces the arbitrariness problem in a new form.31 The worry is that the distinction between atomic and non-atomic sentences is itself an arbitrary, language-relative feature. To illustrate this point, Thomason 28 Two claims are contraries if they cannot both be true. Two claims are contradictories if they cannot both be true, and they cannot both be false. 29 See Horn (2001), pp. 117–18. 30 Of course, the rule in itself is insufficient to determine the truth-values of all complex category mistakes, only those that are built out of atomic sentences and a negation operator. 31 See Thomason (1972), 218–19. A very similar argument is suggested in Brady & Routley (1973), p. 214.

96

ARGUMENTS IN FAVOUR OF THE MBT V IE W envisages a language, call it ‘Schminglish’, which is very similar to English, except that it contains the word ‘schmeiny’, which means the same as the English expression ‘not shiny’. The idea is then that ‘The velocity of light is schmeiny’ in Schminglish means the same as ‘The velocity of light is not shiny’ in English, and thus both sentences should receive the same truth-value. The problem, however, is that TAR delivers conflicting recommendations with respect to the above English and Schminglish sentences: it recommends that the English sentence should be true, while the Schminglish sentence should be false. But since the two sentences were supposed to be translations of each other, this is untenable. There are two problems with Thomason’s argument. The first is that it is not clear that he is correctly applying TAR to this case. The second is that he is wrong to think that the ‘falsidalist’ must be committed to TAR. Two somewhat delicate issues arise in applying TAR to the Schminglish sentence ‘The number two is schmeiny’. First, note that in order to determine that the sentence is genuinely atomic we would need to know more about the syntax of Schminglish. For all we have been told, it is possible that Schminglish syntax represents ‘schmeiny’ as composed of two separable morphemes, one representing the property of being shiny and the other representing negation (perhaps somewhat like the contractions ‘don’t’ or ‘can’t’ in English). This possibility is perhaps a bit convoluted, but I mention it because on views that take meanings to be sufficiently fine-grained, this syntactic hypothesis may be required in order to respect Thomason’s stipulation that ‘The velocity of light is schmeiny’ in Schminglish and ‘The velocity of light is not shiny’ in English mean the same thing. The second complication is this: even if we accept that ‘The velocity of light is schmeiny’ is atomic, in order to apply TAR we must also show that the sentence is a category mistake in Schminglish. That is to say, we must show that Schminglish speakers would find sentences such as ‘The velocity of light is schmeiny’ infelicitous in the relevant kind of way. It is obviously hard to predict the linguistic intuitions of imaginary speakers of an imaginary language. But in so far as we can make such predictions, it is not at all clear that the sentence would seem infelicitous to Schminglish speakers. After all, given the truth-conditions dictated by 97

THE MBT V IE W the corresponding English translations, we can infer that ‘schmeiny’ denotes a property which is false of mirrors and silverware, and true of opaque pieces of wood and of abstract objects such as the velocity of light. Since Schminglish contains a predicate referring to this property, then plausibly Schminglish speakers have some purpose for expressing this property. For example, they might have a strange religious belief according to which things possessing this property (e.g. the velocity of light) are sacred, while things lacking it (e.g. mirrors) are not. So it is not implausible that Schminglish speakers might find sentences such as ‘The theory of relativity is schmeiny’ entirely natural. Thomason takes TAR to entail that the Schminglish sentence ‘The velocity of light is schmeiny’ is false. But this entailment holds only on the assumptions that the sentence is atomic and that the sentence is a category mistake. It is not obvious, however, that either of these assumptions is correct. Putting aside the issue of whether or not TAR applies in this case, I think Thomason is simply wrong to think that in order to avoid assigning truth-values to category mistakes arbitrarily, the falsidalist should accept TAR (or some other variant of it). Just as with any other (truth-valued) sentence, a category mistake ‘p’ should be assigned the value true if it is true that p, and the value false if it is false that p. Those who propose the arbitrariness argument, expect a more informative rule for assignments of truth-values to category mistakes. Routley, for example, complains that “the no-typers provide no general recipe for whether a prima facie non-significant sentence is to be reckoned in their book as true or false”.32 But just as there is no general recipe to tell us whether ‘Water boils at 100 degrees’, ‘Lisbon is the capital of Portugal’, or ‘Goldbach’s conjecture holds’ are true or false, one should not expect such a recipe in the case of category mistakes. The thought that there ought to be some special rule for ‘assigning’ truth-values to category mistakes or else such assignments are arbitrary is motivated only from the point of view that, fundamentally, takes category mistakes to have no truth-values, and sees the assignment of such values 32

Routley (1969), p. 368.

98

THE SUPERVALUATIONIST TR E ATMENT OF CATEGORY MISTAK ES as a mere technical exercise or game.33 But a proponent of the view that category mistakes are genuinely truth-valued need not think of ‘assigning truth-values’ as a technical exercise, but rather as an attempt to describe the world as it is.34

§4

The Supervaluationist Treatment of Category Mistakes

As noted at the start of this chapter, the literature features various proposals for treating the phenomenon of category mistakes using nonclassical logics. Each of these proposals can naturally underlie an account of category mistakes which falls within the bounds of the MBT view. Discussing every one of these proposals would require far too much space, so I will restrict myself to what is probably the most sophisticated proposal: Thomason’s supervaluationist account of category mistakes.35 I will examine Thomason’s account in some detail, and point to some serious difficulties that it faces.

§4.1 The formal details Thomason’s logical treatment of category mistakes is based on two ideas, both suggested independently by van Fraassen: supervaluationist logic and intensional logic.36 The main ideas behind supervaluationist logics are now widely familiar from the supervaluationist treatment of vagueness, and I will not repeat them here. The main idea behind inten-

33 In particular, note that the need to assign arbitrary truth-values is more likely to arise for someone developing a many-valued logic in which the philosophical significance of the different truth-values is less clear. See for example Martin’s admission that some of his truth-value assignments are arbitrary (Martin (1975), p. 78). 34 Unfortunately, the main response to the arbitrariness argument was given by Haack who bites the bullet and claims that in cases where no general rule applies, the assignment of truth-values should indeed be arbitrary (Haack (1971), p. 74). This is unacceptable, as is rightly pointed out by Brady & Routley (1973). 35 Thomason (1972). 36 The first was initially suggested in van Fraassen (1966), the second in van Fraassen (1967). The term ‘intensional logic’ here follows the terminology of van Fraassen.

99

THE MBT V IE W sional logic is to capture, within logic, necessary relations between certain predicates, such as the fact that ‘green (all over)’ and ‘red (all over)’ cannot be co-instantiated. The problem is how to capture such relations without predetermining which objects fall under which predicates (that is presumably a contingent matter which should be determined by particular models). With this aim in mind, one defines a ‘logical space’ containing ‘points’. Each predicate is allocated a set of points in the logical space. In each model, every object is allocated a point in the logical space, and an object is only allowed to satisfy a predicate if the point it occupies is one of the points allocated to that predicate. For example, we can allocate to ‘red’ and ‘green’ disjoint sets of points in the logical space. This ensures that in every model, an object falling under the extension of ‘red’ will occupy a point allocated to ‘red’, and therefore could not fall under the extension of ‘green’. So relative to every model, no object satisfies both ‘green’ and ‘red’ and ‘Nothing is both red and green’ comes out as logically valid. With these ideas in place, we can turn to Thomason’s account of category mistakes. One starts with a standard language L for first-order logic, enriched with set of S of ‘points’ (call this set the logical space).37 A sortal specification for L is a function E, which assigns to each i-place predicate P of L a subset of Si. For example, if R is a two-place relation, then E(R) will be a set of ordered-pairs of points in the logical space, and if P is a one-place predicate of L, then E(P) is a set of points. The idea is that E(P) will represent the set of points that are allocated to objects for which it is sortally correct (i.e. not a category mistake) to apply P. For example, if P is the predicate ‘blue’ then the points in E(P) are ones which may be assigned tomatoes, but may not be assigned numbers. An interpretation for L relative to E is a function I that assigns to each predicate P a subset of E(P). I(P) is supposed to represent the set of points allocated to objects of which it will be true to predicate P. The constraint

37 Thomason’s account also incorporates sortally restricted quantifiers. For simplicity of exposition, I ignore this feature of the account and treat the quantifiers as unrestricted, but see Lappin (1981), pp. 53–6 for criticisms of Thomason’s treatment of the quantifiers.

100

THE SUPERVALUATIONIST TR E ATMENT OF CATEGORY MISTAK ES that I(P) ⊆ E(P) ensures that it will only be true to predicate P of an object if this predication is sortally correct. Given a domain D, a reference function is a function ref which assigns each singular term in L a member of D. A location function is a function loc that assigns each member of D a point in the logical space. Thus, if c is a constant of L then ref(c) will be an object o to which c refers, loc(o) will be the point in the logical space assigned to o, and thus loc(ref(c)) will be the point in the logical space assigned to the referent of c. A bivalent valuation (relative to E, I, loc, and ref ) is a (total) function f from formulas of L to the set {true, false}, such that it assigns to every atomic formula P(t1, . . . ,tn) the value: true if ∈ I(P) false if ∈ E(P)/I(P) An arbitrary value (true or false) if ∉ E(P)

A bivalent valuation thus assigns an atomic sentence its ‘standard’ value if it is not a category mistake, and an arbitrary value if it is. Non-atomic formulas are assigned truth-values according to the usual recursive truthclauses. A valuation (relative to E, I, loc, and ref ) is a (possibly partial) function V from formulas of L to the set {true, false} that assigns to a formula φ the value: true if for all bivalent valuations f (relative to E, I, loc, ref ), f(φ) = true. false if for all bivalent valuations f (relative to E, I, loc, ref ), f(φ) = false. V is undefined otherwise.

The valuation of φ corresponds to what is usually called a ‘supervaluation’ and is supposed to represent the ultimate truth-value of φ. It is easy to see that for any atomic formula φ, φ has a defined truth-value if and only if φ is sortally correct.

§4.2 Validity and implication As his primary notions of validity and implication, Thomason opts for EI-validity, and EI-implication: Given a sortal specification E, and an

101

THE MBT V IE W interpretation I, a formula φ is EI-valid if for every domain D, for every functions ref, loc relative to D, for every valuation V relative to E, I, loc, and ref, V(φ) = true. Informally: a formula φ is EI-valid if it is (super)true in every model compatible with the specification EI. One can prove that if φ is a classically valid formula, then for any sortal specification E and interpretation I, φ is EI-valid. For example, for any formula ψ the formula ψ ∨¬ψ is EI-valid. This implies that even the sentence ‘The theory of relativity is eating breakfast or the theory of relativity is not eating breakfast’ is EI-valid. Thomason views the fact that his logic is conservative over classical logic (in the sense that all classically valid formulas are valid on his logic) as an essential feature of the account.38 (Note that although this feature is an old-time favourite of supervaluationists, its motivation is somewhat suspect: if one is willing to reject bivalence, why insist on maintaining the law of excluded middle?). The notion of EI-implication is defined as follows: given a sortal specification E, and an interpretation I, φ EI-implies ψ (denoted: φ╞EI ψ), if and only if, for every domain D, for every functions ref, loc relative to D, for every valuation V relative to E, I, loc, ref, if V(φ) = true then V(ψ) = true. Informally: φ EI-implies ψ if for every model compatible with the specification EI in which φ is (super)true, ψ is (super)true as well. The notion of EI-implication is particularly important for Thomason. He claims that any adequate theory of category mistakes should explain not only why category mistakes are infelicitous but also account for “valid inferences arising from sortal phenomena”.39 He never explicitly defines this notion, but his discussion suggests examples of the following sort: Inference 1: (7) Every person has a birthplace. (8) Virginia is an excellent cook.

38 Thomason (1972), p. 231. However, I argue below that there is a sense in which Thomason’s system is not fully classical. 39 Thomason (1972), p. 209.

102

THE SUPERVALUATIONIST TR E ATMENT OF CATEGORY MISTAK ES Therefore, (9) Virginia has a birthplace. Inference 2: (10) Every number is divisible by itself. (11) Bob is prime. Therefore, (12) Bob is divisible by itself.

Thomason’s proposal is that Inference 1 is valid because in any possible world in which the premise (8) is true, Virginia is a person, so by premise (7), she must have a birthplace. Similarly, Inference 2 is valid because in every possible world in which premise (11) is true, Bob is a number and so by premise (10), Bob must be divisible by itself. The claim is that Thomason’s formal system, equipped with the notion of EI-implication, has the advantage of being able to account for the validity of such inferences. There are several things to say about Thomason’s discussion of implication. First, one needs to proceed with caution before deeming inferences such as 1 and 2 to be valid. Many such inferences may seem valid at first glance, but turn out to be invalid on closer reflection. For example, Virginia in premise (8) could be an excellent cooking robot, and Bob in premise (11) could be a prime polynomial. Second, even if the inferences are valid, it is not clear that it is a task of a semantic theory of category mistakes to account for their validity. The third thing to note is this. Thomason argues that it is an advantage of his system over classical logic that his system manages to capture the validity of a range of inferences arising from ‘sortal phenomena’. But closer reflection shows that for many of the inferences Thomason has in mind, it is not really the treatment of category mistakes or the supervaluationist logic that is doing the work in accounting for the apparent validity of the inferences. Rather, what accounts for the validity of the inferences is simply the fact that Thomason is using the resources of intensional logic.

103

THE MBT V IE W Let us focus on a particular case. Suppose our language L contains the name ‘Bob’, and three predicates: ‘likes dancing’ (LD), ‘dislikes dancing’ (DD) and ‘is a person’ (PER). Suppose our sortal specification E and our interpretation I are such that E(LD) = E(DD) ⊆ I(PER), corresponding to the (simplified) assumption that it is only sortally correct to assert of someone that they like or dislike dancing if they are a person. (Note that this in turn means that I must satisfy I(LD) ⊆ I(PER) and I(DD) ⊆ I(PER), corresponding to the thought that if someone likes dancing or dislikes dancing, they must be a person). Now it is easy to see that on Thomason’s system the following two claims will hold: LD(Bob) ╞EI PER(Bob), and DD(Bob) ╞EI PER(Bob), which correspond to the claim that the following two inferences are valid: ‘Bob likes dancing. Therefore, Bob is a person’ and ‘Bob dislikes dancing. Therefore, Bob is a person’. And while it is true that the corresponding inferences will not be valid in pure classical logic, this is not really the apt comparison. Thomason’s use of intensional logic allows him to take into account the particular meanings of the predicates and entailment relations among them, and this fact is essentially orthogonal to the issue of the sortal specification or his treatment of category mistakes. The correct comparison should thus be to classical intensional logic. An easy way to set up a suitable comparison is to consider a system that is just as Thomason’s, except that the sortal specification is defined to be E*, where for every n-place predicate Q, E*(Q) = Sn. This means that no sortal restrictions are in place, and the sortal specification is essentially redundant. It is also easy to see that any valuation relative to E* and I will be bivalent and that the corresponding notion of E*I-implication will be classical. The interesting point is that on this classical notion (keeping the same interpretation I as above) it will also be correct that LD(Bob) ╞E*I PER(Bob) and that DD(Bob) ╞E*I PER(Bob), i.e. the inferences noted above would still come out as valid. The reason is that I satisfies I(LD) ⊆ I(PER) and I(DD) ⊆ I(PER), and thus on any model (compatible with E*, I) on which it is true that, for example, LD(Bob), it would also be true that PER(Bob). It turns out then, that at least in these cases, Thomason’s particular sortal specification and treatment of category mistakes plays no 104

THE SUPERVALUATIONIST TR E ATMENT OF CATEGORY MISTAK ES role in explaining the validity of the relevant inferences. Where Thomason’s system does diverge from classic intensional logic, is when one considers inferences containing negation. So for example, it is true that ¬LD(Bob) ╞EI PER(Bob), but not that ¬LD(Bob) ╞E*I PER(Bob). Which system is preferable in this respect depends on whether or not one takes the corresponding inference (‘It is not the case that Bob likes dancing. Therefore, Bob is a person’) to be intuitively valid.40 Finally, I wish to turn to what I take to be the most problematic feature of Thomason’s account of validity. As we have seen, given a suitable interpretation function I and sortal specification E, the inference from ‘Bob likes dancing’ to ‘Bob is a person’ is valid on Thomason’s system (that is, LD(Bob) ╞EI PER(Bob)). Surprisingly, however, the corresponding conditional, ‘LD(Bob)→PER(Bob)’, is not EI-valid. To see why, consider a model in which loc(ref(Bob))∉ I(PER) (i.e. where Bob is not a person). For all bivalent valuations in the model ‘PER(Bob)’ receives the value false, but since LD(Bob) is sortally incorrect, it receives an arbitrary truth-value—i.e. on some bivalent valuations it receives the value true and on some the value false. This in turn means that there are some bivalent valuations on which the antecedent of the conditional is true but its consequent is false, and thus the conditional is false. It follows that there is a model on which the conditional is not (super)true, and hence the conditional is not EIvalid. Similarly, assuming that the interpretation function is specified so that the interpretations of ‘dislikes dancing’ and ‘likes dancing’ are disjoint, Thomason’s system validates the inference from ‘Bob dislikes dancing’ to ‘It is not that case that Bob likes dancing’ (that is, DD(Bob) ╞EI ¬LD(Bob)). But for similar reasons as above, the conditional ‘If Bob 40 It should be noted in this context, that the pragmatic theory of category mistakes I defend in Chapter 5 delivers an interesting verdict about this inference. Although the inference isn’t strictly speaking valid (it can have a true premise and a false conclusion), the theory nevertheless predicts that the inference is reasonable, in the sense that there are no contexts in which the premise can be felicitously asserted but the conclusion denied. (This is so because in contexts where it is felicitous to utter the premise, the participants must presuppose the conclusion, perhaps after accommodation). On this view the inference is similar to inferences such as ‘The Queen of England isn’t coming to the ceremony. Therefore, England has a queen’.

105

THE MBT V IE W dislikes dancing then it is not the case that Bob likes dancing’ is not EI-valid. (On a model in which Bob is not a person both ‘DD(Bob)’ and ‘LD(Bob)’ receive arbitrary truth-values in each bivalent valuation, so on some such valuations both claims receive the value true.) This result is highly problematic for two reasons. First, it seems that in keeping with the spirit of Thomason’s views, conditionals such as ‘If Bob likes dancing then Bob is a person’ or ‘If Bob dislikes dancing then it is not the case that Bob likes dancing’ should come out as valid. (Note, by the way, that on the classical intensional logic of implication—that is, E*Iimplication—both conditionals are indeed valid.) Second, since Thomason’s system validates the relevant inferences, but not the corresponding conditionals, the system violates one of the most prominent classical derived rules of inference: Conditional Proof (CP). According to CP, we can infer from φ ╞ ψ that╞ φ→ψ. As we have seen, on Thomason’s system there are cases where φ ╞EI ψ holds but ╞EI φ→ψ does not hold, so CP fails. Thus although it is true that all classically valid inferences and formulas (that is, those that are valid on pure, non-intensional classical logic) are still valid on Thomason’s system, the system is nevertheless not fully conservative over classical logic, in the sense that there are derived rules of inference (e.g. CP) that are classically valid but not valid on Thomason’s system. Even putting aside any general attachment to classical logic, CP is a highly plausible principle which captures an important feature of the way we reason with conditionals. Thomason’s system thus suffers from a serious deficiency in representing our inference patterns.41

§4.3 The problem of complex category mistakes According to some versions of the MBT view, not only are atomic category mistakes truth-valueless, but any sentence that has a category mistake as its constituent is truth-valueless as well.42 An interesting feature of Thomason’s account is that this is not always the case: many sentences 41

This criticism is closely related to Fara’s criticism of supervaluationism about vagueness (see Fara (2004), pp. 205–15). Fara also points out other such departures from classical logic which equally apply to Thomason’s system. 42 See e.g. Goddard (1968) and Lappin (1981).

106

THE SUPERVALUATIONIST TR E ATMENT OF CATEGORY MISTAK ES which contain category mistakes as constituents can nevertheless be (super)true or (super)false. The general thought that it would be far too coarse-grained to treat any sentence with a categorically-mistaken constituent as truth-valueless, is on the right track. After all, on the MBT view, truth-value gaps are supposed to match up with the phenomenology of infelicity, but as we have noted in the previous chapters (and will see to a greater extent in the next chapter), not every sentence with a categorically-mistaken constituent is infelicitous in the relevant way. For example, ‘John said that the number two is green’, or ‘If numbers were coloured, the number two would be blue’ are felicitous sentences. The problem, however, is that while Thomason’s system allows for some finer-grained distinctions between different sentences which have categorically-mistaken constituents, these distinctions do not correctly track the relevant infelicity data. Consider for example: (13) The number three is blue and the number three is even. (14) Either the number two is blue or the number two is prime.

Both sentences are infelicitous in the way that is characteristic of category mistakes.43 But on Thomason’s system (13) is deemed false (because on any bivalent valuation the second conjunct is false, which is sufficient to make the conjunction false), and (14) is deemed true (because on any bivalent valuation the second disjunct is true, which is sufficient to make the disjunction true). Thus some sentences that exhibit the relevant kind of infelicity nevertheless turn out to receive bivalent truth-values, against the spirit of the MBT view.44

43 The data is perhaps slightly less clear-cut with respect to (14), though note that Thomason himself acknowledges that such disjunctions are infelicitous (see Thomason (1972), p. 236). 44 Note that a similar problem plagued the logical positivists: a conjunction such as ‘2 + 2 = 5 and φ’ seems falsifiable and a disjunction such as ‘2 + 2 = 4 or φ’ verifiable, even if φ is some statement that is neither falsiable nor verifiable. But there was also a pull towards classifying such disjunctions and conjunctions as meaningless, due to their allegedly meaningless constituents. See Hempel (1950), pp. 46–51.

107

THE MBT V IE W Thomason recognizes this problem, and makes several remarks in response.45 First, he notes that in general one need not accept that any sentence with a truth-valueless constituent is itself truth-valueless. For example, one might argue that ‘The king of France is bald’ is truth-valueless, but ‘It would be odd to say that the king of France is bald’ is true. Even if we accept this,46 the remark is beside the point: the worry was not that (13) and (14) should be deemed by the theory to be truth-valueless because they have a truth-valueless constituent, but rather because they are infelicitous in the relevant sort of way. Responding to precisely this worry, Thomason goes on to claim that in his theory “truth-value gaps were introduced not in order to render truthvalueless all those sentences that will be judged deviant by a native speaker, but to avoid having to give truth-values to formulas arbitrarily. Truth for no reason is ruled out, but trivial or misleading truth is not”.47 He goes on to claim that although sentences such as (14) are infelicitous, one can perhaps explain their anomaly via a pragmatic account that is consistent with the claim that such sentences are true. Thomason’s appeal at this stage to the idea that the infelicity of category mistakes could be explained via a pragmatic account is quite surprising, given that earlier in the paper he argues at length against this approach.48 Moreover, once such a pragmatic account is provided, it is hard to see any remaining motivation for the claim that (some) category mistakes are truth-valueless. (If there is an account of the infelicity of category mistakes that is consistent with their having bivalent truth-values, why not apply the account across the board to all category mistakes?) The only motivation Thomason mentions in this context is the arbitrariness argument, which I have already discussed and rejected above (§3.3). Moreover, note that 45

Thomason (1972), pp. 235–7. Note that it is not clear that we should accept this. Generally, if one thinks that a sentence is truth-valueless if and only if it fails to express a proposition, and that a sentence that has a constituent which fails to express a proposition itself fails to express a proposition, then one will deny that there are cases where a sentence with a truth-valueless constituent is truth-valued. 47 Thomason (1972), p. 236. 48 Thomason (1972), pp. 214–19. 46

108

THE SUPERVALUATIONIST TR E ATMENT OF CATEGORY MISTAK ES given an interpretation I, one can appeal to the E*I models presented in §4.2 above in order to systematically assign bivalent truth-values to sentences of our language, so the arbitrariness argument does not really get off the ground. The final note Thomason makes in defence of his treatment of complex category mistakes is that this treatment is essential in order to preserve certain important aspects of classical logic. It seems highly plausible that any sentence of the form ‘A or not A’ is true. But this requires, Thomason argues, that sentences such as (15) must be true, even though they are infelicitous: (15) Either the number two is blue or the number two is not blue.

Sentences such as (15) present defenders of the MBT view with a dilemma. On the one hand, (15) seems infelicitous in a very similar way to its constituent disjuncts, and since the latter are treated as truth-valueless, there is pressure to treat the whole disjunction as truth-valueless as well. On the other hand, the most basic principles of logic pull us towards taking (15) to be true. Thomason acknowledges that both considerations have their appeal, but ends up opting for the second horn of the dilemma. It is crucial to realize, however, that the dilemma may be a false one. By rejecting the MBT view altogether and providing an account of category mistakes that is consistent with assigning them standard bivalent truthvalues, one can fully accept the laws of classical logic, while at the same time respect the intuition that the infelicity of (15) should be explained by appeal to the same mechanisms which explain the infelicity of its constituents. The next chapter is devoted to defending such an account.

109

5

2

The Pragmatic Approach

§1

The Pragmatic Approach

In the previous chapters I have argued against the syntactic approach, the meaninglessness view, and the MBT view of category mistakes. The question thus remains: why are category mistakes infelicitous? The obvious suggestion is to turn to the realm of pragmatics. The question of what exactly pragmatics is and how it should be distinguished from semantics is tricky and controversial1, but roughly, I take pragmatics to deal with those linguistic phenomena that go beyond the contributions of expressions to the truth-conditional contents of sentences.2 Since pragmatics deals with aspects of language that go beyond specifying the truth-conditions of sentences, it is a particularly suitable domain for seeking an explanation for why a certain sentence might be infelicitous despite being grammatical, meaningful, and truth-valued. Indeed, the literature in pragmatics provides some well-known examples for sentences that have precisely this profile. The sentence ‘Today is Tuesday and today is not Tuesday’ is grammatical, meaningful, and has a truth-value (false). Nevertheless, in most contexts, uttering this sentence would be infelicitous, because the sentence is trivially false, and thus it is hard to see what the point of uttering it would be. Similarly, the sentence ‘Jane is happy but

1

See e.g. Bach (1999), and King & Stanley (2005). This rough characterization allows us to count as non-pragmatic some features that have to do with context-dependence (e.g. the contribution of indexicals to truth-conditional contents). It also allows us to count as pragmatic some aspects of literal meaning, namely those aspects of meaning that do not contribute directly to the truth-conditions of a sentence (e.g. conventional implicatures). 2

110

THE NAÏV E PR AGM ATIC APPROACH satisfied’ is grammatical, meaningful, and (assuming Jane is both happy and satisfied) true. However, the sentence is nevertheless infelicitous. The familiar Gricean explanation of its infelicity is that although the sentence has the same truth-conditional content as ‘Jane is happy and satisfied’, the use of ‘but’ generates a conventional implicature (roughly, that there is some tension between being happy and being satisfied). Since this implicature is hard to accept, the sentence is infelicitous. A pragmatic approach to category mistakes thus seems like a promising path. Yet merely claiming that category mistakes are pragmatically inappropriate is insufficient. What we need is a specific pragmatic account which explains the infelicity of category mistakes. In this chapter, I offer such an account. In a nutshell, I argue that category mistakes are infelicitous because they suffer from (pragmatic) presupposition failures. Before turning to present my own account, in §2 I discuss and reject an alternative pragmatic account (‘the naïve pragmatic approach’), according to which category mistakes are infelicitous because they are either trivially true or trivially false. In §3, I discuss the notion of pragmatic presuppositions which forms the foundation of my own account. In §4, I show how the infelicity of category mistakes can be accounted for by a presuppositional theory. And in §5, I point to various advantages my proposal has in accounting for the rich range of facts concerning category mistakes. §6 contains some final reflections on the implications of my account.

§2 The Naïve Pragmatic Approach Perhaps the first suggestion that comes to mind when searching for a pragmatic account of category mistakes is the following: what is wrong with sentences such as ‘The number two is green’ is that they are trivially false. (Call this ‘the naïve pragmatic approach’). It is widely accepted that, at least in most contexts, trivially false sentences are infelicitous: in standard contexts, an utterance of ‘London is in England and London is not in England’ seems odd. Moreover, it is not hard to find theoretical explanations for why utterances of trivially false sentences are infelicitous. One 111

THE PR AGM ATIC APPROACH can explain why such sentences are infelicitous using Grice’s theory of conversation: according to Grice’s sub-maxim of quality, one ought not to assert what one believes to be false.3 But presumably, if a sentence is trivially false, then the speaker believes it is false, and thus asserting the sentence would violate the maxim. One can also explain why trivially false sentences are infelicitous on Stalnaker’s theory of conversation. According to Stalnaker, conversation takes place against a set of possible worlds (‘the context-set’) which represents those worlds which are compatible with what speakers in the conversation take for granted. The role of a successful assertion that p, on this view, is to rule out from the context-set those worlds which are incompatible with p. But presumably, if p is trivially false then the participants in the conversation take for granted that p is false, and thus removing all worlds incompatible with p from the context-set would result in an empty context-set, which would prevent any further fruitful communication.4 The naïve pragmatic approach can also be extended to explain why sentences such as ‘The number two is not green’ are infelicitous. According to the extended explanation, the sentence is infelicitous because it is trivially true. It is also widely accepted that, at least in most contexts, trivially true sentences are infelicitous: in standard contexts, an utterance of ‘London is in England or London is not England’ seems odd. And as above, it is not hard to find a theoretical explanation for why trivially true sentences are infelicitous. Appealing to Grice’s theory of conversation, such sentences violate the first sub-maxim of quantity (“Make your contribution as informative as required”),5 because arguably, trivially true sentences are not informative at all. On Stalnaker’s theory of assertion, asserting trivially true sentences involves no update to the context-set, and is thus pointless.6

3

See Grice (1989), p. 27. See Stalnaker (1999), p. 49. Note, though, that Stalnaker’s explanation here depends on the assumption that one can only subtract, but never add, worlds to the context-set—an assumption which is in itself suspect. 5 6 Grice (1989), p. 26. Stalnaker (1999), p. 49. 4

112

THE NAÏV E PR AGM ATIC APPROACH The naïve pragmatic approach claims that category mistakes are infelicitous because they are either trivially true or trivially false. The approach thus offers an account of the infelicity of category mistakes which is compatible with the claim that category mistakes are meaningful and truthvalued. The problem, however, is that the approach does not capture the phenomenon of category mistakes correctly. For a start, the kind of infelicity that is associated with category mistakes seems very different in character than that associated with sentences which are otherwise trivially false or trivially true. An utterance of the sentence ‘London is in England and London is not in England’ may indeed be odd, but it is odd in very different manner than ‘Green ideas sleep furiously’. This can be attested to by noting that many have taken the infelicity of category mistakes to entail that they are meaningless, or at the very least truth-valueless, but sentences such as ‘London is and is not in England’ rarely illicit such reactions. More concretely, there are a range of cases in which being infelicitous in the way that is typical of category mistakes simply does not match up with being trivially true or trivially false. First, it seems that being trivially false (trivially true) requires being false (true). But consider an utterance of the following sentence, in a context where John is in fact thinking of a green chair, but where all participants in the conversation mistakenly assume that he is thinking of a number: (1) The thing John is thinking of is green.

An utterance of (1) would certainly exhibit the kind of infelicity associated with category mistakes, even though the sentence is true in this context, and hence not trivially false. Underlying this example is an observation we encountered in the previous chapter: whether a sentence is infelicitous in the way that is typical of category mistakes depends not on the actual facts, but on what speakers in the conversation presume the facts to be. Admittedly, this is not a particularly grave problem for the naïve pragmatic approach. The approach can be amended to say that category mistakes are infelicitous because they are presumed to be trivially false (true), 113

THE PR AGM ATIC APPROACH or taken for granted to be false (true). (In fact, further reflection shows that the theoretical explanations provided above are more satisfactory as explanations for why sentences which are presumed to be trivially false (true) are infelicitous.) But the naïve pragmatic approach suffers from further problems, which are not solved by this amendment. Utterances can exhibit the infelicity associated with category mistakes even if they are not (presumed to be) trivially false or trivially true. For one thing, as we have seen, many philosophers and linguists believe that all category mistakes are truth-valueless, and thus certainly do not take it for granted that ‘The number two is green’ is false, or that ‘The number two is not green’ is true.7 To be clear: the fact that many theorists believe that category mistakes are neither true nor false does not entail that category mistakes are in fact neither true nor false. But this fact does provide a strong reason to think that they are not trivially false or trivially false. Putting aside those that take category mistakes to be truth-valueless, the following cases also present a problem for the naïve pragmatic approach: (2) *Either the temperature in London is green or the temperature in London is 5 degrees. (3) *The temperature in London isn’t green and the temperature in London is 5 degrees.

Both sentences are infelicitous in the way that is typical to category mistakes. But since both sentences are true if and only if the temperature in London is 5 degrees, and since it is neither trivially true nor trivially false that the temperature in London is 5 degrees, then (2) and (3) are neither trivially true nor trivially false. A defender of the naïve pragmatic approach might shift to a different explanation for why (2) and (3) are infelicitous: perhaps they are inappropriate because the first disjunct in (2) and the first conjunct in (3) are redundant for calculating the truthvalue of the respective sentences, thus violating some further maxim of 7 It is perhaps also worth noting that this is not merely an abstract theoretical commitment: those theorists who take category mistakes to be truth-valueless seem to typically have pre-theoretic intuitions that support this position.

114

THE NAÏV E PR AGM ATIC APPROACH conversation (e.g. the maxim of manner, which requires one to be as brief as possible). But first, note that this constitutes yet a further complication of the approach. Second, even if the fact that the first constituent of each sentence is redundant constitutes some reason for why (2) and (3) are pragmatically inappropriate, this explanation hardly seems to account for the kind of infelicity associated with category mistakes (note for example, that ‘Either the temperature in London is 100 degrees or the temperature in London is 5 degrees’ does not exhibit the relevant kind of infelicity, even though the first disjunct is taken for granted to be false). Finally, a similar problem to that exhibited by (2) and (3) can be seen in sentences where the truth-value of the sentence does not obviously depend on that of only one of its constituents. Consider for example counterfactuals such as (4) and (5): (4) *If this toothbrush were pregnant, it would have very cute toothbrushes as babies. (5) *If the temperature in London were green, it wouldn’t be the same as the temperature in Paris.

Such counterfactuals are infelicitous in the relevant way, but are neither trivially true nor trivially false.8 The upshot is that a sentence can exhibit the infelicity associated with category mistakes despite failing to be trivially true or trivially false, contra the naïve pragmatic approach. An additional problem for the naïve pragmatic approach is that it fails to capture the contrast between (6) and (7): (6) Numbers are coloured and the number two is green. (7) *The number two is green and numbers are coloured.

8 One might argue that these are counter-possibles (i.e. counterfactuals with necessarily false antecedents) and that counter-possibles are always true. But at least in the case of (4) it is not clear that this is a counter-possible, and at any rate the suggestion that counter-possibles are always true is controversial enough to mean that such sentences are not trivially true. (Note that although it is sometimes said that on the Lewis-Stalnaker view of counterfactuals counter-possibles are ‘trivially true’, this is a different sense of ‘trivially’ than I intend here. In the sense I intend, ‘trivially’ means something like ‘obviously’, while on its other use, it means something like ‘a limit case of a definition’).

115

THE PR AGM ATIC APPROACH Both sentences consist of two trivially false conjuncts and are thus trivially false. And while perhaps both are not entirely felicitous, (7) seems considerably worse, and in particular seems to exhibit the kind of infelicity associated with category mistakes to a much greater extent than (6). Now in keeping with the general aims of my project, I do not expect the naïve pragmatic approach to give a precise criterion for which trivially false (true) sentences are category mistakes and which are not. Still, the contrast between (6) and (7) is striking because both sentences are composed of the exactly same conjuncts, and would hope that our theory of category mistakes would have at least something to say by way of accounting for such contrasts. While the naïve pragmatic approach seems to provide an initially compelling account for the infelicity of category mistakes (one that has the advantage of being compatible with the claim that category mistakes are truth-valued), the account nevertheless turns out to be too coarsegrained, failing to take into account some of the more subtle facts about when sentence do or do not exhibit the infelicity associated with category mistakes. An alternative pragmatic account is thus needed.

§3 The Background Framework: Pragmatic Presuppositions Since the account of category mistakes that I go on to propose relies crucially on the notion of presupposition, in this section, I provide some necessary background concerning this notion.9 Presupposition is a widely recognized phenomenon. Roughly put, a range of lexical items trigger certain presuppositions in sentences in which they appear. For example, the use of ‘stop’ in ‘Jane stopped smoking’ triggers the presupposition that Jane used to smoke, and the use of the it-cleft in ‘It was Jill who murdered Jack’ triggers the presupposition 9 This discussion is not intended to provide a comprehensive survey of the literature on presupposition: it merely serves to highlight those aspects that will be important to my own discussion. For helpful surveys on the topic see Heim (1990), Beaver (1997), Simons (2006), and Schlenker (2008c).

116

THE BACKGROUND FR AME WOR K that someone murdered Jack. The presupposition generated by uttering a sentence is information that is in some sense implied by the utterance, but is not normally intended to be the main point of the assertion. Indeed, at the point in which the sentence is asserted, participants in the conversation are normally supposed to already take the relevant information for granted, or else an assertion of the sentence would be infelicitous. Thus for example, when a speaker asserts ‘It was Jill who murdered Jack’, the main point of the assertion is not to convey that Jack was murdered (participants are already expected to have this information), but rather that the murderer was Jill. If participants in the conversation were not previously aware that Jack was murdered, uttering the sentence would normally be infelicitous. Although the phenomenon of presupposition is widely recognized, its precise nature, foundational underpinnings, and linguistic behaviour are highly contested. In §3.1, I introduce some of the tests that have been proposed for recognizing presuppositions. While no test is on its own entirely uncontroversial or unproblematic, taken together they provide a powerful tool for deciding which presuppositions a sentence generates. In §3.2, I offer a partial overview of some of the foundational issues concerning presupposition. This discussion will serve to clarify some of the theoretical assumptions that underlie my own account of category mistakes in §4.

§3.1 Tests for presupposition Several tests have been proposed in the literature for determining which presuppositions a sentence generates. Below I introduce some of these.

§3.1.1 Infelicity The most obvious test for presupposition is this: if s generates the presupposition p, then an utterance of s would be infelicitous, unless p is taken for granted by participants in the conversation. One problem with the infelicity test is that a sentence might be infelicitous for reasons other than presupposition failure. In particular, note that typical cases where an atomic sentence s is claimed to generate a 117

THE PR AGM ATIC APPROACH presupposition p, are ones where s also entails p.10 But this means that given a context in which it is taken for granted that p is false, it would also be taken for granted that s is false (at least assuming that the context is closed under entailment). Thus the contexts in which the presupposition fails, are also contexts where s is trivially false (or at least presumed to be trivially false), and hence there is a natural alternative explanation for the infelicity of the utterance (namely, that it is trivially false). This problem, though, is not entirely intractable: as we have seen in §2 of this chapter, this kind of alternative explanation is not always successful. Infelicity due to presupposition failure seems to have a different phenomenology than that generated by mere trivial falsity (for example, the infelicity due to presupposition failure is often associated with the intuition s is neither true nor false), and the ‘trivial falsity’ explanation is usually insufficient to explain the complex projection behaviour of presuppositions. Another problem for the infelicity test is the mechanism of accommodation. It is commonly agreed that the sentence ‘I am picking my sister from the airport’ presupposes that I have a sister. Nevertheless, there are contexts where it is perfectly acceptable to utter this sentence, even if, prior to my utterance, participants in the conversation were not aware that I have a sister. Participants will simply infer from the fact that I used the possessive phrase ‘my sister’ that I have a sister, or in other words, they will accommodate the context so as to include the claim that I have a sister. The upshot is that, due to the mechanism of accommodation, many utterances which apparently suffer from presupposition failure (that is, they are uttered in contexts where the relevant presupposition is not taken for granted) are nevertheless felicitous. A promising theoretical account of accommodation is proposed by von Fintel.11 According to von Fintel, accommodation is a process that occurs after a sentence s has been uttered, but before the context is updated with the content of the relevant assertion. Thus in the above

10 Indeed, Abbott (2006), §2 argues that all presuppositions have this feature, and this is what distinguishes them from conventional implicatures. 11 von Fintel (2000).

118

THE BACKGROUND FR AME WOR K example, although prior to the utterance it was not taken for granted in the conversation that I have a sister, once the sentence has been uttered the context is updated accordingly, and by the time my assertion that I am picking my sister from the airport is evaluated, it is evaluated against the accommodated context, one which already entails the presupposition that I have a sister. Since the presupposition is satisfied in the relevant context, the case does not pose a counterexample to the principle that presupposition failure entails infelicity. Unfortunately, even if von Fintel is right that accommodation does not make for counterexamples to the correctness of the infelicity test, it does pose a problem to its applicability: if presuppositions are usually accommodated, the infelicity associated with their failure would rarely exhibit itself. On the other hand, it is worth noting that in many cases it is impossible or at least very hard to accommodate presuppositions. For example, it is extremely hard to accommodate a presupposition p if it is already take for granted that ¬p is true. (Consider the infelicity of: ‘I don’t have any siblings. My sister is (isn’t) coming to dinner tomorrow’.) And more interestingly, it seems hard to accommodate presuppositions that are highly surprising, even if participants have no reason to assume they must be false. (Consider the inappropriateness of saying out of the blue ‘When I pick up my Nobel Prize tomorrow, I’ll need a good suit’.)

§3.1.2

The HWM test

Another test, proposed by von Fintel, is the ‘Hey, wait a minute’ (HWM) test.12 The suggestion is that s presupposes that p, just in case it would be felicitous to respond to an utterance of s with something along the lines of ‘Hey, wait a minute—I had no idea that p!’. For example, consider the following two conversations: (8) A: The king of France is bald. B: Hey, wait a minute. I had no idea France has a king!

12

Von Fintel (2004).

119

THE PR AGM ATIC APPROACH (9) A: The king of France is bald. B: *Hey, wait a minute. I had no idea that he is bald!

B’s utterance is perfectly felicitous in (8), but infelicitous in (9). This shows, according to von Fintel, that while the claim that France has a king is a presupposition of A’s utterance, the claim that he (the king of France) is bald, is not. The test thus promises to be particularly helpful in distinguishing what a sentence presupposes from what it merely entails. The problem with the HWM test is that it vastly over-generates. Even if passing the test is a necessary condition for being a presupposition, it certainly is not a sufficient condition. In fact, most contextual entailments of an assertion (roughly, those that are not equivalent or nearly equivalent to the original assertion) pass the test. Consider for example the following: (10) A: John is meeting me in London tonight. B: Hey, wait a minute—I had no idea John is back in the UK! Last I heard he was in China. (11) A: Jane is meeting me at the movies tonight. B: Hey, wait a minute—I had no idea Jane has the evening off tonight! Last time we talked she told me she has to stay late at work every day of the week.

B’s utterances in both conversations seems perfectly felicitous, even though it does not seem that A’s utterance in (10) generates the presupposition that John is in the UK, or her utterance in (11), the presupposition that Jane has the evening off tonight. The HWM is thus not an adequate sufficient condition for presuppositions, and should be used with caution. It is, however, nevertheless useful as a necessary condition (that is, failing to pass the test indicates that p is not a presupposition).

§3.1.3

The projection tests

No doubt the most robust tests for presuppositions are the projection tests. As we have seen in §3.1.2, it is often hard to distinguish between what s pre-

120

THE BACKGROUND FR AME WOR K supposes and what it merely entails. However, an important fact about presuppositions is that they project in systematic ways when embedded in certain environments. For instance, it is standardly accepted that the sentence ‘John knows that it is raining’ presupposes that it is raining. This may be initially hard to justify, precisely because the sentence also entails that it is raining (thus, for example, the proposed presupposition might pass the HWM test simply because it is an entailment). A crucial observation, however, is that the negation of the original sentence (‘John doesn’t know that it is raining’) also seems to presuppose that it is raining, even though the negated sentence does not entail that it is raining. There is a sizeable body of literature debating both the form and content of the projection rules, but the following are fairly widely accepted projection properties: a. Negation: If s presupposes p, then ¬s presupposes p as well. For example, ‘John doesn’t know that it is raining’ presupposes that it is raining. It is worth noting, though, that there is a special use of negation, under which ¬s does not seem retain the presupposition p. This is the use we get when, for example, in response to an utterance of ‘The king of France is bald’, one exclaims ‘The king of France is not bald—there is no king of France!’. According to Horn’s prominent theory, such uses involve ‘meta-linguistic negation’, and there are several syntactic and phonetic marks (e.g. a special intonation and a particular behaviour under negative polarity items) which help to distinguish meta-linguistic negation from ordinary uses of negation.13 b. Conditionals: If s1 presupposes p, then ‘If s1 then s2’ presupposes p. For example, ‘If John stopped smoking then he will pass his medical exam’ presupposes that John used to smoke. If s2 presupposes p, then ‘If s1 then s2’ presupposes that if s1 then p (with the latter conditional interpreted materially). Note in particular, that where s1 entails p the presupposition is trivially satisfied, and is thus in effect cancelled. For example, ‘If John used to smoke then he stopped smoking’, does not presuppose that John used to smoke.14 13

See Horn (1985). Though it has also been noted that in some special cases we simply get the presupposition p, rather than the weaker conditional presupposition (if s1 then p). For example, 14

121

THE PR AGM ATIC APPROACH c. Conjunctions: If s1 presupposes p, then ‘s1 and s2’ presupposes p. For example, ‘The king of France is bald and the queen of England is not’ presupposes that France has a king. If s2 presupposes p, then ‘s1 and s2’ presupposes that if s1 then p. Note that in particular, where s1 entails p the presupposition is trivially satisfied, and is thus in effect cancelled. For example, ‘France has a king and the king of France is bald’, does not presuppose that France has a king.15 d. Questions: If s presupposes p, then asking a yes/no question of the form ‘s?’, also presupposes p. For example, ‘Has John stopped smoking?’ and ‘I wonder whether John has stopped smoking’ presuppose that John used to smoke.

For several other connectives and operators, it has been much more difficult to determine what their projection properties are (even though it is clear that they have some non-trivial projection properties). For example, although it seems fairly clear that disjunctions inherit the presuppositions of their disjuncts in some form or other, it is still an open question what the precise projection properties for disjunctions are. A common proposal is that if s1 and s2 generate the presuppositions p1 and p2 respectively, then ‘s1 or s2’ presupposes that (p1 or s2) and (p2 or s1), but Soames has argued that this proposal runs into trouble in cases where p1 and p2 contradict each other (‘John has either just stopped or just started smoking’).16 Another area where the projection properties are still very much under debate are the presuppositions of quantifiers. What, for example,

the sentence ‘If John was at the conference then it was him that solved the problem’ seems to presuppose that someone solved the problem, rather than that if John was at the conference then someone solved the problem. This problem (which generalizes to other connectives, such as conjunction) is known as the ‘proviso problem’. (see Geurts (1996)). 15 One notable aspect of this proposed projection rule is that, even though conjunction is truth-conditionally commutative, the rule is not symmetric (‘s1 and s2’ generates different presuppositions than ‘s2 and s1’). For an argument in favour of alternative, symmetric projection rules see e.g. Rothschild (2008). 16 See Soames (1979) and Magidor (2010) for a detailed discussion of this issue.

122

THE BACKGROUND FR AME WOR K are the presuppositions generated by ‘Every student parked their car’, ‘Some student parked their car’, and ‘At least five students parked their car’? Heim (1983) predicts that all three sentences presuppose that every student has a car. Beaver (2001) predicts that all three sentences presuppose that some student has a car. And other theories predict more complex presuppositions, in particular ones that differ depending on the quantifier in question.17 Note, though, that these competing views have some predictions in common. Thus for example, all the above mentioned theories predict that ‘Every student parked their car’ and ‘No student parked their car’ suffer from presupposition failure where the context entails that there are some students, and that no student has a car. The projection properties make for a compelling set of tests for presuppositional hypotheses. Suppose an atomic sentence s is claimed to generate the presupposition p. The projection properties give us a range of predictions concerning the presuppositions of sentences in which s is embedded. Whether those predictions turn out to be correct is vital evidence in determining whether the hypothesis that p is a presupposition of s is correct. Thus, for example, the hypothesis that ‘John knows that it is raining’ presupposes that it is raining, is strongly supported by the fact that ‘John doesn’t know that it is raining’ and ‘Does John know that it is raining?’ seem to retain this presupposition. On the other hand, the hypothesis that ‘John knows that it is raining’ presupposes that John believes that it is raining can be refuted by noting that there is no temptation whatsoever to think that the above negated sentence or question presuppose that John believes that it is raining. As should be clear from the preceding discussion, the projection tests are not problem free either. For one thing, they depend on what the correct projection properties are, and as we have seen, this is a controversial matter. For another thing, in order to apply a projection test one needs to 17

See e.g. Fox (2008). See also Chemla (2009) for some experimental data which suggests that the presuppositions of quantifiers indeed depend on the quantifier in question.

123

THE PR AGM ATIC APPROACH determine whether a certain sentence in which s is embedded generates the predicted presupposition or not. These predictions are usually tested using other tests for presuppositions such as the infelicity or HWM tests (§3.1.1–2), but as we have seen these tests also suffer from limitations. But although no test for presupposition is in itself infallible or entirely conclusive, taken together they provide a range of intricate predictions that offer a compelling tool for testing presuppositional hypotheses. In §4, I will apply this tool to justify a presuppositional account of category mistakes.

§3.2 Foundational issues The notion of presupposition also raises a range of more foundational questions: what exactly are presuppositions, what gives rise to them, and what does their failure involve? According to the traditional account, a sentence s presupposes p if and only if both s and ¬s entail p. Under standard assumptions, this entails that if s presupposes p, then whenever p is false, s must be neither true nor false. While the traditional approach located presupposition primarily as a logical-semantic phenomenon, a pragmatic approach to presupposition emerged following the seminal work by Stalnaker in the 1970s.18 According to Stalnaker, “communication . . . normally takes place against a background of beliefs or assumptions which are shared by the speaker and his audience, and which are recognized by them to be so shared”.19 (Call this set of background assumptions ‘the context’.) According to the pragmatic approach, presuppositions are constraints on the context: if a sentence s generates a presupposition p, an assertion of s cannot proceed smoothly unless the context already entails p (that is to say, unless speakers already take it for granted that p is true). As Stalnaker notes, thus characterized the pragmatic approach is neutral with respect to the logical and semantic status of presuppositions. One could argue for a version of the pragmatic

18 See Stalnaker (1973), and chs. 1 and 2 of Stalnaker (1999), originally published in 1970 and 1974 respectively. 19 Stalnaker (1999), p. 48.

124

THE BACKGROUND FR AME WOR K approach that goes hand in hand with a logical-semantic one: when s is evaluated against a context which does not entail p, s is neither true nor false, and—perhaps because of this—the conversation cannot proceed smoothly.20 On the other hand, one could opt for a version of the pragmatic approach on which even in contexts in which the presupposition p of s fails, s receives a bivalent truth-value. On this approach, the pragmatic constraints on the context merely lead to the assertion of s being infelicitous, not truth-valueless. Taking the second version of the pragmatic approach, Karttunen and Peters suggested a way to systematically incorporate the phenomenon of presupposition into one’s compositional semantic theory.21 Their suggestion was that each lexical item has two separate facets to its meaning: the first is its standard truth-conditional content (i.e. its contribution to the truth-conditions of sentences in which it appears). The second is its presuppositional component, i.e. its contribution to the presuppositions of sentences in which it appears. Thus for example, ‘stopped’ can receive the following two-dimensional semantic value: , where the first member indicates the truth-conditional component of the verb, and the second member indicates the presuppositional component. By standard rules of composition, one can then infer that ‘John stopped smoking’ has the content that John stopped smoking, and generates the presupposition that John used to smoke. As uttered in a context where it is not taken for granted that John used to smoke, an assertion of ‘John stopped smoking’ will receive a standard 20 Stalnaker (1973), p. 452 argues that least some pragmatic presuppositions can be explained in this way. For criticism of this idea see Soames (1989) and Magidor (2010), pp. 169–71. A particular issue to note is that, at least the traditional semantic approach requires that presupposition failure occurs when the presupposition is in fact false, while the pragmatic approach takes presuppositions to fail when they are taken for granted to be false. Since these two notions may come apart, it is not clear that the pragmatic approach to presuppositions can be fully reconciled with the semantic one. 21 Karttunen and Peters (1979). Note that Karttunen and Peters use the term ‘conventional implicature’ for what I here call ‘presupposition’. For simplicity of the discussion, I will adopt the fairly common assumption that one can identify presuppositions and conventional implicatures (see e.g. Sauerland and Stateva (2007), p. 2 for support of this identification, and Abbott (2006), §2 for opposition to it).

125

THE PR AGM ATIC APPROACH bivalent truth-value (if John never used to smoke, this value would presumably be false), but an utterance of the sentence would nevertheless be infelicitous. In addition to the presuppositions of basic lexical items, Karttunen and Peters postulated that certain connectives and operators have what they called ‘heritage values’, which ensure that each connective and operator has the relevant projection properties (cf. §3.1.3 above). While Karttunen and Peters’s work provides an impressive systematic account of how presuppositions of complex sentences are generated compositionally from the presuppositional components of their constituents, their account has been criticized for being descriptive rather than explanatory.22 The objection concerns in particular the heritage or projection properties: on Karttunen and Peters’s theory, the projection properties are specified explicitly as additional lexical information which is independent from the truth-conditional contribution of the relevant connectives. Many theorists maintain, however, that the projection properties of a connective are closely related to its truth-conditional contribution, and thus that the projection properties should be explained or predicted from the truth-conditional aspect of the meaning of the connectives, rather than simply stipulated as an extra lexically encoded facet of meaning. The first proposal for how the projection properties might be predicated in this manner, appears in Stalnaker’s explanation of the projection properties of ‘and’.23 Stalnaker starts with the general thesis that the role of a successful assertion of s is to update the context by adding to it the content of s. Next, he argues that one could think of an assertion of ‘s1 and s2’ as if it involved two successive assertions—first of s1 and then of s2. In order for the assertion of s1 to be successful, the context must entail its presupposition p1. Assuming this presupposition is indeed satisfied by the context, the assertion is successful, and the context is updated so as to entail s1. Now, consider the assertion of the second conjunct s2. In order for this assertion to be successful, the context must entail its presupposition p2. However, s2 is not asserted against our original context c, but 22

See Heim (1983).

23

See Stalnaker (1999), pp. 59–60.

126

THE BACKGROUND FR AME WOR K against the updated context which includes s1. Since the updated context satisfies p2 if and only if the original context satisfies the conditional ‘if s1 then p2’, the result is that the original context needs to satisfy this conditional constraint. The overall effect is that an assertion of ‘s1 and s2’ presupposes p1 and if s1 then p2, which means that we have obtained precisely the same projection properties that Karttunen and Peters stipulate for ‘and’, but this time by inferring them from general maxims of conversation rather than merely stipulating them. Stalnaker’s explanation is certainly compelling for the case of ‘and’, but it is much harder to provide a similar explanation for the projection properties of other connectives such as ‘if’ and ‘or’. However, in her renowned 1983 paper, Heim generalized Stalnaker’s insights by proposing a radical revision of standard semantics. On the new conception, the basic semantic-values of sentences are not truth-conditions, but rather context change potentials (CCPs). The CCP of a sentence s is a rule which determines, for each context c, what the effect of uttering s would be on the context: that is to say, what the new, updated context c + s would be following the assertion of s. On Heim’s theory, presuppositions of atomic sentences are still stipulated by the lexicon (as in Karttunen and Peters’s theory), and the context change potential of an atomic sentence s is specified so that c + s is undefined if the presupposition p of s is not satisfied by c. The CCPs of complex sentences are defined recursively using previously defined CCPs, in a manner which ensures the connectives have precisely the right projection properties. For example, c + (s1 and s2) is defined to be (c + s1) + s2. As Heim shows, this simple suggestion ensures, on the one hand that c + (s1 and s2) is defined if and only if c satisfies both p1 and if s1 then p2, and on the other hand, that when c + (s1 and s2) is defined, the context is updated with the logical conjunction of s1 and s2, thus entailing that the assertion has the correct truth-conditional contribution. The upshot is that a single semantic property (the context change potential) explains both the presuppositional and truth-conditional behaviour of the connective, and is in this sense seems more explanatory to Karttunen and Peters’s theory. Heim’s revisionary semantics, as well as other proposals in a similar spirit, are known as ‘dynamic semantics’. (The term derives from the idea 127

THE PR AGM ATIC APPROACH that the role of semantics is to specify the effect of an assertion on the dynamics of conversation). Although the dynamic semantics framework is very much thriving (in part due to its success in accounting for other phenomena such as anaphora), its motivation as providing an explanatory theory of presupposition has come into serious doubt. A simple yet compelling objection by Soames shows that Heim’s context change semantics is much less explanatory than it was initially presumed to be.24 Soames notes, that if the context change potential of ‘s1 and s2’ is stipulated to be (c + s2) + s1 instead of (c + s1) + s2 as Heim proposed, then the truth-conditional effect of the conjunction on the context would remain the same, while the presuppositional effect would change. The upshot is that, on Heim’s context change semantics, not every connective with the same truth-conditional contribution as ‘and’, has the projection properties that ‘and’ is predicted to have. Thus there is a sense in which even on Heim’s theory, the projection properties of ‘and’ are simply stipulated rather than predicted from its truth-conditional contribution. Moreover, in a series of recent influential papers Philippe Schlenker has argued that, given a traditional non-dynamic and bivalent semantics, the projection properties of connectives can be nevertheless be explained from general pragmatic principles.25 While the status of some of the specific pragmatic principles Schlenker appeals to might be subject to debate, it is at the very least clear that his account is highly explanatory in that it provides a way to infer a very wide range of projection data from a small set of very general principles. The above discussion brings into relief at least three central issues in the theory of presupposition. The first is whether an assertion suffering from presupposition failure must be truth-valueless. (That is, when s is uttered against a context c which does not satisfy a presupposition p of s, will the utterance be truth-valueless, or might it nevertheless receive a bivalent truth-value). As we have seen, according the traditional account of presupposition, s is always truth-valueless when its presupposition 24 25

See Soames (2009), pp. 119–20 (originally published in 1989). See Schlenker (2008a), Schlenker (2008b), Schlenker (2008c), and Schlenker (2010).

128

THE BACKGROUND FR AME WOR K fails. And it is important to realize that dynamic semantics accounts of presupposition also maintain that presupposition failures result in truthvalue gaps. On the other hand, other accounts such as Karttunen and Peters’s or Schlenker’s, allow that s has a standard bivalent truth-value even in contexts in which its presupposition fails. The second question is whether the projection properties need to be stipulated, or can they be explained from the semantic-values of lexical items and some general principles of conversation. As we have seen both the dynamic account and Schlenker’s account aim to explain rather than stipulate the projection properties (although there is serious doubt as to whether dynamic accounts achieve this aim). On the face of it, Karttunen and Peters’s account maintains that the projection properties are simply stipulated in the lexicon rather than explained. This interpretation may, however, be too quick. On another way to interpret their theory, Karttunen and Peters provide an apparatus for systematically describing the projection properties. While as it stands the apparatus does not offer any explanation for the description it offers, it does not preclude an explanation to be provided by a further theory.26 The final issue, which we have not yet touched upon, is the triggering problem. Just as we might ask why certain connectives have the projection properties they do, we might also ask why a range of lexical items (e.g. ‘stop’ or ‘know’), trigger the presuppositions they do. Most of the theories we have discussed (Karttunen and Peters, Schlenker, Heim) simply stipulate the triggering properties, and do not attempt to explain them. Stalnaker suggests that at least some triggers can be explained on general conversational grounds. For example, he offers the following explanation for the fact that ‘x knows that p’ presupposes that p: “Suppose a speaker were to assert that x knows that p in a context where the truth of p is in doubt or dispute. He would be saying in one breath something that could be challenged in two different ways. He would be leaving it unclear 26

Indeed, as I see it, even if it turns out that the projection (or triggering) properties can be explained or predicted, there is still an open question as to whether these properties should nevertheless be represented in the lexicon or not. (Cf. the discussion of the supervenience of syntactic properties on semantic ones in Chapter 2).

129

THE PR AGM ATIC APPROACH whether the main point was to make a claim about the truth of p, or to make a claim about the epistemic situation of x (the knower), thus leaving it unclear what direction he intended the conversation to take”.27 The problem, however, is that Stalnaker’s reasoning does not explain why ‘x knows that p’ presupposes in particular that p, rather than, for example, that x believes that p, or that x is justified in believing that p. (By parity of reasoning, the same worry applies to these entailments as well). Providing a successful and general account that would explain why different lexical items trigger their presuppositions is a difficult open problem.28 But at any rate, as with the projection properties, one can offer a descriptive account of presupposition triggers, leaving the question of if and how the suggested triggers can be explained for independent inquiry. It is time to take stock. As we have seen, some theories of presupposition (in particular those based on dynamic semantics) associate presupposition failure with truth-value gaps. Combining a presuppositional theory of category mistakes with a view on which presupposition failure entails truth-value gaps, leads one to a version of the MBT view. But my rejection of the MBT view in Chapter 4 should give us good reason not to appeal to such theories of presupposition—at least not in the current context.29 Moreover, as we have seen there is no need to appeal to 27

Stalnaker (1999), p. 55. Though see Abrusán (2011a) and Abrusán (2011b) for a recent attempt to resolve at least some instances of this problem. 29 Do the arguments I raise in Chapter 4 against standard semantic versions of the MBT view apply equally to theories which combine a presuppositional account of category mistakes with the claim that presupposition failures make for truth-value gaps (as with, e.g., dynamic semantics)? The argument in Chapter 4, §2, which raised general problems for truth-value gaps and partial propositions certainly applies here as well. Whether the argument in Chapter 4, §3.1—concerning the incorrect infelicity predictions—applies here, depends on a subtle question concerning how the notion of context is interpreted by the relevant theories. If contexts consist of those claims which are taken for granted in the conversation, the relevant objection can be avoided. On the other hand, in so far as dynamic semantics attempts to use the framework to account not only for presupposition failures, but also for standard semantic phenomena such as the context-sensitivity of demonstratives and indexicals, it is far from clear that they can hold on to this kind of pragmatic notion of context (at least not without accepting a highly revisionary semantics). Finally, consider the problems concerning complex category mistakes, discussed in relation to Thomason’s view (Chapter 4, §4.3). Since the current theories rely on the projection rules for presuppositions, 28

130

A PR ESUPPOSITIONAL ACCOUNT OF CATEGORY MISTAK ES theories that associate presupposition failures with truth-value gaps, as there are compelling alternative theories.30 As far as the projects of explaining both the projection properties and triggering properties are concerned, I wish to simply remain neutral. As suggested above, one can offer a descriptive account of the basic triggers and projection properties, without committing to any particular theory of whether and how these properties should be explained. Admittedly, there is some interaction between the descriptive and explanatory issues: suggesting different projection or triggering properties places different burdens on one’s explanatory analysis, and conversely certain explanations might entail specific descriptive hypotheses. But it seems fair to suppose that, except in borderline cases, the descriptive question should be pursued prior to the explanatory one: that is to say, one should determine what presuppositions are generated by various sentences by analysing their linguistic behaviour, and only then attempt to explain why they behave in the way they do.

§4

A Presuppositional Account of Category Mistakes

§4.1 A basic example In this subsection I would like to consider one simple and paradigmatic case of a category mistake: ‘Two is green’. My proposal is the following: the predicate ‘green’ is a presupposition trigger. In a sentence of the form ‘x is green’, the predicate triggers the presupposition that x is coloured. To put things in the terminology of Karttunen and Peters, the truthconditional content of ‘green’ is the property of being green and its

they succeed in predicting the infelicity of complex category mistakes. But note that this is achieved at the high price (one which Thomason’s system avoids): not all classically valid sentences are deemed to be true. For example, ‘It’s not the case that two is both green and not green’, suffers from presupposition failure, and is thus deemed by the relevant theories to be neither true nor false, rather than true (as my own theory predicts). 30 In addition to Schlenker’s work mentioned above, in recent work Danny Fox has argued that successful quantifier projection predictions of trivalent systems can also be derived in an explanatory bivalent system (see Fox (MS1), and Fox (MS2)).

131

THE PR AGM ATIC APPROACH presuppositional content is the property of being coloured.31 In particular, ‘Two is green’ generates the presupposition that the number two is coloured. But in most contexts of conversation, participants in the conversation do not take it for granted that the number two is coloured, and moreover, take it for granted that the number two is not coloured, making it very hard to accommodate this presupposition. The upshot is that in most contexts of conversation, an utterance of the sentence ‘Two is green’ suffers from presupposition failure, and is thus infelicitous.32 Now it is worth bearing in mind that this proposal should be interpreted against the backdrop of the discussion in §3. First, recall that I am not committed to any particular solution to the triggering problem (or indeed to the claim that there is such a solution). As a simplifying working hypothesis, we can assume that the presuppositional aspect of ‘green’ is stipulated in the lexicon. However, if a compelling solution to the triggering problem is offered and the stipulation can be avoided, then all the better for my theory. Second, as explained in §3, I am assuming that presupposition failures do not make for truth-value gaps. Thus even in those contexts in which the presupposition of ‘The number two is green’ fails and the utterance is infelicitous, it nevertheless receives a bivalent truthvalue (presumably, false). Third, the notion of context in play is a Stalnakerian one: that is, the context consists of those propositions that are taken for granted in the conversation.33 This point affords the theory quite a bit more flexibility than semantic accounts of category mistakes. For example, although in most contexts it is not taken for granted that numbers are coloured, we can consider special contexts, where this proposition is taken for granted for the purpose of conversation. Consider for example 31 Thus, assuming that the presuppositional component is represented in the lexicon, ‘green’ should receive the lexical entry . 32 In addition, when one considers the sentence in the abstract—apart from any particular context of conversation—it also seems infelicitous, because it is infelicitous in most or in typical contexts. (Cf. the oddness of ‘I’m bringing my unicorn to work’ considered out of context). 33 Though we need not follow Stalnaker in assuming that the context (or indeed that propositions more generally), are represented by a set of possible worlds. One could instead think of the context as a set of finer-grained propositions.

132

A PR ESUPPOSITIONAL ACCOUNT OF CATEGORY MISTAK ES Jane, the philosopher of mathematics discussed in Chapter 3, §4. Suppose Jane is giving a series of lectures in which she lays out her views. In her first lecture, she explains the basics of the theory, and argues that numbers are coloured. In future lectures, she presents further work that relies on this initial conjecture. Thus for the purposes of the lecture series, participants take it for granted that numbers, and in particular the number two, are coloured.34 This explains, why, when in her second lecture Jane argues that the number two is green, the utterance does not sound infelicitous. (It may strike the audience as obviously false, or highly implausible, but at least not infelicitous in the way that is typical of category mistakes). Similarly, the notion of context in play can help explain the intricate infelicity facts concerning utterances such as ‘The thing John will be talking about is green’. If participants in the conversation take it for granted that John will be talking about the number two, then (assuming the context is otherwise standard), the utterance would be infelicitous—even if as a matter of fact, John will be talking about a table. Conversely, if participants in the conversation take it for granted that John will be talking about a table, the utterance would be felicitous, even if as a matter of fact, John will be talking about the number two. The above remarks already provide some initial reasons for accepting that ‘Two is green’ generates the presupposition that two is coloured. To be somewhat more systematic, however, it would help to return to the tests for presuppositions presented in §3.1. Start with the infelicity test: we have already noted that ‘Two is green’ is infelicitous in standard contexts, where the claim that two is coloured is not taken for granted (or easily accommodated). Indeed, the kind of infelicity generated by the sentence, exhibits some obvious similarities to that generated by uncontroversial cases of presupposition-failure. Note for example, that category mistakes, just like standard cases of presupposition-failure, bring out in many the intuition that the sentence is neither true nor false, or that the

34 It is important to keep in mind here that, as Stalnaker notes, ‘taking for granted for the purposes of conversation’ does not require that speakers actually believe the proposition (see e.g. Stalnaker (1999), pp. 39–40).

133

THE PR AGM ATIC APPROACH question of whether the sentence is true “simply doesn’t arise”.35 On the other hand, just as the infelicity test would predict, in those special contexts where the proposed presupposition is taken for granted, ‘Two is green’ no longer seems infelicitous in the relevant way. Next, consider the HWM test. B’s utterances in both (12) and (13) strike me as felicitous, indicating that the proposed presupposition at least satisfies this necessary condition for presuppositionhood: (12) A: The number two is green. B: Hey wait a minute—I didn’t think the number two even had a colour! (13) A: The thing I will discuss in my lectures today is green. B: Hey wait a minute—I had no idea the thing you were going to discuss in your lecture was coloured! (I thought you were going to discuss the theory of relativity!)

Finally, let us consider the various projection tests in turn: Negation: In standard contexts (assuming one isn’t employing the special meta-linguistic use of negation) the sentence ‘The number two isn’t green’ is infelicitous. Conversely, just as we would expect, (14) creates a special context where uttering ‘Two isn’t green’ is felicitous: (14) According to the theory I will be advancing in this lecture series numbers have colours: The number three is green, but the number two isn’t green.

Moreover, the proposed presupposition passes the HWM test. (Note that passing the HWM test is a much more telling indication for the presence of presupposition in the environment of negation than unembedded, because in the environment of negation the proposed presupposition is no longer also entailed by the relevant sentence): (15) A: The number two isn’t green. B: Hey, wait a minute—I didn’t think the number two even had a colour!

35 Strawson (1950), p. 330. To be maximally clear: I am certainly not suggesting that these intuitions are correct, I am simply reporting a fact about the phenomenology of category mistakes, and the way it resembles the phenomenology of other presupposition failures.

134

A PR ESUPPOSITIONAL ACCOUNT OF CATEGORY MISTAK ES (16) A: The thing I will discuss in my lectures today is not green. B: Hey wait a minute—I had no idea the thing you were going to discuss in your lecture was even coloured! (I thought you were going to discuss the theory of relativity!)

Conditionals: as is predicted by the projection properties for the conditional, (17) and (18) below are infelicitous. (17) *If the number two is green, then it’s Jack’s favourite number.36 (18) *If the number two is everyone’s favourite number, then the number two is green.

More interestingly, we get the correct prediction that (19), where the proposed presupposition of the consequent is entailed by the antecedent, is felicitous: (19) If numbers are coloured, then the number two is green.

A somewhat more difficult case is the following: (20) ?If the number two is green, then it is coloured.

The problem with (20) is that the sentence is not quite as infelicitous as (17) and (18)—that is to say, it is at least arguably felicitous. But according to the projection rules proposed above, the sentence still generates the presupposition that the number two is coloured, and hence it is predicted to be just as infelicitous as ‘The number two is green’. This problem, however, is an instance of a well-known issue concerning the projection rule for conditionals more generally. For example, note that (21) is, in a similar fashion to (20), arguably felicitous: (21) ?If the king of France is bald, then France has a king.

36 One might try to argue that the infelicity here is generated not because of presupposition failure, but because (17) is an indicative conditional with a trivially false antecedent. But note that the infelicity remains even if one reverts to the counterfactual conditional, which seems to have a similar projection property (‘If Jill were to stop beating her dog, I would consider going out with her’ seems to presuppose that Jill is beating her dog).

135

THE PR AGM ATIC APPROACH Examples like these have motivated some to propose different, symmetric projection rules for the conditional (and similarly for other connectives such as conjunction and disjunction). For example, one might argue that ‘If s1 then s2’ presupposes if s1 then p2 and if s2 then p1.37 Given such a symmetric projection rule, when the presupposition of the antecedent is entailed by the consequent (as in (20) and (21)), the presupposition of the antecedent would effectively be cancelled, and will not generate an infelicity. Whether or not one wishes to handle examples such as (21) by reverting to symmetric projection rules or via some other means,38 the crucial point is this: the fact that (20) is arguably felicitous does not challenge the hypothesis that ‘green’ is a presupposition trigger. On the contrary, this observation is exactly in line with a well-known phenomenon that occurs in the case of other presuppositional triggers, as can be seen by examples such as (21). Conjunction: as is predicted by the projection properties for conjunction (22) and (23) are infelicitous: (22) *The number two is green and it is even. (23) *The number two is even and it is green.

More interestingly, we get the correct prediction that (24), where the proposed presupposition of the second conjunct is entailed by that of the first conjunct, is felicitous: (24) Numbers are coloured and the number two is green.

Another interesting example, one which as we have seen caused some trouble for Thomason’s version of the MBT view (Chapter 4, §4.3), is the following:

37

See Karttunen & Peters (1979), f.n.17, and Rothschild (2008). An alternative explanation for the relative felicity of sentences such as (21) is the following. The inference ‘The king of France is bald. Therefore, France has a king’ is a valid one, and in general, we tend to accept conditionals which correspond to valid inferences (cf. the discussion of Conditional Proof in Chapter 4, §4.2). The temptation to allow a conditional which corresponds to a valid inference might be greater than the temptation to disallow a sentence suffering from presupposition failure, thus explaining the relative felicity of (21). 38

136

A PR ESUPPOSITIONAL ACCOUNT OF CATEGORY MISTAK ES (25) *The number two is blue and it is odd.

The problem for Thomason’s view was that the sentence is intuitively just as infelicitous as (22) and (23) but according to his view, the second false disjunct ensures that the sentence would be false rather than truthvalueless. The current theory, on the other hand, gets the correct prediction here: (25) is indeed false, but the presupposition of the first disjunct is projected to the sentence as a whole, ensuring that the sentence suffers from the same presupposition failure as (22) and (23). A more difficult case for the current theory occurs when we reverse the order of the conjuncts in (26): (26) *The number two is odd and it is blue.

The problem is that according to the standard projection rules, (26) generates the conditional presupposition if the number two is odd then it is blue. But since the antecedent of this conditional presupposition is presumably taken for granted to be false, then the presupposition is trivially satisfied, and on the face of it, the theory predicts that (27) does not suffer from presupposition failure. (Of course, (26) might nevertheless be predicted to be infelicitous due to having a trivially false conjunct, but I have argued that the infelicity arising from trivial falsity is distinct from the kind of infelicity arising from presupposition failures—and (26) seems to suffer from this second kind of infelicity). Two related points are relevant here. First, this is a perfectly general problem for the theory of presupposition. Consider the following example: (27) *Barcelona is the capital of France and the king of France is bald.

The sentence has a trivially false first conjunct, but it nevertheless seems to generate the presupposition that France has a king, and indeed suffer from presupposition failure. Second, very plausibly this problem is simply an instance of the proviso problem noted above (f.n. 14). Assuming that in these cases the presupposition of the second conjunct are retained in their original, stronger form (rather than the weakened, conditional

137

THE PR AGM ATIC APPROACH form), the problem is defused.39 At any rate, as with (20), whatever solution one opts for, the crucial point is that this is a perfectly general problem for the theory of presupposition and not one that tells specifically against the current presuppositional hypothesis. Questions: as predicted, the questions (28) and (29) seem infelicitous in standard contexts: (28) *Is the number two green? (29) *I wonder if the number two is green.

Other operators: As I have noted above, the precise projection rules for other operators such as disjunctions and quantifiers are much more debatable. But it does seem that category mistakes exhibit a similar projection profile to that of other presupposition triggers. For example, (30) seems at the very least to suffer from presupposition failures relative to contexts where it is taken for granted that John did not use to smoke and where it is not taken for granted that John is a fool: (30) Either John stopped smoking or he’s a fool.

Analogously, (31) seems at the very least to infelicitous in contexts where it is not taken for granted that two is not Jane’s favourite number, and where (as is standard) it is taken for granted that two is not coloured: (31) Either the number two is green or it’s not Jane’s favourite number.

Concerning the projection properties of quantifiers, it is standardly agreed that both (32) and (33) at least suffer from presupposition failure relative to contexts which entail that there are some students, but no student used to smoke:

39 In fact, one might be able to extend Schlenker’s recent account of the Proviso problem (Schlenker (2011)) to show that (26) retains the stronger presupposition. On Schlenker’s account, ‘s1 and s2’ presupposes p2 (rather than the conditional if s1 then p2), when s1 is irrelevant to s2. Obviously, a lot depends on how one cashes out the notion relevance, but assuming it is cashed out so that the first conjunct of (26) is irrelevant to the second conjunct, the desired prediction would follow.

138

A PR ESUPPOSITIONAL ACCOUNT OF CATEGORY MISTAK ES (32) Every student stopped smoking. (33) Some student stopped smoking.

Thus whatever the precise projection rules for the quantifiers, we can at least explain why, in standard contexts, the following sentences are infelicitous: (34) *Every number is green. (35) *Some number is green.

And also:40 (36) *Some prime is green.

We have seen that the hypothesis that ‘green’ is a presupposition trigger (and in particular that ‘x is green’ triggers the presupposition that x is coloured) passes the standard tests for presupposition. The projection tests are particularly important: on the one hand, they provide the most robust tests for presupposition. Thus the fact that ‘The number two isn’t green’, or ‘If the number two is green then so is the number three’ retain the relevant infelicity, while ‘If numbers are coloured, then the number two is green’ does not, provides strong evidence that the atomic category mistake ‘The number two is green’ in fact suffers from presupposition failure, as I propose. On the other hand, the projection properties also ensure that the current theory not only accounts for simple, atomic category mistakes such as ‘Two is green’, but also for complex ones such as ‘Two isn’t green’ or ‘Some prime is green’. Given the general theory of presupposition projection, the simple hypothesis regarding ‘green’ is thus sufficient to explain not only the infelicity of atomic category mistakes containing this predicate, but also that of a wide range of complex category mistakes. A final word is in order. One might wonder why I have stipulated that the specific presupposition triggered by ‘x is green’ is that x is 40 The explanation of the infelicity of (36) is that we require the context to at least entail that some prime is coloured, but this is taken for granted to be false (since one takes it for granted that all primes are numbers, and thus not coloured).

139

THE PR AGM ATIC APPROACH coloured, rather than for example that x is a concrete object. Two things are worth noting in this context. First, the crucial claim that I am keen to defend in this section is that ‘green’ is a presupposition trigger, and that this is sufficient to account for the infelicity of category mistakes such as ‘Two is green’. It is less important for my theory precisely which presupposition ‘green’ triggers (as long as it is one that accounts for the relevant data). Nevertheless, I do think there are some reasons to prefer the weak presupposition I propose above, over a more substantive one (e.g. that x is concrete). Consider for example the following two discourses: (37) Physicist A: It turns out that photons are not concrete objects, but they are nevertheless coloured. This photon, for example, is green. (38) Physicist B: Photons are concrete objects, but they are not coloured. *This photon, for example, is green. (/*This photon isn’t green).

I take it that the second sentence Physicist A utters in (37) is felicitous, while the second sentence Physicist B utters in (38) is infelicitous. This tells against the hypothesis that ‘This photon is green’ presupposes that this photon is a concrete object, and in favour of the hypothesis that the presupposition generated is that the photon is coloured. (As we shall see, similar arguments motivate a preference for such weaker presuppositional hypotheses over substantive ones in the case of other triggers as well).

§4.2 Other cases It should now be fairly clear how I would like to account for category mistakes more generally. Consider the following category mistakes: (39) (40) (41) (42) (43) (44) (45)

My toothbrush is pregnant. The chair is dreaming. The theory of relativity is prime. John drinks procrastination. Procrastination drinks relativity. Jane sleeps furiously. Jill ducked under the theory of relativity.

140

A PR ESUPPOSITIONAL ACCOUNT OF CATEGORY MISTAK ES My claim is that the infelicity of all these category mistakes can be explained as instance of presupposition failure. The presupposition triggers in question are the predicates ‘pregnant’, ‘dreaming’, and ‘prime’ in (39)–(41); the relation ‘drinks’ in (42)–(43); the adverb ‘furiously’ in (44); and the preposition ‘under’ in (45). The question of defining precisely which presupposition each of these expressions triggers turns out to be somewhat tricky. It is important to realize, however, that this problem is not specific to the triggers involved in category mistakes. For example, it is commonly accepted that ‘manage’ is a presupposition trigger.41 But which presupposition does it trigger precisely? At a first pass we might propose that ‘x managed to φ’ triggers the presupposition that it was difficult for x to φ. But this cannot be quite right, as can be seen from the felicity of (46): (46) Jill is such a great athlete—she managed to run a 5 minute mile without even making an effort!

Presumably, when one utters (46) one does not presuppose that it is difficult for Jill to run a 5 minute mile without even making an effort. Nor does the sentence trigger the presupposition that it is difficult for most, or generally difficult to φ, as can be seen by the felicity of (47): (47) John is so stupid—he worked for two hours until he finally managed to solve this really simple problem.

Presumably, when one utters (47), one does not presuppose that it is generally difficult to solve a really simple problem. Despite the fact that it is hard to precisely formulate the presupposition triggered by ‘managed’, it is commonly accepted that some such presupposition is in fact triggered, and that we are generally good at telling whether the relevant presupposition is satisfied or not relative to a variety of contexts. (Indeed, our inability to give a precise specification of the relevant presupposition is not very different than our inability to give precise analyses of the truth-conditional content of pretty much any 41

See e.g. Beaver (2001), p. 12.

141

THE PR AGM ATIC APPROACH expression using other terms). As with the case of ‘managed’, I will offer only a rough indication, rather than a precise specification, of the presuppositions triggered in the case of (39)–(45). With this caveat in mind, consider (39). One might start with the proposal that ‘x is pregnant’ triggers the presupposition that x is female. But this is proposal is too strong, as can be seen by the felicity of (48): (48) Biologist: We’ve now discovered a method that enables men to be pregnant. This man, for example, is pregnant.

On the other hand, the proposal that the presupposition triggered is that x is an animal, seems too weak (it cannot account for why, in standard contexts, ‘This man is pregnant’ is infelicitous). If English had a word such as ‘unpregnant’ (the analogue of ‘unhappy’), we would probably have been able to specify the presupposition as ‘x is either pregnant or unpregnant’. But a quick reflection shows just how difficult it is to specify the precise truth-conditions of ‘unhappy’, so it is not clear that this analogy is of much help. Perhaps a good approximation is the proposal that ‘x is pregnant’ triggers the presupposition that x can be pregnant. One needs, of course to take care with the interpretation of the modality in question. It cannot be so restrictive as to make (49) suffer from a presupposition failure: (49) In spite of trying for many years, my aunt isn’t pregnant. In fact, the doctors told my aunt that she cannot become pregnant.

On the other hand, it cannot be so permissive as to include any metaphysical possibility. After all, it may be that sex is not an essential property of humans, and thus that it is metaphysically possible for someone who is actually a man to be a woman (in particular a pregnant woman). Yet it nevertheless seems that ‘This man is pregnant’ is, in standard contexts, infelicitous. Even without pinning down the precise force of the modality in question we can assume that in the relevant sense, it is standardly presupposed that men, as well as numbers or tooth brushes cannot be pregnant—thus accounting for the infelicity of category mistakes

142

A PR ESUPPOSITIONAL ACCOUNT OF CATEGORY MISTAK ES such as (39).42 Conversely, the hypothesis explains why (39) is much more felicitous when embedded in special contexts (e.g. ‘If toothbrushes were able to be pregnant, then this one would be’; or ‘Members of my tribe brush their teeth with this type of hairy worm. I had to stop using my toothbrush recently, though, because it is pregnant’). The presupposition triggered by ‘x is dreaming’ is roughly that x has mental states (or perhaps that x is able to dream—appealing to a similar modality as above). With respect to ‘x is prime’, a good hypothesis is that it triggers the presupposition that x is either prime or composite. This not only explains why (in standard contexts) sentences such as ‘The theory of relativity is prime’ or ‘The theory of relativity is not prime’ are infelicitous, but also passes other tests for presupposition, such as the following instances of the HWM test: (50) A: 2.25 is prime B: Hey, wait a minute! I didn’t think that non-natural numbers are prime or composite! (I thought only natural numbers are . . . ). (51) A: The thing I will be lecturing about today is prime. B: Hey, wait a minute! I didn’t think the thing you were lecturing on today is prime or composite! (I thought you were going to lecture about the theory of relativity!).

Given the presupposition I propose, some of the projection tests turn out difficult to apply. For example, one might note that (52) is still arguably infelicitous: (52) ?If my chair is prime or composite, then it is prime.

However, this infelicity can presumably be explained by the fact that presupposition generated by the use of ‘prime’ and ‘composite’ in the antecedent, projects through the conditional.43 On the other hand, imag42

Kratzer’s notion of circumstantial modality (see Kratzer (1991)) may certainly be helpful here, though one would need to say more about the set of facts that constitute the circumstantial modal base. 43 Although given that the antecedent is a disjunction, it is not easy to pin down precisely which presupposition is projected. As I do not want to address the difficult question of the projection rules for disjunctions, I will leave this problem aside.

143

THE PR AGM ATIC APPROACH ine that English had a word ‘primed’ (by analogy to ‘coloured’) which meant roughly ‘prime or composite’. Then it is plausible to think that (53) would be felicitous: (53) If the theory of relativity is primed, then it is composite.

Again, one might wonder why I haven’t simply proposed that ‘x is prime’ triggers the presupposition that x is a number. The reason is that (54) strikes me as perfectly felicitous: (54) Mathematician: you know, not only numbers but also polynomials are prime or composite. This polynomial, for example, is prime.

Since the second sentence in (54) is uttered against a context in which the polynomial is taken for granted to be prime or composite, yet not taken for granted to be a number, the felicity of (54) tells against the hypothesis that the presupposition triggered is that x is a number.44 There is nothing in the theory of presupposition that requires presupposition triggers to be monadic predicates. (Indeed, words that have been proposed to be presupposition triggers belong to diverse syntactic categories, including, among others, determiners such as ‘the’, implicatives such as ‘managed’, iterative adverbs such as ‘too’, and adverbs of manner such as ‘slowly’).45 Thus the infelicity of (42) and (43) can be explained by assuming that the relational predicate ‘drinks’ is a presupposition trigger.

44 Another argument against the ‘x is a number’ hypothesis is the following. It is not clear whether the hypothesis is that the presupposition triggered is that x is a natural number, or that it is a number more generally. Both suggestions face a problem: in a context where the participants in the conversation are mathematicians, ‘2.25 is prime’ would be infelicitous— telling in favour of the proposal that the presupposition is that 2.25 is a natural number. On the other hand, in a context involving lay people, ‘2.25 is prime’ might be perfectly felicitous, telling in favour of the proposal that the presupposition is the more general one. Note that this variable data is straightforward to explain using my own presuppositional hypothesis: in both instances the presupposition triggered is that 2.25 is prime or composite. The mathematicians, however, take it for granted that 2.25 does not satisfy this presupposition (entailing that the utterance is infelicitous) while the lay people do not take this for granted (enabling the utterance to be felicitous). 45 See e.g. Beaver (2001), pp. 10–11, and Abbott (2006), p. 3.

144

A PR ESUPPOSITIONAL ACCOUNT OF CATEGORY MISTAK ES We can propose that ‘x drinks y’ presupposes (roughly) that x is capable of drinking, and y is capable of being drunk. Thus (42) is infelicitous because it generates the presupposition that John is capable of drinking and that procrastination is capable of being drunk—the second conjunct of which is taken for granted to be false. Similarly, (43) is infelicitous because it generates the presupposition that procrastination is capable of drinking and that relativity is capable of being drunk—both conjuncts of which are taken for granted to be false (thus generating a kind of ‘double category mistake’).46 The infelicity of (44) can be explained by supposing that ‘x φs furiously’ presupposes (roughly) that φ-ing is performed with some emotion (or to use Karttunen and Peters’s terminology, and assuming adverbs are generally of type , the lexical entry for ‘furiously’ would be ); and the infelicity of (45) can be explained by assuming that ‘x φ-ed under y’ triggers (roughly) the presupposition that y is located in space (or, assuming prepositions are generally of type , the lexical entry for ‘under’ would be ); Since in standard contexts speakers take it for granted that sleeping is performed with no emotion and that theory of relativity has no spatial location both (44) and (45) suffer from presupposition failure and are thus infelicitous. Leaving aside the question of precisely which presupposition is triggered in each instance, it is easy to see how my account generalizes. Atomic category mistakes are accounted for by the claim that a wide range of expressions (including most verbs, adjectives, adverbs, and prepositions) are presupposition triggers. Since the relevant presupposi46 An interesting question in this context is whether there are any category mistakes of the form xRy, where for some z, xRz is not a category mistake, and for some w, wRy is not a category mistake. One suggestion is that ‘My table is smaller than the number three’ is such a case, because neither ‘My table is smaller than my bed’ nor ‘The number two is smaller than the number three’ are category mistakes. (This example depends, however, on assuming that ‘smaller’ is used with the same meaning in all three sentences). Assuming ‘smaller’ indeed has a uniform meaning in all of the aforementioned examples, we can handle the relevant category mistake by proposing that ‘x is smaller than y’ triggers the presupposition that x and y are comparable in size.

145

THE PR AGM ATIC APPROACH tions are ones that are standardly taken for granted to be false, atomic category mistakes are, in typical contexts, infelicitous. More complex category mistakes, in particular ones involving negation, conjunctions, quantifiers, and so forth, are accounted through via the standard projection properties for presuppositions. The hypothesis of a wide set of triggers, combined with the background theory of presupposition projection thus provides a powerful theory that accounts for the wide variety of category mistakes.

§4.3 Characterizing category mistakes? Having argued that category mistakes are infelicitous because they suffer from presupposition failure, a natural question one might raise at this point is what separates category mistakes from other instances of presupposition failure? It is worth realizing that this is really the question of how to characterize the class of category mistake brought in a slightly different guise. As I have noted at the outset, my task in this book is not to address this question, and I am generally sceptical that a fully satisfactory answer (that is, informative necessary and sufficient conditions for being a category mistake) can be given. Moreover, the competing theories I rejected were not concerned with addressing this question either, and my objections to these theories did not involve their refraining to address it, but rather their failure to give a compelling account of the infelicity of category mistakes. Nevertheless, let me offer a few brief remarks on the issue of characterizing the relevant class. Given the presuppositional account of category mistakes, there are three general directions one could take if attempting to characterize the class of category mistakes. First, one might try to offer a uniform characterization of the presuppositions that give rise to the phenomenon. Perhaps the most promising way to achieve this aim would be to try and rephrase all the relevant presuppositions in modal terms (‘x is able to be pregenant’, ‘x is able to be green’, and so forth). But one would need to be convinced that such an approach can be generalized to

146

A PR ESUPPOSITIONAL ACCOUNT OF CATEGORY MISTAK ES all cases of category mistakes, and ideally, one would also have more to say about the force of the relevant modality. A second approach is to try to locate some distinctive feature of the presupposition failures that are involved in the case of category mistakes. In previous work (Magidor (2007), Chapter 5), I have considered one proposal of this sort: in the case of category mistakes the presupposition is triggered by functionally applying some expression F (the presupposition trigger) to some expression k; this generates a presupposition of the form G(k); Moreover, in contexts where the relevant phenomenology of category mistakes is present, one not only takes it for granted that G(k) is false, but in addition one first, takes for granted that k is a C (for some category C), and second, takes for granted the generic statement that Cs are not G. Thus, for example, in the case of ‘Two is green’ one not only takes it for granted that two is not coloured, but also that two is a number, and that numbers are (generally) not coloured; and in the case of ‘John sleeps furiously’ one takes it for granted not only that (say) sleeping is performed with no emotion, but also (say) that sleeping is an unconscious activity and that unconsciousness activities are not performed with emotion. But while this proposal may be in the right direction, I am not convinced that it ultimately generates accurate necessary and sufficient condition for being a category mistake.47 The final proposal is to identify category mistakes according to the presupposition triggers that give rise to the phenomenon (e.g. ‘green’, ‘prime’, ‘very’, and so forth). The question, however, is how to delineate the relevant class of triggers. One could simply provide a list of all the relevant triggers but this would not be very informative, and also has the disadvantage of excluding new words from triggering presuppositions which give rise to category mistakes. An alternative suggestion assumes that there is some natural subclass of presupposition triggers: though we 47 The sufficiency of the condition is particularly doubtable. For example, consider the sentence ‘The unicorn is tall’. The existential presupposition (that unicorns exist), is generated here by applying the trigger ‘The’ to the predicate ‘unicorn’. But one might not only take it for granted that unicorns don’t exist, but also that (say) unicorns are fictional animals, and fictional animals don’t exist.

147

THE PR AGM ATIC APPROACH cannot give informative non-circular conditions for belonging to this class, we can nevertheless identify the class by ostension and have a reasonably clear idea of which triggers do or do not belong to it. While this final proposal is no doubt more modest in its aims, it may also be the most promising. Ultimately, there may not be much more one can say by the way of characterizing category mistakes. But it is crucial to separate the question of how to analyse the concept of a category mistake from the question of how to explain the infelicity of those sentences that fall under the concept. Whether or not one of the above proposals can succeed in addressing the former question, my interest in the current project is focused on addressing the latter.48

§5

Merits of the Account

In previous sections, I have presented and argued for a presuppositional account of category mistakes. In this section, I would like to highlight some of the advantages of the account, in particular in comparison to the competing accounts I have discussed throughout the book. On the theoretical side of things, the presuppositional account enjoys solid foundational underpinnings. For a start, the account assumes that category mistakes are meaningful. This entails that no revisions of standard principles of compositionality are needed (cf. Chapter 3, §2). It also entails that the account can straightforwardly accommodate the fact that category mistakes in one language can be directly translated into other languages (cf. Chapter 3, §3), or that category mistakes have metaphorical uses (cf. Chapter 3, §5). The account also assumes that category mistakes receive standard bivalent truth-values, and thus no revision of classical

48

It is well worth noting, though, that even if the presuppositional account is not extended so as to address the question of characterization, it can nevertheless offer quite a lot by way of deciding whether particular case is or is not a category mistake: if a certain infelicitous sentence is to be deemed a category mistake, one must show that it is infelicitous because it suffers from some kind of presupposition failure, and as we have seen above there are substantive tests for determining this question.

148

MER ITS OF THE ACCOUNT logic is required (cf. Chapter 4). Finally, the account relies on the independently motivated and widely accepted theory of presupposition. The account thus fits into our more general linguistic theory, and does not require any ad-hoc machinery, introduced exclusively to explain the phenomenon of category mistakes. On the technical side, the presuppositional theory of category mistakes accounts for the many intricate aspects of the phenomenon (in particular aspects that competing theories had trouble accommodating). Consider the issue of embedding atomic category mistakes into more complex sentences. Many theories of category mistakes take the flatfooted approach that any sentence which contains a category mistake as a constituent must be infelicitous as well: If category mistakes are syntactically ill-formed, then sentences of which they are constituents must be syntactically ill-formed as well (cf. Chapter 2, §4). Similarly, if category mistakes are meaningless, then sentences of which they are constituents must be meaningless too (cf. Chapter 3, §2 and §4). And at least on many versions of the MBT view, if category mistakes are truth-valueless, then so are sentences of which they are constituents. But this flat-footed approach does not do justice to the phenomenon of category mistakes. As we have seen, there are a range of environments where one can felicitously embed a category mistake (‘John said that the number two is green’; ‘Jill dreamt that her toothbrush was pregnant’; ‘Numbers are coloured and the number two is green’). Some of the alternative theories I have discussed do take a more fine-grained approach to the issue of embedding (one according to which not every embedding of a category mistake results in the relevant kind of infelicity). Nevertheless, even those theories are not able to fully deal with the subtleties of the data. For example, on Thomason’s supervaluationist approach, the sentence ‘The number three is blue and the number three is even’ was deemed acceptable (because the second conjunct and hence the entire conjunction is false, rather than truth-valueless); Similarly, on the naïve pragmatic approach the sentence ‘The temperature in London isn’t green and the temperature in London is 5 degrees’ was deemed acceptable (because the second conjunct ensures that the entire conjunction is neither trivially 149

THE PR AGM ATIC APPROACH true nor trivially false). On the other hand, my own account correctly predicts of both of these examples that they are infelicitous (because they generate the presuppositions that the number three is coloured and the temperature in London is coloured respectively). Another crucial observation that we have encountered throughout the book concerns the role of context in generating the infelicity associated with category mistakes. The sentences ‘That is green’ or ‘Jill’s best friend is pregnant’ may be perfectly felicitous in some contexts (e.g. where ‘that’ refers to a chair or where Jill’s best friend is a woman), while they may be deemed to be category mistakes in other contexts (e.g. where ‘that’ refers to the number two, or where Jill’s best friend is a man). As the syntactic structure and meaning of a sentence are context-invariant features of it, both the syntactic approach and the meaninglessness view have a hard time accounting for these observations (cf. Chapter 2, §7, and Chapter 3, §2). As we have seen, the MBT view, which places the phenomenon of category mistakes at the level of content or reference, is better suited to account for the context sensitivity of category mistakes. However, the MBT view relies on the wrong notion of context: the content and referent of demonstratives such as ‘that’ or of definite descriptions such as ‘Jill’s best friend’ relies on a semantic notion of context, one that is determined by the actual facts (e.g. who is in fact Jill’s best friend). On the other hand, whether or not a sentence exhibits the phenomenology typical of category mistake depends on what speakers presuppose to be the facts. Thus for example, against a context where it is taken for granted that Jill’s best friend is a man (even if unbeknownst to the participants in the conversation the friend is a woman), ‘Jill’s best friend is pregnant’ would be infelicitous. Since the presuppositional theory of category mistakes I propose relies on a pragmatic Stalnakerian notion of context rather than on a semantic one, the theory is well-placed to account for such infelicities. The fact that the presuppositional account of category mistakes relies on a pragmatic notion of context (one which ties the context to participants’ beliefs and assumptions), has a further advantage in dealing with the relativity which the phenomenology of category mistakes exhibits: 150

MER ITS OF THE ACCOUNT sentences can exhibit the relevant phenomenology relative to some speakers, while seeming felicitous to other speakers (ones who possess different background information, beliefs or assumptions). Recall, for example, cases of the following sort (cf. Chapter 2, §6): (55) This woman fathered my children. (56) This machine is thinking about the theory of relativity (57) Space is curved.

To sufficiently ignorant speakers, (55)–(57) are likely to exhibit the kind of infelicity associated with category mistakes. But to speakers who realize that this woman might have previously been a man who later underwent a sex-change operation, that this machine might be a sophisticated and intelligent computer that is capable of thinking, or that space has a curvature-value, the sentences can seem entirely acceptable. The crucial point is that any view which explains the phenomenon of category mistakes purely in terms of fixed linguistic features has difficulties accounting for such a diversity of intuitions. If it follows from the rules of our language that ‘curved’ cannot be properly applied to the subject term ‘space’ (whether because such an application would be syntactically ill-formed, meaningless, or truth-valueless) then (57) is a category mistake, even if some speakers who are well-versed in modern physics judge it to be felicitous. If, on the other hand, such an application is after-all properly allowed by the rules of the language, then (57) is not a category mistake, even if most speakers judge it to be highly infelicitous.49 The presuppositional account of category mistakes avoids this dilemma, precisely because the phenomenon is not explained purely 49 Of course, such theories can have some fall-back responses to this problem. They might, for example, maintain that the phenomenology of category mistakes is often associated with sentences that are not ultimately category mistakes (presumably, the theory would then need to be supplemented with a further explanation for the infelicity of these apparent category mistakes). Or they might argue that the word ‘curved’ is ambiguous— with one meaning which does properly apply to ‘space’ and one which doesn’t. But these fall-back responses come at a price, one that my own account avoids paying.

151

THE PR AGM ATIC APPROACH in terms of predetermined linguistic features. While one aspect of the phenomenon may plausibly be labelled ‘linguistic’ (that is, which presuppositions are generated by the respective sentences), another aspect—namely, whether or not the relevant presuppositions are taken for granted in context—depends on the beliefs and assumptions of the participants of the conversation. Just as ‘John stopped smoking’ can be felicitous or infelicitous depending on whether John is taken for granted to have smoked in the past, so (56) can be felicitous or infelicitous depending on whether speakers take it for granted that machines cannot think. The same point plays a role in explaining another aspect of the phenomenology of category mistakes. It has been noted that sentences can exhibit the relevant phenomenology to varying degrees. Consider for example, the following examples, introduced by Drange:50 (58) (59) (60) (61) (62) (63)

Englishmen like coffee better than tea. Squirrels like coffee better than tea. Bacteria like coffee better than tea. Stones like coffee better than tea. Electrons like coffee better than tea. Quadratic equations like coffee better than tea.

Although (58) is obviously false, it is clearly not a category mistake; On the other hand, (63) clearly is a category mistake; Sentences (59)–(62) seem to exhibit the relevant phenomenology to an increasing degree; Accounts of category mistakes that rely merely on fixed linguistic features, at least face the non-trivial task of explaining this gradedness: it is hard to see how (59)–(63) could be syntactically ill-formed or meaningless to varying degrees.51 The current theory, on the other hand, can account for this data rather straightforwardly. The use of ‘likes coffee 50

Drange (1966), p. 16. Note that Chomsky’s notion of ‘degrees of grammaticalness’ (Chomsky (1961), and Chomsky (1965), pp. 148–53) will not help account for this gradedness. On Chomsky’s view sentences are grammatical to varying degrees depending on which sorts of syntactic rules were being violated by the sentence, and this does not seem to apply in the current case. 51

152

MER ITS OF THE ACCOUNT better than tea’ triggers the same form of presupposition in each of (58)–(63)—let us assume this is the presupposition that the subject has some preference-ordering over coffee and tea. In (58), the presupposition (that Englishman have some preference-ordering over coffee and tea) is taken for granted to be true which means that the sentence does not at all seem to be a category mistake. In most contexts, speakers do not take for granted the presupposition generated by each of (59)–(63). But these sentences differ with respect to how easily the relevant presupposition can be accommodated: one would normally be quite open-minded to the possibility that squirrels have a preference for tea over coffee or vice versa; While one normally assumes that no such thing can be true of stones, if Nature were to publish a paper by a leading scientist claiming that there is now compelling evidence that stones have mental states, and in particular preferences between coffee and tea, then perhaps the presupposition can nevertheless be accommodated; And it is much harder to imagine the kind of evidence that would compel one to accept that quadric equations have a preference between coffee and tea.52 Thus, here too, the fact that my account of category mistakes partially involves what speakers believe or assume (a feature that naturally comes in degrees), rather than relying exclusively on standard linguistic features (features that do not naturally come in degrees), affords the current account a flexibility that other accounts lack. The presuppositional account of category mistakes is thus both foundationally solid, as well as highly successful in accounting for the subtle complexities of the phenomenon of category mistakes.

52 In fact, it seems that having varying degrees of infelicity depending on how easy or hard it is to accommodate the relevant presuppositions is a general feature of presupposition accommodation. Consider for example the increasing infelicity of saying out of the blue: ‘I’m bringing my sister/my Nobel prize/my 6-legged son/my talking donkey to the office tomorrow’.

153

THE PR AGM ATIC APPROACH

§6

Postscript: Some Final Reflections on the Implications of the Account

As has already been noted, the discussion throughout the book has implications to various foundational and methodological questions in the philosophy of language, in particular concerning the interface between syntax, semantics, and pragmatics. To bring back a few examples: on the foundational side, we have seen that one can plausibly accept strong principles of compositionality that go hand in hand with the view that syntactic well-formedness is sufficient for meaningfulness. We have also seen that theories of meaning which deem category mistakes to be meaningless ought to be rejected. On the methodological side, we have observed that a helpful test in deciding whether a phenomenon applies at the level of meaning, is to replace some relevant expression with one that has the same content but a different meaning. So, for example, the recognition that relative to contexts in which ‘that’ and ‘two’ have the same content, ‘Two is green’ and ‘That is green’ are both deemed to be category mistakes, suggests that the phenomenon should not be explained at the level of meaning. Similarly, the fact that with ‘The man is pregnant’, replacing the syntactically simple ‘man’ with the syntactically complex ‘the person who I proved to Jane wasn’t a woman’ still results in a category mistake, makes trouble for a syntactic account of category mistakes. We have also noted that in deciding whether a context-sensitive phenomenon is ultimately semantic or pragmatic, it is instructive to consider whether the notion of context in play is a semantic one (i.e. depends on the actual facts) or a pragmatic one (i.e. depends on what participants take for granted in the conversation). With my positive account of category mistakes in place, it is worth highlighting three further areas to which my account has some implications. First, it is important to realize that my account entails that the phenomenon of presupposition is much more prevalent than is commonly assumed: It follows from my account that practically any predicative expression is a presupposition trigger. This is bound to have repercussions

154

POSTSCR IPT: SOME FINAL R EFLECTIONS to the triggering problem of presuppositions. Those who maintain that presupposition triggers ought to be explained rather than stipulated53 might conclude from this prevalence—especially if it the various triggers are deemed to be uniform across a wide range of languages54—that addressing the triggering problem is all the more pressing (though, of course, the wide range of presupposition triggers stipulated by my account also makes the triggering problem considerably more difficult). Conversely, those who insist that presupposition triggers cannot be ultimately explained and must be lexically encoded, might take my account to show that a two-dimensional theory of meaning (along the lines proposed by Karttunen and Peters55) is not as ad-hoc as it might perhaps initially seem: It may be theoretically unappealing to posit a second dimension of meaning if this dimension is introduced only in order to accommodate the semantics of a very small number of words such as ‘too’, ‘stop’, or ‘know’, but much less so if a very wide range of lexical items receive a non-trivial presuppositional component of meaning. Either way, if my account is correct, the wide-range of presupposition triggers it posits must be taken into account in the debate surrounding the triggering problem. Second, my account also has some implications to issues concerning other linguistic phenomena. Metaphor, metonymy, and fictional discourse are three linguistic settings which often involve category mistakes. 53

See e.g. Levinson & Annamalai (1992), Simons (2001), and Abrusán (2011b). It is interesting to consider whether all presuppositions triggered in the context of category mistakes are in fact cross-linguistic. Consider the various fruit-picking verbs in Hebrew, discussed in Chapter 2, §3. The verb ‘livtzor’ for example, only felicitously applies to grapes. Now, it is clear that English does not have an analogous verb. A more interesting question, though, is whether the truth-conditional contribution of ‘livtzor’ is ‘to pick grapes’ or merely ‘to pick’ (with grapes figuring only in the presupposition of the verb). If we adopt the latter approach, then we get the intriguing result that English and Hebrew have two verbs with the same truth-conditional meaning but different presuppositional properties. However, I am not sure which of the two proposals is correct, or even how to decide between them. 55 Although, as I have noted before, it is not entirely clear that Karttunen & Peters (1979) should really be interpreted as taking this anti-explanatory line. An alternative interpretation is to merely treat their theory as a descriptive device, which can then be supplemented with a pragmatic account of triggering. 54

155

THE PR AGM ATIC APPROACH (Consider ‘The silence is liquid’ intended metaphorically; ‘The ham sandwich left without paying’, uttered by a waiter in reference to the customer who ordered the sandwich; and ‘The tree was sleeping’ as told in a fairy-tale). It is worth observing that certain accounts of each of these linguistic phenomena require the relevant sentences to have a literal meaning or even a literal content. For example, a Gricean theory of metaphor or of metonymy requires the relevant sentences to literally express a proposition, which can then serve to generate conversational implicatures. Or consider a view of fictional discourse which posits a propositional operator, ‘fictionally’, so that when ‘The tree was sleeping’ is uttered in the context of a fiction, the sentence expresses the proposition that fictionally, the tree was sleeping. But if ‘fictionally’ is a propositional operator it requires a proposition, and the claim that fictionally, the tree was sleeping can only be true if the embedded statement expresses a (literal) content. The upshot is that if category mistake are meaningless or even fail to express propositions, then the above-mentioned theories of these linguistic phenomena are all ruled out.56 On the other hand, if my account is correct then category mistakes are both meaningful and contentful, and a much wider range of theories remain as open possibilities. Finally, one might wonder what implications my account has to questions in metaphysics. The modern interest in category mistakes started with the expectation that the concept will play a key role in metaphysics. My pragmatic account of category mistakes does not lend support this hope: it is highly doubtful that the presuppositions associated with category mistakes reveal anything about the fundamental nature of ontological categories. Nevertheless, my account might still leave room for the notion of category mistakes to play a more modest role in metaphysical theorizing. An extremely common style of argument in metaphysics relies on Leibniz’s Law:57 if a and b differ with respect to some property, then they 56 See Stern (2006), pp. 252–3 for the suggestion that some such theories of metaphor ought to be rejected because category mistakes are meaningless. 57 For a detailed discussion of uses of arguments by Leibniz’s Law in metaphysics see Magidor (2011).

156

POSTSCR IPT: SOME FINAL R EFLECTIONS are not identical. This suggests a template for a wide-range of arguments in metaphysics. For example, one might argue that a statue is distinct from the coincident lump of clay, because the statue is Romanesque, but the lump is not Romanesque. Or one might argue that I must be distinct from my body, because I like philosophy, but my body does not like philosophy. There is range of strategies available for those who wish to resist such arguments. One such strategy involves insisting that the argument in question relies on false premise, albeit one which seems to be true. In particular, one might argue that, for example, ‘The lump of clay is Romanesque’ is true but pragmatically misleading, and hence speakers mistakenly take it to be false. Consequently, speakers mistakenly infer that the negation of the sentence (‘The lump of clay is not Romanesque’) must be true, and can thus figure as a legitimate premise in the argument.58 It is clear that since many of the relevant sentences (‘The lump of clay is Romansque’, ‘My body likes philosophy’, etc.) are category mistakes, a pragmatic account of the phenomenon is key to pursuing this strategy, though the details are somewhat delicate.59 The presuppositional account I defend thus has a range of implications in a variety of areas of philosophy and linguistics. But most importantly, it provides a compelling explanation for the significant and intriguing phenomenon of category mistakes. 58 Compare this with the direct-referentialist strategy of claiming that ‘Lois Lane believes that Clark Kent can fly’ is true but pragmatically misleading, which explains why speakers mistakenly take it to be false, and correspondingly, take ‘Lois Lane does not believe that Clark Kent can fly’ to be true. 59 See Magidor (2011), §4 for a detailed discussion.

157

This page intentionally left blank

REFERENCES

Abbott, B. (2006): ‘Where have some of the presuppositions gone?’, in Betty Birner & Gregory Ward (eds.), Drawing the Boundaries of Meaning: Neo-Gricean Studies in Pragmatics and Semantics in Honor of Laurence R. Horn, Philadelphia: John Benjamins, pp. 1–20. Abrusán, M. (2011a): ‘Triggering verbal presuppositions’, Proceedings of SALT 20: 684–701. —— (2011b): ‘Predicting the presuppositions of soft triggers’, Linguistics and Philosophy 34: 491–535. Arad, M. (1996): ‘A minimalist view of the syntax-lexical semantics interface’, UCL Working Papers Linguistics 8: 215–42. Aristotle (1984): The Complete Works of Aristotle, edited by Barnes, J., New Jersey: Princeton University Press. Armstrong, S., Gleitman, L., and Gleitman, H. (1983): ‘What some concepts might not be’, Cognition 13: 263–308. Asher, N. (2011): Lexical Meaning in Context: a Web of Words, Cambridge: Cambridge University Press. Bach, K. (1999): ‘The semantics pragmatics distinction: what it is and why it matters’, in Turner, K. (ed.) The Semantics-Pragmatics Interface from Different Points of View, Oxford: Elsevier, pp. 65–84. Baker, A.J. (1956): ‘Category mistakes’, Australasian Journal of Philosophy 34:13–26. Baker, M. (1997): ‘Thematic roles and syntactic structure’, in Liliane Haegeman (ed.), Elements of Grammar, Dordrecht: Kluwer, pp. 73–137. Barwise, J., & Perry, J. (1981): ‘Semantic innocence and uncompromising situations’, Midwest Studies in Philosophy 6: 387–404. Beall, J.C., & van Fraassen, B.C. (2003): Possibilities and Paradox, Oxford: Oxford University Press. Beaver, D.I. (1997): ‘Presupposition’, in Johan van Benthem and Alice ter Meulen (eds.), Handbook of Logic and Language, Amsterdam: Elsevier, pp. 939–1008. —— (2001): Presupposition and Assertion in Dynamic Semantics, Stanford: CSLI Press. Benacerraf, P. (1965): ‘What numbers could not be’, Philosophical Review 74: 47–73. Bergmann, M. (1977): ‘Logic and sortal incorrectness’, Review of Metaphysics 31: 61–79. Bezuidenhout, A. (2001): ‘Metaphor and what is said: a defence of a direct expression view of metaphor’, Midwest Studies in Philosophy 25: 156–86.

159

R EFER ENCES Black, M. (1962): ‘Metaphor’, in Black, M., Models and Metaphors, Ithaca: Cornell University Press, pp. 25–47. Block, N. (1998): ‘Semantics, conceptual role’, in E. Craig (ed.), Routledge Encyclopaedia of Philosophy, London: Routledge. Retrieved December 1, 2006, from: . Boisvert, D., & Lubbers, C. (2003): ‘Frege’s commitment to an infinite hierarchy of senses’, Philosophical Papers 32: 31–64. Bouchard, D. (1995): The Semantics of Syntax: A Minimalist Approach to Grammar, Chicago: University of Chicago Press. Brady, R., & Routley, R. (1973): ‘Don’t care was made to care’, Australasian Journal of Philosophy 51: 211–25. Camp, E. (2004): ‘The generality constraint and categorical restrictions’, Philosophical Quarterly 54: 210–31. —— (2005): ‘Critical Study of Joseph Stern, Metaphors in Context’, Nous 39: 715–31. —— (2006a): ‘Metaphor in the mind: the cognition of metaphor’, Philosophy Compass 1:154–70. —— (2006b): ‘Contextualism, metaphor, and what is said’, Mind & Language 21: 280–309. Campbell, J. (1993): ‘A simple view of colour’, in John Haldane & Crispin Wright (eds.), Reality: Representation and Projection, Oxford: Oxford University Press, pp. 257–68. Carnie, A. (2002): Syntax: A Generative Approach, Oxford: Blackwell. —— (2011): Modern Syntax: A Course Book, Cambridge: Cambridge University Press. Carroll, J. (1994): Laws of Nature, Cambridge: Cambridge University Press. Chemla, E. (2009): ‘Presuppositions of quantified sentences: experimental data’, Natural Language Semantics 17: 299–340. Chomsky, N. (1957): Syntactic Structures, The Hague: Mouton. —— (1961): ‘Some methodological remarks on generative grammar’, Word 17: 219–39. Reprinted in Fodor & Katz (1964). —— (1965): Aspects of the Theory of Syntax, Cambridge, MA: MIT press. —— (1975): The Logical Structure of Linguistic Theory, New York: Plenum Press. Cooper, R. (1983): Quantification and Syntactic Theory, Dordrecht: Reidel. Copi, I. (1971): The Theory of Logical Types, London: Routledge & Kegan. Davidson, D. (2001a): ‘On saying that’, in Davidson, D., Inquiries into Truth and Interpretation, Oxford: Clarendon Press, pp. 93–108. —— (2001b): ‘What metaphors mean’, in Davidson, D., Inquiries into Truth and Interpretation, Oxford: Clarendon Press, pp. 245–64.

160

R EFER ENCES Denham, K., & Lobeck, A. (2010): Linguistics for Everyone, Boston: Wadsworth. Diamond, C. (2001): ‘What nonsense might be’, in Diamond, C., The Realistic Spirit, Cambridge, MA: MIT press. Drange, T. (1966): Type Crossings, The Hague: Mouton. Dummett, M. (1981): Frege: Philosophy of Language (2ed ed.), London: Duckworth. Evans, G. (1982): The Varieties of Reference, Oxford: Clarendon Press. Ewing, A.C. (1937): ‘Meaningfulness’, Mind 46: 347–64. Fara, D. (2004): ‘Gap principles, penumbral consequence and infinitely higherorder vagueness’, in J.C. Beall (ed.), Liars and Heaps, Oxford: Oxford University Press, pp. 195–221. Originally published under ‘Delia Graff’. Field, H. (2003): ‘No fact of the matter’, Australasian Journal of Philosophy 81: 457–80. Fine, K. (2003): ‘The non-identity of a material thing and its matter’, Mind 112: 195–234. Fodor, J., & Katz, J. (1963): ‘The structure of a semantic theory’, Language 39: 170–210. Reprinted in Fodor & Katz (1964). Fodor, J., & Katz, J., eds. (1964): The Structure of Language: Readings in the Philosophy of Language, New Jersey: Prentice Hall. Fodor, J.D. (1977): Semantics: Theories of Meaning in Generative Grammar, New York: Harvester Press. Fogelin, R. (1988): Figuratively Speaking, New Haven: Yale University Press. Fox, D. (2008): ‘Two short notes on Schlenker’s theory of presupposition projection’, Theoretical Linguistics 34: 237–53. Fox, D. (MS1) ‘Presupposition, trivalence, and relevance’, HUJI and TAU syntax and semantics reading group 2011 (available on ). Fox, D. (MS2) ‘Presupposition projection from quantificational sentences: trivalence, local accommodation, and presupposition strengthening’, unpublished manuscript, Hebrew University of Jerusalem and MIT. Frege, G. (1952): ‘On sense and reference’, in Max Black and Peter Geach (eds.), Translations from the Philosophical Writings of Gottlob Frege, Oxford: Blackwell, pp. 56–78. Fromkin, V., ed. (2000): Linguistics: An Introduction to Linguistic Theory, Oxford: Blackwell. Geurts, B. (1996): ‘Local satisfaction guaranteed: a presupposition theory and its problems’, Linguistics and Philosophy 19: 259–94. Givón, T. (2001): Syntax: An Introduction (vol. 1), Philadelphia: John Benjamins. Glanzberg, M. (2004): ‘Against truth-value gaps’, in in J.C. Beall (ed.), Liars and Heaps, Oxford: Oxford University Press, pp. 151–94.

161

R EFER ENCES Goddard, L. (1966): ‘Predicates, relations, and categories’, Australasian Journal of Philosophy 44: 139–71. —— (1968): ‘Towards a logic of significance’, Notre Dame Journal of Formal Logic 9: 219–41. Goldstick, D. (1974): ‘Against ‘Categories’’, Philosophical Studies 26: 337–56. Grice, P. (1989): Studies in the Way of Words, Cambridge, MA: Harvard University Press. Grimshaw, J. (1979): ‘Complement selection and the lexicon’, Linguistic Inquiry 10: 279–326. —— (2005): Words and Structure, Stanford: CSLI. Haack, R.J. (1971): ‘No need for nonsense’, Australasian Journal of Philosophy 49:71–7. —— (1975): ‘Still No need for nonsense’, Australasian Journal of Philosophy 53:161–4. Hallden, S. (1949): The Logic of Nonsense, Uppsala: Uppsala Universitets Arsskrift. Harris, R.A. (1993): The Linguistics Wars, Oxford: Oxford University Press. Hawthorne, J. & Magidor, O. (2009): ‘Assertion, context, and epistemic accessibility’, Mind 118: 377–97. Heim, I. (1983): ‘On the projection problem for presupposition’, Proceedings of the Second Annual West Coast Conference on Formal Linguistics: 114–26. Reprinted in Portner & Partee (2002). —— (1990): ‘Presupposition projection’, in Rob van der Sandt (ed.), Reader for the Nijmegen Workshop on Presupposition, Lexical Meaning and Discourse Processes, University of Nijmegen. Heim, I., & Kratzer, A. (1998): Semantics in Generative Grammar, Oxford: Blackwell. Hempel, C.G. (1950): ‘Problems and changes in the empricist criterion of meaning’, Revue Internationale de Philosophie, 41: 41–63. Hodges, W. (2001): Logic, London: Penguin. Horn, L.R. (1985): ‘Metalinguistic negation and pragmatic ambiguity’, Linguistics 61: 121–74. —— (2001): A Natural History of Negation, Stanford: CLSI Press. Horwich, P. (1990): Truth, Oxford: Blackwell. Hughes, G., & Cresswell, M. (2003): A New Introduction to Modal Logic, London: Routledge. Jackendoff, R.S. (1972): Semantic Interpretation in Generative Grammar, Cambirdge, MA: MIT Press. Johnson-Laird, P.N. (1983): Mental Models, Cambridge: Cambridge University Press. Karttunen, L., & Peters, S. (1979): ‘Conventional implicatures’, Syntax and Semantics 11: 1–56.

162

R EFER ENCES Katz, J.J. (1964): ‘Semi-sentences’, in Fodor & Katz (1964), pp. 400–16. —— (1970): ‘Interpretative semantics vs. Generative semantics’, Foundations of Language 6: 220–59. King, J. & Stanley, J. (2005): ‘Semantics, pragmatics and the role of semantic content’, in Zoltan Szabo (ed.), Semantics vs. Pragmatics, Oxford: Oxford University Press, pp. 111–64. Kratzer, A. (1991): ‘Modality’, in Arnim von Stechow and Dieter Wunderlich (eds.), Semantik: Ein internationales Handbuch der zeitgenössischen Forschung, Berlin: De Gruyter, pp. 639–50. Lakoff. G. (1971): ’Pressuposition and relative well-formedness’, in Leon Jakobovitz and Danny Steinberg (eds.), Semantics: an Interdiciplinary Reader, Cambridge: Cambridge University Press, pp. 329–40. Lambert, K. (1968): ‘On the no type theory of significance’, Australasian Journal of Philosophy 46: 79–86. Lappin, S. (1981): Sorts, Ontology, and Metaphor, Berlin & New York: de Gruyter. Levin, B., & Rappaport Hovav, M. (1995): Unaccusativity: At the Syntax-Lexical Semantics Interface, Cambridge, MA: MIT Press. —— (2005): Argument Realization, Cambridge: Cambridge University Press. Levinson, S., & Annamalai, E. (1992): ‘Why presuppositions aren’t conventional’, in Ashok Klekar and Ravindranatha Srivastava (eds.), Language and text: Studies in honour of Ashok R. Kelkar, Delhi: Kalinga, pp. 227–42. Ludlow, P. (2011): The Philosophy of Generative Linguistics, New York: Oxford University Press. Lyons, J. (1995): Linguistic Semantics: an Introduction, Cambridge: Cambridge University Press. Maddy, P. (1980): ‘Perception and mathematical intuition’, Philosophical Review 89:163–96. Magidor, O. (2007): Category Mistakes, D.Phil dissertation, University of Oxford. —— (2009a): ‘The last dogma of type confusions’, Proceedings of the Aristotelian Society 109:1–29. —— (2009b): ‘Category mistakes are meaningful’, Linguistics & Philosophy 32: 553–81. —— (2010): ‘Natural language and how we use it: psychology, pragmatics, and presupposition’, Analysis 70: 160–74. —— (2011): ‘Arguments by Leibniz’s Law in metaphysics’, Philosophy Compass 6:180–95 Martin, J. (1975): ‘A many valued semantics for category mistakes’, Synthese 31: 63–83.

163

R EFER ENCES Martinich, A.P. (1984): ‘A theory of metaphor’, Journal of Literary Semantics 13: 35–56. McCawley, J. (1968): ‘The role of semantics in grammar’, in Emmon Bach and Robert Harms (eds.), Universals in Linguistic Theory, New York: Holt, pp. 124–69. —— (1970): ‘Where do noun phrases come from?’, in Roderick Jacobs and Peter Rosenbaum (eds.), Readings in English and Transformational Grammar, Washington: Georgetown University Press, pp. 166–83. —— (1971): ‘Interpretative semantics meets Frankenstein’, Foundations of Language 7: 285–96. McDaniel, K. (MS): ‘Categories of being’, unpublished manuscript, Syracuse University. Montague, R. (1970): ‘Universal Grammar’, Theoria 36: 373–98. —— (1974): ‘The proper treatment of quantification in ordinary English’, in Richmond Thomason (ed.), Formal Philosophy: Selected Papers by Richard Montague, New Haven: Yale University Press, pp. 247–70. Moran, R. (1997): ‘Metaphor’, in Bob Hale and Crispin Wright (eds.), A Companion to the Philosophy of Language, Oxford: Blackwell, pp. 248–68. Pap, A. (1960): ‘Types and meaninglessness’, Mind 69: 41–54. Parsons, T. (1981): ‘Frege’s hierarchies of indirect senses and the paradox of analysis’, Midwest Studies of Philosophy 6: 35–57. —— (1995): ‘Thematic relations and arguments’, Linguistic Inquiry 26: 635–62. Partee, B. (interview): ‘Is this a good time for formal semantics?’, video interview, retrieved on March 1, 2011 from: Portner, P., & Partee, B., eds. (2002): Formal Semantics: The Essential Readings, Oxford: Blackwell. Prior, A.N. (1954): ‘Entities’, Australasian Journal of Philosophy 32: 159–68. Quine, W.V.O. (1953): ‘Mr. Strawson on logical theory’, Mind 62: 433–51. —— (1960): Word and Object, Cambridge, MA: MIT Press. Recanati, F. (2001): ‘Literal/nonliteral’, Midwest Studies in Philosophy 25: 264–74. Reimer, M., & Camp, E. (2006): Metaphor’, in Ernest Lepore and Barry Smith (eds.) (eds.), The Oxford Handbook of Philosophy of Language, Oxford: Oxford University Press, pp. 845–63. Resnik, P.S. (1993): Selection and Information: a Class Based Approach to Lexical Relationships, PhD dissertation, University of Pennsylvania. Rothschild, D. (2007): ‘Presuppositions and scope’, Journal of Philosophy 104:71–106.

164

R EFER ENCES Rothschild, D. (2008): ‘Presupposition projection and logical equivalence’, Philosophical Perspectives 22: 473–97. Routley, R. (1966): ‘On significance theory’, Australasian Journal of Philosophy 44: 172–209. —— (1969): ‘The need for nonsense’, Australasian Journal of Philosophy 47: 367–84. Russell, B. (1903): The Principles of Mathematics, London: Routledge. —— (1905): ‘On denoting’, Mind 14: 530–8. —— (1908): ‘Mathematical logic as based on the theory of types’, American Journal of Mathematics 30: 222–62. Ryle, G. (1938): ‘Categories’, Proceedings of the Aristotelian Society 38: 189–206. —— (1949): The Concept of Mind, London: Hutchinson’s University Library. Sauerland, U., & von Stechow, A. (2001): ‘The syntax-semantics interface’, in Neil Smelser & Paul Baltes (eds.), International Encyclopaedia of the Social & Behavioural Sciences, Oxford: Pergamon, pp. 15412–18. Sauerland, U., & Stateva, P. (2007): ‘Introduction’, in Uli Sauerland and Penka Stateva (eds.), Presupposition and Implicature in Compositional Semantics, Basingstoke: Palgrave Macmillan, pp. 1–11. Schlenker, P. (2008a): ‘Be articulate: a pragmatic theory of presupposition projection’, Theoretical Linguistics 34: 157–212. —— (2008b): ‘Presupposition projection: explanatory strategies’, Theoretical Linguistics 34: 287–316. —— (2008c): ‘Presupposition projection: the new debate’, Proceedings of SALT 18: 655–92. —— (2010): ‘Presuppositions and local contexts’, Mind 119: 377–91. —— (2011): ‘The proviso problem: a note’, Natural Language Semantics 19: 395–422. Schnieder, B. (2006): ‘By Leibniz’s Law: remarks on a fallacy’, Philosophical Quarterly 56: 39–54. Searle, J. (1979): ‘Metaphor’, in Andrew Ortony (ed.), Metaphor and Thought, Cambridge: Cambridge University Press, pp. 83–111. Seuren, P. (1988): ‘Lexical meaning and presupppsition’, in Werner Hiillen and Rainer Schulze (eds.), Understanding the Lexicon. Meaning, Sense and World Knowledge in Lexical Semantics, Tubingen: Niemeyer, pp. 170–87. Shapiro, S. (1997): Philosophy of Mathematics: Structure and Ontology, Oxford: Oxford University Press. Simons, M. (2001): ‘On the conversational basis of some presuppositions’, Proceedings of SALT 11: 431–48. —— (2006): ‘Foundational issues in presupposition’, Philosophy Compass 1: 357–72.

165

R EFER ENCES Smart, J.J. (1953): ’A note on categories’, British Journal for the Philosophy of Science 4: 227–28. Smiley, T. (1960): ‘Sense without denotation’, Analysis 20: 125–35. Soames, S. (1979): ‘A projection problem for speaker presuppositions’, Linguistic Inquiry 10: 623–66. Reprinted in Soames (2009). —— (1989): ‘Presupposition’, in Gabbay, D., and Guenthner, F. (eds.), Handbook of Philosophical Logic, vol. 4, pp. 553–16. Reprinted in Soames (2009). —— (1999): Understanding Truth, Oxford: Oxford University Press. —— (2009): Philosophical Essays vol. 1, New Jersey: Princeton University Press. Sommers, F. (1963): ‘Types and ontology’, Philosophical Review 72: 327–63. Sorensen, R. (2001): Vagueness and Contradiction, Oxford: Oxford University Press. Stalnaker, R. (1973): ‘Presuppositions’, Journal of Philosophical Logic 2: 447–57. —— (1976): ‘Possible worlds’, Nous 10: 65–75. —— (1978): ‘Assertion’, Syntax and Semantics 9: 315–32. Reprinted in Stalnaker (1999). —— (1999): Context and Content, Oxford: Oxford University Press. Stern, J. (2000): Metaphor in context, Cambridge, MA: MIT Press. —— (2006): ‘Metaphor, literal, literalism’, Mind & Language 21: 243–79. Steward, H. (1997): The Ontology of Mind, Oxford: Oxford University Press. Strawson, P.F. (1950): ‘On referring’, Mind 59: 320–44. —— (1952): Introduction to Logical Theory, London: Methuen. —— (1970): ‘Categories’, in Oscar Wood, and George Pitcher (eds.), Ryle: A Collection of Critical Essays, New York: Doubleday, pp. 181–212. Thomason, R. (1972): ‘A semantic theory of sortal incorrectness’, Journal of Philosophical Logic 1: 209–58. van Fraassen, B. (1966): ‘Singular terms, truth-value gaps and free logic’, Journal of Philosophy 63: 481–95. —— (1967): ‘Meaning relations among predicates’, Nous 1: 161–79. van Valin, R. (2001): An Introduction to Syntax, Cambridge: Cambridge University Press. von Fintel, K. (2000): ‘What is presupposition accommodation’, unpublished manuscript, MIT. —— (2004): ‘Would you believe it? The king of France is back! (Presuppositions and truth-value intuitions)’, in Marge Reimer and Anne Bezuidenhout (eds.), Descriptions and Beyond, Oxford: Clarendon Press, pp. 315–41. Westerhoff, J. (2005): Ontological Categories, Oxford: Oxford University Press. Williamson, T. (1994): Vagueness, London: Routledge. —— (2000): Knowledge and its Limits, Oxford: Oxford University Press. Ziff, P. (1964): ‘On understanding “understanding utterances” ’, in Fodor & Katz (1964), pp. 390–9.

166

I NDEX

and metaphor 13, n. 39, 66–74, 77 n. 60, 78, 148, 155–6 and translation 33, 58–9, 97–8, 148 atomic sentences 3, 7 n. 13, 12, 46–8, 57 n. 29, 95–8, 106, 117–8, 139, 145–6, 149–50 characterization of 3–4, 146–8 complex sentences 3, 56–7, 96–7, 106–9, 114–6, 130 n. 29, 134–40, 146, 149 conditionals 107, 135–6, 139, 143–4 conjunctions 56–7, 62 n. 35, 107–8, 114–6, 136–8, 145–6, 149–50 context-sensitivity of 3, 18, 23, 41–2, 54–5, 81, 85–6, 92–3, 113, 132–4, 143–4, 150–2, 154 counterfactuals 115, 135 n. 36, 143 definite descriptions 42, 53–5, 85–6, 88–9, 92–4, 113, 133–4 demonstratives 3, 23, 39–41, 54, 56, 150–1, 154 disjunctions 3, 107–9, 114–5, 138, 143–4 embedding 38–9, 59–66, 84, 134–40, 149–50 falsidal view 12–3, 94 n. 23, 95–9 in computational linguistics 20–1 in different languages 3, 33–6, 45 n. 7, 58–9, 60, 73 n. 54, 96–8, 155 in metaphysical debates 11–2, 13 n. 39, 156–7 infelicity of 1–5, 25, 31–3, 35–42, 44, 51 n. 17, 56, 75, 80–1, 91–3, 95–8, 105 n. 40, 107–16, 132–53 negations 3, 11, 94–7, 105, 134–5, 139, 146 no-type view 12–3, 95 n. 26, 98 (see also falsidal view) predicates 2, 11, 20, 26–7, 39–40, 43, 47–56, 71, 84–5, 93 n. 22, 95 n. 26, 98, 100–4, 131–2, 139, 140–1, 144–5, 147, 154–5

Abbott, B. 20 n. 59, 118 n. 10, 125 n. 21, 144 n. 45 Abrusán, M. 19 n. 54, 130 n. 28, 155 n. 53 ambiguity 5, n. 9, 16–7, 20–1, 51, 70, 72 n. 50, 151 n. 49 Annamalai, E. 155 n. 53 Arad, M. 32 n. 14, 40 n. 31 arbitrariness 29–30, 34, 95–9, 101, 105–9 argument realization 28–30 Aristotle 7 Armstrong, S. 62 n. 36 Asher, N. 13 n. 40, 44 n. 6, 61 n. 33, 76 n. 59 assertion 87 n. 12, 112, 117–20,124–9 Bach, K. 110 n. 1 Baker, A.J. 11 n. 25, 29 n. 9 Barwise, J. 63 n. 37 Beall, J.C. 14 n. 41, 44 Beaver, D.I. 20 n. 59, 116 n. 9, 123, 141 n. 41, 144 n. 45 Benacerraf, P. 2 n. 3, 44 n. 6 Bergmann, M. 13 n. 37 Bezuidenhout, A. 74 n. 55 Black, M. 68 n. 44, 68 n. 45, 69 n. 48 Block, N. 77 n. 61 Boisvert, D. 65 n. 38 Bouchard, D. 19 n. 56 Brady, R. 12, 95 n. 25, 96 n. 31, 99 n. 34 Camp, E. 13 n. 40, 46 n. 11, 67 n. 42, 68, 72 n. 50, 73 n. 54, 74 n. 55, 77 n. 60 Campbell, J. 4 n. 5 Carnie, A. 19 n. 55, 19 n. 56, 19 n. 58, 28 n. 6, 44 n. 6 Carroll, J. 4 n. 5 category mistakes adjectives 2, 26–7, 36, 145 adverbs 2, 50–1, 62 n. 63, 140–1, 144–5, 147

167

INDE X category mistakes (cont.) prepositions 2, 140–1, 145 quantifier phrases 56–7, 138–9, 156–7, 122–3, 138–9, 146 questions 138 verbs 2, 20, 26–30, 35–6, 38 n. 26, 57 n. 29, 70–2, 145–6, 155 n. 54 Chemla, E. 123 n. 17 Chomsky, N. 5 n. 9, 15–20, 25–8, 31 n. 13, 32 n. 15, 33–5, 37–8, 44, 152 n. 51 Colourless green ideas sleep furiously 1–2, 5 n. 9, 15, 44, 113, 140–1, 145, 147 compositionality 5, 45–57, 64, 67, 78, 84, 125–6, 148, 154 concepts 60–2 conceptual analysis 3–4, 146–8 conditional proof 106, 136 n. 38 context context change potentials (CCPs) 127–30 context-sensitivity 71–4, 110 n. 2, 154 (see also category mistakes – context sensitivity of ) notion of 112, 124, 130 n. 29, 132–3, 150, 154 conventional implicatures 110–1, 118 n. 10, 125 n. 21 conversational implicatures 72–3, 156 Cooper, R. 37 n. 25 Copi, I. 8 n. 15 Cresswell, M. 87 n. 14 Davidson, D. 57 n. 29, 63 n. 37, 65–6, 68–9, 70 n. 49, 73 deep structure 15–7, 25–6 Denham, K. 19 n. 56 Diamond, C. 14 n. 41, 44 n. 6, 75 n. 56 disambiguation 16–7, 20, 70 (see also ambiguity) Drange, T. 12, 44 n. 6, 152 Dummett, M. 65 n. 38 dynamic typing 21–3 Evans, G. 47 n. 12 Ewing, A.C. 11–2

Fara, D. 106 n. 41 fictional discourse 147 n. 47, 155–6 Field, H. 89 n. 17 Fine, K. 13 n. 39, 43 Fodor, J. 16–7, 20–1 Fodor, J.D. 32 n. 16, 39 Fogelin, R. 68 n. 45 Fox, D. 123 n. 17, 131 n. 30 Frege, G. 54, 63–4, 84 Fromkin, V. 19 n. 56, 19 n. 58 functions 8, 21, 36, 48–57, 85–6, 100–6, 147 partial 52–6, 85, 101 generality constraint 47 n. 12 generative semantics 18 Geurts, B. 121 n. 14 Givón, T. 19 n. 55, 28 n. 6, 44 n. 6 Glanzberg, M. 87 n. 12 Goddard, L. 12, 13 n. 37, 82 n. 3, 106 n. 42 Goldstick, D. 11 n. 29, 12 n. 36 grammaticality 8–9, 11, 15, 17, 25–42, 46–7, 71–2, 73 n. 54, 93 n. 21, 110–1, 152, 154 Grice, P. 68, 72–3, 111–2, 114–5, 156 Grice’s theory of conversation 111–2, 114–5 Grimshaw, J. 28 n. 6, 36 n. 23, 38 n. 27, 40 n. 31 Haack, R.J. 11 n. 29, 12, 95 n. 25, 99 n. 34 Hallden, S. 82 n. 2, 82 n. 3 Harris, R.A. 18 n. 51 Hawthorne, J. 92 n. 20 Heim, I. 37 n. 25, 50 n. 16, 53–4, 85 n. 10, 116 n. 9, 123, 126 n. 22, 127–31 Hempel, C.G. 107 n. 44 Hodges, W. 44 n. 6 Horn, L.R. 19 n. 54, 96, 121 Horwich, P. 87 n. 13 Hughes, G. 87 n. 14 indirect speech 60, 65–6, 72 n. 50 infelicity 1 n. 1, 18, 110–1, 117–20, 152 (see also category mistakes – infelicity of)

168

INDE X information theory 20 interpretative semantics 17–8 Jackendoff, R.S. 25 n. 1, 37 n. 25, 38 n. 27, 39 n. 29, 40 n. 31 Johnson-Laird, P.N. 15 n. 42, 40 n. 31 Julius Caesar problem 2 n. 3 Karttunen, L. 125–32, 136 n. 37, 145, 155 Katz, J.J. 16–7, 18 n. 51, 20–1, 25 n. 2 kernel sentences 15–7 King, J. 110 n. 1 Kratzer, A. 37 n. 25, 50 n. 16, 53–4, 85 n. 10, 143 n. 42 Lakoff, G. 18, 37 n. 25, 62 n. 36 Lambert, K. 11 n. 29, 12, 95 n. 26 Lappin, S. 13, 44 n. 6, 56 n. 25, 75 n. 57, 82 n. 3, 82 n. 6, 93 n. 22, 100 n. 37, 106 n. 42 law of excluded middle 102 Leibniz’s Law 156–7 Levin, B. 29 n. 8, 30 n. 10, 30 n. 11, 34 n. 20 Levinson, S. 155 n. 53 Lewis, D.K. 115 n. 8 lexical categories 34–5 liar paradox 4 n. 6, 8–9, 90 n. 18 literal meaning 45, 66–74, 79, 110 n. 2, 156 Lobeck, A. 19 n. 56 logic four-valued 13 n. 37, 82 intensional 99–106 Strong Kleene 57 n. 28, 85 n. 10 supervaluationist 13 n. 37, 81, 82, 85 n. 10, 99–109, 136–7, 149 three-valued 13 n. 37, 57 n. 28, 82, 85 n. 10, 131 n. 30 Weak Kleene 57 n. 28, 85 n. 10 logical positivism 9, 77, 107 n. 44 Lubbers, C. 65 n. 38 Ludlow, P. 14 n. 41, 44 n. 6 Lyons, J. 19 n. 55, 44 n. 6 Maddy, P. 61 Magidor, O. 6 n. 11, 13 n. 39, 13 n. 40, 32, 43 n. 1, 92 n. 20, 93 n. 21, 122

n. 16, 125 n. 20, 147, 156 n. 57, 157 n. 59 Martin, J. 13 n. 37, 82 n. 2, 82 n. 4, 82 n. 6, 91–9 Martinich, A.P. 72 n. 51 MBT view of category mistakes 7, 23, 80–109, 130, 136–7, 149–50, 154 McCawley, J. 18, 35 n. 21, 36–7, 38 n. 27, 40 n. 31, 44 n. 6, 59 n. 31 McDaniel, K. 45 n. 7 meaning see theories of meaning meaninglessness view of category mistakes 4 n. 6, 6–7, 11, 14, 17 n. 47, 23, 43–79, 81, 92, 93 n. 22, 149–50, 154 metaphor 66–74 and translation 73 n. 54 contextualist theories 74 n. 55 expansion theory 68–72, 74 n. 55 Gricean theories 68, 72–3, 156 in cognitive science 67 interaction theories 68 non-cognitivist theories 68, 73 simile theories 68–9, 73 substitution theory 68–72 metonymy 155–6 Montague Grammar 48–56 Montague, R. 5 n. 10, 48 n. 13, 48–50 Moran, R. 69 n. 48 nonsense 38–9, 60 n. 32, 79 Object Oriented Programming 21–2 Pap, A. 11, 75 n. 56, 82 n. 2 Parsons, T. 28 n. 7, 65 n. 38 Partee, B. 5 n. 9 Perry, J. 63 n. 37 persistence 13 n. 39, 157 Peters, S. 125–32, 136 n. 37, 145, 155 phrase markers 25–6, 31 n. 13 Postal, P. 18 pragmatic approach to category mistakes 6–7, 82, 105 n. 40, 110–57 pragmatics 5–6, 31–2, 110, 154–6

169

INDE X presupposition 6, 13 n. 40, 18, 20, 37 n. 25, 82, 91–3, 116–57 accommodation 13 n. 40, 105 n. 40, 118–9, 153 and truth-value gaps 6, 82, 91–3, 125–33, 148–50 conditionals 121, 135–6, 143–4 conjunctions 122, 126–8, 136–8, 145–6, 149–50 disjunctions 122, 138, 143–4 heritage values 126 (see also presupposition projection) logico-semantic approach 13 n. 40, 91–2, 124–5 negations 121, 123, 134–5, 146 pragmatic approach 124–31, 155 presupposition failure 6, 20, 54, 82, 91–3, 117–48 presupposition projection 118, 120–31, 134–40, 143–6 presupposition tests 117–24, 133–40, 143–4 presupposition triggers 19 n. 54, 116–7, 129–48, 154–5 proviso problem 121 n. 14, 137–8 quantifier phrases 122–3, 131 n. 30, 138–9, 146 questions 122, 138 triggering problem 19 n. 54, 129–32, 154–5 presuppositional account of category mistakes 7, 13 n. 40, 82, 130–57 principle of referentiality 93 n. 22 Prior, A.N. 11–2 propositional attitude ascriptions 38–9, 59–66, 72 n. 50, 78, 84, 123, 130, 157 n. 58 propositions 6, 53, 63–6, 80–91, 108 n. 46, 130 n. 29, 156 partial propositions 6, 83–91, 130 n. 29 Quine, W.V.O. 11–2 Rappaport Hovav, M. 29 n. 8, 30 n. 10, 30 n. 11, 34 n. 20 Recanati, F. 74 n. 55

reference 3, 7, 23, 54, 62 n. 35, 70, 80–1, 84, 88 n. 15, 92–3, 101, 150, 157 reference failure 81, 83–4 Reimer, M. 68, 77 n. 60 Resnik, P.S. 20, 21 n. 61 Ross, H. 18 Rothschild, D. 85 n. 9, 122 n. 15, 136 n. 37 Routley, R. 2 n. 3, 12, 44 n. 6, 82 n. 3, 82 n. 6, 94–9 Russell, B. 7–11, 44 n. 6, 53 n. 21, 85 n. 10, 94 n. 23, 96 Russell’s paradox 7–9 Russell’s theory of types 8–10 Ryle, G. 9–11, 44 n. 6 s-selection 19–20 same-saying 65–6 Sauerland, U. 19 n. 55, 28 n. 6, 44, 125 n. 21 Schlenker, P. 116 n. 9, 128–31, 138 n. 39 Schminglish 96–8 Schnieder, B. 13 n. 39 Searle, J. 44 n. 6, 70 n. 49, 72 n. 51 selectional features 26–7, 33–4, 35 selectional restrictions 15 n. 42, 19–20 selectional violations 15 n. 42, 59 n. 31 semantic anomalies 16, 19–20 semantics 5–6, 15–8, 31–2, 35–7, 127–8, 154–5 dynamic semantics 127–30 lexical semantics 28–9 semantics-pragmatics interface 5–6, 31–2, 110, 154–6 type-theoretic semantics 13 n. 40, 45 n. 9, 48–56, 84–6 sense 64–5 Seuren, P. 20 n. 59 Shapiro, S. 2 n. 3, 13 n. 39, 14 n. 41, 44 n. 6 Simons, M. 116 n. 9, 155 n. 53 Smart, J.J. 11 n. 25 Smiley, T. 44 n. 6, 82 n. 3, 82 n. 6 Soames, S. 86 n. 11, 89 n. 17, 122, 125 n. 20, 128 Sommers, F. 11 n. 25 Sorensen, R. 14 n. 41, 44 n. 6

170

INDE X sortal phenomena 102–6 sortal presuppositions 15 n. 42 Stalnaker, R. 4 n. 5, 92 n. 20, 112, 115 n. 8, 124–33, 150 Stalnaker’s theory of conversation 112, 124–7, 129–30 Stanley, J. 110 n. 1 Stateva, P. 125 n. 21 static typing 21–3 Stern, J. 13 n. 39, 14 n. 41, 44 n. 6, 71–2, 74 n. 55, 75 n. 57, 156 n. 56 Steward, H. 14 n. 41, 43–4 Strawson, P.F. 11, 25 n. 2, 44 n. 6, 53 n. 21, 94, 134 n. 35 structuralism in mathematics 13 n. 39 supervenience 6, 29–30, 34, 129 n. 26 synaesthesia 62 n. 36 synonymy 34, 58–9, 65–6 syntactic approach to category mistakes 6, 17–8, 25–42, 44, 59 n. 31, 66 n. 40, 93 n. 22, 149–50, 152, 154 syntax generative grammar 25–6 statistical account of syntax 15 n. 44 syntax-semantics interface 5–6, 15–8, 27–32, 35–7, 154–5 thematic roles 20, 28–9, 57 n. 29 Tarskian truth schema 87–90 theories of meaning 45–6, 75–8, 154 conceptual role semantics 77–8

truth-conditional 75–6 verificationism about meaning 77–8 Thomason, R. 13 n. 37, 30–1, 42 n. 34, 82 n. 5, 83, 85 n. 10, 93 n. 22, 95 n. 25, 96–109, 130 n. 29, 136–7, 149 transformations 15–6 trivial sentences 110–6, 118, 135 n. 36, 149–50 truth-value gaps 6, 80–109, 128–33 (see also logic) universal grammar 33 n. 17 vagueness 2, 13 n. 39, 44, 81, 99, 106 n. 41 validity 82 n. 7, 101–6, 130 n. 29, 136 n. 38 van Fraassen, B. 14 n. 41, 44, 99 van Valin, R. 19 n. 58 von Fintel, K. 86 n. 11, 118–20 von Stechow, A. 19 n. 55, 28 n. 6, 44 Westerhoff, J. 11 n. 25 Whitehead, A.N. 8 n. 14 who/which distinction 27, 32 n. 15, 37 Williamson, T. 4 n. 5, 13 n. 39, 44, 87–9 Wittgenstein, L. 9 Woolf, V. 69 n. 47 Ziff, P. 25 n. 1, 25 n. 2

171