Pragmatics, Truth and Underspecification: Towards an Atlas of Meaning 9789004365445, 9004365443

Citation preview

Pragmatics, Truth and Underspecification

Current Research in the Semantics/Pragmatics Interface Series Editors Klaus von Heusinger (University of Cologne) Ken Turner (University of Brighton)

Editorial Board Nicholas Asher (Université Paul Sabatier) Johan van der Auwera (University of Antwerp) Betty Birner (Northern Illinois University) Ariel Cohen (Ben Gurion University) Paul Dekker (University of Amsterdam) Regine Eckardt (University of Constance) Markus Egg (Humbolt University Berlin) Donka Farkas (University of California, Santa Cruz) Brendan Gillon (McGill University) Jeroen Groenendijk (University of Amsterdam) Yueguo Gu (Chinese Academy of Social Sciences) Larry Horn (Yale University) Yan Huang (University of Auckland) Manfred Krifka (Humboldt University Berlin)

Chungmin Lee (Seoul National University) Claudia Maienborn (University of Tübingen) Alice ter Meulen (University of Geneva) Jaroslav Peregrin (Czech Academy of Sciences and University of Hradec Králové) Allan Ramsay (University of Manchester) Rob van der Sandt (Radboud University Nijmegen) Kjell Johan Sæbø (University of Oslo) Robert Stalnaker (Massachusetts Institute of Technology) Martin Stokhof (University of Amsterdam) Henk Zeevat (University of Amsterdam) Thomas Ede Zimmermann (University of Frankfurt)

volume 34

The titles published in this series are listed at brill.com/crispi

Pragmatics, Truth and Underspecification Towards an Atlas of Meaning

Edited by

Ken Turner Laurence Horn

LEIDEN | BOSTON

The Library of Congress Cataloging-in-Publication Data is available online at http://catalog.loc.gov LC record available at http://lccn.loc.gov/2018010826

Typeface for the Latin, Greek, and Cyrillic scripts: “Brill”. See and download: brill.com/brill‑typeface. ISSN 1472-7870 ISBN 978-90-04-34199-9 (hardback) ISBN 978-90-04-36544-5 (e-book) Copyright 2018 by Koninklijke Brill NV, Leiden, The Netherlands. Koninklijke Brill NV incorporates the imprints Brill, Brill Hes & De Graaf, Brill Nijhoff, Brill Rodopi, Brill Sense and Hotei Publishing. All rights reserved. No part of this publication may be reproduced, translated, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission from the publisher. Authorization to photocopy items for internal or personal use is granted by Koninklijke Brill NV provided that the appropriate fees are paid directly to The Copyright Clearance Center, 222 Rosewood Drive, Suite 910, Danvers, MA 01923, USA. Fees are subject to change. This book is printed on acid-free paper and produced in a sustainable manner.

Whatever happened to meaning? Jay D. Atlas

∵

Contents An Underspecified Preface ix Ken Turner and Laurence Horn

Part 1 On the Landscape of Negation 1

An (Abridged) Atlas of Negation: Polar Landscape in an Era of Climate Change 3 Laurence Horn

2

Dispelling the Cloud of Unknowing: More on the Syntactic Nature of Neg Raising 54 Chris Collins and Paul Postal

3

Presuppositions, Negation, and Existence 82 Barbara Abbott

4

More Ado about Nothing: On the Typology of Negative Indefinites 107 Johan van der Auwera and Lauren Van Alsenoy

Part 2 On Sense-Generality and the Semantics/Pragmatics Landscape 5

Distinguishing Ambiguity from Underspecificity 149 Una Stojnić, Matthew Stone and Ernie Lepore

6

Metaphor, Minimalism, and Semantic Generality: Seeing Things in Context 169 Michiel Leezenberg

7

A Radically Pragmatic Account of Number Words and the Reversibility of Scales 195 Jerrold Sadock

viii 8

contents

Utterances and Expressions in Semantics and Logic 212 David Braun

Part 3 On Grammar, Inference, and Truth 9

Grammar as Procedures: Language, Interaction, and the Predictive Turn 261 Ruth Kempson and Ronnie Cann

10

Illusory Inferences in a Question-Based Theory of Reasoning 300 Philipp Koralus and Salvador Mascarenhas

11

A Commitment-Theoretic Account of Moore’s Paradox 323 Jack Woods

12

Remarks on Davidson’s Polymorphous Concept of Truth and Its Role in a Theory of Meaning 349 Ken Turner Index of Names 399 Index of Subjects 400

An Underspecified Preface Very general remarks may be helpful, and are not always untrue. It is no doubt perilous to make them. But that cannot be helped. Warnock 1958: 52

… I was relying upon a reader who would be ready to meet me half-way— who does not begrudge a pinch of salt. Frege 1984: 193

… Better to leave matters a little semantically underdeterminate … than make the language impractical to use or impossible to learn. Atlas 2005: 16

∵ The philosophy of language is an old subject, and since 1879 it has been a great one. There is room in it for many books. This is another one. But if each of several books on the philosophy of language, or, indeed, on semantics and/or pragmatics, is concerned, at least in its early pages, with the elements of the subject, then space must be found in the latest for some context and, perhaps, an apology or two.1 Such is the purpose for which the present few paragraphs have been set aside. Let us begin at the start of the Fregean Era. From 1879 quantifier/variable and argument/function structures (slowly, and through the intellectual filters of others who often neglected to recognise their source) became constituents of logical and philosophical thinking and they (again, slowly) dissolved the 1 The first apology goes to Quine, of course. Cf. Quine (1952).

x

turner and horn

inhibiting restraints of Aristotelian and Stoic reasoning. The new ‘system’ was the first modern logical language, with the universal quantifier, the material conditional, negation and identity as the primitive symbols—with modus ponens and an implicit rule of substitution as the inference rules—and the existential quantifier, conjunction and disjunction as the derived symbols. The crucial innovation was to recognise that universally quantified statements were not semantically related to names and other nominal expressions but were, instead, composed from two functions, joined by a propositional connective, and this move enabled the representation of multiple generality and constituted a huge conceptual advance in formal inference. The coarse-grained notion of undifferentiated content was in time replaced, largely as a result of problems having to do with the informativeness of identity statements, by a distinction between sense and reference. This distinction was accompanied by the notion of force, which was to receive a thorough examination at the hands of certain Oxford ordinary language philosophers, and the notion of tone (or ‘colouring’), which was developed, later, by certain other Oxford ordinary language philosophers as part of the work of attempting to design, with a clever and natural exploitation of the Kantian categories of the understanding, a conversational logic. So, although the seeds were planted by Frege at the University of Jena, the nascent green shoots of contemporary pragmatics most vigorously appeared in and around Corpus Christi and St. John’s Colleges in Oxford (and later at the University of California Berkeley). The sense/reference distinction eased the problem of identity (think Morning Star and Evening Star; Alice Cooper/Vincent Furnier; John Wayne/Marion Robert Morrison; Lewis Carroll/Charles Dodgson, etc.) but introduced other difficulties to do with the precise nature of sense: (a) its location, in an abstract and timeless and unchanging ‘third realm’ (somewhere beyond the first, the physical, and the second, the mental) didn’t help to convince logicians and philosophers that it was a serviceable tool and (b) ‘grasping’ a sense, from somewhere presumably in the second realm, was always a vague and questionable action. The timelessness of the inhabitants of the third realm calls to mind certain modern preoccupations having to do with indexicality and other contextually dependent expressions. The issue would seem to be this. On the one hand there is a ‘content’ that fails to be truth-conditional. These cases require that the contextually dependent items must be expanded and made explicit—‘cashed out’ in some of the more modern idioms—so that the context-sensitivity is expunged. The assumption is that all context-sensitivity can be severed, and this assumption can be challenged (think ‘the essential indexical’; cf. Perry 1979). On the other hand there is a ‘content’ that is more than truth-conditional.

an underspecified preface

xi

The original 1879 revolutionary was later to say, probably in 1897, ‘The distinction between what is part of the thought expressed in a sentence and what only gets attached to the thought is of the greatest importance for logic’ (Frege 1979: 141; emphasis added), and a bit later, in 1918, ‘conversational suggestions make no difference to the thought’ (Frege 1977: 9), thus setting part of the agenda for more than the next one hundred years. These ‘attachment’ cases have generated increasingly delicate classifications of such inferences as implicatures (conversational, conventional and the others), explicatures, higherorder explicatures, implicitures and the rest. These two sets of issues leave the modern philosopher of language, and those in collateral disciplines, with The Goldilocks Problem, as they seek accounts that avoid too little ‘content’ (indexicals and their relations), or accounts of too much ‘content’ (implicatures and their neighbours), in search of theories of just the right amount of content. There is another problem. Consider a typical example from the recent literature: (i)

He handed her the key and she opened the door.

This, it is claimed, really means (ii).2 (ii) He handed her the key and she opened the door with key that he had handed to her. (i), it is claimed, though meaningful, is meaningful in an ‘incomplete’, or ‘wrong’, way. (ii), on the other hand, with the addition ‘with the key that he had handed to her’ is meaningful in the right way. But, let’s ask: What makes the addition in (ii) the proper difference that converts being meaningful in the wrong way to being meaningful in the right way? Why not also add (iii): (iii) … to go from the back garden to the kitchen … and/or (iv): (iv) … to get the electric carving knife … and (v): 2 Crucially, as Bach 2001 and others have noted, (i) doesn’t SAY this; the question is whether (i) MEANS (ii) or whether all we can assert is that the speaker of (i) is likely to have intended to COMMUNICATE (ii).

xii

turner and horn

(v) … in an attempt to fight off the invaders … and (vi): (vi) … who had come from the future to kidnap their children … and (vii): (vii) … because their eldest son, Johan, would become an evil warlord who would one day subjugate with his armies the civilised Universe … Etc. So, we must ask: What is a complete thought (a.k.a., in some of the more modern idioms, full propositionality)? Without an answer to this question the matter of what is now called pre-semantic pragmatic enrichment quickly descends into a version of The Mein Herr Problem: “What a useful thing a pocket-map is!” I remarked. “That’s another thing we’ve learned from your Nation,” said Mein Herr, “map-making. But we’ve carried it much further than you. What do you consider the largest map that would be really useful?” “About six inches to the mile.” “Only six inches!” exclaimed Mein Herr. “We very soon got to six yards to the mile. Then we tried a hundred yards to the mile. And then came the greatest idea of all! We actually made a map of the country, on the scale of a mile to the mile!” “Have you used it much?” I enquired. “It has never been spread out, yet,” said Mein Herr: “the farmers objected: they said it would cover the whole country, and shut out the sunlight! So we now use the country itself, as its own map, and I assure you it does nearly so well.” Carroll 1893: 169

The right amount of content becomes important because that is the source of truth-conditions and an interest in the notions of truth condition and truth value is a pervasive one throughout the Fregean Era. The idea is that a theory of truth can serve as, or be, a contribution to delineating the outlines of some theory of meaning. But there is also a discussion to be had about the absence of truth-value. Some non-declarative sentence types are not truth valued. Some declarative sentence types are not truth valued (the performatives and other non-assertives). And some non-performative declarative sentences

an underspecified preface

xiii

are not truth valued (those which contain constituents that lack a reference or are presuppositionally disabled). There have been a number of proposals advanced to accommodate these truthvalueless constituents in larger compositions. On the one hand, it is proposed that all such compositions are false; on the other, that all such compositions are truthvalueless. Both of these proposals betray the logic of the classical semantics of the elementary connectives. On one view, the consistent logic and the amplified truth tables can be given as follows, although there are competing positions. In the case of negation, the negation of a truth-value gapped proposition is a truth-value gapped proposition. In the case of conjunction, the conjunction of a falsehood and a truth-value gapped proposition is false, otherwise it is truth-value gapped. For disjunction, the disjunction of a truth and a truth-value gapped proposition is true, otherwise it is truth-value gapped. And in the case of the single headed arrow, a false antecedent and a gapped consequent, or a gapped antecedent and a true consequent, result in a truth, otherwise the composition is gapped. Note that these are compositions for a gapped bivalent logic, not a trivalent logic. The notion of a trivalent logic doesn’t, to some, make sense. A truth-value gap is the absence of a truth-value, not a third truth value. Whatever the outcome of these later discussions, especially the last one, concerning the nature and consequences of truthvaluelessness (where the present authors themselves have yet to reach consensus) the introduction of quantifier/variable constructions marks the distinction between the old and the modern worlds.3 In the wake of this innovation, Aristotelian and Stoic (and Boolean) reasoning are left receding in the rear-view mirror of intellectual history. Enter the theory of denoting. And, in next to no time, in historical terms, exit the theory of denoting. Enter the theory of descriptions. One abrupt difference between the two theories is in the logician’s attitude to natural language syntax. In 1903 it was possible to claim that ‘grammar seems … to bring us much nearer to a correct logic than the current opinions of philosophers; … grammar, though not our master, will yet be taken as our guide’ (Russell 1903/1937: 42). By 1905 the guide has been emphatically dismissed, whence Russell’s unpacking of (1) into (1’) below, which has been described as resulting from an act of butchery (Gareth Evans). The story here is by now a familiar one. The motivation this time was the problem of empty names. This problem can be approached

3 There is at least one, and perhaps at most one, object such that that object is a co-author of this Preface and that object is convinced of the truth of this statement. (Apologies to Dummett: cf. any of Dummett 1973, 1981, 1991, 1994.)

xiv

turner and horn

in several ways. Either an appeal is made to the third realm and sense, or it can be assumed that such expressions function referentially and then form a conception of reality such that their denotata ‘exist’ in some extended sense of that term or one can attempt to find an extensional interpretation of these phrases such that the problem of their non-existent denotata does not arise. Bluntly stated, the choice is between (i) going intensional, and falling back on one of the consequences of the 1879 revolution, or (ii) embracing ontological extravagance, and opening to all comers the gates of ‘reality’ or, perhaps most sensibly, (iii) exploiting the quantificational capacity of the new logic Sinnlos. The architect of the theory of descriptions chose the last option and interpreted such expressions not as troublesome subject-predicate structures but as the complex existentially quantified statement corresponding to the conjunction of (1’a-c): (1) The F is G. (e.g. The king of France is bald.) (1’) There is at least one object x such that a. x Fs, and b. For any object y, if y Fs then y is identical to x, and c. x Gs This analysis of the sentence entails, by conjunction elimination, that there is an object that Fs so the conjunction is false if (1’a) is false. The negative of (1) can be represented by placing the negative operator before (c), or before the entire conjunction. If true, the two negative sentences are true for different reasons. The former is true if there is a unique, extant object which Fs which fails to G and the latter is true if there is no object which Fs or because there is more than one or because a unique, extant object fails to G. The main consequence of this analysis is that sentences containing empty names or defective descriptions can now, without appeal to the third realm, be said to be meaningful and have truth values and the law of the excluded middle shown to be preserved. The architect of the theory that generated analyses like (1’) was not modest about the virtues of this solution. He said that it ‘clears up two millennia of muddleheadedness about “existence”’ (Russell 1961: 785). Some more modern readers are not impressed. Jay David Atlas, for example, makes two claims: (i) not is not scopally (much less lexically), ambiguous, but rather sense-general and so, because logical forms cannot be the semantic representations of sense-general sentences, (ii) sense-general sentences do not have underlying logical representations of the kind in (1) or indeed of any other kind. The punch line for these claims is that speakers do not select from the logico-linguistically given readings of a syntactically or lexically ambiguous

an underspecified preface

xv

sentence but that, from a meaningful but radically sense-general sentence, they construct a contextually determined interpretation of an utterance. This can be put in the Kantian-esque slogan: ‘Pragmatic inference without sense-generality is blind; sense-generality without pragmatic inference is empty’ (Atlas 1989: ix). In short, and more colloquially: ‘one is reading meaning into the words, not reading meaning out of them’ (Atlas 1989: 26). There is some merit in examining the arguments for these claims in more detail to see what kind of run we get from them for our money. The argument is principally founded on some diagnostics to distinguish between ambiguity and generality. Philosophers have been none too careful in attending methodologically to this distinction but linguists, on the other hand (in particular Zwicky and Sadock 1975), have designed a battery of tests to facilitate the distinction and the required basis is provided by one of them. Consider (2a) which has possible interpretations (2b–c) and impossible interpretations (2d–e): (2) a. b. c. d. e.

A likes visiting relatives and so does B. A likes going to visit relatives and B likes going to visit relatives. A likes relatives who are visiting and B likes relatives who are visiting. A likes going to visit relatives and B likes relatives who are visiting. A likes relatives who are visiting and B likes going to visit relatives.

The parallel interpretations in (2b–c) and the impossibility of the crossed interpretations in (2d–e), under the legislation provided by the ConjunctionReduction Identity Test (CRIT), are taken as evidence for the ambiguity of (2a). The CRIT consists of the propositions (A) and (B): (A) The impossibility of a crossed, literal paraphrase for a conjunction-reduced sentence S entails the ambiguity of S. (The distinct, parallel paraphrases express distinct senses.) (B) The possibility of a crossed, literal paraphrase for a conjunction-reduced sentence S entails the non-ambiguity of S. (The distinct parallel paraphrases do not express distinct senses.) In parentheses we note that the following sentences (3–5) permit crossed interpretations and are, therefore, general with respect to whether A or his or her circumstances change: (3) A became as strident as any of the Brexiteers and then the same thing happened to B.

xvi

turner and horn

(4) A became more strident than the average Brexiteer and then the same thing happened to B. (5) A became the most strident Brexiteer in the party and the next year the same thing happened to B. The CRIT is then applied to the theory of descriptions’ most iconic examples: Given (6a) we can derive (6b): (6) a. The King of France is not bald and the Queen of England is not bald. b. The King of France is not bald and the same goes for the Queen of England. And given (7a) we can derive (7b): (7) a. It is not true that the King of France is bald and it is not true that the Queen of England is bald. b. It is not true that the King of France is bald and the same goes for the Queen of England. And now it is asked whether there are crossed, in addition to the parallel, interpretations: (8) a. The King of France is not bald (because France is not, currently, a monarchy), and the same goes for the Queen of England (because she is a typical hirsute Windsor). b. It is not true that the King of France is bald (because France is not, currently, a monarchy), and the same goes for the Queen of England (because she is a typical hirsute Windsor). The argument hinges on whether there are native speakers who accept (8a– b). If there are such speakers then there is evidence that “ ‘Not’-sentences are semantically less specified, and theoretically more complex, than the twothousand-year tradition in logical theory has heretofore recognized” (Atlas 1989: 91).4

4 There is a complication that might be noted. Zwicky and Sadock (1975) regard their CRIT and other identity of sense diagnostics for ambiguity as inapplicable to cases of “privative opposition” in which one understanding unilaterally entails the other. The opposition between internal (narrow-scope) and external (wide-scope) negation is unfortunately just such a case,

an underspecified preface

xvii

The second, more general, claim, that sense-general sentences do not have underlying logical representations of the kind in (1) or indeed of any other kind is a very interesting one. The question may be about whether formal methods have comprehensive application on natural phenomena. The 1879 analysis was confined to providing a semantics and a proof theory for arithmetical statements. The formalism and especially the proposals about multiple generality were perfectly adequate to this principal purpose. But there were no proposals then, nor subsequently with the theory of descriptions, for phenomena from the natural world that can be found outside the relatively narrow bandwidth of arithmetical discourse: demonstratives, indexicality, opacity, modality, temporality, information structure and more. The few remarks about first person pronouns seemed to result in parts of the objective third realm becoming private, non-objective, places, inaccessible to others, and the attempts to deal with opacity, and, with multiple embeddings, the potential infinity of senses, make natural languages, on certain assumptions, unlearnable. Frege was quite clear that (i) formal methods have little purchase on natural language and (ii) natural language is no guide to a formal language for pure thought: ‘It cannot be the task of logic to investigate language and determine what is contained in a linguistic expression. Someone who wants to learn logic from language is like an adult who wants to learn how to think from a child’ (Frege 1980: 67–68); and this because ‘the excessive variety of logical forms that has gone into the shaping of our language makes it difficult to isolate a set of modes of inference which is both sufficient to cope with all cases and easy to take in at first glance’ (Frege 1953: 103e). The papers that follow are all ‘attached’, in one way or another, with the thought—or topics themselves ‘attached’ to the thought—of Jay David Atlas (e.g. Atlas 1989, 2005). This branch of the philosophical tree has, over the years, become a large one, with its own branches (with branches). So, the papers treat such themes as the ambiguity/underspecification distinction, dynamic syntax and social interaction, erotetic reasoning, metaphor, Moore’s Paradox, presupposition, radical pragmatics, the semantics/logic interface, the syntax, semantics and pragmatics of negation, and truth-meaning determination (modulo Davidson). We thank Jay for giving us so much to learn from, to think about, to reflect upon, and to argue with. The purpose of prefatory and introductory remarks is to ease the reader’s transition from the all too frequent great bloomin’, buzzin’ confusion in the

given that the former entails the latter but not vice versa (Horn 1985: 126–127). Atlas (1989: § 2.2) rejects this part of the Zwicky/Sadock thesis.

xviii

turner and horn

outside world provoked by, for example, unpredictable world leaders, weak selfdestructive governments, price inflation, commodity shrinkflation, climate change, cybercrime and sudden and seemingly inexplicable currency fluctuations in the planet’s most powerful and prosperous nations, to the tranquil Elysium where real reflection and thought, one might almost say timeless Fregean thought, can take place. We hope that these remarks have assisted—in some perhaps small way— that transition. Ken Turner and Laurence Horn

References Atlas, Jay David, 1989. Philosophy without Ambiguity: A Logico-Linguistic Essay. Oxford: Clarendon Press. Atlas, Jay David, 2005. Logic, Meaning, and Conversation: Semantical Underdeterminacy, Implicature, and their Interface. Oxford: Oxford University Press. Bach, Kent, 2001. You Don’t Say? Synthese 128: 15–44. Carroll, Lewis, 1893. Sylvie and Bruno Concluded. (With forty-six illustrations by Harry Furniss.) London: Macmillan and Co. Dummett, Michael, 1973. Frege: Philosophy of Language. London: Duckworth. Dummett, Michael, 1981. The Interpretation of Frege’s Philosophy. London: Duckworth. Dummett, Michael, 1991. Frege: Philosophy of Mathematics. London: Duckworth. Dummett, Michael, 1994. Origins of Analytical Philosophy. Cambridge, Massachusetts: Harvard University Press. Frege, Gottlob, 1953. The Foundations of Arithmetic: A Logico-Mathematical Enquiry into the Concept of Number. Second Revised Edition. (English Translation by J.L. Austin.) Oxford: Basil Blackwell. Frege, Gottlob, 1977. Logical Investigations. (Edited with a Preface by P.T. Geach. Translated by P.T. Geach and R.H. Stoothoff.) Oxford: Basil Blackwell. Frege, Gottlob, 1979. Posthumous Writings. (Edited by Hans Hermes, Friedrich Kambartel and Friedrich Kaulbach with the assistance of Gottfried Gabriel and Walburga Rodding. Translated by Peter Long and Roger White with the assistance of Raymond Hargreaves.) Oxford: Basil Blackwell. Frege, Gottlob, 1980. Philosophical and Mathematical Correspondence. (Edited by Gottfried Gabriel, Hans Hermes, Friedrich Kambartel, Christian Thiel and Albert Veraart. Abridged for the English Edition by Brian McGuinness and translated by Hans Kaal.) Oxford: Blackwell. Frege, Gottlob, 1984. Collected Papers on Mathematics, Logic, and Philosophy. (Edited

an underspecified preface

xix

by Brian McGuinness. Translated by Max Black, V.H. Dudman, Peter Geach, Hans Kaal E.-H.W. Kluge, Brian McGuinness and R.H. Stoothoff.) Oxford: Basil Blackwell. Horn, Laurence, 1985. Metalinguistic Negation and Pragmatic Ambiguity. Language 61: 121–174. Perry, John, 1979. The Problem of the Essential Indexical. Noûs 13: 3–21. Quine, Willard van Orman, 1952. Methods of Logic. London: Routledge & Kegan Paul. Russell, Bertrand, 1903/1937. The Principles of Mathematics. 2nd Edition. London: George Allen & Unwin Ltd. Russell, Bertrand, 1961. History of Western Philosophy, and its Connection with Political and Social Circumstances from the Earliest Times to the Present Day. 2nd Edition. London: Allen and Unwin. Warnock, Geoffrey, 1958. English Philosophy Since 1900. London: Oxford University Press. Zwicky, Arnold and Jerrold Sadock, 1975. Ambiguity tests and how to fail them. In John P. Kimball, ed. Syntax and Semantics 4. 1–35. New York: Academic Press.

Jay David Atlas

part 1 On the Landscape of Negation

∵

chapter 1

An (Abridged) Atlas of Negation: Polar Landscape in an Era of Climate Change Laurence R. Horn*

1

Atlases and Atlases

For the ancient Greeks, Atlas was the greatest of the Titans, with shoulders supporting the world (or on some accounts the celestial spheres). On the broad shoulders of our own Atlas stand the twin edifices of neo-Gricean pragmatics (Horn 1989, Levinson 2000) and Relevance theory (Sperber & Wilson 1986/1995, Carston 2002). The reinterpretation of Grice (1967)’s maxims in Atlas & Levinson 1981 directly underlies the dualistic model of non-logical inference that I have been purveying since Horn 1984 (see also Horn 1989, 1993, 2007, as well as the related tripartite variant in Levinson 2000) and it was Atlas’s emphasis on the underdetermination of truth-conditional content by linguistic semantics (Atlas 1977, 1979) that provides a foundation stone of the Relevance-theoretic line on the semantics/pragmatics interface. In particular, the Atlas advocacy of the semantic non-specificity of natural language negation inspired both the widespread retreat from logical presupposition and truth-value gaps (Kempson 1975, Wilson 1975, Gazdar 1979; see Burton-Roberts 1989a for a spirited defense of the Strawsonian notions) and the arguments for metalinguistic or echoic negation (Horn 1985, 1989, Carston 1996).1 But we must begin by sorting out our Atlases. Despite the various burdens shouldered by each—and notwithstanding the principle of parsimony embodied in the Modified Occam’s Razor (Grice 1989: 47)—the Titanic Atlas,

* Versions of some of this material were presented in Tel Aviv, Berlin, Tokyo, Reading, Amherst, New Haven, Cambridge, Chicago, Swarthmore, and Columbus, as well as at the 2005 Pomona Jayfest. I am indebted to those offering comments at those presentations and to Patricia Amaral, Barbara Abbott, Kent Bach, Bart Geurts, Anastasia Giannakidou, Jack Hoeksema, Michael Israel, Jerry Sadock, Victor Sánchez Valencia, Scott Schwenter, and Debra Ziegeler for pointers and comments—and especially to Jay Atlas, sine quo non. 1 See however Pitts (2011: 363–364) for useful commentary on the relation between Atlas’s radical underspecification view and the pragmatic ambiguity line endorsed in Horn 1985.

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_002

4

horn

Charles Atlas, and Jay David Atlas must be recognized as distinct individuals, although some confusion persists:2 Earlier this month while sitting in a waiting room, I saw in a magazine a picture of Charles Atlas. It was a picture of the statue at 630 Fifth Avenue in New York City. There he was with his body of great muscles, bearing the weight of the entire world on his shoulders … Many times in this life we feel like Charles Atlas. So often we feel like we have the weight of the world on our shoulders. We do not have to bear that weight. Everything is not meant to be on our shoulders. http://www.addisville.org/artdecember2002.htm

I want you to go with me to New York City and take a walk up Fifth Avenue. Our first stop is going to be the RCA building. Inside [sic] the RCA building, there is a statue of Charles Atlas. Charles Atlas is straining every muscle, and he is carrying the world upon his shoulders. That’s one way that you and I can live. Now let’s walk across the street and let’s go to St. Patrick’s Cathedral. Inside St. Patrick’s Cathedral there is another statue. This statue is of Jesus … http://www.amazingrace.net/NotMyWill.html

In fact, the massive 15-foot, 7-tonne bronze in question—commissioned by John D. Rockefeller, sculpted by Lawrie Lee, and on view at http://tinyurl.com/ h5qnusa—depicts not Charles (let alone Jay) Atlas but the eponymous Titan. The celebrated body-builder and erstwhile 97-pound weakling Charles Atlas, né Angelo Siciliano, is profiled at http://en.wikipedia.org/wiki/Angelo_ Siciliano. It is in fact Charles Atlas, seen impressively displaying his patented leopard-skin bikini briefs at http://tinyurl.com/nbsj9t7, and not the Titan or the Pomonan, who was the recipient of Physical Culture magazine’s $ 1000 prize as the World’s Most Handsome Man. This award is not to be confused with the $500 Atlas Prize offered by Jay (Atlas 1996: 307–308) to “the young man or woman who comes up with a good theory [on why the pseudo-anti-additivity of only John licenses weak NPIs]”; see von Fintel (1999: 109) and below for why this award is unlikely to be claimed.

2 The two blog entries below are no longer accessible, the links (active on April 1, 2005 at the time of the Jayfest) having passed over into the great digital beyond over the intervening decade. Reassuringly, the internet still offers photos of “the Charles Atlas Statue at 630 Fifth Avenue”. The later three links in the text are still active as of August 15, 2016.

an (abridged) atlas of negation

5

My own debt to Atlas arises from his exploits as a sparring partner on (inter alia) the semantics of only. To what can one attribute our respective series of papers on the topic (Atlas 1991, 1993, 1996, 1997, 2002, 2005, 2006, 2007, 2013; Horn 1969, 1970, 1979, 1989, 1992, 1996, 1997a, 2002, 2005, 2006, 2009, 2011), treatises devoted in no small part to attacks by each of us on the other’s most recent formulation? A bit of an etymological detour may be helpful in appreciating this legacy. The American Heritage Dictionary of Indo-European Roots includes an entry for the telǝ- root, complete with some perhaps speculative etymological forays, which I have abridged in (1) below. Note in particular the zero-grade form in 6 for our world-supporting Titan and its suffixed counterpart in 3 relating to retaliation; we return to “reciprocal punishment in kind” via lex talionis at the end of this study. (1) (from Watkins 2011: 92–93) telǝENTRY: DEFINITION: To lift, support, weigh; with derivatives referring to measured weights and thence to money and payment. 1. Suffixed form *telǝ-mon-. telamon, from Greek telamōn, supporter, bearer. 2. Suffixed form *tel(ǝ)-es-. a. toll; philately, from Greek telos, tax, charge; b. tolerate, from Latin tolerāre, to bear, endure. 3. Suffixed zero-grade form *tlǝ-i-. talion; retaliate, from Latin tāliō, reciprocal punishment in kind, possibly “something paid out,” from *tali- (influenced by tālis, such). 4. Suffixed variant zero-grade form *tala-nt-. talent, from Greek talanton, balance, weight, any of several specific weights of gold or silver, hence the sum of money represented by such a weight. 5. Perhaps intensive reduplicated form *tantal-. tantalize, Tantalus, from Greek Tantalos, name of a legendary king, “the sufferer”. 6. Perhaps zero-grade form *tlǝ-. Atlantic, Atlas, from Greek atlās (stem atlant-), name of the Titan supporting the world. (Pokorny 1. tel- 1060.) As we shall see in our survey of attempts to pin down the nature of negation, negative polarity licensing, and the semantics of only, the presence of Titanic Atlas’s son-in-law Tantalus in subfamily 5 of this group is all too germane.

6 2

horn

Only: Entailment, Assertion, and the Locus of Asymmetry

In another century, I found myself led to conclude from my (then) analysis of only a Fs as implicating rather than entailing its prejacent a Fs that in the paradigms of (2) and (3), (2) a. I love only you. b. I love you. (3) a. It’s YOU that I love. b. I love nobody distinct from you. [2b] does not follow from [2a] by virtue of semantics, just as [3b] does not follow from [3a]. [2a] is not a declaration of love nor [3a] a declaration of fidelity, but the recipient in each case is pragmatically licensed to hope for the best. Horn 1992: 182

Predictably, Atlas (1993: 314, 317) was not convinced: [This] consequence strikes me as so outrageously counter-intuitive as to be a reductio of the theory. The idea that simultaneously I love only you could be true while I love you is false just seems crazy to me … I just have a very hard time with Horn’s theory about the truth-conditions of I love only you. In truth, so do I. We return below to the focus construction in (3a); our first quarry is the inference from (2a) to (2b) and to the task that I have referred to (Horn 2005, with apologies to E.M. Forster) as “Only connect”. Switching to the canonical subject-focus only example (Horn 1969), does (4a) bear parallel meaning connections to its two “exponents” in (4b) and (4c)? (4) a. Only Muriel voted for Hubert. b. Muriel voted for Hubert. c. Nothing distinct from Muriel voted for Hubert.

[prejacent] [exclusion]

(4c) is clearly a necessary condition for the truth of (4a), which is false if someone other than Muriel voted for Hubert. But if it turns out that only Hubert himself voted for Hubert, or that nobody at all did, and hence the prejacent (4b) is false, how do we assess the truth value of (4a)? This is not a simple matter to decide, as history has proven:

an (abridged) atlas of negation

7

The semantics for only says this: it asserts that no proposition from the set of relevant contrasts C other than the one expressed by its sister sentence α is true. There is in addition an implicature [sic] that α is in fact true. There is an industry devoted to the issue of whether the latter ingredient is an implicature (conversational or conventional), a presupposition, or part of the truth-conditions. von Fintel 1994: 133

Following Atlas’s lead on (2a,b) we could simply take (4a) to entail both (4b) and (4c) and hence the conjunction of the two. Peter of Spain was perhaps the most celebrated sponsor of the prejacent entailment (Syncat. Tract 3, cap. 9–10): Tantum homo currit; ergo homo currit (‘only a man is running, therefore a man is running’), where running (currit) is asserted (ponitur) of a man (homo). The general rule is “Omnia exclusiva vera reliquit suam preiacentem veram” (‘Every true exclusive proposition preserves the truth of the prejacent’; Peter of Spain 1992: 110). Similarly, for Ockham (Summa Logicae II.17, 1980: 132– 142), “Every exclusive proposition has two exponents: one affirmative and the other negative. For example, ‘Only a man is an animal’ has these exponents: ‘A man is an animal’ and ‘Nothing other than man is an animal’ ”. This symmetricalist analysis of only has been co-sponsored or endorsed by a wide range of medievals and moderns, from Abelard (1970: 332), Walter Burley (Pinborg 1981), and William of Sherwood (O’Donnell 1941, Kretzmann 1968) to Taglicht (1984), Keenan & Stavi (1986), von Fintel (1993), Giannakidou (2006), and especially Atlas (1991, 1993, 1996, 2005, 2007), but no matter how long the petition on its behalf, the petitioners must cope with the evidence for asymmetry that has been assembled over the last several decades (cf. e.g. Horn 1969, 1992, 1996, 2002, 2009, McCawley 1981, Ippolito 2006, van Rooij & Schulz 2007). This evidence includes the apparent cancellability, or more accurately suspendibility, of the prejacent but not of the exclusion, as reflected in the contrast between the hopeless (5a) and the marginal (5b) and in the ability of fronted only XP constituents to license “negative” inversion, as in (6). (5) a. *Only Ann will pay her taxes on time, but maybe someone else will. b. Only Ann will pay her taxes on time, and #(maybe) even she won’t. (6) {Only in dreams/*Even in dreams/*In dreams} can I fly. Even more significantly, only phrases license negative polarity items (NPIs, as underlined below) outside their focal domains, as in (7); cf. Klima (1964,

8

horn

whence (7a,b)), Horn (1996, 2002, 2009), and Beaver & Clark (2008)—pace Atlas (1993), Progovac (1993), and Giannakidou (2006, 2011). (7) a. Only young writers ever accept suggestions with any sincerity. b. Only his sister will expect him to write any more novels. c. You {only/#even} lift a finger to help anyone when you want something back. d. {Only three/#[Three]F} of the guests drank a drop of the punch. Once we start down the asymmetricalist path, it is tempting to conclude that the prejacent of an only sentence is not entailed after all but rather semantically presupposed (Horn 1969), or perhaps pragmatically presupposed or conventionally implicated (Horn 1979, Karttunen & Peters 1979), or maybe just conversationally implicated (McCawley 1981: 227, Horn 1992, Ippolito 2005, van Rooij & Schulz 2007)—or perhaps it could simply be ignored altogether (Geach 1962: 187). While I’m sympathetic to the spirit of such asymmetricalist proposals— and not just because I’ve endorsed most of them at one time or another—I have gravitated in the current century to Atlas’s position that they are flawed in the letter. For starters, semantic asymmetricalists must explain away Atlas’s observation that simple cancellations of the prejacent in contexts like (8a,b) are unsalvageably bad. (8) a. #Only Hillary trusts Bill, and (even) she doesn’t. b. #I love only you, but I don’t love you either. Further, as Atlas (1993, 1996, 1997) has stressed, only NP cannot be a monotone negative (or, in Ladusaw’s sense, downward-entailing [DE]) operator, given the lack of entailment from (9a) to (9b). (9) a. Only Socrates entered the race. b. Only Socrates entered the race early. But that leaves us with the polarity licensing properties of only to explain— or explain away—and Atlas’s early attempts to do so (“only is not, in general, a trigger for Negative [Polarity] Items”—Atlas 1993: 313) are roughly as persuasive as my own theory of love in (2a,b), as I have argued in some detail (Horn 2005, 2009) and Atlas himself (2007, 2012) now appears willing to concede.3 3 It should be noted that Giannakidou (2006, 2011) has maintained the position that NPIs are

an (abridged) atlas of negation

9

The licensing of NPIs like those in (7c) is problematic for the standard formal approaches, since only XP is veridical, entailing its prejacent, and non-DE, failing to license inferences from sets to subsets. But while only clauses are indeed conjunctive in terms of what they entail—as seen in (8) and argued for by Atlas (1993, 1996)—they exhibit a pragmatic or rhetorical asymmetry that belies their truth-conditional symmetry in that their negative component is more salient and linguistically robust (see Horn 1969, 1970, 1996, 2002, 2009). Note that the same pragmatic asymmetry can be seen in “connected” exceptives of the form {Every/No} X but Y. These unpack semantically into conjunctions of a generalization and an exception to that generalization (von Fintel 1993, García-Álvarez 2011). Thus, for García-Álvarez (10a) entails—and asserts—(10a’), or more formally (10a”). By the same token, the negative exceptive (10b) entails and asserts (10b’/b”). (10) a. a’. a”. b. b’. b”.

Every nurse but Lucy is flirtatious. Every non-Lucy nurse is flirtatious ∧ Lucy is non-flirtatious ((nurse’—{Lucy’}) ⊆ flirtatious’) ∧ ¬ flirtatious’ (Lucy’) No nurse but Lucy is flirtatious. No non-Lucy nurse is flirtatious ∧ Lucy is flirtatious ((nurse’—{Lucy’}) ⊆ ¬ flirtatious’) ∧ flirtatious’ (Lucy’)

But how then are we to explain the licensing differences in (11) and (12)? (11) a. Every nurse but Lucy drank (#a drop of the punch). b. No nurse but Lucy drank (a drop of the punch). (12) a. Every economist but Krugman (*ever) argued for austerity. b. No man but a blockhead ever wrote except for money. [Samuel Johnson] How can we resolve the tension between the semantic symmetry of exclusives and negative exceptives and their asymmetry with respect to NPI-licensing? One possibility suggested for exclusives (Horn 1996; cf. von Fintel 1997) takes the long-established converse relation between exclusive and universal propositions—

not actually licensed by only but may be “rescued” within a domain like those of (7); see Horn (2009, 2016a) for arguments against this approach.

10

horn

(12) ONLY (A,B) [‘only As are Bs’] ↔ ALL (B,A) [‘all Bs are As’] —to predict the fact that the predicate of an only-subject predictably licenses NPIs in the same way that the restrictor of ∀ does:4 (12’) a. γOnly Americans have ever walked on the moon. b. All those who have ever walked on the moon are Americans. Given the interdefinability of only and universals, both corresponding to the subset relation B⊆A, could the status of the prejacent in the former be reduced to that of the restrictor in the latter, i.e. as an instance of existential import? (See Horn 1996, drawing on Walter Burley’s “De exclusivis”, in Pinborg 1981 and de Rijk 1985.) While this seemed like a good idea at the time and does in fact get some predictions right, Ippolito (2005) and van Rooij & Schulz (2007) have independently shown that it is the prejacent and not the corresponding existential proposition that constitutes the positive component of an only statement with conjoined NP focus. Thus, to say that only John and Mary smoke is to imply that John and Mary smoke, not just that someone does. (13) a. Only [John and Mary]F were at the party. (van Rooij & Schulz 2007) b. Only [John and Peter]F smoke. (Ippolito 2005: (24)) (See Ippolito 2006, 2008 for other problems with what she dubs the “weak presupposition” analysis.) We must thus assume that whatever the status of the positive implication of an only sentence may be, it is the prejacent that constitutes the content of that inference. But given the equivalence in (12) (or at least its validity in the left-to-right direction, which is all Atlas (1996) accepts), doesn’t this yield a DE status for only XP after all? Atlas (1996: 285–286) is in fact willing to grant that only CN is DE, while continuing to maintain that only with NP focus (more exactly only a, where a is an individual constant) is non-monotonic and hence not a polarity licenser (nor a “negative item”): it only seems to license NPI minimizers because it’s (i) a focus operator and (ii) a pseudo-anti-additive quantifier. I have argued strenuously (Horn 2005, 2006, 2009: 201–203) that focus in itself does not license

4 Clark (2006) has argued that the NPIs licensed by only are exactly those that are licensed in the restriction of universals, precisely as predicted by conversion principle in (12)—basically any, ever, and the minimizers. In keeping with my usual practice, Google-sourced cites are annotated with a gamma as in (12’a).

an (abridged) atlas of negation

11

NPIs (recall (7) above) and that pseudo-anti-additivity, as defined by Atlas, is neither necessary nor sufficient for licensing NPIs. So where do we stand? It would be nice if we could return to a DE account for all instances of only licensing, but that door is closed and locked behind us, with a familiar titanic figure standing guard: To establish the DE-ness of only, we need to be convinced that in a situation where it is true that only John ate vegetables for breakfast it is also strictly speaking true that only he ate kale for breakfast. It may be misleading and odd to actually assert the conclusion but that does not mean that it isn’t true. The problem with this account is that it is highly controversial that only-sentences can in fact be true in situations where their positive component is not satisfied … It is hard to imagine I could emulate the incredulity of the reaction by Atlas to the asymmetric analysis of only, cf. Atlas (1996: 280), so let me quote a representative passage: [Assume Socrates saw a mangy dog in the agora, which escaped the notice of his companions.] Then Socrates saw an animal is true. On the downward monotonic view, it is entailed—entailed!!—that Only Socrates saw a snow leopard. That is, in every possible model (world) in which Only Socrates saw an animal is true, Only Socrates saw a snow leopard is true in that model … So much the worse for downward monotonicity in only a sentences. While there are, of course, valiant attempts to justify a semantics for only where the inference that so offends Altas’ sensibility is perfectly truthpreserving, one cannot deny the force of the objection. von Fintel 1999: 103–106

What we seek is a quasi-asymmetricalist approach that concedes the nonDE status of only environments. Von Fintel’s (1999) own approach invokes the notion of Strawson-entailment. Essentially, one proposition Strawson-entails another if the entailment would go through modulo satisfaction of its presuppositions (cf. Hoeksema 1986 for a precursor of this idea). Strawson-downward entailment can be defined derivatively: A function f of type ⟨σ, τ⟩ is StrawsonDE iff for all x, y of type σ such that x => y and f (x) is defined, it holds that f (y)=> f (x). Only NP is not DE, since Only Chris ordered Thai curry does not entail Only Chris ordered Thai red curry (perhaps nobody ordered red curry), but it does entail Only Chris ordered Thai red curry if anyone did. Thus, only NP

12

horn

is Strawson-DE in that if the presuppositions of only are satisfied (and hence the prejacent is true), the set => subset entailment is valid. But is only a Fs always undefined when a doesn’t F? (14a) seems false rather than undefined if only Atlas can lift the rock, and if the Red Sox win 101 games and no other team is in triple figures my wager in (14b) is lost and not just trapped in some Strawsonian limbo. (14) a. Only Hercules can lift that rock. b. I bet you that only the Yankees will win over 100 games. On the other hand, if nobody can lift the rock, or if no team wins more than 99 games, it’s less clear how to assess the truth of these statements (see Horn 1996: 16 and Roberts 2005 for discussion of a similar example, but see Beaver & Clark 2008 for complications).5 A more general question is just what sort of theory of semantic presupposition and presupposition failure we are prepared to accept (see Atlas 1989 for some relevant considerations). While I agree with von Fintel (1999: 107) on the desirability of finding a middle ground between the incomplete asymmetric theories for the semantics of only and their empirically inadequate symmetricalist competitors, I remain skeptical that a Strawson DE approach relying on presupposition as a partial function is the best solution for a happy medium. I have been arguing for some time (Horn 1996, 1999, 2002) that even granting entailment status to both (4b) and (4c) (and similarly the exceptives in (10a,b)), not all entailments are created equal. Entailed material that is outside the scope of the asserted, and hence potentially controversial, component of utterance meaning (Stalnaker 1978) counts as assertorically inert and is hence effectively transparent to NPI-licensing and related diagnostics of scalar orientation. In particular, we can agree with Atlas that the only NP statement does indeed entail both the prejacent and the exclusive propositions, while maintaining that only the latter component is asserted or at issue, the prejacent being assertorically inert. On the account of Horn 2002/2009, the entailed prejacent is ignored in the determination of NPI licensing and inversion, since these properties are sensitive to downward assertion, not downward entailment as such. 5 It may be tempting to suggest that the presupposition invoked in the Strawson-DE approach is not the prejacent itself (Hercules can lift that rock for (14a)) as in Horn 1969, but rather the existential premise, as in Horn (1996). While von Fintel himself informally accepts this approach (1999: 104), which directly follows from the conversion principle in (12), we have already seen that such existentialist analyses prove untenable.

an (abridged) atlas of negation

13

Additional support for this line of argument is provided by the behavior of sentences with proximatives almost, barely, and not quite (Horn 1996, 2002, 2011; cf. also Schwenter 2002 and Amaral 2007, 2010). As seen in (15), (15) a. Gore almost won the election. a’. Gore didn’t win the election. b. Bush barely won the election. b’. Bush won the election. almost induces a negative entailment—from (15a) it follows that (15a’)—and barely (≈ almost not) a positive one—from (15b) we can conclude that (15b’) is true. As with only, the most straightforward account of the contribution of almost and barely is a simple conjunction of what Sevi (1998) dubs the proximal and polar components:6 (16) a. Gore almost won. a’. CLOSE-TO [Gore won] (proximal) ∧ ¬[Gore won] (polar) b. Bush barely won. (= Bush almost didn’t win) b’. CLOSE-TO ¬[Bush won] (proximal) ∧ [Bush won] (polar) Yet it’s not almost with its negative polar component, but barely, with a positive polar component, that licenses NPIs: (17) a. #Dana almost {moved a muscle/touched a drop/spoke to anyone}. b. Dana barely {moved a muscle/touched a drop/spoke to anyone}. The contrast in (17) makes life difficult for both DE-based and verdicality accounts of NPI licensing, à la Ladusaw 1979 and Giannakidou 1998, given the fact that barely environments are both non-DE and veridical.7 (17) is also problematic for Linebarger’s (1987) theory of indirect licensing, given the fact that (18a) entails, if it is not logically equivalent to, (18b), and vice versa, yet the intimate semantic relation between almost doing something and not quite doing it (cf. 6 “CLOSE-TO” in the representations in (16) is a place-holder for a more fully fledged representation of the proximal component in terms of what holds at alternative possible worlds or indices; see Sadock (1981), Sevi (1998), and Horn (2011) for elaboration. 7 Atlas (1997) supports the non-monotonicity (non-DE-ness) of barely VP by observing that if John is barely six feet tall, it doesn’t follow that he’s barely [six feet tall and blond], and we might add that if John is totally naked he’s not barely dressed. For more on the rhetorical asymmetries of proximatives, see Horn 1996, 2002, 2009, 2011 and Ziegeler 2000, 2006.

14

horn

also Atlas 2005: 156–158) fails to distinguish their behavior with respect to the distribution of NPIs shown in (19). (18) a. I almost completed my assignments. b. I didn’t quite complete my assignments. (19) a. I almost completed {some/*any} of my assignments. b. I didn’t quite complete any of my assignments. Extending the Strawson-DE theory of licensing (von Fintel 1999) to proximatives would treat statements like (15a,b) as neither true nor false if their “presuppositions” fail, but if Bush barely won, the proposition that he lost is simply false. On our account, proximatives—like exclusives and exceptives—crucially involve the contrast between what is entailed (both proximal and polar components) and what is asserted or at issue (with almost and barely, just the proximal component; with not quite, just the polar). Thus in (20a,b) (20) a. It’s too bad you almost died in the explosion—now you’ll need months of therapy. b. It’s too bad you didn’t quite die in the explosion—now I’ll have to polish you off. both proximal and polar components are entailed (I came close to completing the assignments; I didn’t complete them), but the two statements differ as to which of these is asserted (cf. Schwenter 2002). Thus in (20a), what’s too bad is that you came close to dying, but in (20b) what’s too bad is that you didn’t get there (yet). The dramatic contrast in (20) invokes the scopal diagnostics Karttunen & Peters (1979) originally posited for conventional implicature. Given the distinction between assertion and entailment, I have argued (Horn 1996, 2002) that these should be reconstructed as diagnostics for non-assertion (or non-atissue meaning). As seen in (21), material that is presupposed, conventionally implicated, or (on our account) inertly entailed will scope out of higher factives. Thus the prejacent of the only clause, i.e. the proposition that Jay endorses the theory, is understood as outside the assertive scope of the higher factives; what I just discovered or what is too bad is that nobody other than Jay endorses the theory, not that Jay himself does.8 8 Compare {I just discovered/it’s too bad} that Jay and only Jay endorses the theory, where

an (abridged) atlas of negation

15

(21) a. I just discovered that only Jay endorses the theory. b. It’s too bad that only Jay endorses the theory. By the same token, when the model contains the prejacent, the only statement can be asserted or questioned, but when it contains the exclusion, this is ruled out as an instance of redundancy: (22) a. I know Jay endorses the theory, but does ONLY Jay endorse it? b. #I know nobody besides Jay endorses the theory, but does ONLY Jay endorse it? c. I know nobody besides Jay endorses the theory, but I just realized (#only) he does. The same patterns obtain for proximatives. In advocating for my case that the polar component of almost and barely clauses, like the prejacent of only sentences, represents a non-asserted entailment, I argued (Horn 2002: 73) that (23) is understood as reporting a discovery that my cat came close to proving the theorem, not that she didn’t succeed. (23) I just discovered that my cat almost proved Fermat’s Last Theorem. Atlas (2007: 20) challenges the generality of this claim, on the basis of his intuition that (24a), when “asserted by a mathematician whose speciality is algebraic geometry and who has discovered the subtle error in Wiles’ first proposed proof given by him in a colloquium in Cambridge”, does indeed foreground the polar component not VP, hence asserting not (24b) but (24c). But I submit that it is precisely under those conditions that the algebraic geometer would utter not (24a) but (24d). (24) a. I just discovered that Andrew Wiles almost proved Fermat’s Last Theorem. b. I just discovered that Andrew Wiles ‘came close to’ proving it. c. I just discovered that Andrew Wiles did not succeed. d. I just discovered that Andrew Wiles didn’t quite succeed in proving Fermat’s Last Theorem.

both prejacent and exclusion are asserted and thus fall within the scope of the higher factives.

16

horn

Is almost equivalent to not quite, as Atlas seems to be suggesting here (contra Atlas 1984)? Well, yes, but only in terms of their entailments, not in terms of how those entailments are packaged. Atlas’s intuitions lead him to argue that the scopal diagnostics are unreliable for picking one’s way between the asserted and non-asserted meaning of sentences with exclusives and proximatives when the context is changed: For every example that seems to support a Karttunen-Peters diagnostic for an assertion’s utterance-meaning being just a foregrounded component of its sentence-meaning and a component of its sentence meaning not being asserted, one can, with a little ingenuity, construct an example with no such component of sentence-meaning exhausting the utterancemeaning of the assertion. Atlas 2007: 20

Similarly, in the case of exclusives, Atlas takes issue with the contrast I posit between (25a,b) (based on the pattern of (22) above). (25) a. I know Hillary trusts Bill, but does ONLY Hillary trust Bill? b. #I know no one distinct from Hillary trusts Bill, but does ONLY Hillary trust Bill? According to Atlas (2007: 21), “One merely has to spend time with the English middle classes, or the denizens of Harvard Yard, to hear [25b] as an arch, indirect, but linguistically acceptable way of asking whether Hillary trusts Bill”. No doubt I have lived a sheltered life, and I fully acknowledge the geographical and metaphorical distance between Harvard Yard and the campus of Yale, but I find it hard to share this judgment. Ultimately, we’re waging a war of intuitions and I’m satisfied with the preparation and morale of my troops; I’m sure Jay feels equally confident about his. Be that as it may, let’s assume that under certain circumstances, what is normally non-at-issue can become at issue. This is in fact not a new observation, dating back at least to Karttunen’s (1973) unpublished study of the assertion and presupposition of stop, which can reverse in the appropriate context (“I know John doesn’t smoke now, but judging from his nervousness and yellow teeth, perhaps he has recently stopped smoking”). In my first attempt to recruit Karttunen-Peters style scopal diagnostics to support an asymmetric approach to the contribution of only, I noted the existence of contexts in which the prejacent of only is brought within the scope of what is at issue, such as (26a),

an (abridged) atlas of negation

17

(26) a. Mary will be upset if only Bill makes it to her dinner party. b. Mary will be upset if Bill and only Bill makes it to her dinner party. in which “Mary may just be apprehensive about an evening alone with Bill; if no one shows up at all she will happily pop a meal into her microwave and a tape into her VCR,” thus essentially amounting to the predication in (26b).9 But this is precisely what is to be expected if the asymmetry of only, almost, or barely is a matter of pragmatics, subject to the vicissitudes of context and speaker’s local goals, rather than built into the semantics. Summarizing, then, on the proposed analysis of exponibles (Horn 2009)— only NP, almost VP, barely VP, and other quasi-conjunctives10—two conjuncts are semantically entailed, but only one is normally asserted or at issue. This allows us to capture linguistic and rhetorical asymmetries; NPI licensing, in particular, is linked to downward assertion rather than to downward entailment as such. This combination of semantic symmetry and pragmatic (in particular, speech-act) asymmetry was foreshadowed in a remark of the archconjunctionalist himself (Atlas 1991: 139): “[I]n asserting Only a is F, we do not thereby assert a is F … What we do assert entails a is F, but it does not ‘say’ it” (emphasis in original).11 Perhaps because on his account Only a is F does “say” that exactly one thing is F, Atlas does not draw the same conclusions I have drawn from the assertoric asymmetry we both posit. This may be also

9

10

11

In retrospect, the example would have been clearer with a different tense and complementizer, but the intuition remains the same for, say, Mary was upset that only Bill accepted her invitation. Thus, for example at most 3 as in (i) is an NPI licenser despite unpacking (on the analysis of Geurts & Nouwen 2007) into conjunction of (ii) + (iii): (i) At most 3 people have (ever) solved this problem. (ii) It is possible that 3 people have (*ever) solved this problem. (iii) for n>3, ¬It is possible that n people have (ever) solved this problem. On the present account, this is because the entailment in (ii) is assertorically inert; at most n asserts only the negative proposition in (iii) and is thus effectively downward monotonic. Another precursor of the moderately asymmetric position is William of Sherwood, in his Treatise on Syncategorematic Words XI.6 (Kretzmann 1968: 71–72): “It is asked why ‘alone’ [solus] is called an exclusive rather than an inclusive; for when someone says ‘Socrates alone is running’, Socrates is included under running but the others are excluded. It must be said that it is because the inclusion occurs not as a result of the force of the word but as a result of the statement as it is before the ‘alone’ is inserted into it. The exclusion, on the other hand, … does occur as a result of the force of the word [‘alone’].”

18

horn

attributable to Atlas’s conviction at the time that only a does not serve as a licenser of NPIs, leading him to adopt the seductive but, I would argue, empirically inadequate pure conjunctive theory.12 In any case, Atlas himself seems to have undergone something of an evolution (if not a conversion) as to the viability of assertoric inertia. As we have seen, Atlas (2007) is particularly skeptical of distributional arguments hinging on the scopal diagnostics of Karttunen & Peters (1979). Then there may be a residual skepticism borne of the nature of inert entailments, along the lines of the discomfort evinced in a contemporary treatment of only. Presenting a chronology of my various characterizations of the meaning of only over the decades, Beaver & Clark (2008: 244–245) conclude with this observation: The position he [= yours truly-LH] ended up supporting is that both the positive and negative components form part of the entailed meaning of only. However, the positive component suffers from a peculiar impediment Horn calls inertia, an impediment which apparently prevents the positive component from getting caught up in the hustle and bustle that normal entailments face in their everyday life. For example, inert entailments do not like to be the main target of an assertion, a negation, or an emotive attitude. In fact, they behave suspiciously like presuppositions. … as if we knew what THOSE are. While on my view, the prejacent of only clauses (like the corresponding non-at-issue entailments of exceptives and proximatives) is indeed “part of the entailed meaning,” presuppositions on most accounts are not (or not necessarily) entailed, a fact that might help dispel the suspicion that the view defended here is simply a shift in nomenclature. Inertia is matter of pragmatics, not semantics, so the observations in Atlas (2007) constitute support rather than a challenge for the proposal. If Atlas 2007 is doubtful as to whether assertoric inertia (while “a genuine pragmatic advance”) is really adequate to account for the distribution of negative polarity items in the kinds of contexts we have been considering, Atlas 2012 is more favorably inclined. He suggests that in harnessing an account of NPI licensing, my invocation of assertoric inertia has obviated earlier critiques (by me and others) of what he calls the “Atlas-Kempson thesis” on the semantic non-specificity of negative sentences (Atlas 2012: 352, 361–363). 12

One additional problem I have with Atlas’s conjunctive unpacking of (i) as (ii) is my sense that the latter is an extremely implausible sentence of English. (i) Only Muriel voted for Hubert. (ii) Exactly one individual, and at most Muriel, voted for Hubert.

an (abridged) atlas of negation

19

At the same time, Atlas would seem to be wedded to his objections to the application of Karttunen-Peters diagnostics to the dispute over the linguistic asymmetry of only and, by extension, of other “exponibles”, e.g. no … but, almost, barely, et al., as practiced in Horn 2002, 2009, 2011. Furthermore, there remains the question of whether this asymmetry, once admitted, might or might not extend to the issue of whether the normally backgrounded component of meaning (the prejacent of only clauses, the polar component of almost and barely clauses) is in fact entailed after all, and thus not subject to contextual override. In the case of exclusives, this issue has often boiled down to the possibility of epistemic cancellation.

2

On the Status of “Epistemic Cancellation”

We know, because the Internet tells us so, that Georg Wilhelm Friedrich Hegel’s last words were as follows:13 (27) †Only one man ever understood me, and he didn’t understand me. We can forgive Hegel his rather desperate attempt to cancel the prejacent of an exclusive given that he was half-dead at the time—and given that, after all, he was Hegel. For those not similarly in extremis, cancellation per se is beyond the pale. But what of suspension, a.k.a. epistemic cancellation? The putative possibility of sentences like (28a,b) constitutes a central argument in the (independent) defenses by Ippolito (2005, 2006) and van Rooij & Schultz (2007) of conversational implicature-based analyses of the prejacent. (28) a. Only [Hillary]F trusts Bill, and maybe not even she does. (Ippolito 2005, (76); cf. Ippolito 2006, (26b)) b. Only [Hillary]F trusts Bill, and perhaps even she does not. (van Rooij & Schulz 2007: (25a)) The claim that the prejacent is easier than the exclusive component to cancel (with an epistemic rider) goes back a while. The examples in (28) are variants of similar ones in Horn (1996), whence also comes the minimal pair in (29):

13

The † diacritic is introduced here to annotate example sentences that may be felicitously uttered only by those on their deathbed.

20

horn

(29) a. Only Kim can pass that test, and #(it’s possible) even she can’t. b. #Only Kim can pass that test, and/but (it’s possible) someone else can. Setting our time machine dial a bit farther back—to Horn (1970: 322)—we arrive at the data in (30), fragrantly redolent of their era. (30) a. Only John smokes pot, and even he may have quit. b. Nobody but Nixon is worthy of contempt, and possibly even he isn’t, either. c. Everybody but Nixon is worthy of salvation, and possibly even he is, too. Even if suspension of the prejacent of exclusives and the exception clause of exceptives as in (30) is not impeccable, there is a clear contrast with cases in which we attempt to suspend the asserted component: (31) a. #Only John smokes pot, and/but Mary may too. b. #Nobody but Nixon is worthy of contempt. Agnew possibly is too. c. #Everybody but Nixon is worthy of salvation, and possibly Agnew isn’t either. This contrast is especially vivid in cases in which the suspender is distinct from the original utterer; while the acceptability of B1’s response below may vary (as indicated by the diacritic), B2’s version is totally out. (32) A: B1: B2:

—Only Hillary trusts Bill. —You’re right. %And what’s more, maybe (even) she doesn’t. —You’re right. #But maybe Chelsea does too.

Atlas (1996: 310–311; 2002) remains unconvinced of the significance of the subtle distinctions drawn here and in the earlier literature: despite what Horn “incorrectly thinks” and “mistakenly believes”, “there is no epistemic cancellation at all”. On the other side, we have not only the intuitions of pragmatic and semantic asymmetricalists from Horn and von Fintel to van Rooij & Schulz and Ippolito to rely on, but the testimony of the man on the street, or rather the blogger and chatter on the web, as in (33). (33) γSince turbulence in the forecast volume is patchy, you can expect to encounter it only intermittently and possibly not at all. γBut the social roles in this experiment were not the only thing and possibly not the most important thing that was different about the two objects.

an (abridged) atlas of negation

21

γ … comforts and conveniences previously enjoyed by only the most wealthy and pampered, and perhaps, not even by them γFor some reason known only to Microsoft and maybe not even them, using the Orientation constants requires you to be in a different universe. γ … the Assembly can do only one thing at a time, and possibly not even that γCode so arcane, complex, or ill-structured that it cannot be modified or even comprehended by anyone but its author, and possibly not even by him/her. Additional instances of epistemic cancellation of the prejacent of a range of only phrases with different foci are provided in Horn 2009: 215; cf. also Beaver & Clark 2008: §9.6 for additional examples and argumentation and a consideration of the role of aspect and modality in facilitating suspension. But in each case, it’s again only the prejacent that can be suspended (if that!), never the exclusion.14 Thus, when sportscaster Colin Cowherd, who obviously hadn’t consulted his Atlas, opined on ESPN on April 9, 2006, “I think the only team that can beat those guys [the Pistons] are the Spurs, and I don’t think THEY can beat them”, he could not have instead stated “I think the only team that can beat those guys are the Spurs, {and/but} I think Miami can beat them too”—at least not without changing his mind mid-prediction.

14

Nor is suspension possible when both entailed conjuncts are directly asserted, i.e. in the frame *p and only p, but possibly (even) not-p. Similar observations hold for only if conditionals, where epistemic cancellation—recruited by Horn (1996) on behalf of a compositional treatment of only if, contra Atlas (1996)—is supported by examples such as these: γThe statement will be excluded from evidence only if the officers interrogated you in a custodial situation without first reading you your rights—and perhaps not even then. γpedophilia is one of those conditions that can be “cured” only if the patient desperately wishes to recover from it (and perhaps not even then.) γThe illusion of effective phrasebooks is possible only if trivial items are being exchanged, and possibly not even then. γ[A contractor is …] Someone who will return your phone call only if he’s signed a contract. And maybe not even then. Alongside the readily attested instances of only if p but maybe not even (if p), we find no cases of only if p … and/but maybe if not-p, nor is epistemic cancellation possible in the frame if and only if p … and maybe not even if p.

22

horn

So what can we can conclude from this brief for asymmetry? For Ippolito and van Rooij & Schulz, the ability to suspend or epistemically cancel the prejacent forces the adoption of an implicature-based analysis of the relation between only clauses and the prejacent, à la McCawley (1981) and Horn (1992, 1996). But the neo-McCawleyan analysis, despite its appeal to current and past cardcarrying asymmetricalists, cannot account for the fact that the prejacent in only sentences simply does not behave like clear instances of implicature. This is seen most clearly in exchanges in which a speaker whose utterance (at least potentially) implicates something that the hearer’s response cancels or suspends. Thus, in (34) the upper-bounding scalar implicature of A’s utterance is cancelled by B without an epistemic rider, while in (35) the conventional implicature of the eventuality end-point associated with A’s use of (not) yet is suspended by B. But in (36), the prejacent is entailed rather than merely implicated, and consequently it cannot be cancelled by B1 or even for some speakers suspended by B2. (34) A: Are some dogs mammals? B: Yes, (some dogs are mammals) and indeed all dogs are mammals. (35) A: You haven’t fully recovered yet. B: No, and what’s more, (I’m beginning to think) I never will. (36) A: B1: B2:

3

Does only Bush trust Cheney? #Yes, and what’s more, even he doesn’t. %Yes, and indeed it’s possible even he doesn’t.

Entailment or Implicature? Evidence from the Griceogloss

In the previous two sections, we have reviewed the conflicting evidence for and against entailment- and implicature-based treatments of the meaning contributions of only, almost, barely, and related exponibles. There is, however, additional evidence we can bring to bear, in the form of the subset diagnostic— or “Griceogloss”—involving mixed-outcome situations such as those as in (37) and (38). (37) a. 20 students tried to solve the problem. [includes any who solved it] b. 20 students solved most of the problems. [includes any who solved all]

an (abridged) atlas of negation

23

c. 20 students don’t drink much. [includes any teetotalers] With clear instances of conversational implicature, we get inclusive readings for the subset described. Thus, if 5 of the students under consideration in (37a) solved the problem, they constitute 5 of the 20 who attempted a solution. If 5 members of the set in (37c) are teetotalers, there are 15 imbibers, again making 20 in all. But now compare predications based on proximatives, exclusives, and exceptives: (38) a. 20 students almost solved the problem. [excludes any who solved it] b. 20 students barely passed the test. [excludes those who failed] c. 20 students solved only the last problem. [excludes those solving none] d. 20 students solved all but the last problem. [excludes those who solved them all] If (38a) is true and 5 students solved the problem, we have a total of (at least) 25 students—the 5 who succeeded and the 20 who almost did. For (38b), similarly, the set of students who actually failed is disjoint from the set of the 20 who barely passed (and thus did pass). In (38c), we again must consider those students who solved no problems to be outside the set of 20 who solved just the last one. And in (38d), we must add those students who solved every problem but the last one to those who solved all the problems. Note also the minimal contrast between only in (38c) and the downwardentailing value at most in (39), as well as the patterning of scalar only (asserting ‘no more than’ and inertly entailing ‘at least’) and its counterparts illustrated in (40). (39) 20 students solved at most the last problem. [includes those solving none] (40) a. 20 of the students are sophomores. [excludes freshmen + juniors] b. 20 of the students are at least sophomores. [includes juniors] c. 20 of the students are at most sophomores. [includes freshmen]

24

horn

d. 20 of the students are only sophomores. [excludes freshmen + juniors] We must conclude that as indicated in §1, and in spite of the complications posed by epistemic cancellation facts in §2, Atlas is correct in arguing that despite the evidence for asymmetries obtaining on a pragmatic level between the components of exponibles, conversational implicature does not suffice for describing the status of the prejacent component of exclusives or the polar component of proximatives.

4

Clefts and Exhaustivity Revisited

We now turn to the exploration of one more front in the symmetricalist wars. Unlike focus constructions built on only, pure focus constructions (clefts, pseudo-clefts, bare focus) do not entail exhaustiveness. I argued in Horn 1981 that while (41a) presupposes or conventionally implicates the existential premise in (41b), the upper bound in (41c) is not a conventional implicature (contra Halvorsen 1978), nor is it—or its counterpart in (41d), as Atlas & Levinson (1981) maintain—an entailment of (41a), but rather a generalized conversational implicature. (41) a. It was [a pizza]F that Mary ate. b. Mary ate something. c. Mary ate nothing (within a contextually inferrable set) other than the pizza. d. Mary ate at most one thing (within a contextually inferrable set). This conclusion is supported by the cancellability and non-detachability of the upper bound (see also the supporting evidence in Vallduví 1992 and Blok 1993). For Atlas (1992, 2005), on the other hand, clefts—at least those with definite foci—are essentially equivalent to overtly exclusive propositions, so that It was the pizza that Mary ate unpacks as Mary ate something and Mary ate at most the pizza. Crucially, it entails that Mary ate exactly one thing (contrary to the observations in Horn 1981) and essentially asserts that Mary ate only the pizza. More recently, Atlas (2005, Appendix 3) takes both definite and indefinite clefts to entail an at most exhaustivity condition. It’s a that Fs is logically equivalent to only a Fs, though the two are non-synonymous,15 and both of these are 15

Cf. also Atlas 2005: 195: “Only John hit Brian is logically equivalent to It was John who hit Brian”.

an (abridged) atlas of negation

25

presumably also equivalent to it’s only a that Fs. But in fact, only a and it’s only a both entail and assert exhaustivity, while clefts and other focus constructions assert the assignment of a value for a given variable and only implicate exhaustivity. Such a conclusion is inconsistent not only with Atlas’s claims about focus constructions but with those of much (though not all) of the previous literature. More specifically, it has been claimed that only-type readings are associated with the category of exhaustive focus (Szabolcsi 1981) or identificational focus (É. Kiss 1998), expressed in Hungarian by preverbal placement or by clefts in English, as opposed to simple new-information focus as typically signaled by prosody. In Krifka’s words, [Exhaustive focus] indicates that the focus denotation is the only one that leads to a true proposition … Kiss (1998) has pointed out that focus movement in Hungarian triggers this specific meaning, and it appears that cleft constructions in English trigger it as well: It’s [JOHN and BILL]F that stole a cookie. This example SAYS that nobody else but John and Bill stole a cookie. Krifka 2008: 259, emphasis added

Can we empirically confirm or disconfirm the claim that structural focus constructions, as reflected by fronted focus in Hungarian, clefts in English and French, and preverbal focus in languages like Greek and Italian, assert and entail exhaustivity, as this view suggests? The first step is to show that whatever It’s John and Bill that stole a cookie says, it’s not that nobody else did. Crucially, we find that clefts, like other focus constructions, fail to license NPIs, absent a dedicated exhaustivity marker:16 (42) a. b. c. d.

It’s Jay who has (#ever) modeled negation with Necker cubes. It’s only Jay who has (ever) modeled negation with Necker cubes. [Jay]F has (#ever) modeled negation with Necker cubes. Only [Jay]F has (ever) modeled negation with Necker cubes.

Linebarger (1991: 184) notices the unacceptability of NPIs in the scope of clefts but offers the wrong diagnosis for the problem, maintaining that NPIs can’t occur in the “background portion” of a focus construction like (42a) or (43a).

16

As far as I know, the argument from NPI licensing for the non-at-issue status of exhaustivity in clefts appears for the first time in my Jayfest presentation, Horn 2005.

26

horn

But background triggering is impeccable if there’s an overt only in the higher clause, as seen in (42b) and (43b): (43) a. *It was John who contributed a red cent to the ACLU. [Linebarger 1991, (42a)] b. It was only John who contributed a red cent to the ACLU. The same contrast obtains in th-clefts, as seen in what I have called, in homage to Dusty Springfield and Aretha Franklin, the SPM construction: (44) a. The only one who could ever reach me was the son of a preacher man. b. #The one who could ever reach me was the son of a preacher man. This contrast is quite general; as of June 15, 2015, there were 342 actual Google hits for the sequence “I’m the only one who ever” compared to just 13 for “I’m the one who ever” (several of them likely not from native speakers). As with itclefts, NPIs are permitted in th-clefts only if maximality (and not just identity) is asserted.17 A second counterargument for the non-at-issue status of exhaustivity in clefts and focus constructions and for the sharp demarcation of such constructions from explicitly marked exclusive clauses (contra Atlas 2005, among others) is provided by the distribution of Karttunen-Peters scopal diagnostics we introduced in § 2. For Atlas & Levinson 1981, the analysis of It was a pizza that Mary ate is essentially conflated with that of Mary ate a pizza and only a pizza (or perhaps, if this is logically distinct, to It was a pizza and only a pizza that Mary ate). The problem with that conflation, as noted in Horn 1981, is the existence of data like those in (45), indicating that “a cleft sentence is pointless to assert or to question, and idiotic to deny, if the corresponding simple [proposition] is already established” in the common ground (Horn 1981: 130). With the insertion of only, exhaustivity becomes at issue; the continuation is then no longer pointless or idiotic, and the infelicity disappears.

17

Other negative expressions within a cleft can similarly allow for NPIs, as in (i), (i) Small reptiles and rodents burrow or slide below the surface or cling to the shaded side of an outcropping. Movement is slow to preserve energy, and it is a rare animal which can or will defy the sun for long. (Steinbeck 1961: 164) As in (43) or (44), it is not the focus itself that functions as the actual licenser here, but the modifier rare.

an (abridged) atlas of negation

27

(45) a. I know Mary kissed John, but it was {#JOHN/✓only JOHN} she kissed. b. I know Mary kissed John, but was it {#JOHN /✓only JOHN} she kissed? c. Mary kissed John, but it wasn’t {#JOHN/✓only JOHN} she kissed. The evidence presented so far bears on the assertoric status of the exhaustivity implication and confirms the view that exhaustivity is not at issue in the relevant English clefts (or, as Wedgwood 2005 and Onea & Beaver 2009 stress, in the corresponding examples with preverbal focus in Hungarian). What has not been shown is whether this non-at-issue component of meaning is a conversational implicature, as maintained in Horn 1981, or is truth-conditionally relevant but inert in the manner of the prejacent of only/no … but sentences or the polar component of almost and barely approximatives. As we have seen, such inert entailments—while failing to affect NPI licensing and scoping out of higher factives—are also uncancellable, which motivates their treatment as (non-at-issue) entailments rather than implicatures. To show that exhaustivity is neither asserted nor entailed in structural focus constructions thus depends on showing that this implication can indeed be canceled. To this end we can begin with some attested examples that tend to suggest that exhaustiveness implications associated with clefts in positive contexts can be overridden in the right circumstances (emphasis added), while as we saw in §2 exhaustivity in only sentences cannot be overridden: (46) a. He was just a burned-out little man with tired eyes and a drained smile, who had a gift that was too big for his soul, and it was the gift that killed him, as much as anything. (Robert Silverberg, Stochastic Man, via Ellen Prince; Horn 1981, (13b)) b. It’s the ideas that count, not just the way we write them. (Richard Smaby, lecture, via Ellen Prince; Horn 1981, (13d)) c. McGwire has been hurt, and while the Cardinals have been under .500 since he went down and struggled to mirror their record for home runs in April, it has been their bullpen and other malfunctions as well as the loss of McGwire’s presence in the lineup that has led to the Cards’ struggles. (Peter Gammons, espn.com, 26 April 2000) d. As we go marching, marching unnumbered women dead, Go crying though our singing their ancient call for bread. Small art and love and beauty their drudging spirits knew. Yes, it is bread we fight for—but we fight for roses too! (1911 poem by James Oppenheim and women’s workers’ anthem; cf. http://en.wikipedia.org/wiki/Bread_and_Roses)

28

horn

The constructed examples in (47)–(49) make the same point: (47) A: It’s Ruth who posited semantically underspecified negation. B: Yes, you’re right—Ruth, and Jay, and Robyn. (48) A: It’s only Ruth who posited semantically underspecified negation. B: #Yes, you’re right—Ruth, and Jay, and Robyn. (49) A: Only Ruth posited semantically underspecified negation. B: #Damn straight—Ruth, and Jay, and Robyn. The literature on clefts contains a variety of arguments against the semantic exhaustivity of clefts (or other so-called “exhaustive listing” constructions) in English, Japanese, French, Greek, and Hungarian; see e.g. Shibatani 1990; Vallduví 1992; Heycock 1994; Wedgwood 2005; Gryllia 2009; Onea & Beaver 2009; Zimmermann & Onea 2011; Drenhaus et al. 2011; Byram Washburn et al. 2013; Horn 2013a, 2016b; Destruel et al. 2015. The possibility of non-contradictory extensions in which the exhaustivity implication of both new information and identificational focus is explicitly cancelled has been noted for a range of other cases.18 In particular, it is worth noting that even in the case of Hungarian structural focus that motivated the Szabolcsi-É. Kiss account there have been alternative treatments that challenge this view. Wedgwood (2005) argues at length that exhaustivity is not part of the encoded meaning with either identificational focus or informational focus. Rather, an exhaustivity inference can be derived pragmatically from the nature of alternative sets arising in discourse contexts and the function of preverbal focus. Edgar Onea and his colleagues (Onea & Beaver 2009, Zimmermann & Onea 2011) maintain that exhaustivity is an epiphenomenon arising from the interaction of focus with the question under discussion. More specifically, “immediately preverbal focus is grammatically constrained to be a question-answering constituent” (Onea & Beaver 2009: 343), and as has often been remarked (cf. e.g. Schulz and van Rooij 2006),

18

The other side of the coin is the well-formedness of examples in which an identificational focus claim is denied on the grounds that exhaustivity is not satisfied: It wasn’t a PIZZA that Mary ate, it was a pizza and a calzone. Horn 1981: fn. 7 takes such cases to involve metalinguistic negation and Onea & Beaver (2009) invoke the same analysis for the counterparts of such sentences in Hungarian. Of course, metalinguistic negation is a suspect device for Atlas (2005), so it’s not clear that this appeal would be congenial to him.

an (abridged) atlas of negation

29

answers to wh-questions tend to be taken exhaustively at least by default (modulo the mention-all vs. mention-some distinction). At the same time, Onea’s work, along with that of other researchers on Hungarian, does confirm the original Szabolcsi/É. Kiss observation that immediately preverbal focus has a stronger implication of exhaustivity than does in situ focus. In the same way, it is undeniable, contra Horn 1981, that the hearer is more likely to infer speaker-intended exhaustivity in the case of a cleft than in the case of bare focus. One relevant factor may be the status of the existential proposition standardly taken as a presupposition or conventional implicature in the case of focus. It is undeniable that this proposition is often part of the common ground.19 In any case, recent experimental studies (see earlier cites of work by Onea & Beaver, Drenhaus et al., Destruel, and Byram Washburn et al.) have supported the position that exhaustivity, while an uncancellable component of what is said in only sentences, can be overridden (more or less easily) in the case of ex situ as well as in situ focus constructions with no explicit exclusive marker. Given the cancelability data, the only conclusion is that exhaustivity implication in clefts is not part of at issue content, nor (unlike the prejacent of only clauses) is it part of entailed content. Rather, it is indeed—as suggested although not coherently argued for in Horn 1981—a conversational implicature. Based on these findings, the apparent symmetry between focus constructions and exclusives, as described by Velleman et al. 2012 and Coppock & Beaver 2014, is striking but incomplete. (50) a. It’s α that F’s b. Only α F’s (50’) a. F(α) b. ¬∃x, x∈D (x≠α ∧ Fx) What is asserted in an utterance of (50a), i.e. (50’a), is not at issue in an utterance of (50b); what is asserted in an utterance of (50b), i.e. (50’b), is not at issue in an utterance of (50a). But (50’a) is an entailment of (50b), while (50’b) is not an entailment of (50a). 19

Dryer (1996) maintains that the existential proposition is indeed given in the sense of pragmatically presupposed in the case of clefts; in bare focus constructions, on the other hand, the relevant sense of givenness is closer to “activated” or “under discussion”. It is to this distinction that he attributes the well-known difference between NOBODY solved the problem and * It is nobody who solved the problem.

30 5

horn

Bullshit and Other Refutations

For a final diagnostic for distinguishing truth-conditionally relevant components of meaning (including non-at-issue entailments) from merely conversational implicatures, we turn to refutations and objections (see Smith et al. 2013). The refutation marker bullshit (cf. Ward 2003, Horn 2013b) seems for most speakers to be restricted to truth-conditional-based objections and thus does not readily serve to reject conventionally implicated material, as seen in (51) where the at-issue prejacent but not the implicatures of non-uniqueness and surprisingness (or noteworthiness; cf. Karttunen & Peters 1979, Kay 1990) can be the target of refutation. (Some speakers with a more lenient distribution for bullshit might find the results clearer by employing “That’s false” or “That’s wrong” as a refutation marker.) (51) A: B1: B2:

Even Hercules can lift that rock. Bullshit—he can’t lift it. #Bullshit—{he’s the only one who can/he’s the most likely to lift it}.

But non-at-issue material can qualify for a bullshit refutation provided that it’s entailed, as shown by sequences involving only or almost: (52) A: B1: B2:

I love only you. Bullshit, you love Chris too. Bullshit, you don’t love anybody.

(52’) A: B1: B1:

I almost cheated on you. Bullshit, you would never dare even think of cheating on me. Bullshit, you DID cheat on me.

Consider, now, the possible responses to clefts with and without only, assuming a monogamous context in which Juliet considers Romeo’s former love Rosaline as a potential romantic rival: (53) R: J1: J2: J3:

It’s you I love. Yes/Maybe so, but you love Rosaline too. #Bullshit, you love Rosaline too. Bullshit, it’s Rosaline you love.

(54) R: J1: J2:

It’s only you I love. #Yes/Maybe so, but you love Rosaline too. Bullshit, you love Rosaline too.

an (abridged) atlas of negation

31

While there is some variation in judgments, akin to those found by the exhaustivity-cancelling studies reported by Onea & Beaver (2009) and Destruel (2013), the evidence from the contrast between (54) and (55) in the viability of “Yes, but” and “Bullshit” responses, though not definitive, tends to support the conclusion that exhaustivity is not only external to the at-issue content of clefts but is also a non-truth-conditional aspect of the meaning contributed by structural focus. To further appreciate the power of the evidence from refutation (as well as from the Griceogloss diagnostic illustrated in §3), we turn to another bone of continued contention, the proper analysis of few.

6

Right Church, Wrong “Few”?

Finally, let us take up once again the gauntlet of assertoric inertia. We have noted the gradual softening in Atlas’s attitude toward the invocation of nonasserted entailment as we move from the skepticism toward this construct in Atlas 2007 to the peaceful coexistence favored in Atlas 2012. In more recent work, Atlas (2013) seeks to generalize the assertoric inertia style analysis of only (see §1 above and references therein) to the treatment of few. Atlas concludes the first section of his disquisition ends with this capsule statement of the subject of inquiry: So if Zwarts, Atlas, Horn, and untold others construed few trees as not many trees, the judgment that few trees was monotone-decreasing was a justifiable one for the preferred interpretations of the sentence-type containing ‘not many trees’. The question is whether ‘not many trees’ is the correct construal of ‘few trees’. Atlas 2013: 3

While the thrust of Atlas’s paper argues for a negative answer to this question, I do not find the argumentation convincing. To be sure, few and not many are not always intersubstitutable salva grammaticalitate, even if they are (as I would maintain, absent counterevidence) intersubstitutable salve veritate. The essential difference between them is that not many is, while few is not, morphologically negative, a factor relevant for the behavior of these expressions with respect to the “Klima diagnostics” (Klima 1964) for sentence negation, as illustrated in the minimal pairs in (55): (55) a. {Not many/#Few} of my friends will be joining us, not even for dessert. b. {Not many/#Few} senators will vote for the bill, will they?

32

horn

Along the same lines (cf. Horn 1996: 31, fn. 12), consider the contrasts in (56); pleonastic negative parentheticals like I don’t think co-occur with morphologically negative quantifiers and quantificational adverbs (not many, not often) and exceptives (no … but) but not with their implicitly negative counterparts ( few, seldom, only): (56) a. {Not many/#Few} people, I don’t think, really know what it means. b. {Not often/#Seldom} do you wear a seat belt, I don’t think. c. {Nobody but/#Only} Trump, I don’t suppose, would have put it that way. Thus not many is more negative syntactically than few, given its n-initial form, but what of the semantic evidence? Evidence from the distribution of the standard scalar diagnostics (Horn 1972, 1989) indicates that while few and not many are stronger values on the negative scale than not all and weaker ones than no/none, as seen in (57), they cannot be positioned relative to each other, as seen in (58): (57) a. {Few/Not many} and possibly none of the students will fail the test. b. γIt is not at all unreasonable to predict that few or even no felons will be caught by the polygraph. c. Not all the guests will try the poached eel, and indeed few of them will. d. γNot all, and indeed, not many officially “X” rated motion pictures would commonly be considered to be pornographic. (58) a. #Few and possibly (even) not many of the students will fail the test. b. #Not many and possibly (even) few of the guests will try the poached eel. To further demonstrate the systematic nature of these patterns, note that they generalize fully to little and not much, the mass counterparts of few and not many respectively: I expect there will be {little/not much} or (even) no wine left over after the symposium vs. #I expect there will be {little or (even) not much/not much or (even) little} wine left over after the symposium. Thus we obtain the quantity scales in (58a,b), versions of which date back to Horn 1972. (58) a. {not all, few/not many, no(ne)} b. {not all, little/not much, no(ne)}

[for count DPs] [for mass DPs]

an (abridged) atlas of negation

33

We now turn to a positive argument for the putative non-negative character of few. Atlas follows Geach (1980: 108) in endorsing a positive entailment for few CN clauses: (59) “Few As are so-and-so” = “Most As are not so-and-so but some are” But what is Geach’s actual evidence for treating few As are so-and-so as an exponible that incorporates a positive existential component? The unpacking appears to be based on the behavior of anaphora: [A] proposition of the form f (few As) & g(the same As) is a portmanteau proposition into which two distinct propositions are packed … For example, “Few M.P.s spoke against the bill, and they were Tories” unpacks as: “Most M.P.s did not speak against the Bill; but some M.P.s did speak against the Bill, and they were Tories”. Geach 1962: 88; Geach 1980: 118

Some commentary on Geach’s observations is apropos here. As Atlas (2013: 4) insightfully recognizes, this analysis of few is “importantly unlike his suggestion for ‘only’”. Indeed, Geach’s analysis of only is radically asymmetric: Only a Fs is true iff nothing distinct from a Fs, and similarly in the case of general terms Only an A Fs is vacuously true if F is empty: F “only an A” is true iff “F(x)” is not true for any interpretation of “x” as a proper name, i.e.: iff “F(x)” is not true of anything. Thus, for example, “Only a dragon breathes fire” turns out to be true iff nothing at all breathes fire. This result appears harmless. Geach 1962: 188

It is formally much more convenient to treat the exclusive proposition as having precisely the exclusive force of its supposed second component, and not to read “F (only Θ)” as implying “F(some Θ)” (i.e. in the degenerate case where “Θ” is taken to be a single proper name, as implying “F(Θ)”). Geach 1980: 208

The evidence I have presented in this study and in previous work leads me to conclude that Atlas is correct in rejecting Geach’s purely negative analysis-ofconvenience of only but incorrect in following Geach’s non-negative analysis of few. Let us begin by unpacking the portmanteau.

34

horn

While the anaphoric follow-up in Geach’s example, “… and they were Tories”, is indeed licensed by the Few M.P.s subject expression, it is not clear why this would lead Geach to see the pronoun as a marker of the presence of a positive semantic entailment; after all, Only three M.P.s spoke against the Bill and they were Tories would then refute his analysis of only as failing to entail a positive proposition. Just what are the facts about anaphora associated with subsets determined under quantification? As it happens, this is a question that has been systematically investigated in empirical work by Anthony Sanford, Linda Moxey, and their associates. Moxey & Sanford (1986, 1993) distinguish refset (reference set) from compset (complement set) anaphora, and demonstrate that while a few X do Y highlights “the small subset of X of which ‘do Y’ is true (the ‘refset’)”, few X do Y highlights “the large, complement, subset of which ‘do Y’ is taken not to be true” (Moxey & Sanford 1986: 189). They also explore the behavior of only a few clauses, which share the intuitively negative character of few clauses but sometimes pattern with the positive (indeed, if Atlas 2013: fn 11 is correct, upward entailing) determiner a few, for example in allowing negative oppositepolarity tags in contexts like {*Few/ ✓ A few/ ✓Only a few} of the guests enjoyed the party, didn’t they? Among the data Moxey and Sanford examine is the reaction their subjects give to sentences like those in (60), which seems to be, but isn’t, patterned on Geach’s locus classicus for illustrating his portmanteau analysis of few: (60) {Few/Very few/A few/Only a few} MPs went to the meeting. They … The responses of their subjects can be summarized in this table, an abridgment of their Table 1 (Moxey & Sanford 1986: 195).20

(61)

refset interpretation compset interpretation few very few a few only a few

20

14 8 38 38

25 29 0 2

The (very few) responses in which subjects construed they with neither the “refset” (= the MPs who attended the meeting) nor “compset” (= the MPs who didn’t) are omitted here.

an (abridged) atlas of negation

35

As Moxey & Sanford (1986: 195–196) summarize the results: The expression few biases people toward the compset, and this tendency is marginally enhanced by the intensifier very (although this is not a statistically reliable effect). In contrast, a few and only a few put focus on the refset. However, when the second sentence in the prompt appears with an overt connective (DET MPs went to the meeting and they …) rather than in simple juxtaposition, the bias is overridden so that few patterns with a few and only a few, inducing refset and blocking compset construal; not a single subject favored compset construal with few in the presence of an overt and. Thus Geach’s example tells us more about the discourse contribution of and than it does about the semantic contribution of few. It might also be noted that while the intensifying effect of very (in rendering few even more refset-friendly than it otherwise is) cannot be directly applied to the contrast between few and a few, given the impossible of * very a few, shifting to a different intensifier yields the relevant results: quite a few moves us upward, increasing the size of the subset in the manner of quite a lot or very many, whereas quite few, like very few, picks out a smaller subset than few itself does. This supports the case for the downward monotone character of few. Given the analysis of Atlas (2013), few CN should pattern essentially with only n CN or more generally only DP. But closer inspection shows that this prediction is not fulfilled. To begin, note that only-initial subjects, with a variety of possible foci, do not readily permit refset anaphora in the absence of an overt connective, as Moxey & Sanford already showed for only a few: (62) a. {Few/#A few/#Only a few} of the students came to the party. They stayed home studying for midterms (instead). b. #Only {three students/the most prepared of the students/Sam} came to the party. They stayed home studying for midterms (instead). It is true that only a few, like few but unlike a few, licenses NPIs. The paradigm in (63) comes from Moxey & Sanford (1986: 205). (63) a. Few people have any manners anymore. b. #A few people have any manners anymore. c. Only a few people have any manners anymore.

36

horn

At the same time, it should be noted that few CN (which is downward entailing, although not anti-additive, on the standard view I am defending) is a better licenser of strict NPIs than is only: (63’) a. b. c. d.

γFew of my friends wore bras [in the 70s], either. γFew tourists have been inside in years. #Only one of my friends could make it here either. #Only Jay has worked on that problem in years.

As we observed in §4, however, neither arguments based on NPI licensing nor those based on the propensity to scope out of higher predicates à la Karttunen & Peters (1979) can be truly decisive in determining whether a given proposition is an inert entailment (like the prejacent of only) or a mere conversational implicature (like the exhaustiveness premise of clefts). Rather, what is crucial is behavior with respect to the “Griceogloss” of § 3, cancelability contexts, and rejection devices. Let us take up each of these in sequence. First, we see that few CN patterns with the downward entailing at most and not with the non-monotonic (although pragmatically asymmetric) only DP, while only a few patterns with latter. (Compare the data in (37) and (38) above.) (64) a. 20 students solved at most one problem. [includes those solving none] b. 20 students solved only one problem. [excludes those solving none] c. 20 students solved few problems. [includes those solving none] d. 20 students solved only a few problems. [excludes those solving none] This supports the thesis that, contra Atlas 2013, few CN differs from only a few CN in merely implicating (as opposed to inertly entailing) the positive existential counterpart (here, that 20 students solved some problem(s)). On the proposed account, the relation between few A are B and some A are B is one of scalar implicature, and hence an affirmation of the former should be consistent with the denial (and not merely the “epistemic cancellation”) of the latter. In fact this relation has been under the microscope for a good halfcentury, give or take. Chomsky (1972: 112) in fact depicts the inference from not many/much to some in contexts like (65a,b) as a presupposition, while presupposing an equivalence between few and not many (and between little and not much).

an (abridged) atlas of negation

37

(65) a. Not many arrows hit the target. b. Some arrows hit the target. For Chomsky, “Sentence [(65a)] (equivalently, Few arrows hit the target) presupposes that some arrows hit the target”. Similarly, Chomsky takes (66a) to presuppose (66b): (66) a. {Not much/Little} enthusiasm was shown for that project. b. At least some enthusiasm was shown. Horn 1972: 77 argues that we are dealing in each case with a Quantity (scalar) implicature, not a presupposition. In either case, if we are correct in equating few and little semantically to not many and not much respectively, it seems implausible that we can avoid monotonic analyses of any of these values. If Few arrows hit the target and Little enthusiasm was shown are monotone decreasing and merely scalar-implicate the corresponding positive existentials, the implicatures should be cancelable—and they are. Contextual cancellation of scalar implicatures often involves either epistemic insecurity or attempts to mislead that fall short of lying (cf. Adler 1997, Horn 2017). The latter is illustrated by the fate of Picasso’s “Guernica”, the celebrated and moving canvas of the horrors of the Spanish Civil War, exiled to New York Museum of Modern Art during Franco’s dictatorship that became a target for repatriation once democracy returned to Spain. Here is an account (Markham 1981; emphasis added) of how that repatriation was accomplished: Since the long-exiled “Guernica” finally came to Madrid on Sept. 10 [1981], Rafael Fernández Quintanilla, the witty diplomat who dealt with the Museum of Modern Art in New York and Picasso’s heirs, has felt free to disclose some of the secrets of his protracted negotiations. One is an elaborate bluff. To demonstrate that the Spanish Government had in fact paid Picasso to paint the mural in 1937 for the Paris International Exhibition, Mr. Fernández Quintanilla had to secure documents in the archives of the late Luis Araquistain, Spain’s Ambassador to France at the time. But Araquistain’s son, poor and opportunistic, demanded $ 2 million for the archives, which Mr. Fernández Quintanilla rejected as outrageous. He managed, however, to obtain from the son photocopies of the pertinent documents, which in 1979 he presented to Roland Dumas, the Paris lawyer named by Picasso to determine when “public liberties” had been re-established in Spain, permitting delivery of the “Guernica” to the Prado.

38

horn

“This changes everything,” a startled Mr. Dumas told the Spanish envoy when he showed him the photocopies of the Araquistain documents. “You of course have the originals?” the lawyer asked casually. “Not all of them,” replied Mr. Fernández Quintanilla, not lying but not telling the truth, either. What makes the diplomat’s move an “elaborate bluff” rather than a lie is his exploitation of a Quantity implicature to convey a falsehood that he did not utter. Now consider an attested case of the contextual cancellation of the positive implicature of few. In response to an earlier op-ed column in the Times, Martin Kimel wrote a letter to the editor that echoes in the highlighted passage below the few predication of the original while affirming the stronger no statement in his added parenthetical. [T]here is something offensive about her complaining that the “shadow of AIDS” has caused her to be monogamous, while the disease is killing so many. Like the author, I have lost “few friends” to AIDS. (In fact, I have lost none.) Yet one need not have suffered any personal losses from AIDS to recognize that complaining about the lost possibility of sexual promiscuity demeans the true suffering of the disease’s real victims. It is not unlike watching famine victims on television while lamenting that the neighborhood grocery store is out of mint chocolate-chip ice cream. Kimel 1990, also cited in Horn 1992: 165

This context is essentially that of (64c); what holds for both writers is “losing few friends to AIDS”, the strongest property truly predicable to both. As Gricean theory predicts, the implicature from an assertion of the weaker “W” to “¬Ka(S)” (i.e. the stronger competitor doesn’t hold as far as the speaker knows) can be cancelled if there’s no simpler or equally simple way to assert W in a given context. But entailments (even inert ones) cannot be overridden by context in the same way. The actual passage in (67a) could not have replaced by either of the sequences in (67b) or (67c) without the contradiction. (67) a. Like the author, I have lost “few friends” to AIDS. (In fact, I have lost none.) b. #Like the author, I have lost “only 3 friends” to AIDS. (In fact, I have lost none.) c. #Like the author, I have lost “only a few friends” to AIDS. (In fact, I have lost none.)

an (abridged) atlas of negation

39

Additional confirmation of the contrast between few and only is not hard to locate. Thus, alongside the attested examples of the compatibility of few and none in (68)— (68) [On how Columbus navigated the ocean by measuring the height in degrees of the sun and the North Star:] γHe was the first to practice the art, for, before his time, though it was taught in the schools, few (in fact none) had dared to try it out at sea. https://books.google.com/books?isbn=1442607149 [From a history of Elkhart County, Indiana:] γFew, in fact none, of the old legal pioneers reside in this county. https://books.google.com/books?id=50QvAAAAYAAJ γI’ve written a book called “Basic Survival and Communications Skills in the Aftermath” … Because, while there are tons and tons of books out there on communications, on prepping, on survival, and so forth, few (in fact none I can find) take the topics and put them into basic terms for beginners. [From Henry Aubin’s The History of Jerusalem] γThe visitors would also have encountered little (possibly no) worship of Yahweh to the exclusion of other gods. https://books.google.com/books?isbn=0385672276 —we find no parallel examples of the compatibility of only or only a few and none: (69) #Only one {in fact none/in fact none that I can find} … #Only a few (in fact none) … #Only a little (possibly no) … While the data in (69) support Atlas’ line (1991, 1884, 1996, 2007, 2012, 2013) on only, the cites in (68) (and in (67a)) are incompatible with Atlas’s line (2013) on few.21 (And both data sets are incompatible with Geach 1962/1980). 21

The problematic data can easily be multiplied. There are, for example, over 100 actual (not estimated) Google hits for “little, possibly no” or “little (possibly no)” and another 60 plus for “little (and) possibly none”. On the other hand, there is not a single Google hit for “only a little (and) possibly no/none”.

40

horn

Precisely the same arguments—several from Atlas himself—that show that only is non-monotonic and not DE, given the prejacent entailment, show that few CN (along with its mass counterpart little CN) is in fact “negative” or DE, extending to empty sets. Atlas’s semantics works well for only a few and only a little, which do entail some, but not for few/little themselves. Finally, we can apply the refutation markers earlier surveyed in § 5. It will be recalled that as shown in (52)–(54), repeated here for convenience, implicated material—like the exhaustivity associated with clefts—cannot be the target of refutation, while entailed material, even if it’s not at issue—like the prejacent of only—can be. (52) A: B1: B2:

I love only you. Bullshit, you love Chris too. Bullshit, you don’t love anybody.

(53) R: J1: J2: J3:

It’s you I love. {Yes/Maybe so}, but you love Rosaline too. #Bullshit, you love Rosaline too. Bullshit, it’s Rosaline you love.

(54) R: J1: J2:

It’s only you I love. #Yes/Maybe so, but you love Rosaline too. Bullshit, you love Rosaline too.

Like the Griceogloss argument, refutation thus serves as a diagnostic to mediate between inert entailments and implicated material. It is thus significant that few As B cannot be refuted or falsified by empty set contexts, while only a few As B can be, as with other expressions with only (although it should be acknowledged that judgments on (72) may be less clear than those for its counterparts with DP focus as in (52) and (54)). (71) A: Few of those senators are reasonable. B: #{Bullshit/That’s false}, none of them are. (72) A: Only a few of those senators are reasonable. B: (?){Bullshit/That’s false}, none of them are. Confirmation (as in the “Yes, but” responses in (53)–(54)) works similarly to refutation, again distinguishing the behavior of few from that of only. In this respect, once again, the relation between few As are B and some As are B behaves

an (abridged) atlas of negation

41

like a standard scalar implicature, not an entailment. Let’s assume that I predict today that few (or not many) of your students will get an A on tomorrow’s exam, and that we discover tomorrow that none of them got an A. My prediction is thereby confirmed rather than disconfirmed. In the same way, your claim that not all of them will fail is verified as soon as we discover that one of them has passed—we don’t to make sure that at least one failed. But if you predict that only Albert will get an A and in fact nobody does, your prediction is not confirmed. The diagnostics we have surveyed show that with only, almost, and barely there is an entailment of the prejacent or polar component, although this entailment is not asserted or at issue, motivating the category of assertoric inertia. In the case of clefts of the form it’s a that Fs, exhaustivity (the proposition that nothing distinct from a Fs) is not only not at issue, it is not entailed. In this section we have presented evidence showing that the positive existential proposition associated with few or little is a conversational implicature (as argued in standard Gricean and neo-Gricean work dating back to Horn 1972); it is not entailed, even inertly. Thus, few As are B, unlike only n As are B or only a few As are B, is semantically compatible with no As are B; few CN is indeed a monotone decreasing expression, as posited by Barwise & Cooper (1981: 185). While in principle I am delighted by the rapprochement between Jay Atlas’s position and mine with respect to the usefulness of the doctrine of assertoric inertia, I have argued that in the case of few Atlas has extended the olive branch a bridge too far.

7

Lex Talionis

In this essay, I have surveyed some significant sallies made by Jay David Atlas into the realms of negation and offered my ripostes to these forays. As is customary, in saying I have surveyed some sallies, I implicate not all. To the catalogue raisonné offered here—the analysis of only and proximative adverbs and their relation to the distribution of polarity items, the status of epistemic cancellation, the semantic vs. pragmatic nature of the exhaustivity premise associated with clefts, the semantic and pragmatic character of few predications— there are two additional points worth touching on here that emerge from Atlas’s (2012) conspectus on the pragmatics of negation, in which he extends his purview of negation to targets old and new. We observed above the skepticism toward metalinguistic negation first voiced in Atlas 1983, a manuscript widely circulated two years prior to the appearance of its target (Horn 1985). Since then, the viability of, and need for,

42

horn

admitting a metalinguistic use of negation has been a ripe, if not overripe, topic of debate and Atlas himself has returned on occasion to pick another bone or two (e.g. in Atlas 2012: §18.7). In between, there has been considerable ink spilled by others on the status of metalinguistic (or echoic) negation; for a sampler, see Burton-Roberts (1989b, 1999), Foolen (1991), van der Sandt (1991), McCawley (1991), Chapman (1996), Carston (1996, 1999), Geurts (1998), Yoshimura (1998), Horn (2002), and especially Pitts (2009, 2011), who offers a comprehensive overview of and bibliography on this debate. Another subdomain of my work on negation evoking Atlas’s rich skepticism (though not outright hostility) concerns my program for an “extended term logic” (Horn 1989: §7.2) based on Aristotle’s treatment of negation (Ackrill 1963) and its elaboration by George Englebretsen (1981) and Fred Sommers (1982). On this analysis, the ordinary “sentential negation” of S isn’t P is taken to involve predicate denial: P is denied to hold of S. Ordinary sentence-scope negation is thus not an iterating one-place connective that applies to fully formed propositions or sentences but a mode of predication; a predicate term can either be affirmed or denied to hold of the subject. In addition, a predicate term itself may be negated (S is not-P), in which case a negative term is affirmed of the subject. This has truth-conditional consequences, as Aristotle points out: if Socrates fails to exist (or is, say, the name of a pet rock), Socrates isn’t ill will be true while Socrates is not-ill is false. Thus, the system allows for the differentiation of “sentence” and “constituent” negation, pre-figuring Jespersen, Klima, Jackendoff et al., while not modeling the one-place negative operator of their systems or those of (most) post-Fregean logics.22 I am grateful for Atlas’s valiant attempt to sort out the unclarities in my original presentation of the neo-Aristotelian account. Given space limitations, I will not here attempt any wholesale defense of extended term logic.23 I will content myself with touching on some of the consequences of a non-classical treatment of the logic of negation. As Atlas (2012: 374–375) observes, the inference from Chris won to Chris didn’t not win (or to It is not the case that Chris did not

22 23

On my treatment of “Aristotle as a Montague grammarian” (Horn 1989: 468–472), see Atlas’s objections (2012: § 18.8, especially p. 369). It is hard to resist a cavil or two, however. Atlas (2012: 371) attempts to “exhibit a syntactically clause-external negation of a disjunction, expressed in ordinary English”. This exhibition involves the acceptability of both (i) and the “rhetorically more natural” (ii): (i) Neither did Chris win or Sandy lose, nor did Chris win or Sandy lose. (ii) Neither did Chris win or Sandy lose, nor did Sandy lose or Chris win. I am afraid that I cannot admit either of these sentences into my own idiolect of ordinary English.

an (abridged) atlas of negation

43

win) is valid, while the inferences in the other direction do not necessarily go through; indeed, while intuitionist logic accepts what Atlas terms “a principle of modest double negation” allowing inflation (A |= ¬¬A), it rejects cancellation (¬¬A |= A). But whether or not one follows Horn 1989 in excluding one-place truth-functional operators and in particular endocentric sentential negation or instead accepts Atlas’s robust defense of such operators, it is worth clarifying what is at stake here. As Atlas points out (2012: 369), “Horn’s primary objection to syntactically external, sentential negation is a syntactic one”; the adjudication of this issue does not bear on the semantic relation of double negations to simple affirmations. Indeed, as acknowledged in the passage from (Horn 1989: 469) that Atlas cites, the promise is that “if ordinary (auxiliary) negation is treated as a mode of predication, a recipe for combining subject and predicate to form a proposition or sentence, rather than an operation on a fully formed proposition or sentence”, we can arrive at the juncture of both worlds: “[W]e get the semantic effect of Fregean negation without the syntactic commitment to an external pre-sentential operator”. But it should be noted that there remain significant differences between the account of sentential (“wide-scope”) negation as a mode of presentation and as a traditional iterating Fregean unary connective. One particularly insightful application of the neo-Aristotelian approach is due to Ladusaw (1996), who presents evidence from negative concord for the conclusion that “the best analysis of clausal descriptive negation in natural language is one which treats affirmation and negation symmetrically as ‘modes of judgement’ rather than asymmetrically, with negation treated as a propositional operator” (Ladusaw 1996: 127). As Ladusaw also points out (1996: 142), the choice to treat canonical negation as a mode of judgment does not require us to accept Aristotle’s assumption that all propositions are categorical (of subjectpredicate form). Rather, we can follow Kuroda (1972) in reviving the distinction due originally to Brentano and Marty between true categorical predications— in which a presupposed subject of judgment is recognized and something predicated of it—and thetic predications, simple unpartitioned judgments involving a simple affirmation or denial. Standard examples of thetic predications include existential statements and other state descriptions.24 Ladusaw 24

For more on categorical/thetic distinction see Sasse 1987 and Horn 1989: §7.3, among others. In another influential paper, Ladusaw himself (1994) puts the distinction to work in connection with Milsark’s generalization and the semantics of individual-level and stage-level predication; cf. also Horn (1997b) for an application of the categorical/thetic dichotomy to the role of existential import in quantified sentences.

44

horn

observes that negation in non-partitioned thetic descriptions as well as in categorical (logical subject-predicate) judgments must also be seen as exocentric, failing to apply to its own output. Finally, on the relation between contradictory and contrary negation, my position is somewhat different from what Atlas (2012: 375) suggests. I have recently argued in some detail (Horn 2014, Horn & Wansing 2015) that while a contrary understanding of apparent contradictory negation is robustly attested in natural language, it typically reflects “virtual contrariety”, the output of pragmatic (R-based) strengthening. Examples include the litotic interpretation of X isn’t good as a contrary rather than contradictory of X is good or the “neg-raised” interpretation of I don’t believe that p as a diffident way of expressing I believe that not-p. The result is that the contradictory negation of a virtual contrary— We’re not not friends; I don’t not want to go to the party—does not reduce to the doubly negated positive proposition.25 I conclude by returning to one of the -tal- cognates that emerges from the Atlas/retaliation family of etyma, the one constituting the black sheep of that family. Lex Talionis (from subfamily 3 in (1) above) is not, as the name may suggest, a mad Latvian scientist from a failed Superman spin-off, but the law of equal and direct repayment adapted by the Torah from the code of Hammurabi: [T]hou shalt give life for life, eye for eye, tooth for tooth, hand for hand, foot for foot, burning for burning, wound for wound, stripe for stripe Exodus 21: 23–25

And if a man cause a blemish in his neighbor; as he hath done, so shall it be done to him; breach for breach, eye for eye, tooth for tooth: as he hath caused a blemish in a man, so shall it be done to him again. Leviticus 24: 19–20

25

On the account of Horn 2014 and Horn & Wansing 2015, They didn’t not like the dinner and She’s not not happy don’t involve the contrary of a contradictory but the contradictory of a virtual (pragmatic) contrary; sentences like They didn’t like it and they didn’t not like it and She’s neither happy nor not-happy are thus coherent rather than contradictory. I confess that I can’t figure out if this is consistent or inconsistent with Atlas’s (2012: 375) revision of the analysis of Sommers (1982), but it should be noted that virtual contrariety is more easily obtained with dispositional predicates (believe, like, happy, be friends with) than with actions like win as in Atlas’s paradigms, whence the relative oddity of #She neither won nor didn’t win.

an (abridged) atlas of negation

45

Whence the maxim and bumper sticker variously attributed to Mahatma Gandhi or Kahlil Gibran (cf. also Cohen & Matson 2002): “An eye for an eye makes the whole world blind”. Fortunately, the effect of Lex Talionis can be more salutary in the intellectual sphere. Jay concludes the preface to Logic, Meaning, and Conversation with these remarks (emphasis mine): The final version [of Atlas (2005)] was written in the summer of 2001 after the appearance of Levinson’s (2000) Presumptive Meanings. Stephen, like Larry Horn in his Natural History of Negation (1989), provided essential stimulus and matter for reflection as I reconsidered neo-Gricean views that in concert with, and in reaction to, them I had been developing since the mid-1970s. My admiration for their work is surpassed only by their cosmic patience with my criticizing, not to say needling, them. Atlas 2005: xi

I have tried in this essay to demonstrate the recursive nature of this process of needling—or, as I prefer to think of it, intellectual acupuncture.

References Abelard, Peter (1970). Dialectica, 2d edition, ed. by L.M. de Rijk. Assen: Van Gorcum. Ackrill, J.L. (1963). Aristotle’s Categories and De Interpretatione, ed. and trans. Oxford: Clarendon Press. Adler, Jonathan (1997). Lying, deceiving, or falsely implicating. Journal of Philosophy 94: 435–452. Amaral, Patrícia (2007). The Meaning of Approximative Adverbs: Evidence from European Portuguese. Ohio State University dissertation. Amaral, Patrícia (2010). Entailment, assertion, and textual coherence: The case of almost and barely. Linguistics 43: 525–545. Atlas, Jay D. (1977). Negation, ambiguity, and presupposition. Linguistics and Philosophy 1: 321–336. Atlas, Jay D. (1979). How linguistics matters to philosophy: Presupposition, truth, and meaning. C.-K. Oh & D.A. Dinneen (eds.), Syntax and Semantics 11: Presupposition, 265–281. New York: Academic Press. Atlas, Jay D. (1983). Comments on ‘Metalinguistic Negation and Pragmatic Ambiguity’ by Larry Horn. Unpublished manuscript, Institute for Advanced Study, Princeton. Atlas, Jay D. (1984). Comparative adjectives and adverbials of degree. Linguistics and Philosophy 7: 347–377.

46

horn

Atlas, Jay D. (1989). Philosophy Without Ambiguity: a Logico-Linguistic Essay. Oxford: Clarendon Press. Atlas, Jay D. (1991). Topic/comment, presupposition, logical form and focus stress implicatures: The case of focal particles only and also. Journal of Semantics 8: 127–147. Atlas, Jay D. (1992). On exhaustiveness and the semantics of clefts: A reply to Horn. Unpublished ms., Pomona College. Atlas, Jay D. (1993). The importance of being ‘only’: Testing the neo-Gricean versus neoentailment paradigms. Journal of Semantics 10: 301–318. Atlas, Jay D. (1996). ‘Only’ noun phrases, pseudo-negative generalized quantifiers, negative polarity items, and monotonicity. Journal of Semantics 13: 265–332. Atlas, Jay D. (1997). Negative adverbials, prototypical negation, and the De Morgan taxonomy. Journal of Semantics 14: 349–368. Atlas, Jay D. (2002). NPI licensing and overcoming assertoric inertia. Paper presented at One Day Only workshop, Groningen, May 8, 2002. Atlas, Jay D. (2005). Logic, Meaning, and Conversation: Semantical Underdeterminacy, Implicature, and their Interface. Oxford: Oxford U. Press. Atlas, Jay D. (2006). Pseudo-anti-additivity, pseudo-anti-multiplicativity, and the future of L.R. Horn’s asymmetry: remarks on only ProperName and even ProperName. Paper presented at Swarthmore Workshop on Negation and Polarity, April 2006. Atlas, Jay D. (2007). On a pragmatic explanation of negative polarity licensing. N. Burton-Roberts (ed.), Pragmatics, 10–23. Houndmills: Palgrave Macmillan. Atlas, Jay D. (2012). Negation. K. Allan & K. Jaszczolt (eds.), Cambridge Handbook of Pragmatics, 351–376. Cambridge: Cambridge University Press. Atlas, Jay D. (2013). ‘Few’, ‘a few’, and ‘only’ noun phrases, non-monotonic quantifiers, and negative polarity items. Paper presented to NYU Semantics Workshop. Posted at https://nyusemantics.github.io/handouts/atlas‑fall‑2013_ms.pdf. Atlas, Jay David & Stephen C. Levinson (1981). It-clefts, informativeness, and logical form. P. Cole (ed.), Radical Pragmatics, 1–51. New York: Academic Press. Barwise, Jon and Robin Cooper (1981). Generalized quantifiers and natural language. Linguistics and Philosophy 4: 159–219. Beaver, David and Brady Clark (2008). Sense and Sensitivity: How Focus Determines Meaning. Oxford: Wiley-Blackwell. Blok, Peter (1993). The Interpretation of Focus. U. of Groningen dissertation. Burton-Roberts, Noel (1989a). The Limits to Debate: A Revised Theory of Semantic Presupposition. Cambridge: Cambridge University Press Burton-Roberts, Noel (1989b). On Horn’s dilemma: Presupposition and negation. Journal of Linguistics 25: 95–125. Burton-Roberts, Noel (1999). Presupposition-cancellation and metalinguistic negation: A reply to Carston. Journal of Linguistics 35:347–364. Byram Washburn, Mary, Elsi Kaiser & Maria-Luisa Zubizarreta (2013). The English it-

an (abridged) atlas of negation

47

cleft: no need to get exhausted. LSA poster and unpublished ms., USC. http://tinyurl .com/q8emwz9. Carston, Robyn (1996). Metalinguistic negation and echoic use. Journal of Pragmatics 25: 309–330. Carston, Robyn (1999). Negation, ‘presupposition’ and metarepresentation: a response to Noel Burton-Roberts. Journal of Linguistics 35: 365–389. Carston, Robyn (2002). Thoughts and Utterances: The Pragmatics of Explicit Communication. Oxford: Blackwell. Chapman, Siobhan (1996). Metalinguistic negation: Some theoretical implications. Journal of Linguistics 32: 387–402. Chomsky, Noam (1972). Some empirical issues in the theory of transformational grammar. S. Peters (ed.), Goals of Linguistic Theory, 63–130. Englewood Cliffs, NJ: PrenticeHall. Clark, Brady (2006). Polarity items and the interdefinability of universals and exclusives. Paper presented at annual meeting, Linguistic Society of America, Albuquerque. Cohen, Allen and Clive Matson (eds., 2002). An Eye for an Eye Makes the Whole World Blind: Poets on 9/11. Oakland: Regent Press. Coppock, Liz and David Beaver (2014). Principles of the exclusive muddle. Journal of Semantics 31: 371–432. Destruel, Emilie (2013). The French c’est-cleft: Empirical studies of its meaning and use. U. of Texas dissertation, downloadable at http://www.emiliedestruel.org/ publications.html. Destruel, Emilie, Daniel Velleman, Edgar Onea, Dylan Bumford, Jingyang Xue and David Beaver (2015). A cross-linguistic study of the non-at-issueness of exhaustive inferences. F. Schwarz (ed.), Experimental Perspectives on Presupposition, 135–156. Berlin: Springer. Drenhaus, Heiner, Malte Zimmermann, and Shravan Vasishth (2011). Exhaustiveness effects in clefts are not truth-functional. Journal of Neurolinguistics 24: 320–337. Dryer, Matthew (1996). Focus, pragmatic presupposition, and activated propositions. Journal of Pragmatics 26: 475–523. Ducrot, Oswald (1973). La preuve et le dire. Paris: Maison Mame. É. Kiss, Katalin (1998). Identificational focus versus information focus. Language 74: 245–273. Englebretsen, George (1981). Logical Negation. Assen: Van Gorcum. von Fintel, Kai (1993). Exceptive constructions. Natural Language Semantics 1: 123–148. von Fintel, Kai (1994). Restrictions on Quantifier Domains. PhD dissertation, U. of Massachusetts. von Fintel, Kai (1997). Bare plurals, bare conditionals, and ‘only’. Journal of Semantics 14: 1–56.

48

horn

von Fintel, Kai (1999). NPI-licensing, Strawson-entailment, and context-dependency. Journal of Semantics 16: 97–148. Foolen, Ad (1991). Metalinguistic negation and pragmatic ambiguity: Comments on a proposal by Laurence Horn. Pragmatics 1: 137–157. García-Álvarez, Ivan (2011). Generality and Exception: A Study in the Semantics of Exceptives. Stanford University dissertation. To be published, Oxford U. Press. Gazdar, Gerald (1979). Pragmatics: Implicature, Presupposition, and Logical Form. New York: Academic Press. Geach, P.T. (1962). Reference and Generality. Ithaca: Cornell U. Press. Geach, P.T. (1980). Reference and Generality, third edition. Ithaca: Cornell U. Press. Geurts, Bart (1998). The mechanisms of denial. Language 74: 274–307. Geurts, Bart and Rick Nouwen (2007). At least et al.: The semantics of scalar modifiers. Language 83: 533–559. Giannakidou, Anastasia (1998). Polarity Sensitivity as (Non)Veridical Dependency. Amsterdam: John Benjamins. Giannakidou, Anastasia (2006). Only, emotive factives, and the dual nature of polarity dependency. Language 82: 575–603. Giannakidou, Anastasia (2011). Negative and positive polarity items: Variation, licensing, and compositionality. C. Maienborn et al. (eds.), Semantics: An International Handbook of Natural Language Meaning, Vol. 2, 1660–1712. Berlin: de Gruyter. Grice, H.P. (1967). Logic and conversation: The William James lectures. In Grice 1989. Grice, H.P. (1989). Studies in the Way of Words. Cambridge: Harvard U. Press. Gryllia, Stella (2009). On the Nature of Preverbal Focus in Greek, U. of Leiden dissertation. Utrecht: LOT (Netherlands Graduate School of Linguistics). Halvorsen, Per-Kristian (1978). The Syntax and Semantics of Clefts. U. of Texas dissertation (= TLF [Texas Linguistic Forum] #11). Heycock, Caroline (1994). Focus projection in Japanese. NELS 24, 157–171. Hoeksema, Jacob (1986). Monotonie en superlatieven. C. Hoppenbrowers et al. (eds), Proeven van taalwetenschap, TABU: 38–49. Horn, Laurence (1969). A Presuppositional Analysis of only and even. CLS 5, 97–108. Horn, Laurence (1970). Ain’t it hard (anymore). CLS 6, 318–327. Horn, Laurence (1972) On the Semantic Properties of Logical Operators in English. UCLA dissertation, reprinted by IULC, 1976. Horn, Laurence (1979). Only, even, and conventional implicature. Paper presented at annual meeting of the Linguistic Society of America, Los Angeles. Horn, Laurence (1981). Exhaustiveness and the semantics of clefts. NELS 11: 125–142. Horn, Laurence (1984). Toward a new taxonomy for pragmatic inference: Q-based and R-based implicature. D. Schiffrin (ed.), Meaning, Form, and Use in Context (GURT ’84), 11–42. Washington: Georgetown University Press. Horn, Laurence (1985). Metalinguistic negation and pragmatic ambiguity. Language 61: 121–174.

an (abridged) atlas of negation

49

Horn, Laurence (1989). A Natural History of Negation. Chicago: U. of Chicago Press. Horn, Laurence (1992). The said and the unsaid. Proceedings of SALT 2, 163–192. http:// journals.linguisticsociety.org/proceedings/index.php/SALT/article/view/3039 Horn, Laurence (1993). Economy and redundancy in a dualistic model of natural language. S. Shore & M. Vilkuna (eds), SKY 1993: 1993 Yearbook of the Linguistic Association of Finland, 33–72. Horn, Laurence (1996). Exclusive company: Only and the dynamics of vertical inference. Journal of Semantics 13: 1–40. Horn, Laurence (1997a). Negative polarity and the dynamics of vertical inference. D. Forget et al. (eds.), Negation and Polarity: Syntax and Semantics, 157–182. Amsterdam: John Benjamins. Horn, Laurence (1997b). All John’s children are as bald as the king of France: Existential import and the geometry of opposition. CLS 33, 155–179. Horn, Laurence (1999). Asymmetric conjunction and assertoric inertia. Paper presented at PRAGMA conference, Tel Aviv, July 1999. Horn, Laurence (2002). Assertoric inertia and NPI licensing. CLS 38, Part 2, 55–82. Horn, Laurence (2005). ONLY connect: How to unpack an exclusive proposition. Paper presented at Jayfest, Pomona College, April 2, 2005. Horn, Laurence (2006). Only browsing: Looking for values on the exclusive shopping network. Paper presented at Swarthmore Workshop on Negation and Polarity, Swarthmore College, April 15, 2006. Horn, Laurence (2007). Neo-Gricean pragmatics: a Manichaean manifesto. N. BurtonRoberts (ed.), Advances in Pragmatics, 158–183. London: Palgrave. Horn, Laurence (2009). ONLY XL: The assertoric asymmetry of exponibles. SALT 19, 198– 221. At http://journals.linguisticsociety.org/proceedings/index.php/SALT/article/ view/2542. Horn, Laurence (2011). Almost forever. E. Yuasa et al. (eds.), Pragmatics and Autolexical Grammar in Honor of Jerry Sadock, 3–21. Amsterdam: Benjamins. Horn, Laurence (2013a). Focus and exhaustivity revisited. Invited plenary talk, DGfS conference, Potsdam, Germany, March 2013. Horn, Laurence (2013b). I love me some datives: Expressive meaning, free datives, and Fimplicature. D. Gutzmann & H.-M. Gärtner (eds.), Beyond Expressives: Explorations in Use-Conditional Meaning, 153–201. Leiden: Brill. Horn, Laurence (2014). The Singular Square: Contrariety and double negation from Aristotle to Homer. J. Blochowiak et al. (eds.), Collection of Papers Dedicated to Jacques Moeschler. http://www.unige.ch/lettres/linguistique/moeschler/ Festschrift/Festschrift.php. Horn, Laurence (2016a). Licensing NPIs: some negative (and positive) results. P. Larrivée & C. Lee (eds.), Negation and Polarity: Experimental Perspectives, 281–306. Berlin: Springer.

50

horn

Horn, Laurence (2016b). Information structure and the landscape of (non-)at-issue meaning. C. Féry & S. Ishihara (eds.), Oxford Handbook of Information Structure, 109– 128. Oxford: Oxford U. Press Horn, Laurence (2017). Telling it slant: Toward a taxonomy of deception. D. Stein and J. Giltrow (eds.), The Pragmatic Turn in Law, 23–55. Mouton Series in Pragmatics. Berlin: de Gruyter. Horn, Laurence and Heinrich Wansing (2015). Negation. E. Zalta (ed.), Stanford Encyclopedia of Negation. http://plato.stanford.edu/entries/negation/. Ippolito, Michela (2005). An implicature analysis of only. Unpublished paper, Boston U. Ippolito, Michela (2006). Remarks on only. Proceedings of SALT 16, 77–87. http://journals .linguisticsociety.org/proceedings/index.php/SALT/article/view/2947/2687. Ippolito, Michela (2008). On the meaning of only. Journal of Semantics 25: 45–91. Karttunen, Lauri (1973). STOP: is there a presupposition? Unpublished ms., Ann Arbor. Karttunen, Lauri and Stanley Peters (1979). Conventional implicature. C.-K. Oh & D.A. Dinneen (eds.), Syntax and Semantics 11: Presupposition, 1–56. New York: Academic Press. Kay, Paul (1990). EVEN. Linguistics and Philosophy 13: 59–111. Keenan, Edward and Jonathan Stavi (1986). A semantic characterization of natural language determiners. Linguistics and Philosophy 9: 253–326. Kempson, Ruth (1975). Presupposition and the Delimitation of Semantics. Cambridge: Cambridge U. Press. Kimel, Martin (1990). Confusion on who are victims of AIDS. Letter to the editor of the New York Times, October 19, 1990. http://tinyurl.com/q67k8bj. Klima, Edward (1964). Negation in English. J.A. Fodor & J.J. Katz (eds.), The Structure of Language, 246–323. Englewood Cliffs: Prentice-Hall. Kretzmann, Norman (1968). William of Sherwood’s Treatise on Syncategorematic Words. Minneapolis: U. of Minnesota Press. Krifka, Manfred (2008). Basic notions of information structure. Acta Linguistica Hungarica 55: 243–276. Kuroda, S.-Y. (1972). The categorical and the thetic judgment (evidence from Japanese). Foundations of Language 9: 153–185. Ladusaw, William (1979). Polarity Sensitivity as Inherent Scope Relations. U. of Texas dissertation. [Published version, New York: Garland, 1980.] Ladusaw, William (1994). Thetic and categorical, stage and individual, weak and strong. SALT 4, 220–229. Republished in L. Horn & Y. Kato (eds.), Negation and Polarity: Syntactic and Semantic Perspectives, 232–242. Oxford: Oxford U. Press, 2000. Ladusaw, William (1996). Negative concord and “mode of judgement”. H. Wansing (ed.), Negation: A Notion in Focus, 127–143. Berlin: de Gruyter. Levinson, Stephen (2000). Presumptive Meanings: The Theory of Generalized Conversational Implicature. Cambridge: MIT Press.

an (abridged) atlas of negation

51

Linebarger, Marcia (1987). Negative polarity and grammatical representation. Linguistics and Philosophy 10: 325–387. Linebarger, Marcia (1991). Negative polarity as linguistic evidence. Papers from the Parasession on Negation. CLS 27, 165–188. McCawley, James (1981). Everything that Linguists Have Always Wanted to Know About Logic But Were Ashamed to Ask. Chicago: U. of Chicago Press. McCawley, James (1991). Contrastive negation and metalinguistic negation. CLS 27, Part 2, The Parasession on Negation, 189–206. Markham, James (1981). For Spain, ‘Guernica’ stirs memory and awe. New York Times, Nov. 2, 1981. http://tinyurl.com/omlwl8a. Moxey, Linda and Anthony Sanford (1986). Quantifiers and focus. Journal of Semantics 5: 189–206. Moxey, Linda and Anthony Sanford (1993). Communicating Quantities. Hove: Lawrence Erlbaum. Ockham, William of (1980). Ockham’s Theory of Propositions: Part II of the Summa Logicae, tr. A.J. Freddoso & H. Schuurman. Notre Dame: U. of Notre Dame Press. O’Donnell, J.R. (1941). The Syncategoremata of William of Sherwood. Mediæval Studies 3: 46–93. Onea, Edgar and David Beaver (2009). Hungarian focus is not exhausted. Proceedings of SALT 19, 342–359. Peter of Spain (1992). Syncategoreumata, L. M de Rijk, ed.; J. Spruyt, trans. Leiden: E.J. Brill. Pinborg, Jan (1981). Walter Burley on exclusives. H.A.G. Braakhuis et al. (eds.), English Logic and Semantics, 305–326. Ingenium: Nijmegen. Pitts, Alyson (2009). Metamessages of Denial: The Pragmatics of English Negation. University of Cambridge dissertation. Pitts, Alyson (2011). Exploring a ‘pragmatic ambiguity’ of negation. Language 87: 346– 368. Progovac, Ljiljana (1993). Negative polarity: Entailment and binding. Linguistics and Philosophy 16: 149–180. de Rijk, L.M., ed. (1982). Some 14th Century Tracts on the Probationes Terminorum. Nijmegen: Ingenium. de Rijk, L.M. (1985). Walther Burley’s tract De exclusivis: An Edition. Vivarium 23: 23–54. Roberts, Craige (2005). Only and conventional presupposition. Unpublished ms., The Ohio State University. van Rooij, Robert and Katrin Schulz (2007). Only: meaning and implicature. M. Aloni et al. (eds.), Questions in Dynamic Semantics, 193–223. London: Elsevier. (CRiSPI 17.) Sadock, Jerrold (1981). Almost. P. Cole, ed., Radical Pragmatics, 257–271. New York: Academic Press. van der Sandt, Rob (1991). Denial. CLS 27, Part 2, The Parasession on Negation, 331–344.

52

horn

Sasse, Hans-Jürgen (1987). The thetic/categorical distinction revisited. Linguistics 25: 511–580. Schulz, Katrin and Robert van Rooij (2006). Pragmatic meaning and non-monotonic reasoning: The case of exhaustive interpretation. Linguistics and Philosophy 29: 205– 250. Schwenter, Scott (2002). Discourse context and polysemy: Spanish casi. C. Wiltshire & J. Camps, eds., Selected Papers from the 30th Annual Linguistic Symposium on Romance Languages, 161–175. Amsterdam: John Benjamins. Sevi, Aldo (1998). A semantics for almost and barely. Unpublished M.A. thesis, Tel Aviv U. Shibatani, Masayoshi (1990). The Languages of Japan. Cambridge: Cambridge U. Press. Smith, E. Allyn, Laia Mayol & Elena Castroviejo-Miro (2013). Felicity of direct denial by meaning type. LSA presentation, Boston. Sommers, Fred (1982). The Logic of Natural Language. Oxford: Clarendon Press. Sperber, Dan & Deirdre Wilson (1986). Relevance: Communication and Cognition. Cambridge: Harvard U. Press. (Second edition, 1995. Oxford: Blackwell.) Stalnaker, Robert (1978). Assertion. P. Cole (ed.), Syntax and Semantics 9: Pragmatics, 325–332. New York: Academic Press. Steinbeck, John (1962). Travels with Charley in Search of America. New York: Viking. Szabolcsi, Anna (1981). The semantics of topic-focus articulation. J. Groenendijk et al. (eds.), Formal Methods in the Study of Language, 513–540. Amsterdam: Mathematisch Centrum. Taglicht, Josef (1984). Message and Emphasis in English. London: Longman. Vallduví, Enric (1992). The Informational Component. New York: Garland. Velleman, Daniel, David Beaver, Emilie Destruel, Dylan Bumford, Edgar Onea, and Liz Coppock (2012). It-clefts are IT (inquiry terminating) constructions. Proceedings of SALT 22, 441–460. http://journals.linguisticsociety.org/proceedings/index.php/ SALT/article/view/2640. Ward, Gregory (2003). Comments on “Felicity and presupposition triggers” by Michael Glanzberg at Linguistics & Philosophy Workshop, Ann Arbor. Watkins, Calvert (ed., 2011). The American Heritage Dictionary of Indo-European Roots, 3rd edition. Boston: Houghton Mifflin Harcourt. Wedgwood, Daniel (2005). Shifting the Focus. Amsterdam: Elsevier. (CRiSPI 14.) Wilson, Deirdre (1975). Presupposition and Non-Truth-Conditional Semantics. New York: Academic Press. Yoshimura, Akiko (1998). Procedural semantics and metalinguistic negation. R. Carston & S. Uchida (eds.), Relevance Theory: Applications and Implications, 105–122. Amsterdam: John Benjamins. Ziegeler, Debra (2000). What almost can reveal about counterfactual inferences. Journal of Pragmatics 32: 1743–1776.

an (abridged) atlas of negation

53

Ziegeler, Debra (2006). Interfaces with English Aspect. Amsterdam: Benjamins. Zimmermann, Malte and Edgar Onea (2011). Focus marking and focus interpretation. Lingua 121: 1651–1670.

chapter 2

Dispelling the Cloud of Unknowing: More on the Syntactic Nature of Neg Raising Chris Collins and Paul Postal*

1

Background: Classical NEG Raising

Classical NEG Raising (henceforth: Classical NR) is a grammatical phenomenon traditionally defined in the simplest cases by the main clause presence of an overt negative element NEG nonetheless interpreted as having scope in a complement clause. See Horn (1978, 2001: 5.2) and Collins and Postal (2014) (hereafter: CP(2014)) for much background discussion and references to the relevant literature. English examples are provided in (1a–d): (1) a. b. c. d.

Harriet did not seem to be agitated. Vincent is not likely to win the election. They didn’t believe that Rodney would testify. I don’t suppose Helen can afford a new smartphone.

Example (1a) has a reading equivalent to the claim that Harriet seemed not to be agitated, while (1b) has a reading equivalent to a claim that Vincent is likely not to win the election. Sentence (1c) has a reading equivalent to a statement that they believed that Rodney would not testify and (1d) has a reading where the speaker supposes that Helen cannot afford a new smartphone. These are the Classical NR readings. Classical NR readings are known to depend on the choice of the main clause predicate. The following sentences corresponding to those in (1) lack the low scope readings of the high clause NEGs. (2) a. b. c. d.

Harriet did not claim to be agitated. Vincent is not afraid to win the election. They didn’t predict that Rodney would testify. I don’t insist that Helen can afford a new smartphone.

* We thank Andrew Radford and Pieter Seuren for many helpful comments on an earlier version of this article. We are also indebted to Larry Horn for extensive discussion of the issues.

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_003

dispelling the cloud of unknowing

55

Following CP(2014) we refer to those predicate elements taken in the preCP(2014) literature to permit Classical NR as Classical NEG Raising Predicates (CNRPs). In CP(2014), Classical NR is taken to be defined by syntactic raising of a NEG from a complement clause point of origin into a containing main clause. As noted there, this is a highly controversial view; most current approaches to Classical NR take it to be a purely semantic or pragmatic phenomenon. In these views, e.g., Horn (1989 [2001:308–330]), Horn and Bayer (1984), Tovena (2001), Larrivée (2004: 103–105), Gajewski (2005, 2007, 2011), Sailer (2005, 2006), Jacobson (2006), Boškovič and Gajewski (2008), Homer (2010) and Romoli (2012, 2013a, 2013b), syntactic NEG raising plays no role whatever. CP(2014) develops a powerful argument for a syntactic view of Classical NR based on the existence of subordinate clause constitutents that we called Horn clauses. First attested in Horn (1975: 283), these are illustrated by the italicized segments of such examples as (3): (3) a. b. c. d. e.

We don’t believe that in any sense has she met the minimal requirements. I don’t think that any of those space aliens will we be able to catch. I do not think that ever before have so many space aliens visited Earth. No one supposes that any of those proposals will the director sign off on. One would never have imagined that any such people could be elected to the senate.

The basic defining characteristics of Horn clauses are (i) the presence of a leftextracted nominal NPI phrase, e.g. in any sense in (3a), in the embedded clause, (ii) the obligatory occurrence of subject auxiliary inversion in the embedded clause and (iii) the occurrence in the main clause of some form of negation. The argument in CP(2014) hinged on a claimed restriction on Horn clauses (in addition to the properties in (i)–(iii) above) due originally to Horn (1975: 283; 1978: 169). The claim was that Horn clauses are limited to the complements of verbs independently permitting Classical NR, that is, complements of CNRPs. However, since the appearance of CP(2014), Horn (2014) has attested a new data class illustrating Horn clauses, but ones occurring in the complements of nonCNRP predicates. These include in particular a nonfactive use of the verb know, which Horn calls know-NF, a terminology we adopt. An example of the type at issue is given in (4): (4) I don’t know that ever before had the Army commander been in charge of the Naval forces. (www.americancivilwarforum.com)

56

collins and postal

We agree that the complement here as in other cases Horn cites are Horn clauses. Based on this newly noticed data, Horn constructs a critique of the Horn clause argument developed in CP(2014) for a syntactic view of Classical NR. We will call the argument he advances on this basis the anti-Horn clause argument, hereafter, Anti-HC. Since the data Horn uncovered were unknown to the authors of CP(2014), Anti-HC was of course not treated in CP(2014). The present study seeks to show that contra Anti-HC, the important facts documented in Horn (2014) fail to undermine CP(2014)’s Horn clause argument for a syntactic conception of Classical NR. We argue this by indicating how a viable Classical NR analysis of the newer data can be provided within the overall syntactic view developed in CP(2014).

2

The CP(2014) Treatment of Classical NR

The examples in (1a–d) illustrate the most straightforward type of Classical NR. Following previous syntactic treatments, CP(2014) assumed that in such examples a NEG has (optionally) raised from the complement clause of a CNRP and appears overtly in the post-Aux position of the immediately higher main clause. It was equally assumed there that Classical NR could take as input NEGs in various different complement clause positions, subject to certain syntactic constraints. The most important of the latter was that the original position of the raised NEG in the complement clause could not be separated from its main clause position by an island boundary. Since Classical NR is taken in CP(2014) to be a syntactic raising phenomenon, that is, one of the same order as the various raisings in wh question clauses, relative clauses, topicalizations, etc., it follows that Classical NR should obey basic constraints on syntactic raising. A great deal of evidence that this is the case was presented in CP(2014), especially chapters 11 and 12. The island constraint was taken to account for such facts as (5d) given that, as illustrated in (5b), clauses with complement internal topicalization are islands: (5) a. When did Ted believe [that Marjorie had visited her mother ⟨When⟩]? b. *When did Ted believe [that her mother, Marjorie had visited ⟨When⟩]? c. Ted did not (= NEG1) believe [that Marjorie had visited her mother [⟨NEG1⟩ in years]]. d. *Ted did not (= NEG1) believe [that her mother, Marjorie had visited [⟨NEG1⟩ in years]].

dispelling the cloud of unknowing

57

In (5c,d) and hereafter, following CP(2014), the notation ⟨X⟩ is utilized to represent covert occurrences of the syntactic constituent X. Hence ⟨NEG1⟩ in (5c,d) represents the lower, unpronounced occurrences of NEG1. In (5c,d), NEG1 starts out in the structure [NEG in years], which CP(2014) call a unary NEG structure. These are distinguished from binary NEG structures, which originate with two NEGs in representations of the form: [NEG1[NEG2 X]]. Binary NEG structures are not of relevance to this work. While NEG1 raises in both (5c,d), the island created by the topic her mother in (5d) renders the result ungrammatical. The fact that the Classical NR phenomenon is subject to island constraints is a powerful argument for its syntactic nature. No existent nonsyntactic approach to Classical NR, including that in Horn (2014), has ever provided an account of this property.

3

The CP(2014) Classical NR Analysis of Horn Clauses

3.1 Basics In this section we briefly sketch the treatment of Horn clauses within the syntactic view of Classical NR found in CP(2014). Consider Horn’s (1975: 238) original Horn clause example (6a): (6) a. I don’t think that ever before have the media played such a major role in a kidnapping. b. I think that never before have the media played such a major role in a kidnapping. For CP(2014), the analysis of (6a) is closely related to that of its paraphrase (6b), which uncontroversially instantiates the Negative Inversion construction. More precisely, the key assumption about Horn clauses in CP(2014) is (7): (7) Horn clauses are Negative Inversion clauses. The ideas just mentioned yield an analysis of (6a) like (8a), taken to represent the common underlying syntactic form of both (6a,b). We suppress from our representations the irrelevant phrase in a kidnapping. (8) a. I think that the media have [NEG1 SOME ever before] played such a major role. b. I think that [NEG1 SOME ever before] have the media played such a major role.

58

collins and postal

c. I do NEG1 think that [⟨NEG1⟩ SOME ever before] have the media played such a major role. Structure (8b) represents the claim that both (6a,b) exist as a function of Negative Inversion. By positing a covert NEG internal to the fronted NPI phrase in (6a), the requirements of Negative Inversion are satisfied. That covert negative occurrence then represents the same NEG having an overt occurrence in the main clause, as illustrated in (8c). This is possible because the main verb think, a CNRP, permits Classical NR. The negative analysis of the NPI in a Horn clause is only possible because CP(2014) analyzes NPIs in general as NEG-containing phrases, a position argued for at length in that work. In these terms, the difference between (6a,b) is that only in the former, as represented in (8c), does NEG1 raise to the matrix clause. Since (6a) is then analyzed as involving Classical NR as well as Negative Inversion, the supposed fact that Horn clauses are limited to complements of CNRPs was taken to follow from the assumption that only CNRPs permit Classical NR. CP(2014) contains much discussion of the grammatical conditions under which Negative Inversion can occur. A highly simplified version of their condition is as follows: (9) The NI Negation Condition The fronted constituent (NI Focus) in a Negative Inversion clause is either syntactically negative or denotes a decreasing function. In the example in (6b), never is both negative and decreasing. In (6a), ever before can meet the NI Condition only under the CP(2014) assumption that it is a unary NEG structure: [⟨NEG⟩ ever before]. In other views of nominal NPIs, they are taken to be existential or indefinite expressions, hence increasing constituents. The fact that they can be extracted under Negative Inversion is thus anomalous. Associating the relevant NPIs with a single NEG, rendering them decreasing negative expressions, dissolves this anomaly. 3.2

Independent Evidence that Horn Clauses are Negative Inversion Clauses CP(2014, chapters 13 and 14) provide much independent evidence that the subject-aux inversion found in Horn clauses is an instance of Negative Inversion. We will not replicate that material here, but limit ourselves to a new argument. This stands on the fact that the modal might is incompatible with uncontroversial instances of Negative Inversion and equally bars the formation of Horn clauses.

dispelling the cloud of unknowing

59

The examples in (10) show that might is incompatible with Negative Inversion. (10) a. b. c. d.

She might never finish her thesis. *Never might she finish her thesis. Vera might rent none of her cottages. *None of her cottages might Vera rent.

The examples in (11) show that a similar constraint holds of Horn clauses, which are incompatible with the presence of might. (11) a. *I don’t believe that ever might she finish her thesis. b. *I don’t believe that any of her cottages might Vera rent. Because we take NPIs like ever in (11a) and any of her cottages in (11b) to have the same negative structures as respectively never in (10b) and none of her cottages in (10d), the constraint in (10b,d) carries over to (11a,b) without special statement. But in any other treatment of Horn clauses we can conceive of, such correlations are inexplicable accidents. The argument from might may be even stronger than at first appears. Probably (10b,d) are bad because of a more general fact that might cannot be under the scope of a local NEG: (12) She might not finish her thesis. (scope: might > NEG; *NEG > might) For, as discussed at length in CP (2014: section 14.2), the fronted phrase in a Negative Inversion clause must have the highest scope in its clause. Therefore, (10b,d) and (11a,b) are ungrammatical because a local negative expression scopes over might. Moreover, there is no general constraint on subject-aux inversion with might, indicating that the constraints in (10) are not functions of a general ban on subject-aux inversion with this modal. (13) a. How might they approach that problem? b. What might be a possible solution? c. Might he be willing to lend some support to that project? Only an analysis taking Horn clauses to be Negative Inversion clauses permits reducing the violations in (10) and (11) to a generalization about negation scope since only on such an analysis is the fronted constituent in (11) a negative quantifier (a unary NEG structure).

60

collins and postal

Even the limited evidence we have given here linking Horn clauses to Negative Inversion then strongly supports the conclusion of CP(2014) that Horn clauses are simply a subtype of Negative Inversion clause.

4

Horn’s New Data Class and Anti-HC

4.1 The New Data Class Horn illustrates know-NF (non-factive know) with examples like (14): (14) I don’t know that I can trust you. Such examples lack the implication obligatorily found with the standard, factive usage of know that the embedded clause is true (see Horn 2014: 184). Horn’s critical discovery relevant here is the existence of many know-NF sentences whose complements appear to be Horn clauses, as (4) above. Additional examples include: (15) a. I don’t know that EVER before had all three boys napped simultaneouly. (www.vrbo.com/194521/reviews) b. I don’t know that ever before had the Army commander been in charge of the Naval forces. (www.americancivilwarforum.com) We find such examples perfectly grammatical and accept Horn’s claim that such examples contain Horn clauses. Based on such examples, the core of Horn’s Anti-HC argument is the claim that know-NF clauses like those in (15) cannot be instances of Classical NR because of such nonequivalences as the following non-Horn clause and Horn clause cases: (16) a. I don’t know that she ever before visited Cyprus. ≠ b. I know that she did not ever before visit Cyprus. (17) a. I don’t know that ever before did she visit Cyprus. ≠ b. I know that never before did she visit Cyprus. Critical to Anti-HC is that not only are cases like (16a) grammatical, but also that (16) and (17) contrast with the corresponding CNRP cases.

dispelling the cloud of unknowing

61

(18) a. I don’t believe that she ever before visited Cyprus. = b. I believe that she did not ever before visit Cyprus. (19) a. I don’t believe that ever before did she visit Cyprus. = b. I believe that never before did she visit Cyprus. Examples (18) and (19) represent, of course, the kind of semantically equivalent pairs which originally motivate talk of Classical NR In illustrating the problem for CP(2014)’s Horn clause argument, we have focused on know-NF, as Horn did. However, Horn (2014) observed that other forms are by the same reasoning apparently also not CNRPs because of the lack of equivalences like those in (16) and (17). These, namely, can’t say and not aware, likewise permit Horn clauses, as in the additional Web examples in (20): (20) a. I can’t say that at any time did I have a problem with any of the customer service team. (www.reviewcentre.com/reviews182288.html) b. I’m not aware that ever in the history of New York State has something like this happened. (issuu.com/downtownexpress/…/113011_de.indd) We can add, that in our dialects, the expressions, not certain, not confident, not evident (to me), not convinced, not persuaded, not obvious (to me), not positive, not recall, not sure, and can’t swear also appear to have the relevant properties that Horn documents with know-NF and the forms in (20); see too section 10 on doubt, which has related properties and which receives a structurally parallel syntactic analysis to CU predicate cases in our terms. That is, for this class of predicates, there are no semantic equivalences of the sort typical of traditionally studied Classical NR cases and Horn clause examples like (21) are grammatical: (21) a. b. c. d.

I am not aware that at any time did they insult any of the applicants. I am not certain that ever before have I met that person. It is not obvious to me that in any sense is he qualified for that position. I do not recall that at any point in time did I consent to their reprinting my article.

Given facts like (17a), (20) and (21), Horn concluded: (22) Horn 2014: 190: “The problem for this otherwise elegant argument for a syntactic rule of NR is that ‘dichotomous’ status is not a necessary property for triggering embedded subject-aux inversion …”

62

collins and postal

By ‘otherwise elegant argument’, Horn refers to the CP(2014) argument from Horn clauses; by ‘dichotomous status’ here he refers to equivalences like those in (18). We will call non-CNRP predicates which, when negated, permit Horn clause complements, including know-NF, Cloud of Unknowing (CU) predicates, and sentences where they occur CU-sentences. So in this terminology, Horn’s key discovery relevant to the overall argument in CP(2014) is that negated CU predicates, although failing to manifest the semantic equivalences usually taken as diagnostic of the simple cases of Classical NR, can take Horn clause complements. Given that, the claim of CP(2014) that Horn clauses provide an overwhelming argument for a syntactic conception of Classical NR is seemingly undermined. That follows because, as first claimed by Horn himself (1975:283; 1978: 169) and endorsed in CP(2014: chapter 14, section 4), Horn clauses are supposedly uniformly linked to main clause CNRPs. This was taken to be explicable under two assumptions: (i) only CNRPs permit Classical NR and (ii) Horn clauses have a syntactic Classical NR analysis as in (8) above. Thus the core of Anti-HC is that CP(2014)’s Horn clause argument for a syntactic view of Classical NR falsely predicts the ungrammaticality of CU predicate cases like (17a), (20) and (21), etc. 4.2 The Apparent Paradox An apparent paradox then arises from two at first convincing arguments. First, there is the seemingly solid argument in CP(2014) sketched above that Horn clauses justify a syntactic view of Classical NR. That clashes with Anti-HC based on Horn’s discovery that Horn clauses are possible with CU predicates. But taking the existence of Horn clauses with CU predicates to justify Anti-HC fails to recognize the range of analyses made possible by the framework of CP(2014). And, actually, Anti-HC itself is not compelling at all. While its internal reasoning is correct, it depends entirely on the assumption that Classical NR can be correctly recognized only in the presence of the sort of semantic equivalences in (18). If that claim is unfounded, Anti-HC is groundless. And it is unfounded. That such equivalences are not a sufficient condition is already shown by the fact, first noted in Horn (1975: 291–292; 1978: 203) and discussed further in CP (2014: 113–114), that truth predicates, which guarantee the relevant equivalences, nonetheless do not permit Classical NR. (23) a. It not true that she ever before visited Bulgaria. = b. It is true that she did not ever before visit Bulgaria. c. *It is not true that ever before did she visit Bulgaria.

dispelling the cloud of unknowing

63

d. It is true that never before did she visit Bulgaria. e. *It is not true that she arrived until Monday. f. It is true that she did not arrive until Monday. Here, despite the unquestionable equivalence of (23a,b), Horn clauses (and strict NPIs like until; see section 7) are impossible. One of the main goals of this paper is to show that equivalences like those in (18) are not a necessary condition for Classical NR either. We argue that the semantic equivalences and Classical NR are independent. Our conclusion so far then is that the discovery that Horn clauses can occur in the complements of CU predicates cannot as such falsify the CP(2014) argument that Horn clauses strongly support a syntactic view of Classical NR. However, Horn’s discovery could in principle undermine the CP(2014) view if it could be shown that there is no syntactic Classical NR analysis of the new data. But nothing like that has been shown and by advancing and justifying just such an analysis, the following sections argue that it cannot be shown.

5

The Inadequacy of the Simplest Classical NR Analysis of CU-Sentences

Perhaps the simplest approach in the framework of CP(2014) would be that a CU predicate-based Horn clause example such as (24d) is a result of Classical NR with (24b) as its input (parallel to the analysis in (8) of section 3 for CNRPs): (24) a. I know-NF that the media have [NEG1 SOME ever before] played such a major role. b. I know-NF that [NEG1 SOME ever before] have the media played such a major role. c. I do NEG1 know-NF that [⟨NEG1⟩ SOME ever before] have the media played such a major role. d. I don’t know-NF that ever before have the media played such a major role. The coindexing between the two occurrences of NEG1 in (24c) then represents the raising of NEG1. In (24b), NEG1 modifies SOME and the extraction via Negative Inversion of [NEG1 SOME ever before] to the clause initial position satisfies the NI Negation Condition and requires subject-aux inversion. Then, (24c) is the result of Classical NR taking as argument the NEG1 of (24b).

64

collins and postal

However, there are at least three reasons to reject such a straightforward Classical NR analysis for CU predicate instances of Horn clauses. First, it predicts the existence of exactly the semantic equivalences found with CNRPs, just the situation Horn (2014: 190) observed does not exist with CU predicates; see (25): (25) a. b. c. d. e. f.

I don’t know that Carol is the best candidate. ≠ I know that Carol is not the best candidate. I can’t say that ever before have the media played such a major role. ≠ I can say that never before have the media played such a major role. I can’t swear that he embezzled the money. ≠ I can swear that he didn’t embezzle the money.

Second, if (25a,c,e) represented instances of a Classical NR analysis in which the overt main clause NEG raised from the complement clause, that NEG would in each case function semantically only in the complement clause. But each such example clearly involves main clause semantic negation. A third and even stronger basis for rejecting the simple Classical NR analysis represented in (24) can be based on the analysis of parenthetical clauses in chapter 17 of CP(2014). This is discussed in section 9 below.

6

A Covert NEG-Based Approach to Horn Clauses in CU-Sentences

6.1 Motivation We suspect that Horn’s (2014) conclusion that the occurrence of Horn clauses in CU-sentences undermines the support from Horn clauses for a syntactic view of Classical NR is based implicitly on the following assumption: (26) The only possible syntactic Classical NR analysis of CU-sentence Horn clauses is that rejected in section 5. But assumption (26) is not sound. We can show this by displaying a distinct syntactic Classical NR treatment of the CU-sentence data Horn discovered which suffers from none of the flaws mentioned in section 5. Beyond accounting for the fact that CU-sentences can take embedded Horn clauses, the analysis we propose will meet two further criteria. First, it is consistent with the fact that Horn clauses are Negative Inversion clauses. Second, as discussed in the previous section, the overt matrix NEG in CU-sentences originates in the main clause and thus correctly functions semantically in that clause, not in the complement clause.

dispelling the cloud of unknowing

65

These assumptions are consistent with the following representation for (24d): (27) I NEG1 know that [[[⟨NEG2⟩ SOME ever] before] have the media played such a major role] Structure (27) contrasts with (24c) in that NEG1, which modifies know-NF, is distinct from the NEG2 modifying SOME. Hence in (27), NEG1 has not been raised from the embedded clause. But the presence of NEG2 as a modifier of SOME determines that the resulting negative adverbial phrase satisfies the NI Negation Condition and can hence undergo Negative Inversion in the embedded clause. And the presence of unraised NEG1 in the main clause correctly accounts for the semantically negated character of the higher clause. But (27) cannot be the correct representation of (24d), since it yields an incorrect interpretation equivalent to that of (28): (28) I don’t know that never before have the media played such a major role. That is, (27) has the wrong number of negative elements to represent the meaning of (24d). The actual Classical NR analysis we propose for (24d) remedies the defect in (27) by adding an additional NEG to the complement, and is given in (29). (29) I [NEG1 know-NF] that [⟨NEG2⟩ [[[⟨NEG3⟩ ever] before] have the media played such a major role]] This representation contains three semantically and syntactically distinct NEGs. NEG2 is a clausal modifier, negating the embedded clause whose initial element is the phrase extracted under Negative Inversion. The covert NEG2 and NEG3 are taken in the framework of CP(2014) to have been deleted. We return to the deletion of the NEGs in the next section. Contrasting with many views about the distribution of syntactic negation, the general position of CP(2014), is that a NEG can modify any category whose semantics is compatible with semantic negation. Thus the presence of elements like NEG2 and NEG3 in the positions where they are found in (20) requires no special stipulations. In particular, NEG2 modifies the clause (denoting a proposition), and NEG3 modifies a covert SOME (denoting a generalized quantifier). We observe that (29) arguably yields the correct interpretation of (24d), one essentially equivalent to (30a) (in turn equivalent to (30b)):

66

collins and postal

(30) a. I don’t know that it is not the case that the media haven’t ever before played a major role. b. I don’t know that the media have played a major role at some time before. This conclusion holds because the occurrence of one negation under the scope of a distinct one yields a nonnegative result semantically. So the outer NEG in the embedded clause of (29) in effect cancels the semantic effect of the lower NEG associated with ever. In predicate logic notation: ¬[¬∃xP(x)] ↔ ∃xP(x). Structure (29) provides both the NEG needed to account for the negative status of the main clause and the NEG needed in the fronted phrase of the Horn clause. As argued in detail in CP(2014) and above, the latter is necessary to satisfy the condition on Negative Inversion, namely, the NI Negation Condition. 6.2 NEG Deletion The previous section crucially appealed to NEG deletion. Representation (29) contains two deleted NEGs. In this section, we consider how they are deleted. CP(2014, chapters 7 and 8) propose the general view that NEG deletion is based on a primitive binary relation between phrases called NEG Deletion (NDEL). (31) The NDEL Interpretation Condition If NDEL(X,Y), then Y = NEG and Y is not pronounced. This states that NDEL is a relation that determines the covertness (in other words, the nonphonetic realization) of a NEG. Phrases instantiating the variable X in (31) are called NEG deleters. CP(2014) further distinguishes between lexical and general NEG deleters. All the examples in the present paper involve general NEG deleters. A fuller representation of structure (29) would indicate that the NEG deleter of NEG2 is the negated matrix verb [NEG1 know], while the NEG deleter of NEG3 is NEG2. These deletions form an object called a NEG Deletion Chain, a sequence of phrases meeting various conditions. For (29), the NEG deletion chain would be (32): (32) ⟨[NEG1 know], NEG2, NEG3⟩ The basic idea is that NDEL(X,Y) holds between each successive pair in a NEG deletion chain and that any deleted NEG occurs in such a chain. The initial element of the chain in (32) for (29) is an example of a NEG deleter. These are claimed to be subject to an additional general condition,

dispelling the cloud of unknowing

67

which is the analog in the system of CP(2014) of the various semantic conditions on NPI licensers posited in standard approaches to NPIs. CP(2014) requires, via (33), that general NEG deleters define nonincreasing functions (see also CP(2014:242, n. 6)). (33) The General NEG Deletion Condition If C = ⟨A, NEG1, …, NEGn⟩ is a NEG deletion chain, and A is a general NEG deleter, then A defines a function which is nonincreasing with respect to the origin position of each NEG in C. Just as standard views of NPIs commonly appeal to a c-command condition on NPI licensing, CP(2014) posits a c-command condition on NEG deletion: (34) The NDEL C-Command Condition If NDEL(X, Y) then X c-commands Y. The NPI which triggers Negative Inversion in Horn clauses exemplifies unary NEG structures since they are represented as [[NEG SOME] N]. Deletion of unary NEGs is taken to be subject to condition (35): (35) The NDEL Clausemate Condition If NDEL(X,Y) and Y is a unary NEG, then X and Y are clausemates. Moreover, CP(2014) posits a condition, whose foundational idea is due to Szabolcsi (2004). This serves to rule out massive overgeneration of structures with covert NEGs. We here, however, present an oversimplified version, since the actual one involved several complex issues not relevant in this work. (36) The NEG Deletion Evenness Condition (simplified version) If G is a NEG Deletion Chain, then G contains an even number of deleted NEGs. All the principles just mentioned were motivated entirely independently of CUsentences, which, as indicated earlier, we were unaware of. Finally, even though CP(2014) allows NEG to modify any syntactic constituent whose semantics can be negated, we know of no cases in English (or indeed in other languages) where an overt NEG is adjoined to a clause. That is there are gaps like:

68

collins and postal

(37) a. *Not Vincent left. b. *I think that not Vincent left. This parallels the fact that even can modify a wide range of constituents, but not clauses: (38) *Even it was raining. “It was even true that it was raining.” Based on very limited knowledge, we therefore boldly suggest a general principle, to be added to the collection of principles governing NEG deletion in CP(2014), as follows: (39) The Clausal NEG Deletion Condition For every structure [Clause NEGx [Clause Y]], there exists a Z such that NDEL(Z, NEGx). Consider then how the CU-sentence (24d) represented in (29) satisfies the general idea that NEG deletion is a function of the NDEL relation satisfying the set of constraints just given. The Clausal NEG Deletion Condition requires NEG2 to be deleted. The General NEG Deletion Condition then requires the NEG deleter of NEG2 to define a nonincreasing function. The only forms in the complement clause of (29) with this property other than NEG2 itself are NEG3 and [NEG3 ever before]. These phrases do not, however, c-command NEG2. Therefore, the General NEG Deletion Condition and the NDEL C-command Condition combine to entail that the NEG deleter of NEG2 cannot be internal to the complement clause. So it must be in the main clause. The NDEL Clausemate Condition requires that the NEG deleter of NEG2 must be its clausemate. Therefore, NDEL(X, NEG2) in (29) can only satisfy the overall set of conditions on NEG deletion if NEG2 raises into the main clause. There it finds a decreasing phrase, [NEG1 know-NF] (the matrix verb modified by negation), which is a c-commanding clausemate and hence a licit NEG deleter. The exact position to which NEG2 raises in the main clause depends on various assumptions that space does not permit considering here. This set of assumptions has the key entailment that Horn clauses and strict NPIs embedded under negated CU-predicates will show island effects. This claim is shown to hold in section 8 below. The NEG Deletion Eveness Condition blocks a large number of unwanted structures. For instance, consider (40a), which differs from (29) in violating the condition:

dispelling the cloud of unknowing

69

(40) a. I NEG1 know that [⟨NEG2⟩ [[[NEG3 ever] before] have the media played such major role]] b. I don’t know that never before have the media played such a major role. In this structure, only NEG2 is deleted, leaving NEG3 overt. The resulting interpretation would be identical to that of (29) (which is a representation of (24d)) since in the current system, NEG deletion has no semantic consequences. However, (40b) cannot have the same interpretation as (24d). But that follows since (40a), with only one deleted NEG, violates the NEG Deletion Evenness Condition. Essentially the same argument shows why a structure differing from (40) in that NEG2 is not deleted but NEG3 is deleted is also ungrammatical. It too would manifest a NEG deletion chain with only a single deleted NEG, and would also conflict with the NEG Deletion Evenness Condition.

7

Strict Negative Polarity Items and CU-Sentences

In arguing for the syntactic nature of Classical NR, CP(2014) appealed not only to Horn clauses but to the properties of various so-called strict NPIs. The idea was the traditional one that certain NPIs require a local relation to their licenser. Relevant then are contrasts like the following involving the strict NPI adjectival modifier all that. Horn cites the following paradigm from CP(2014: 85): (41) a. Arnold is *(not) all that intelligent. b. Lucinda doesn’t believe/think that Arnold is all that intelligent. c. *Lucinda doesn’t know/realize that Arnold is all that intelligent. For CP(2014), the negation in (41b) originates in the embedded clause, as a sister to all that, and raises to the matrix clause where it is realized overtly. Such raising is impossible in (41c), since know/realize are not CNRPs. However, Horn (2014: 192–193) rightly observes that the internet example (42) is well-formed, when the instance of know is know-NF: (42) I don’t know that it was all that easy even back then. Further examples of strict NPIs (highlighted) in clauses embedded under negated CU-predicates are given in (43):

70

collins and postal

(43) a. But she doesn’t know that it’s all that bad. (http://crookedkat.tumbir .com/) b. Bruschetta is quite delicious as well, but I can’t say I’ve indulged in ages. (www.realkidseatspinach.com/bruschetta‑chicken/) c. No, I can’t say that I think anything of your post-graduate course idea. (sebastianmarshall.com/on‑the‑competitive‑edge‑podcast) d. I was not aware that he was all that annoying until … you wrote it here. (www.sfx.co.uk > Features) While sometimes awkward or ungrammatical in clauses embedded under negated CU-predicates, Web occurrences of strict NPIs in such contexts are common and the present authors find all the above sentences grammatical. Because none of the matrix predicates in (43) is a traditional CNRP, Horn views the existence of such examples as just as problematic for CP(2014) as his newly noted Horn clause data. However, we propose an analysis of such sentences parallel to our treatment of Horn clauses with CU-sentences in (29) above. First, in CP(2014), strict NPIs are analyzed as unary NEG constituents where the NEG has raised away. In these terms, (44a) has analysis (44b): (44) a. Lucinda is not all that intelligent. b. Lucinda is NEG1 [⟨NEG1⟩ all that intelligent] For CP(2014), a parallel analysis extends to all strict NPIs, and in effect to the fronted NPI phrase in a Horn clause as well, as analysis (29) shows. So in these terms, example (42) receives analysis (45): (45) I do [NEG1 know-NF] that [[⟨NEG2⟩ [it was [⟨NEG3⟩ all that easy] even back then]] The properties of (45) are in most respects parallel to those of CU-sentences with Horn clauses. There are three semantically and syntactically distinct NEGs, NEG1 negates the matrix predicate, NEG2 negates the most deeply embedded clause and NEG3 negates the phrase all that. Finally, NEG2 and NEG3 are deleted, with NEG2 having raised into the matrix clause to permit its deletion. As with Horn clauses, we do not show the position of the covert raised NEG2 in (45). From our point of view, the analyses of Horn clauses and strict NPIs in CUsentences are parallel. Both involve a double negation structure in the complement clause with both NEGs ultimately deleted.

dispelling the cloud of unknowing

8

71

CU-Sentences and Islands

Our analysis of CU-sentences with Horn clause complements in section 6 and strict NPIs in section 7 determines that they will be ungrammatical when the raised NEG we posit is forced to cross a clausal island boundary. At issue then are the sort of facts illustrated below with the traditional CNRP believe: (46) a. b. c. d.

I don’t believe that they have ever made such a proposal before. I don’t believe that ever before have they made such a proposal. That they have ever before made such a proposal, I don’t believe. *That ever before have they made such a proposal, I don’t believe.

CP(2014) argues that the ungrammaticality of Horn clause cases like (46d) is a function of the interaction of our syntactic view of Classical NR with the wellknown fact that topicalized clauses are islands. Strikingly then, the same pattern found with the CNRP in (46) is found with a CU-predicate like (not) sure: (47) a. b. c. d.

I am not sure that they have made such a proposal at any point. I am not sure that at any point have they made such a proposal. That they have made such a proposal at any point, I am not sure of. *That at any point have they made such a proposal, I am not sure of.

CU-sentences are also constrained by the island formed by the complement of the fact: (48) a. I cannot swear that they interrogated her thoroughly at any time. b. I cannot swear that at any time did they interrogate her thoroughly. c. I cannot swear to the fact that they interrogated her thoroughly at any time. d. *I cannot swear to the fact that at any time did they interrogate her thoroughly. Here the non-Horn clause case (48a), the Horn clause case (48b) with no island and the non-Horn clause case (48c) with an island are well-formed. But (48d) where the Horn clause is separated from the main clause by the island boundary is ungrammatical. Under the view that Horn clauses require syntactic raising of a NEG from the complement clause into the containing main clause, the NEG raising in (48d) but not that in (48b) must cross the island boundary, leading to the ungrammaticality.

72

collins and postal

As noted in section 7, strict NPIs receive an analysis parallel to the fronted NPI phrase of a Horn clause. Therefore, it follows that strict NPIs, such as those from the set in hours/days/weeks/years …, should also be sensitive to island boundaries in the complement of a negated CU-predicate. And they are: (49) a. I can’t say that I’ve cooked such an expensive steak in weeks. b. *I can’t say that such an expensive steak, I have cooked in weeks. (50) a. I can’t swear that I have eaten such an expensive steak in years. b. * That I have eaten such an expensive steak in years, I can’t swear to. The ill-formedness of example (49b) is due to the island created by a clause internal topic, while that of (50b) devolves on the fact that topicalized clauses are islands. Summarizing, the syntactic Classical NR analysis of CU-sentences in section 6 not only accounts for the basic facts about CU-sentences and Horn clauses, it accounts as well for the sensitivity to island boundaries in CU-sentences of Horn clauses and strict NPIs.

9

Negative Parentheticals

CP(2014: chapter 17) expanded upon ground-breaking observations in Ross (1973) to build an additional argument for a syntactic view of Classical NR. This was based on negative parentheticals like those highlighted in (51): (51) a. The council is not, Ted doesn’t believe, prepared to support that proposal. b. The council is not, no one believes, prepared to support that proposal. Ross suggested that examples like (51) require the main verb of the parenthetical, here believe, to be a CNRP. Cases like (52) with non-CNRP parenthetical verbs are ill-formed: (52) a. *The council is not, Ted doesn’t realize, prepared to support that proposal. b. *The council is not, no one dreams, prepared to support that proposal. Given these considerations, one might a priori expect that CU-predicates would permit negative parentheticals parallel to those in (51). But Horn (personal

dispelling the cloud of unknowing

73

communication of April 19, 2014 to PMP) insightfully observes that this is not the case: (53) a. I don’t know/can’t say that the council is prepared to support that proposal. b. *The council is not, I don’t know-NF/can’t say, prepared to support that proposal. The state of affairs illustrated in (53) for know-NF and can’t say is general. CUsentences never permit corresponding negative parenthetical clauses. Given the treatment of negative parentheticals in CP(2014: chapter 17), the simple Classical NR analysis treated in section 5 and arguably assumed by Horn (2014) to be the Classical NR analysis does indeed wrongly predict that examples like (53) should be grammatical (exactly as the Classical NR examples in (51) are grammatical). And that is a strong additional reason to reject the simplest analysis of section 5. But the more complex Classical NR analysis developed in section 6 rightly predicts cases like (53b) to be ill-formed. To see why, one must first consider the basic structure of a parenthetical clause. We assume the structure involves deletion, as specified in (54): (54) CP(2014: 191) “We will assume that the parenthetical is a reduction of a full clausal structure involving a complement clause that is covert in the parenthetical itself … so a parenthetical like (3a) is taken to realize an underlying structure of the form (3b).” The (3a,b) mentioned in (54) were: (55) a. Sally will, Eugene assumes, take a morning flight. b. [Sally will take a morning flight] [LINK [Eugene assumes that Sally will take a morning flight]] In structure (55b), the embedded clause that Sally will take a morning flight is deleted. This yields the parenthetical. The account was further explicated as follows: (56) CP(2014: 192) “Hereafter, we refer to the italicized structure [in (55b)] as an afterthought. Parenthetical constructions are then taken to have the general underlying form in (4):

74

collins and postal

(4) [A X] + Afterthought (= [LINK [B Y[C Z]]]) We will refer to the clause A, that is, the clause modified by the parenthetical, as the prime …; to clause B as the secondary; and to clause C as the tertiary. The idea of the LINK constituent is that it expresses whatever the precise relation between the prime and the afterthought is.” So example (55a) is analyzed as: (57) [PRIMARY Sally will take a morning flight] [LINK [SECONDARY Eugene assumes that [TERTIARY Sally will take a morning flight]]] A further key idea of the analysis in CP(2014) is that the prime modified by a parenthetical clause P is identical to the tertiary of P. Then, the tertiary of the parenthetical is deleted. A fundamental element of chapter 17 of CP(2014) is a principle called there (p. 197) the Parenthetical Non-decreasingness Condition. This requires the combination of all the elements of origin in the upper part of the secondary (e.g., modifiers, negation, the verb itself), to define, in the appropriate sense (see below), a non-decreasing function with respect to the parenthetical’s (covert) complement clause (the tertiary required independently to be identical in the relevant sense to the clause the parenthetical ‘modifies’). This principle is motivated by contrasts like those in (58) and (59): (58) a. *Terry is not, Jack/everyone/a friend of hers denies, a top level worker. b. Terry is not, no one denies, a top level worker. c. Terry is not, Sandra didn’t deny, a top level worker. (59) a. *Mercury is not, those women/most authorities doubt, too cool to vacation on. b. Mercury is not, none of them doubt, too cool to vacation on. c. Mercury is not, I don’t doubt, too cool to vacation on. In (58a) and (59a), the unmodified negative verbs create a decreasing context. But when the negative verbs are modified by negation or have a negative subject, the context is nondecreasing and the examples are acceptable. However, as noted in CP(2014: 244–245, n. 6) there is an issue here with the definition of non-decreasing as revealed by a putative inference like the following:

dispelling the cloud of unknowing

75

(60) a. Napoleon Bonaparte doubted that his mistress had consumed a sweetened drink. b. Pepsi is a sweetened drink. c. Therefore, Napoleon Bonaparte doubted that his mistress had consumed Pepsi. If the verb doubt defined a literally decreasing function with respect to its complement, (60) would be valid. But conclusion (60c) clearly does not follow from the conjunction of (60a,b) as there is no way, barring time machines, that Napoleon could have known anything about Pepsi. Nonetheless, doubt arguably does have a decreasing aspect seen in the fact that the inference to (60c) is valid if one replaces (60b) by assumption (61): (61) Napoleon Bonaparte believed that Pepsi is a sweetened drink. Let us call statements like ‘Pepsi is a sweetened drink’ in arguments like (60) a subset premise. So our claim is that doubt is downward entailing with respect to its complement clause when all the relevant subset premises are true in the belief worlds of the logical subject of the propositional attitude. Where the subset premise in (60b) fails to justify the decreasingness claim, (61) does justify it. It only makes sense to claim that a verb like doubt (and, moreover, other ‘propositional attitude’ predicates) is decreasing given such a condition. Hence, the Parenthetical Nondecreasingness Condition must ultimately be characterized in terms of this more complex notion of ‘decreasing’ limited to inferences based on appropriate subset premises. But providing a precise version of the latter is beyond the scope of this work. Subject to the refinement just discussed, the Parenthetical Nondecreasingness Condition rightly blocks examples like (58a) and (59a). But the addition of negative elements to the parenthetical clauses as in (58b,c) and (59b,c) reverses that polarity and permits satisfaction of the Parenthetical Nondecreasingness Condition. Briefly stated, that condition also allows negative parentheticals based on CNRPs like Ted doesn’t believe in (51a) because the NEG they contain can be analyzed as the result of raising under a simple Classical NR analysis. The relevant clause is then underlyingly nonnegative and forms an increasing element. On the contrary, negative parentheticals based on non-CNRPs like Ted doesn’t realize cannot be analyzed in terms of Classical NR. Their NEG thus functions semantically in the parenthetical clause, and serves to define a decreasing function, violating the Parenthetical Nondecreasingness Condition. Critically then, the Classical NR analysis of CU-sentences represented by (29) above posits an original (nonraised) NEG in the CU-predicate clause. That

76

collins and postal

means the clause defines a decreasing function in the sense defined above, so examples like (53b) violate the Parenthetical Nondecreasingness Condition for exactly the reasons (58a) and (59a) do. It turns out then that Horn’s observation about the interaction of CU-predicates and parentheticals only provides an argument against the simple Classical NR analysis of CU-sentences (see section 5), an analysis we rejected. But his observation does not conflict with the analysis we actually advocate since the latter combines with the treatment of parentheticals in CP(2014: chapter 17) to correctly predict facts like (53b). In fact, our actual analysis entails the ill-formedness of parentheticals like (53b) which Horn observed.

10

The Verb Doubt and Horn Clauses

As documented by Google-supplied examples such as (62), Horn clauses occur under the verb doubt unaccompanied by any overt instance of NEG: (62) a. I doubt that at any time did a dragonfly think tank get together and decide that the species needed to fly faster, change colors or become smaller. (catbirdscout.blogspot.com/…/life‑without‑error….) b. Yes his actions were incredibly stupid but I doubt that at any time did he think a single punch would result in someone’s death. (forums .scottishfootballforums.co.uk/…/14179-f …) c. I doubt that under any circumstances would we let our defenses down in that regard. (brookelorren.com/blog/page/53/) The analysis of CU-sentences we have advanced leads to a simple account of such Horn clauses parallel to that for CU-sentences. The verb doubt is a negative predicate representing a decreasing function with respect to its complement clause in the same restricted sense (involving belief worlds) discussed in the previous section. We need take no position on whether its negative aspect is just a lexical property of the meaning of the verb or represents the force of a covert instance of NEG. Turn then to a Horn clause case like (63a), which we propose has a structure representable as (63b). (63) a. I doubt that in any real sense at all did Lucy consider dating Greg. b. I doubt that [⟨NEG1⟩ [[PP in [⟨NEG2⟩ SOME sense at all]] did Lucy consider dating Greg]]

dispelling the cloud of unknowing

77

This is almost entirely parallel to our proposed analysis in (29) of the CUsentence (24d). It differs only in that where the former has an explicit NEG in the main clause, (63a) has the negative verb doubt. Just as in (29), there are two NEGs in the embedded clause (which cancel semantically), one of which modifies the embedded clause and the other of which modifies SOME. Parallel to structure (29), NEG1 undergoes Classical NR to the matrix clause; the relevant landing site in the main clause is not relevant to our present concerns. A virtue of analysis (63b) is that it correctly predicts that strict NPIs such as those highlighted in (64), can occur in the complement clauses of doubt: (64) a. Roger: Your report doesn’t surprise me as I doubt that area has been surveyed in years, (http://cruisersnet.net/alert‑region/035‑ef‑nav‑alerts/? sort=geo) b. Without makeup, I doubt she’s all that nice to look at on the outside. (http://www.thetruthaboutguns.com/2014/09/robert‑farago/shannon ‑watts‑stops‑making‑sense/) And a further consequence is that Horn clauses and strict NPIs will be constrained by island conditions, confirmed by the contrasts in (65) and (66): (65) a. I doubt that at any point did he steal money. b. *What I doubt is that at any point did he steal money. c. *I doubt the assertion that at any point did he steal money. (66) a. b. c. d.

I doubt that he has given money to a charity in years. *What I doubt is that he has given money to a charity in years. *I doubt the assertion that he has given money to a charity in years. *I doubt that to a charity, he has given money in years.

Examples (65b) and (66b) illustrate pseudo-cleft islands, while (65c) and (66c) illustrate the Complex Noun Phrase Constraint. Example (66d) illustrates a clause-internal topic island. A last virtue of analysis (63b) is that via the logic of the previous section, it correctly predicts that doubt clauses have no simple parenthetical correspondents, but do have one when an odd number of negative elements appear in the parenthetical clause: (67) a. I doubt that Lois has children. b. *Lois does not, I doubt, have children. c. Lois does not, I don’t doubt, have children.

78

collins and postal

d. Lois does not, no one doubts, have children. e. *Lois does not, no one doesn’t doubt, have children. Example (67e) is ill-formed because the negations in the pair {not, no one} cancel, leaving the clause with the same illicit (decreasing) polarity as that in (67b).

11

Lack of Equivalences in CU-Sentences

Horn’s (2014) Anti-HC argument was fundamentally driven by the lack of equivalences seen in CNRP cases (as in (18) and (19) above) in CU predicate cases. This was taken to undermine the claim that the Horn clauses and strict NPIs possible in CU predicate complement clauses can be analyzed in terms of a syntactic view of Classical NR. This would in turn then refute the idea central to CP(2014) that Horn clauses and strict NPI properties strongly support a syntactic conception of Classical NR. Horn’s view might well be more generally shared, as indicated by a referee’s remark on an earlier version of this paper: (68) “In fact the neg-raising phenomenon consists in the interpretation of a negation in a clause that doesn’t contain an overt negation.” This statement implies that there could be no Classical NR analysis of CUsentences, because it is only willing to recognize the phenomenon in the presence of structures which guarantee the equivalences. We rejected such views, justifying our rejection by showing how one could formulate a syntactic Classical NR analysis perfectly compatible with the lack of equivalences and consistent with all of the relevant assumptions of CP(2014). In fact, we claim that Classical NR is independent of the existence of equivalences. Such equivalences are neither necessary nor sufficient to demonstrate that Classical NEG Raising has taken place. However, even acceptance of our syntactic analysis of the CU data so far leaves an important question unanswered. Why do the equivalences not exist with CU predicates? Since we claim that both CNRPs and CU predicates permit Classical NR, what precise syntactic characteristics distinguishing the two have that consequence? An obvious clue is that the NEG appearing in the main clause in a CUsentence is semantically a constituent of the main clause, while in seemingly parallel CNRP cases, it is not. These considerations suggest the following condition:

dispelling the cloud of unknowing

79

(69) If some NEG raises into a main clause C whose predicate P is a CU predicate, there is a distinct NEG in C scoping over P. Notice that if (69) is sound, doubt cannot be a CU predicate. There would thus be at least three classes of predicates permitting Classical NR (CNRPs, CU predicates and doubt). Condition (69) distinguishes only seemingly parallel pairs like (70a,b), accounting for why the former is equivalent to (71a) but the latter is not equivalent to (71b): (70) a. I don’t believe that in any sense is she a candidate. b. I can’t say that in any sense is she a candidate. (71) a. I believe that in no sense is she a candidate. b. I can say that in no sense is she a candidate. The Horn clauses in the complements of (70a,b) argue that both involve Classical NR. But condition (69) does not allow (70b) to have the simplest type of Classical NR analysis schematized in (72) and permitted for (70a): (72) Subject Aux NEG1 Predicatex [[⟨NEG1⟩ SOME sense] is she a candidate] It requires there to be an original (hence semantic) NEG in the main clause, with consequences discussed in earlier sections. In particular, the structure of (70b) needs to be schematized as follows: (73) Subject Aux NEG1 Predicatex [⟨NEG2⟩ [⟨NEG3⟩ SOME sense] is she a candidate] Condition (69) in effect forces the overt main clause NEG in (70b) to originate in the main clause, entailing that the raised NEG has been deleted. And the NEG Deletion Evenness Condition of CP(2014) then forces there to be a second downstairs NEG. These consequences determine rightly that the NEG in (70b) has semantic force in the main clause, rendering impossible equivalences like (70b)/(71b). But (69), which permits structure (72) for (70a), allows the equivalence (70a)/(71a). A relevant observation though is that (69) does not preclude the possibility that CNRPs like believe could instantiate structure (73) (as well as structure (72)). If so, examples like (70a) might be structurally ambiguous. However, it is extremely difficult to find factual considerations which distinguish the two

80

collins and postal

meanings of CNRPs corresponding to analyses (72) and (73). So at this point we do not know whether CNRPs can also instantiate analysis (73) and we cannot pursue the issue here.

12

Conclusion

Horn (2014) assembled an array of important new data relevant to the understanding of the Classical NR phenomenon. Our reaction to this new material has argued several major points. First, nothing in that body of data is incompatible with the syntactic view of Classical NR in CP(2014). Far from disconfirming the analysis of Classical NR in CP(2014), the syntax of CU-predicates supports many of the basic assumptions of that work, including the analysis of NPIs as negative phrases and the existence of NEG deletion. And, in particular, the study of CU predicates emphasizes the independence of Classical NR from the sort of semantic equivalences which initiated its study. However, Horn’s insightful observations do raise many new questions about Classical NR, e.g. whether ambiguous analyses of CNRP cases are possible. And they also highlight the fact that no characterization of the overall class of predicates permitting Classical NR is yet available.

References Boškovič, Željko and Jon Gajewski. 2008. Semantic correlates of the NP/DP parameter. In NELS 39: Proceedings of the Thirty-Ninth Annual Meeting of the North East Linguistic Society, ed. by Suzi Lima, Kevin Mullin, and Brian Smith, 1: 121–134. Amherst: University of Massachusetts, Graduate Linguistic Student Association. Collins, Chris and Paul M. Postal. 2014. Classical NEG Raising. Cambridge, MA: MIT Press. Gajewski, Jon. 2005. Neg-raising: Polarity and presupposition. Doctoral dissertation, MIT. Gajewski, Jon. 2007. Neg-raising and polarity. Linguistics and Philosophy 30: 298–328. Gajewski, Jon. 2011. Licensing strong NPIs. Natural Language Semantics 19: 109–148. Homer, Vincent. 2010. Neg-raising and positive polarity: the view from modals. Ms., UCLA. http://semanticsarchive.net/cgi‑bin/browse.pl?search=homer Horn, Laurence R. 1975. Neg-raising predicates: toward an explanation. In Papers from the Eleventh Regional Meeting of the Chicago Linguistic Society, ed. by Robin E. Grossman, L. James San, and Timothy J. Vance, 280–294. Chicago: University of Chicago, Chicago Linguistic Society.

dispelling the cloud of unknowing

81

Horn, Laurence R. 1978. Remarks on neg-raising. In Pragmatics, ed. by Peter Cole, 129– 220. New York: Academic Press. Horn, Laurence R. 1989 [2001]. A natural history of negation. Stanford, California: CSLI Publications. Horn, Laurence R. 2014. The cloud of unknowing. In Jack Hoeksema and Dicky Gilbers (eds.), Black Book: a festschrift in honor of Frans Zwarts, 178–196. Groningen, the Netherlands: University of Groningen. Horn, Laurence R. and Samuel Bayer. 1984. Short-circuited implication: a negative contribution. Linguistics and Philosophy 7: 397–414. Jacobson, Pauline. 2006. I can’t seem to figure this out. In: Birner, B.J., Ward, G. (eds.) Drawing the boundaries of meaning: neo-Gricean studies in pragmatics and semantics in honor of Laurence R. Horn, Studies in Language Companion Series, vol. 80. Amsterdam: John Benjamins Publishing Company. Larrivée, Pierre. 2004. L’association négative. Geneva. Librairie Droz. Postal, Paul M. 2005. Suppose (if only for an hour) that negative polarity items are negation-containing phrases. Paper presented at Workshop on Polarity from Different Perspectives. New York University. March 2005. Romoli, Jacopo. 2012. Soft but strong neg-raising, soft triggers and exhaustification. Doctoral Dissertation, Harvard. Romoli, Jacopo. 2013a. A scalar implicature-based approach to neg-raising. Talk given at New York University, May 6. Romoli, Jacopo. 2013b. A scalar implicature-based approach to neg-raising. Ms., Macquarie University. Ross, John Robert. 1967 [1986]. Constraints on variables in syntax. Doctoral dissertation, MIT. Published as Infinite syntax, Norwood NJ: Ablex. Sailer, Manfred. 2005. NEG-raising in an underspecified semantics framework. Paper presented at Colloque de Syntaxe et Sémantique, Paris. http://www.cssp.cnrs.fr/ cssp2005/index_en.html Sailer, Manfred. 2006. Don’t believe in underspecified semantics: Neg raising in lexical resource semantics. Empirical issues in syntax and semantics 6. O. Bonami & P. Cabredo Hofherr (eds.) 375–403. http://www.cssp.cnrs.fr/eiss6 Szabolcsi, Anna. 2004. Positive polarity—negative polarity. Natural Language and Linguistic Theory 22: 409–452. Tovena, Lucia M. 2001. Neg-raising: negation as failure. In Perspectives on negation and polarity items. Jack Hoeksema, Hotze Rullmann, Victor Sanchez-Valencia and Ton van der Wouden (eds.) 331–356. Amsterdam: John Benjamins.

chapter 3

Presuppositions, Negation, and Existence Barbara Abbott*

1

Introduction

This year (2015) marks the 110th anniversary of the publication of Russell’s classic “On denoting.” It should not cast any shadow on that great work to note that the problems it provided solutions to are still the subject of controversy. Two of those problems involved noun phrases (NPs) which fail to denote. Russell’s examples (1a) and (1b) (1) a. The present king of France is bald. b. The present king of France is not bald. are puzzling because they have the form of simple contradictories, and yet we are not inclined to say either one is true. Example (2) (2) Pegasus does not exist. is even more problematic; the lack of denotation for Pegasus, which makes the sentence true, also seems to rob it of a meaningful constituent. Once the king of France is unpacked according to Russell’s analysis, (1b) is revealed to be ambiguous. Its logical forms are given in (3). (3) a. ∃x[Kx ∧ ∀y[Ky ↔ y=x] ∧ ¬Bx] b. ¬∃x[Kx ∧ ∀y[Ky ↔ y=x] ∧ Bx]

* I am grateful to Jay Atlas, David Braun, Noel Burton-Roberts, Elena Guerzoni, Ruth Kempson, Jerry Sadock, Mandy Simons, and Deirdre Wilson for comments on an earlier version of this paper. And I am particularly indebted to Larry Horn for helpful comments on more than one earlier version, including detailed suggestions—some of which have been incorporated verbatim, with no specific credit given. Of course none of these people can be assumed to agree with the finished product.

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_004

presuppositions, negation, and existence

83

(3a) says that there is a unique (French) king, who is not bald (obviously false in the present day, as well as in 1905), but (3b), the logical contradictory of (1a), says that it is not the case that there is a unique king who is bald (which is true). We can apply the analysis to sentence (2) once we recognize Pegasus as a concealed definite description, e.g. the winged horse of Greek mythology. (2) can then be unpacked as (4) (4) ¬∃x[Wx ∧ ∀y[Wy ↔ y=x]] which says that it is not the case that there is a unique winged horse of Greek mythology, and that seems both meaningful and true, as required. Problems solved. Well, not quite. Strawson (1950) challenged the first solution above, arguing that neither (1a) nor (1b) could be used to assert the existence of a king of France. (Actually, his discussion did not use Russell’s examples, but rather very similar ones involving the king’s wisdom rather than his baldness.) Instead, Strawson argued, the use of such sentences presupposes1 the existence of a king of France, and failing that existence, neither (1a) nor (1b) could be used to make either a true or a false statement. In Strawson’s words, “the question of whether it’s true or false simply doesn’t arise” (Strawson 1950, 330).2 In an extended series of essays and two books, Jay Atlas (1977, 1978, 1979, 1989, 2004, 2005) has taken issue with the work of both Russell and Strawson. To summarize his views, a sentence like The king of France is not bald (wise) can be used to express (at least) two thoughts—one of which predicates nonbaldness (or lack of wisdom) of an assumed-to-exist king of France, and the other of which is a blanket denial that the actual world (at present) is one which contains a unique male monarch of France with no hair on his head (or smarts). These correspond to (3a) and (3b) respectively. However, contrary to Russell’s analysis, such sentences are not ambiguous. Atlas uses the term sense-general to describe such sentences semantically, but does not mean that term to imply

1 This is the term in common use today. Strawson did not use it in his 1950 paper, but did two years later. 2 Strawson’s views here are highly reminiscent of those of Frege (1892), though Strawson did not mention that earlier work. For a long time I assumed that Strawson’s neglect of Frege’s work was the consequence of the two world wars in which Germany and England were on opposite sides. However, I have since learned that apparently almost everybody ignored Frege, Germans included (see Dummett 1967, 225; 1981, xliif), although the explanation seems to remain a mystery.

84

abbott

a proposition—something with truth conditions. The latter sort of thing only arises in particular contexts, associated with tokenings of (1b). In those tokenings, either the NP the king of France can serve as a sentence topic or it can serve as part of the sentence focus. In the former case, but not the latter, the existence of a king of France will be presupposed.3 Finally, in the statement made in the utterance of a sentence like (2) (Pegasus does not exist) the NP Pegasus does not serve as sentence topic. Hence such a statement will not presuppose the existence of Pegasus. Atlas’s work has done a great deal to clarify our thinking about these troublesome issues, and he deserves praise for the methodical care he has taken with them. However there remain some aspects of his approach which deserve further examination, and that is the purpose of this paper. We’ll begin in § 2 by looking in general at the nature of the different understandings of negative sentences. (Understandings is the useful general term from Zwicky & Sadock 1975 which we can take to include instances of ambiguity as well as generality, vagueness, pragmatic implication, and whatever else could logically be included.) As discussed at length in Horn 1989, two different kinds of ambiguity of negation have been proposed historically, and it makes a difference to Atlas’s claims which one he is concerned with. Following that, in § 3 we will turn to the arguments Atlas gives for his “sense-general” theory of negative sentences like (1b), containing a definite description. In §4 we’ll look at a Gricean alternative to this view, and in §5 we turn to negative existence statements like (2). The final section contains concluding remarks.

2

Negation: Scope vs. Lexical Ambiguity

As we will see in this section, at least two different kinds of ambiguity of negation, or two different views of what such an ambiguity consists of, have been proposed. Russell’s analysis postulates a simple scope ambiguity—we’ll review this kind of thing in §2.1. But it has also been suggested that this ambiguity is of a different type, which could be viewed as an ambiguity in the word not, and we will look at this in §2.2. 2.1 Scope Ambiguities In typical formal languages of logic, negation is a sentence operator—something which combines with a sentence to make a new sentence. As such it has

3 On this point, Atlas echoes Strawson’s revised view on presuppositions—cf. Strawson 1964.

presuppositions, negation, and existence

85

a semantic scope which is determined by its syntactic scope—the sentence with which it combines. Other sentence operators are (a) the other sentence connectives (conjunction, disjunction, etc.); (b) the existential and universal quantifiers; (c) operators expressing necessity and possibility, in a modal logic; and (d) sentence-taking predicates like believe’ and try’ in an intensional logic like Montague’s (Montague 1973). Each of these operators has a semantic scope, determined by its syntactic scope. In these formal languages, the scopes of any operators in a sentence are fixed by the syntactic structure and the result is the desirable property of univocality. It has been traditional to view natural languages as being different from formal languages in this regard. That is, a natural language sentence with more than one sentence operator may be ambiguous as a result of having different ways to construe the semantic scope of the syntactically fixed operators. Typical examples, with paraphrases intended to suggest the commonly attributed understandings, are given in (5)–(10). (5) Mary will win or Bill will win and Sue will discover an anomaly. a. Mary or Bill will win, and Sue will discover an anomaly. b. Mary will win, or alternatively Bill will win and Sue will discover an anomaly. (6) Many of the arrows didn’t hit the target. [= Jackendoff 1972, ex. 7.46b] a. The arrows hitting the target were few in number. b. Lots of arrows missed the target. (7) The current president might not have been a Democrat. a. Barack Obama might not have been a Democrat. b. It might have been the case that a Republican was elected in 2012. (8) Ralph believes that someone is a spy. [from Quine 1956, 178] a. Ralph believes that there are spies. b. Someone is suspected by Ralph of being a spy. (9) John would like to marry a girl his parents don’t approve of. [= Partee 1972, ex. 1] a. John wants to irritate his parents by marrying somebody they don’t approve of. b. The girl who John wants to marry isn’t approved of by his parents.

86

abbott

(10) Oedipus wanted to marry his mother. a. Oedipus wanted to commit incest. b. Oedipus wanted to marry a woman who, unbeknownst to him, was his mother. Thus it is not uncommon for natural language sentences with two (or more) sentence operators to show these scope ambiguities.4 Natural language sentences with more than one sentence operator are not necessarily ambiguous. It can happen that potentially different understandings turn out to be equivalent, as in the case of (11): (11) Someone likes someone. Or for other reasons an operator can be prevented from taking a bigger scope than another operator within whose syntactic scope it lies, as in example (12): (12) Not many arrows hit the target. where many cannot escape from the clutches of not. But in general we have been led traditionally to expect scope ambiguities in sentences containing more than one sentence operator, and usually we have not been disappointed. 2.2 A Different Kind of Ambiguity of Negation? The phenomenon just scouted should be distinguished from another possibility. As discussed in Horn 1989, Aristotle proposed that a sentence like (13): (13) Socrates is ill. has two negative counterparts. The contradictory (or exclusion) negation of (13) results in a proposition that is true if (13) is not true, regardless of whether there is anybody named Socrates. This might be roughly paraphrased as in (14a) (though see the provisos in Atlas 1989, 71 and elsewhere): (14) a. It is not the case that Socrates is ill.

4 An example from Bäuerle (1983) has brought scope analyses into question. See Abbott 2010, 51 for discussion and references; and Keshet 2010 for a reply.

presuppositions, negation, and existence

87

However negation can also function to deny a term, in this case the predicate, giving the understanding in (14b) (also called the choice negation): (14) b. Socrates has the property of not being ill—i.e. Socrates is well. This distinction was apparently expressed in Greek with two distinct word orders (see the discussion in Horn 1989, §1.1.1), but it has been claimed that both interpretations are expressible with the one negative sentence in English given in (14c) (14) c. Socrates is not ill. Viewed in this way, the distinction would constitute a kind of ambiguity of negation for English.5 Possibly the difference could be represented as one of sentence scope, were Socrates taken to be a sentence operator. Of course that is what Russell in essence proposed, by analyzing proper names as standing for definite descriptions, which he in turn had analyzed as quantificational expressions. Nevertheless it seems useful in principle to distinguish the two sorts of potential ambiguity. Atlas is, of course, very well aware of the difference between the two sorts of potential ambiguity. Nevertheless it seems that he views the difference as of no significance for his project, since he is arguing against either version. So sometimes he seems to blur the two together, as in the following quote: It was Russell’s, and is presently the standard, view that the sentence: [(15)]

The king of France is not wise.

is ambiguous, having two senses and two logical forms, one the narrowscope predicate-negation, usually represented in English by:

5 This suggests that the distinction is between sentence negation and predicate negation, and that is assumed in the quote to be given below from Atlas. (Cf. Jackendoff 1969.) However, according to Horn, Aristotle held that neither was a case of sentence negation. Instead, the contradictory reading involves a negative mode of predication which Horn calls “predicate denial”, while “predicate negation” is a term negation of the predicate. Compare Horn 1989, 15.

88

abbott

[(16)]

The king of France is “non-wise”.

and the other the wide-scope sentence-negation usually represented in English by: [(17)]

It is not {true/the case} that the king of France is wise. (Atlas 1977, 323.)

In this passage we seem to have an identification between narrow scope negation and predicate negation on the one hand, and wide scope negation and sentence negation on the other. However, as noted above, in general, scope ambiguities involve sentence operators and not a variation in an operator taking a sentence vs. a predicate as its argument. 2.3 Why the Difference Matters The difference between the two types of ambiguity becomes relevant when we inquire as to the intended extent of the claims about sense generality. If Atlas’s arguments are aimed at negation qua predicate or sentence level operator, they would seem to have no implications for other sentence operators and the scope ambiguities they seem to be involved in. However if his arguments are aimed at negation simply as a sentence operator, then they can be expected to have much wider repercussions. And the author himself is uncharacteristically indecisive on this point. In Atlas 1977, which was an early introduction of his thoughts about the ambiguity of negation, example (15) above was contrasted with an example involving an explicit quantifier. (The examples in (18) are from Atlas 1977 and Atlas 1989, exx. 1–3 in both cases.) (18) Everyone didn’t show up. a. No one showed up. b. Not everyone showed up. Concerning these sentences he remarks in the earlier work: “This is a genuine ambiguity accountable for by a difference in scope” (Atlas 1977, 323). However, by the time of Atlas 1989, his views had apparently shifted, for this comment is missing there. Instead, in the later work (18) is used as an apparent example of his sense generality view. Thus, in the later work. the text preceding (18) begins: “The paradigm cases of scope ambiguity involving negation have led to a mistaken semantic description of negative sentences standardly thought to possess presuppositions. These mistaken descriptions assign to a single sur-

presuppositions, negation, and existence

89

face structure two distinct underlying forms” (Atlas 1989, 69; boldface added). Example (18) is then introduced as an illustration of this mistake. I would be more certain that the author of Atlas 1989 intended his remarks to hold for all scope ambiguities, and not just the traditional interaction between negation and definite descriptions, were it not for the fact that he does not return to example (18), or others like it, until the very end of the book, and there he seems to have retreated to a less confident position. The example in question at that point is a sentence from one of Abraham Lincoln’s classic remarks— example (19) below: (19) You can fool all of the people some of the time. Concerning this example he says that “if the quantifiers were scope-non-specific in the semantic representation of the sentence, [my claim] would be … that there are no distinct ‘readings’ …” (Atlas 1989, 148; italics in original, boldface added); and his discussion continues briefly in this hypothetical mode before the final summary paragraph of the book. Thus it remains unclear how broadly the concept of sense generality is intended to apply. This makes a difference, because it is much easier to argue against an ambiguity of the negative morpheme (sentence vs. predicate operator) than it would be to argue against negation participating in scope ambiguities. That is because the latter sort of phenomenon is quite common, as noted above in connection with examples (5)–(10).

3

Negative Sentences with Definite Descriptions

We turn now to the arguments that sentences like (1b), repeated here as (20), (20) The present king of France is not bald. are not ambiguous, i.e. do not get assigned two (or more) different logical forms by the grammar of English. Instead, the claim is that (20) is sense-general, and can be used to express different propositions and make different statements in different contexts, only some of which will presuppose that there is a king of France. Atlas puts forward a number of arguments toward this conclusion. Most of them are based on the seminal discussion of ambiguity in Zwicky & Sadock 1975. However the first one we consider below is new.

90

abbott

3.1 Phonetics vs. Grammar? The first kind of argument we’ll consider rests on the fact that the very same sequence of words can be associated with different intonational contours, with different results for what is taken to be the sentence topic, and what is taken to be the new information being conveyed. Compare the following (here I follow Atlas in using italics for what are to be taken as statements rather than sentences; small capitals indicate focal stress): (21) a. The king of France isn’t bald. b. The king of France isn’t bald. c. The king of France isn’t bald. Atlas holds that (21a) actually has two “unmarked” uses, both of which are represented as (21b). These are the two understandings corresponding to (15) and (16), repeated here as (22a,b), respectively (with the property of baldness replacing wisdom). (22) a. The king of France is “non-bald”. b. It is not {true/the case} that the king of France is bald. These are the ones that we are primarily concerned with. The first of these is described as failing “to be a semantically referential, singular assertion”, due to the absence of a king of France, while the other is the “true, non-singular assertion that is the external negation of the proposition that the king of France is bald” (Atlas 1989, 90). On the other hand (21c) is a “true [assertion] about the bald” (90), namely, that the king of France isn’t one of them. “All of these statements employ the same sentence, whose meaning is unspecified for, rather than ambiguous between, predicate and sentence negation or between objectlanguage negation and metalanguage denial” (90). He continues (p. 91): “Our interpretations depend not on the syntactic and semantic but on the possible phonetic representations of the sentence uttered. The range of interpretation includes statements that are choice [i.e. predicate term] negations, exclusion [i.e. sentence] negations, and metalinguistic predications.” The strange part about this claim is the apparent assumption that phonetic form is not part of sentence form. For a linguist this assumption goes against the grain; spoken language, as we are taught early on, is the primary form of language. Furthermore it has become standard in linguistic analyses these days to include some indication of topic-focus structure in representations of sentences, and to assume that such structure is semantically relevant. As Gundel & Fretheim (2004) note: “It is not surprising … that most accounts of topic and

presuppositions, negation, and existence

91

focus have built these concepts into the grammar, as part of the syntax and/or semantics … or as a separate information structural component” (191). Thus for most linguists nowadays sentences (21b) and (21c) are different, because their prosodic structures, which reflects their informational structures, are different. That being the case, it would seem that we don’t really have an argument for the sense generality view yet. 3.2 Argument Types from Zwicky & Sadock The remaining arguments have a rather different character. For these we assume the association of distinct statements with a given sentence, and apply tests found in Zwicky & Sadock 1975 and elsewhere to determine whether these statements count as different readings, or whether they are encompassed within a single, general sense. Even if the evidence points to the latter conclusion (as Atlas believes), it would seem that it does not point specifically to a conclusion that this general sense is not one that could be grammatically encoded in the sentence. We will turn to that issue below, in § 4. 3.2.1 Privative Ambiguity As Atlas notes, there is a potential complication in the application of Zwicky & Sadock’s tests. If we consider just the two “unmarked” understandings of (21a)—the one that presupposes the existence of a king and says he isn’t bald, and the one that is the blanket denial—we can see that the former understanding entails the latter. Thus we have the kind of situation Zwicky & Sadock call privative. Zwicky & Sadock’s example of an expression with a privative ambiguity is the word dog, which they take to be ambiguous on whether it’s specified for the sex of the animal. I have to confess that this word is not ambiguous for me—I don’t have the +male reading. Ignoring that, the test for privative ambiguity, as opposed to generality, is whether the same expression can be both asserted and denied of the same object or situation. Were dog ambiguous in the way Zwicky & Sadock assume, (23) should be o.k. (23) It’s a dog, but it isn’t a dog. [I.e. it’s a bitch.] I find (23) completely impossible, but, as I say, I don’t have the reading in question. The only potential example I’ve been able to think of that might work for me is the word drink, which I think people agree is ambiguous depending on whether the drink is taken to be alcoholic or not. Still, (24) is far from being completely natural:

92

abbott

(24) I’d like a drink, but not a drink, please. The reason is not far to seek—any speaker of (24) would be being unnecessarily obscure. What some obnoxiously coy person might say is (25): (25) I’d like a drink, but not a drink drink [wink wink]. using what Horn (2006, 16) describes as a “value added” type of lexical clone, to specify the R-narrowed, +alcoholic reading. Or alternatively, instead of the doubling, one could contrastively stress the second occurrence of drink, as in (25’): (25’) I’d like a drink, but not a drink. For more on this, see Horn to appear. In any case, Atlas accepts the test, and applies it to (26): (26) The king of France is not wise. The two commonly assumed (following Russell) understandings of (26) are paraphrased unambiguously in (27a) and (27b): (27) a. It’s not the case both that there is a unique king of France and that he is wise. b. There is a unique king of France, but he is not wise. (27a) is of the form not (p and q), while the more specific (27b) is of the form (p and not q). These are shown in (27a’) and (27b’) respectively. (27) a’. not (p and q) b’. (p and not q) If (26) is ambiguous in this way, it should be possible to assert it on the (27a) reading, but deny it on the (27b) reading. Given the equivalence in (28), (28) not (p and q) and not (p and not q) ≡ not p this should yield a net assertion that there is no king of France. Atlas’s applications of this test are given in (29) (= Atlas 1977/1989, 72, ex. 7):

presuppositions, negation, and existence

93

(29) {?The king of France is not wise/?It’s not {true/the case} that the king of France is wise}, but {he/the king of France} is (not non-) wise. None of the variants in (29) is very good; Atlas remarks that they are “semantically out of bounds” (p. 72). However, in addition to the already expected obscurity problem for this test, noted with respect to example (24) above, the applications in (29) have another difficulty. That difficulty comes from Atlas’s assumption that the denial of (26) on the (27b) reading must be a predicate denial rather than a sentence denial. (Observe the parentheses around “not non-” in (29).) If we conjoin an assertion of (27a) with a predicate denial of (27b), we are left with a contradictory statement—something of the form (30a), or equivalently, (30b): (30) a. not (p and q) and (p and not not q) b. not (p and q) and (p and q) Definitely out of bounds for those who like logical consistency, but not the right structure for what we want to test. A better example to use for the application of Zwicky & Sadock’s test for privative ambiguity would be (31): (31) The king of France isn’t wise but the king of France isn’t not wise. Now admittedly (31) isn’t great, but (23) and (24) weren’t that great either. Furthermore I have been assured by a number of informants that there are circumstances in which (31) could be felicitously used to deny the existence of a king of France (though see below, §4). Compare also the discussion by Strawson (1952, 18) of the example in (32): (32) A: Does he care about it? B: He neither cares nor doesn’t care; he’s dead. (I am grateful to Larry Horn for telling me about this example of Strawson’s.) Thus, contrary to Atlas’s conclusion, this test may indicate ambiguity after all. ((31) can also be used in cases of “coerced middles”; cf. Horn 2017.) 3.2.2 Tests of Transformational Potential The next test is one which involves “employing a transformation whose application to underlying structure requires identity of sense of constituents” (1977, 326; 1989, 73). This test uses examples like (33) (cf. Atlas 1989, 73, ex. 8):

94

abbott

(33) The king of France is not wise, and the same goes for the Queen of England. The crucial understanding for (33) would be the one on which the first clause is understood with sentence negation (i.e. not presupposing the existence of a king of France), while the second clause does presuppose the existence of a Queen of England. This is the so-called “crossed” understanding. If we can understand (33) in this way, it is supposed to be evidence against an ambiguity for (26) (The king of France is not wise) and in favor of its generality. However, it’s not clear to me that the results of the identity of sense test will be any more conclusive than one’s intuitions concerning (26) itself. That is, I don’t know why people who find (26) ambiguous won’t find it difficult to get the crossed reading for (33), and vice versa. The identity of sense test is a subcase of Zwicky & Sadock’s tests of “transformational potential”. The idea behind them is that “If the semantic representations for certain sentences lack specification of some piece of meaning, then the applicability of transformations to them cannot possibly depend on whether or not this piece of meaning is present” (Zwicky & Sadock 1975, 14). Atlas leaves other applications of this test as an exercise for the reader, but assures us that the results will confirm that the sentences in question are not ambiguous. However a couple of constructions that were assumed to be derived by transformational rule at the time Zwicky & Sadock were writing— the cleft and pseudocleft constructions—do seem to “disambiguate” e.g. (26) in favor of a nonpresuppositional interpretation. Compare the examples in (34): (34) a. It isn’t the king of France who is wise. b. The one who is wise isn’t the king of France. Both of these sentences move the NP the king of France into predicate position, where it tends to lose its presuppositionality. I am not sure what the response would be to this—perhaps that these constructions should not be considered transformationally related to the canonical sentence form counterpart. Of course the whole basis for this kind of test lies in a model of grammar which is no longer assumed as the default. 3.2.3 Semantic Differentiae The last test we’ll consider is what Atlas, following Zwicky and Sadock, calls “semantic differentia”.6 His discussion of this test is on pp. 72f (in 1989; cf. also 1977, 324f). He says first: 6 There seems to be a small disagreement between Atlas and Zwicky & Sadock over the declen-

presuppositions, negation, and existence

95

When a sentence has relatively similar understandings [as is true in this case] … but these differ only by one’s being sense-specified and the other sense-unspecified for some particular semantic feature, e.g. [+/- fact], the feature must be such that the lexicons of natural languages can plausibly fail to use it. 1989, 72

Note the suggestion here that we are not dealing with a scope ambiguity, but instead with a lexical ambiguity. The discussion continues to talk about the degree of difference between the putative readings of an ambiguous expression: “If the difference in understanding were very great it would point to ambiguity rather than sense-generality” (1989, 72). The implication is that small differences in understanding are more likely to be a reflection of generality rather than ambiguity. (This idea is one for which Zwicky & Sadock cite Richman 1959.7) However this criterion is certainly not absolute. Indeed, we have Jerry Morgan’s example (35): (35) Someone is renting the house. (cited by Atlas elsewhere in the book (1989, 33)) where it is clear that there is an ambiguity, but where the two readings are mutually entailing. As Zwicky & Sadock note: … from the facts that a particular semantic differentia is simple and that it is formally marked in some language we can conclude nothing about the status of this distinction for any particular example in any language; both lack of specification and systematic ambiguity are consistent with these facts. Zwicky & Sadock 1975, 5

sion of differentia—which is a plural for Atlas, but a singular for Zwicky & Sadock. (The plural for them is differentiae; cf. Zwicky & Sadock 1975, 4.) 7 There is a pernicious typo in Zwicky & Sadock’s reproduction of Richman’s conclusion. They quote him (unfortunately under the name “Richmond” rather than “Richman”) thus (where φ and ψ represent two properties expressed by some term T): “The question is this, are φ and ψ sufficiently alike (in some unspecifiable way)? If they are, T is ambiguous; if not, not [p. 91f]” (Zwicky & Sadock 1975, 5, n. 12; italics in original). What’s missing is the “un-”prefix for “ambiguous”; cf. Richman 1959, 92.

96

abbott

Against this background, the systematic scope ambiguities illustrated in (5)– (10) above provide evidence in favor of ambiguity for the sentences we have been considering.

4

The Gricean Alternative8

Let us for the moment ignore the evidence reviewed so far, which is not that conclusive anyway, and assume that Atlas’s arguments go through. It still seems possible to hold the view that sentences like (1b), (21a), and (26), repeated here as (36a–c), (36) a. The present king of France is not bald. b. The king of France isn’t bald. c. The king of France is not wise. do have a logical form, which is that of the more general, sentence level negation, and that an understanding which presupposes or entails the existence of a king of France is the result of a conversational implicature (Grice 1975). The idea would be that, if the speaker knew that there was no king of France, it would be more informative for them to have said that. Given that the king does not exist, the issue of his baldness or wisdom doesn’t even arise. It would be being unnecessarily obscure in almost any context to assert any of (36) knowing that there is no king. The only exception might seem to be when someone else, ignorant of the king’s nonexistence, has for some reason made the corresponding positive assertion, although even then the simple denial of existence would be more informative. Given the generality of this line of reasoning, the implicature in question would be a generalized one, neutralized only in very special contexts, or with an explicit cancellation, as in (37): (37) The king of France isn’t bald (or wise)—there isn’t any! (I have to say that I find (37) somewhat strained. We’ll return to this below.) Something like this is the view of Kempson (1975), whom Atlas frequently cites approvingly, and it seems to have been Atlas’s original thought as well. However he argues against this view in Atlas 1979, and toward the end of his

8 This alternative is “Gricean” rather than “Grice’s” because Grice’s own views on this particular issue were somewhat more complicated. See Grice 1981.

presuppositions, negation, and existence

97

book (Atlas 1989). The arguments given there against the Gricean view are highly theoretical; indeed, they verge on the aesthetic. The situation is summarized as in (38) (Atlas 1989, 144): (38) a. Prag (K*, L-) = L+ b. Prag (K**, L-) = Lwhere “Prag” represents “the Gricean inference”, the Ks are different kinds of contexts, “L+” represents the understanding which assumes the existence of the king and “L-” the understanding which doesn’t. For Atlas, the problem with (38) seems to be its asymmetry: “In the second case the pragmatics adds nothing to the semantic interpretation; in the first case it obviously does” (1989, 144). But this claim can be questioned, since the contextual features which give rise to the neutralization or cancellation of a generalized implicature are certainly affecting the interpretation, and therefore in that sense adding something to it. Atlas schematizes his own view of the situation as in (39): (39) a. Prag (K*, L) = L+ b. Prag (K**, L) = LHere L stands for “the sense-general semantic representation of ‘the F is not G’ ” (1989, 145). The idea is that we have a single semantic representation, and pragmatic factors determine which of the two propositions associated with sentences of the type we’ve been discussing is actually expressed on any occasion of utterance. Thus, “the pragmatic theory does theoretical work in both cases” (1989, 145; small caps in original). The problem of asymmetry is removed. An important assumption throughout the discussion here is that the two understandings represented with L+ and L- are equally natural. This is reflected in Atlas’s assertion that, in effect, both are “unmarked” uses of sentences like those in (36); and it is made explicit when he says the following: Taking a negative sentence in isolation competent speakers know that it has (at least) two uses or understandings. Independently of context, the understandings are phenomenologically of equal status, neither judged less a function of the meaning of the sentence than the other. Atlas 1989, 142; boldface added

And he describes the advantage of his (39) over the Gricean competitor in (38) as follows: “Understandings that phenomenologically are of equal status are

98

abbott

now represented by the theory of being as equal status, produced in the same way by the same mechanism” (1989, 145). However, I do not believe that the two understandings represented with L+ and L- are symmetrical. I want to claim instead that these understandings are not phenomenologically equivalent; rather, it is the L+ understanding of sentences like those in (36) that is overwhelmingly the favorite. That is why the examples in (36) seem so strange. In fact it is very difficult for me to get the Lunderstanding of these sentences at all; I find it very hard to make (37) sound anything but contradictory.9 (And Atlas himself (1975 et seq.) had made a similar point!) It might be thought that one factor promoting the prominence of the presuppositional understanding is the fact that syntactically, negation occurs with the verb of the sentence, as though not became glued onto the verb making it more difficult to become semantically unstuck and go flying to the front of the sentence. On the other hand modals don’t seem to have this problem. Examples like (7) above, repeated here as (40), (40) The current president might not have been a Democrat. a. Barack Obama might not have been a Democrat. b. It might have been the case that a Republican was elected in 2012. are common in the literature illustrating scope ambiguity. And indefinite descriptions can also vary their scope with respect to negation pretty easily. It seems pretty easy to get examples like those in (41) (41) a. Someone didn’t call me up today. b. A book isn’t on the table. with the indefinite subject understood as being inside the scope of the negation (especially when uttered as direct denials of the corresponding positive utterance). Compare also example (18), repeated here as (42), (42) Everyone didn’t show up. a. No one showed up. b. Not everyone showed up.

9 Note that Horn 1985 argues that (37) is an example of metalinguistic negation. I regret that space has prevented inclusion of discussion of this phenomenon with the other varieties of negation being considered here.

presuppositions, negation, and existence

99

where the interpretation in (42b), with wide scope negation, is perhaps easier to get than (42a). That being the case, it is something of a mystery as to why the nonpresuppositional understanding of sentences like those in (36) is not more robust. In any case, leaving this puzzle aside, the Gricean account seems to match the data better than Atlas’s alternative. Given that the implicature (if that’s what it is) of the existence of a referent for a definite description occurring in a negated sentence is a generalized implicature, we expect that interpretation to predominate. Note, though, that we could have a similarly Gricean account were these sentences actually ambiguous; the same line of reasoning would tell us why one interpretation was more plausible and prominent than the other. (Cf. the discussion in Sadock 1978.)10

5

Negative Existence Statements

We turn now to the last group of issues, those concerning negative existence statements like (2), repeated here as (43). (43) Pegasus does not exist. Let’s quickly review the traditional problem such sentences present. Given his assumptions of compositionality of both sense and reference, Frege (1892) would potentially have problems at both levels. If proper names do not have a sense (as suggested by intuition and argued by Mill 1851 and Kripke 1972), then the sense of the whole sentence should be defective or nonexistent. At the level of reference, the truth of a sentence like (43) would mean that the subject term, Pegasus, lacks a referent, and so the whole sentence should lack a referent, which in Frege’s view is its truth value; in brief, the truth of (43) implies its being neither true nor false.11 These are the problems.

10

11

It should be noted that Burton-Roberts (1989, 1991) makes a number of the same points as those in this section. However he does not approve of the Gricean approach. Rather, his own view is something which he schematizes as in (i) (Burton-Roberts 1991, 171ff): (i) a. Prag (K*, L+) = L+ b. Prag (K**, L+) = Lwhere the more specific, presupposing understanding is the semantically assigned reading, and the other is derived pragmatically. Curiously, Atlas remarks that the truth of (43) will imply its falsity rather than its truthvaluelessness on the traditional view: “a statement like Pegasus doesn’t exist will be pre-

100

abbott

Frege solved the problem at the level of sense by postulating senses for proper names, senses equivalent to those expressed by definite descriptions. Russell completed the solution by analyzing definite descriptions into referential parts (plus quantifiers and so forth). However, as noted at the outset, Strawson (1950) raised objections for Russell’s quantificational analysis of definite descriptions, and Kripke (1972) has argued very convincingly for a return to Mill’s position, that proper names are not equivalent to definite descriptions anyway. What to do? Atlas’s idea is that the problem stems from an improper informational analysis of sentences like (43). More specifically, he argues that (43), or a statement made with (43), does not have Pegasus as a topic NP, and thus that the NP Pegasus is not presuppositional. Instead, the topic of (43) is existence. Atlas gives three main criteria for an NP to be considered a topic. (Each of these is given as a necessary, but not a sufficient condition.) They are (a) ability of the containing statement to be an answer to the question What about NP?; (b) the acceptability and equivalence of a statement with an as-for-NP prefix; and (c) lack of primary stress on the NP in question. A fourth, but somewhat more contentious criterion is given as a sufficient condition for topichood: that the existence of a referent of the NP is presupposed (1989, 111f). Atlas first considers criteria (a) and (c). “With normal intonation and stress, to what question is the statement of John exists a felicitous answer? I think it is Who/What exists? (and more generally, What about what exists?). The topic is: what exists …” (Atlas 1989, 112). And notice that the unmarked pronunciation of John exists is with main sentence stress on John rather than exists. Subsequently the reader is given the following pairs (Atlas 1989, (52) & (53); Atlas 2004 gives a similar set (p. 352)): (44) a. John exists. b. As for what exists, John does. (45) a. John exists. b. As for John, he exists. The pairing in (44) is asserted to be correct—i.e. they are paraphrases. However the pairing in (45) is not correct. Instead, (46) is given as an appropriate paraphrase for (45b).

dicted to be false if true …” (Atlas 1988, 382; 1989, 106). Possibly this is a typo. Or I’m missing something.

presuppositions, negation, and existence

101

(46) John exists. (e.g. John doesn’t live, John exists.) [= Atlas 1989, 113, ex. 53a’] Again the conclusion is that the subject NP is not a topic NP in the unmarked case of an existence statement. Instead, “the topic is: what exists” (1989, 113). And if the subject NP is not the topic, then, according to his fourth criterion, the existence of a referent is not presupposed, but is instead asserted. It is significant that these criteria are examined primarily with respect to positive, rather than negative, existence statements. When Atlas comes to consider the case of negative existence statements he says: “Parallel arguments support the same conclusion for explicit negative existence statements. A quick argument to this effect is easily formulated: presuppositions are invariant under main-verb negation” (Atlas 1989, 114; also Atlas 2004, 352). This claim might be seen to conflict with the assumption that negative sentences like those in (36) above have, as one of their uses, a “true, non-singular assertion that is the external negation of the proposition that the king of France is bald” (Atlas 1989, 90). On this use, the presupposition of existence of a king of France is not maintained. Unless the positive sentence also has a nonpresuppositional use, we can’t be certain that negative sentences maintain the presuppositions of their positive counterparts. But possibly by “main-verb negation” Atlas means to exclude the sentence level negation interpretation. In any case there is a more pressing reason not to assume that positive and negative existence statements have the same topic-comment structure, and that is that typically the nonexistence of something is more interesting and newsworthy than its existence. Almost all the time in our ordinary daily lives, the things we are talking about are things of whose existence we are confident, and that being the case, the fact that these things exist is almost always not worth noting. So Atlas may be right about positive existence statements typically being about what exists instead of about the referents of their subject NPs, (except with indefinite subjects, e.g. A solution to this problem exists). It may also be true that 99.99% of the tokenings of such sentences occur in philosophical discussions. However this is not true of nonexistence. Thus, given the general correlation between topichood and old information, we might expect existence to be topical in the case of positive existence statements without holding the same expectation for nonexistence in negative existence statements. Atlas, however, insists that both positive and negative existence statements are about existence. Thus he suggests that (47), (47) As for what exists, Pegasus doesn’t. [= Atlas 1989, 115, ex. 59]

102

abbott

is a natural paraphrase of (43). However I find this suggestion hard to accept (and some remarks in Atlas 2004, 353, indicate that I am not alone in this). To my ear (47) has something of the jarring sensation felt with (48): (48) As for Bill, Jane smokes. On the other hand I find (49) (49) As for Pegasus, he doesn’t exist. to be perfectly natural. Of course another factor playing a role in the determination of sentence topics is what the surrounding discourse concerns. And it would seem that, outside the philosophical discussions of existence alluded to above, the question of something’s not existing would typically most naturally arise in the case that one had assumed that that thing did exist, and when that thing was the subject of discussion. Consider a typical child’s disabusements concerning Santa Claus and other nonexistent notables; these seem much more likely to occur during a conversation about the putative entity in question, rather than one about existence. The scenario in (50) strikes me as more likely than the one in (51). (50) A: B: A: B:

Mommy, is Santa Claus a real person? No dear, Santa Claus doesn’t exist. What about the tooth fairy? Sorry dear, the tooth fairy doesn’t exist either.

(51) A: Mommy, what exists? B: Well, as for what exists, Santa Claus doesn’t. And while we’re on the topic of existence, the tooth fairy doesn’t either. Of course (51) may be a little unfair to Atlas, but on the other hand, he would have to admit that (50) is a perfectly natural discourse. This raises the question of the extent to which the problem of negative existence statements has been solved. Even assuming he were right that the unmarked case of a negative existence statement has existence for its topic, Atlas cannot deny that there are natural conversations like (50), where it would seem that a nonexistent entity is topic. It would seem that he would have to say that such conversations are ill-formed, that Mother B in (50) can’t be making the statements she seems to be making because of the lack of a referent for the topical, hence presupposed, NPs. But that is hard to accept.

presuppositions, negation, and existence

103

Even if Atlas’s claims were right about all negative existence statements, would that solve the problem? I don’t think it’s clear that it does. It seems that we would still have the fundamental problem associated with any statements involving proper names of nonexistent entities, and that is the question of what such proper names contribute semantically to the sentences (or statements) in which they occur. If we accept Russellian direct reference, then we have only one level of meaning (in a medium-sized sense of the word “meaning”, i.e. one not including pragmatics) to worry about, but if we also accept Kripke’s (1972) arguments for the nondescriptionality of proper names then we still seem to have a problem on that one level—namely, that proper names of nonexistent entities have nothing to contribute semantically. It seems that that problem remains, and that it continues to be particularly pressing with the negative existence statements that we want to say are true, and outside of any fiction.

6

Concluding Remarks

In this paper I’ve tried to counter arguments for the “sense generality” of negative sentences, and to provide some support for the alternative Gricean view that negative sentences literally express a simple sentence negation and only conversationally implicate the existence of referents for any singular terms involved. I should make it clear that I’m not ready to adopt this Gricean view—I have to confess that I really don’t know what to think about definite descriptions denoting nonexistent entities in negative sentences! In addition, I’ve pointed out what seem to me to be clear examples of negative existence statements where existence itself is not the topic of the statement, and also tried to suggest that serious problems remain for these kinds of statements no matter how the issues of topicality are ultimately resolved. I should note that I have not questioned the assumption that changing the topic-focus structure of a statement affects the presuppositions of its singular terms, but see von Fintel 2004 for arguments that this is not the case. As always, there is more to be said on all of these topics.

References Abbott, Barbara. 2010. Reference. Oxford: Oxford University Press. Atlas, Jay David. 1977. Negation, ambiguity, and presupposition. Linguistics and Philosophy 1, 321–336.

104

abbott

Atlas, Jay David. 1978. On presupposing. Mind 87, 396–411. Atlas, Jay David. 1979. How linguistics matters to philosophy: Presupposition, truth, and meaning. In David Dineen & Choon-Kyu Oh, eds., Syntax and semantics, volume 11: Presupposition. New York: Academic Press, 265–281. Atlas, Jay David. 1988. What are negative existence statements about? Linguistics and Philosophy 11, 373–394 Atlas, Jay David. 1989. Philosophy without ambiguity: A logico-linguistic essay. Oxford: Clarendon Press. Atlas, Jay David. 2004. Descriptions, linguistic topic/comment, and negative existentials. In Marga Reimer & Anne Bezuidenhout, eds., Descriptions and beyond. Oxford: Clarendon Press, 342–360. Atlas, Jay David. 2005. Logic, meaning, and conversation. Oxford: Oxford University Press. Bäuerle, Rainer. 1983. Pragmatisch-semantische Aspekte der NP-Interpretation. In Manfred Faust, Roland Harweb, Werner Lehfeldt, & Götz Wienold, eds., Allgemeine Sprachwissenschaft, Sprachtypologie und Textlinguistik. Festschrift für Peter Hartmann. Tübingen: Gunter Narr Verlag, 121–131. Burton-Roberts, Noel. 1989. The limits to debate: A revised theory of semantic presupposition. Cambridge: Cambridge University Press. Burton-Roberts, Noel. 1991. Review of Philosophy Without Ambiguity: A Logico-Linguistic Essay, by Jay David Atlas. Mind & Language 6, 161–176. Dummett, Michael. 1967. Gottlob Frege. In Paul Edwards, ed., The encyclopedia of philosophy, vol. 3. New York: Macmillan, 225–237. Dummett, Michael. 1981. Frege: Philosophy of Language, 2nd edition. Cambridge, MA: Harvard University Press. von Fintel, Kai. 2004. Would you believe it? The king of France is back! (Presuppositions and truth-value intuitions). In Marga Reimer & Anne Bezuidenhout, eds., Descriptions and beyond. Oxford: Clarendon Press, 315–341. Frege, Gottlob. 1892. Über Sinn und Bedeutung. Zeitschrift für Philosophie und Philosophische Kritik, 25–50. Trans. as ‘On sense and reference’, in Peter Geach & Max Black, eds., Translations from the philosophical writings of Gottlob Frege. Oxford: Blackwell, 56–78. Grice, H. Paul. 1975. Logic and conversation. In Peter Cole & Jerry L. Morgan, eds., Syntax and semantics, vol. 3: Speech acts. New York: Academic Press, 41–58. Reprinted in H. Paul Grice, 1989, 22–40. Grice, H. Paul. 1981. Presupposition and conversational implicature. In Peter Cole, ed., Radical pragmatics. New York: Academic Press, 183–198. Reprinted in H. Paul Grice, 1989, 269–282. Grice, H. Paul. 1989. Studies in the way of words. Cambridge, MA: Harvard University Press.

presuppositions, negation, and existence

105

Gundel, Jeanette K. & Thorstein Fretheim. 2004. Topic and focus. In Laurence R. Horn & Gregory Ward, eds., The handbook of pragmatics. Malden MA: Blackwell Publishing, 175–196. Horn, Laurence R. 1985. Metalinguistic negation and pragmatic ambiguity. Language 61, 121–174. Horn, Laurence R. 1989. A natural history of negation. Chicago: Chicago University Press. Horn, Laurence R. 2006. Speaker and hearer in neo-Gricean pragmatics. Wai Guo Yu/Journal of Foreign Languages 164: 2–26. Horn, Laurence R. 2007. Toward a Fregean pragmatics: Voraussetzung, Nebengedanke, Andeutung. In Istvan Kecskes & Laurence R. Horn, eds., Explorations in pragmatics: Linguistic, cognitive and intercultural aspects. New York: Mouton de Gruyter, 39–69. Horn, Laurence R. 2017. Lie-toe-tease: Double negatives and unexcluded middles. Philosophical Studies 174: 79–103. Horn, Laurence R. To appear. The lexical clone: Pragmatics, prototypes, productivity. To appear in Rita Finkbeiner & Ulrike Freywald, eds., Exact repetition in grammar and discourse. Berlin: de Gruyter. Jackendoff, Ray. 1969. An interpretive theory of negation. Foundations of Language 5, 218–241. Jackendoff, Ray. 1972. Semantic interpretation in generative grammar. Cambridge, MA: MIT Press. Kempson, Ruth M. 1975. Presupposition and the delimitation of semantics. Cambridge: Cambridge University Press. Keshet, Ezra. 2010. Split intensionality: A new scope theory of de re and de dicto. Linguistics and Philosophy 33, 251–283. Kripke, Saul. 1972. Naming and necessity. In Donald Davidson & Gilbert Harman, eds., 1972, Semantics of natural language. Dordrecht: Reidel, 253–355 and 763–769. Reprinted separately, with a preface added, in 1980. Cambridge, MA: Harvard University Press. Mill, John Stuart. 1851. A system of logic, ratiocinative and inductive, being a connected view of the principles of evidence, and the methods of scientific investigation. Chapter II: Of names. 3rd edition. London: John W. Parker. Montague, Richard. 1973. The proper treatment of quantification in ordinary English. In Jaakko Hintikka, Julius Moravcsik & Patrick Suppes, eds., Approaches to natural language: Proceedings of the 1970 Stanford Workshop on Grammar and Semantics. Dordrecht: Reidel, 221–242. Reprinted in Richmond Thomason, ed., 1974, Formal philosophy: Selected papers of Richard Montague. New Haven: Yale University Press, 247–270. Partee, Barbara H. 1972. Opacity, coreference, and pronouns. In Donald Davidson & Gilbert Harman, eds., Semantics of natural language. Dordrecht: Reidel, 415–441.

106

abbott

Quine, W.V. 1956. Quantifiers and propositional attitudes. Journal of Philosophy 53, 177– 187. Richman, Robert J. 1959. Ambiguity and intuition. Mind 68, 87–92. Russell, Bertrand. 1905. On denoting. Mind 14, 479–493. Sadock, Jerrold M. 1978. On testing for conversational implicature. In Peter Cole, ed., Syntax and semantics, volume 9: Pragmatics. New York: Academic Press, 281–297. Strawson, P.F. 1950. On referring. Mind 59, 320–344. Strawson, P.F. 1952. Introduction to logical theory. Oxford: Oxford University Press. Strawson, P.F. 1964. Identifying reference and truth-values. Theoria 30, 96–118. Zwicky, Arnold M. & Jerrold M. Sadock. 1975. Ambiguity tests and how to fail them. In John P. Kimball, ed., Syntax and semantics, volume 4. New York: Academic Press: 1–36.

chapter 4

More Ado about Nothing: On the Typology of Negative Indefinites* Johan van der Auwera and Lauren Van Alsenoy

1

Introduction

This paper aspires to advance the typology of negative quantification, i.e., expressions that contain a negative indefinite such as nothing in sentences without a clausal negator. Section 2 defines negative quantification; it sets it apart from negative concord, i.e., the use of a negative indefinite in sentences that do contain a clausal negator. This section also describes the variety sample that part of the paper is based on. Section 3 discusses the frequency of negative quantification. Sections 4 to 6 concern the distinction between strict and non-strict negative quantification and its relation to the distinction between strict and non-strict negative concord. Section 7 makes the point that negative quantification is more a property of constructions than of languages. Section 8 is the conclusion.

2

Definition and Method

2.1 What are Negative Indefinites? The term ‘negative indefinite’ seems transparently obvious: it should refer to indefinite constructions that are negative. Consider the examples in (1).

* This paper is loosely based on Van Alsenoy (2014: 109–122) and builds on van der Auwera & Van Alsenoy (2016). The work by Van Alsenoy was supported by the Research Foundation Flanders. We are grateful to Dik Bakker (Amsterdam), Mily Crevels (Leiden), Nora England (Austin), Markku Filppula (Joensuu), Ewald Hekking (Querétaro), Marisa Malvestiti (El Bolsón), Edith Pineda-Bernuy (Canberra) and Umarani Pappuswamy (Shillong) for their help with the data. The subtitle of this paper is most similar to the title of Werle (2002). Like us Werle (2002) discusses the earlier typologies of Kahrel (1996) and Haspelmath (1997), but unlike us Werle (2002) does not pay attention to the strictness of negative concord—nor to that of negative quantification—and he does not extend the database beyond what he found in Kahrel (1996) and Haspelmath (1997).

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_005

108

van der auwera and van alsenoy

(1) a. I saw somebody b. I saw nobody There is universal agreement that somebody is an indefinite pronoun. (1b) seems a straightforward negation of (1a), so one would assume that nobody is no less indefinite and that it is also negative, the negation residing in the nopart of nobody. But there are problems. First, one might question the indefiniteness of nobody. Is it truly an indefinite person that the speaker did not see? Or wouldn’t one rather analyze (1b) as meaning that the speaker saw nobody and that the question as to whether the unseen person was definite or indefinite does not arise? This problem is avoided when nobody is called a ‘negative quantifier’. Furthermore, nobody also contrasts with everybody, as in (1c)—another reason for calling nobody a ‘quantifier’. (1) c. I saw everybody Interestingly, there is an alternative to (1b), viz. (1d), featuring anybody and this anybody makes good sense as being truly indefinite, as illustrated with its nonnegative uses in (2) (the negative polarity uses in (2a–b) and the free choice use in (2c)). (1) d. I didn’t see anybody (2) a. If you see anybody, tell him or her that I am ready b. Did anybody call? c. Anybody can tell you that! Thus, if (1b) and (1d) are in some sense equivalent, then nobody too could be considered to be indefinite. The second problem concerns the negativeness of a word like nobody. Whereas the negative character of (1b) is only expressed by nobody, this is not the case in (3). (3) You ain’t seen nobody yet Picking up loose suggestions by Jespersen (1922: 352) and Mathesius (1937: 81–92) linguists now consider (3) to manifest ‘negative concord’, i.e., the phenomenon that two seemingly negative expressions agree in negativity and only express one semantic negation. But the hedge with ‘seemingly’ is important: linguists do not agree as to whether the occurrence of nobody in (3) is only

more ado about nothing

109

‘seemingly’ negative instead of ‘truly’ negative. This is an important issue, connected with the larger issue of determining the nature of negative polarity and free choice (cf. Giannakidou 2006, de Swart 2010 and Horn 2011 for a general appreciation of the issues and van der Auwera & Van Alsenoy 2016 for our point of view). This paper focuses on the use of words and constructions like nobody in (1b). For our purposes nothing much depends on the terminological choice between ‘negative indefinite’ and ‘negative quantifier’, but we will opt for the terms ‘negative quantifier’ and ‘negative quantification’. The latter abbreviates conveniently as ‘NQ’, similar to ‘NC’, the common abbreviation of ‘negative concord’. The two phenomena are related, we consider both the NQ and the NC uses of nobody to manifest negative indefiniteness and in section 6 we offer an integrated account of both. For this reason also it is the term ‘negative indefinite’ that figures in the title of this paper. The integrated account is not only useful, but necessary. This has to do with the fact that NC comes in various types. One type, so-called ‘non-strict NC’, is illustrated in (4). (4) Spanish (Haspelmath 1997: 201) a. Nadie vino nobody came ‘Nobody came’ b. No vi a nadie neg saw to nobody ‘I didn’t see anybody’ The NC is called ‘non-strict’ because the appearance of the concord pattern is dependent on something else, viz. word order. In (4b) the negative indefinite is postverbal and we get concord. In (4a), however, the negative indefinite is preverbal, concord is impossible, and we in effect get the ‘NQ’ strategy. What we see in Spanish (4a) is therefore just as much an illustration of non-strict NC as of what could be called ‘non-strict NQ’. The latter is indeed non-strict, because in (4b) NQ is impossible, just as NC is impossible in (4a). 2.2 Method The data used in this study is the research literature that relates to negative indefinites, complemented with personal information from linguists and/or native speakers. When we make frequency claims, these will be based on a version of Miestamo’s ‘Restricted Sample’ of 179 languages (Miestamo 2005: 36). This sample is a variety sample, i.e., a sample that attempts to reflect maxi-

110

van der auwera and van alsenoy

mal linguistic diversity. Miestamo first randomly chose one language for every genus (see Dryer 2000), he ordered the genera according to ‘macro-areas’ (see, again, Dryer 2000), he noted that the macro-areas of Australia and New Guinea only have a coverage of 43.2%, and he then randomly suppressed languages for the better described areas so that they too are represented in the same proportion (for full details see Miestamo 2005: 27–39). Note that a variety sample does not take account of the number of languages in any one genus. Thus a language of a single language genus has the same weight as a language of a genus with, say, 20 languages, which could be exactly like the one language chosen with respect to the issue under investigation. When we claim that a phenomenon is rare or frequent, this claim is relative to the language internal variability of language. The numbers cannot thus directly reflect the actual frequencies, but, as is implicit in Miestamo (2005), we take them to be reasonable approximations. To the extent that it was possible we used the very same 179 languages that constituted Miestamo’s ‘Restricted Sample’. However, in roughly one-third of the languages, the descriptions of Miestamo’s languages simply did not supply enough details about negative indefinites. In those cases we replaced the language in question with another language from the same genus from the same macro-area or, when that was not possible, a language from another genus not covered yet, but also from the same macro-area. The resulting sample is described in Appendix 1. As will be clear in what follows, though, the information provided for many languages is still suboptimal, in part because grammarians have not been aware of the parameters of variation in the area of NQ (much less than for NC), and in part because for most languages we rely on just one grammar, and this is typically just based on one dialect or even just the idiolects of a few speakers. A related restriction is that we will only discuss main declarative clauses, so not questions or conditionals, for instance, and also not elliptical declaratives, such as the short answer in (5), even though they are relevant in distinguishing negation from negative polarity. (5) —Who did you see today? —Nobody We also exclude constructions with either more than one negative indefinite or more than one clausal negator. (6) African American Vernacular English (Howe 2005: 189) Nobody don’t want to have nothing to do with nobody

111

more ado about nothing table 4.1

Negative indefinites in Kahrel (1996: 37)

clausal negator & neutral indefinite clausal negator & negatively polar indefinite no clausal negator & negative indefinite clausal negator & negative indefinite other

John did not buy something John did not buy anything John bought nothing John did not buy nothing There is no what John bought

67.7% 22.5% 12.5% 12.5% 17.5%

(7) East Flemish (van der Auwera et al 2006: 308) Ik en heb niemand niet gezien I neg have nobody neg gezien ‘I haven’t seen anyone’

3

Frequency

The frequency of the NQ strategy has been investigated before, viz. by Kahrel (1996). He used a world-wide sample of 40 languages, also a variety sample like the one we will use, though arrived at by a different method (Kahrel 1996: 96– 106). Kahrel (1996: 37) found the NQ strategy in 5 languages (or 12.5 %), viz. in Chukchi, Dutch, Evenki, Manŋarayi and Nama. There are four other strategies. In Table 4.1 the ‘NQ’ strategy is marked in bold. Note that the frequencies add up to more than 100%: the reason is that languages can have more than one strategy. The five strategies are illustrated with (pseudo-)English. The details of the four other strategies will not be dwelt on in this paper. Still, a few remarks are in order. First, a neutral indefinite is one that occurs in all contexts, whether positive, negative or negatively polar. To that extent the pseudo-English something in John didn’t buy something is misleading, for in real English something is a positive polarity item. Second, the neutral indefinite need not be a pronoun. It can also be a generic noun having a meaning like ‘person’ or ‘thing’. This is problematic, for one would suppose that every language has this strategy. However, we list it only if languages have no other strategy. Third, Kahrel (1996: 43–45) calls anything a ‘special indefinite’, but in line with much current research, we relabel it as ‘negatively polar’. Fourth, the strategy with both a clausal negator and a negative indefinite is the negative concord strategy. Finally, the last strategy is here just called ‘other’. Kahrel (1996: 52–55) calls it ‘existential’. We do indeed illustrate the ‘other’ construction with an existential, but we now know (cf. Van Alsenoy 2014: 197–251) that there are still other strategies and therefore the term ‘other’ is more appropriate.

112

van der auwera and van alsenoy

Though Kahrel’s sample is small, it is clear that the ‘NQ’ strategy is relatively rare. This has been confirmed by Haspelmath (1997: 202), but not in a straightforward way. The first problem is that Haspelmath (1997) uses a very wide sense of ‘negative indefinite’, viz. the “deliberately vague sense of ‘indefinite […] that has ‘direct negation’ as an important use’” (Haspelmath 1997: 199). In this way, even the English negatively polar anybody as in (1d), repeated below, counts as a negative indefinite. (1) d. I didn’t see anybody Of course, anybody cannot by itself make the clause negative. Still, in principle, if even with Haspelmath’s wide definition of ‘negative indefinite’ the NQ strategy is rare, then it should be at least as rare with the common and narrower definition. The second concern is that Haspelmath’s own sample, which supports the claim about the rarity of the NQ strategy, is as small as Kahrel’s and furthermore highly biased towards European languages, with 23 European languages of the total of 40. It is in these 23 languages that the NQ strategy occurs in as many as 10 of them, and it never occurs in the remaining nonEuropean languages (Haspelmath 1997: 202). This suggests (Haspelmath 1997: 202) that the NQ strategy is an areal, European phenomenon. In later work, where Haspelmath (2005: 466–469; 2013) uses the same wide definition but a broader sample (206 languages), he again concludes that the NQ strategy is rare and makes a new areal claim to the effect that NQ is typical for Western Europe and Mesoamerica.1 Most importantly, even though Haspelmath’s definition of ‘negative indefinite’ is wide, he offers an explanation for why it would be rare (Haspelmath 1997: 203). Semantically, clausal negation should be marked at the level of the clause, which more or less means on the verb or related to the verb, but definitely not on a participant.2 From this point of view, (1d) is fine, but (1b) is not. (1b) constitutes a form/meaning mismatch, tolerated indeed, but only by a modest number of languages. We find this explanation to be highly plausible, but we would want to be more certain that the NQ strategy is indeed a rarity. What we now do is to combine Kahrel (1996) and Haspelmath (2005/2013). Like Haspelmath (2005/2013) we use a wider sample and we can thus look for an areal effect, but like Kahrel (1996) we use the more common and more 1 Note that Kahrel (1996) does not make an areal claim, which is justified, since 40 languages for the whole world is too small for this kind of claim. 2 A similar point had been made before (e.g. Heidolph 1970: 99, Kefer 1974: 67), though not, as far as we know, in a typological context.

113

more ado about nothing table 4.2

Negative indefiniteness in this study

clausal negator & neutral indefinite clausal negator & negatively polar indefinite no clausal negator & negative indefinite clausal negator & negative indefinite other

49.7 % 47.5 % 11.7 % 19% 13.8 %

restrictive definition of ‘negative indefinite’. Using the very same categories that Kahrel (1996) uses gives us Table 4.2. When one compares Kahrel’s Table 4.1 and our own Table 4.2 one notices differences and similarities. In the large sample the combination of a clausal negator and a neutral indefinite type is still the most frequent strategy, but it is now closely followed by the negative polarity type. In the large sample the NC type is more frequent than in the small sample. But, most importantly for our purposes, the percentages for the NQ type are very similar. So we too confirm that the NQ type is rather rare. At least with respect to the four major strategies, the figures also support Haspelmath’s explanation of the rarity of the NQ strategy. In the two top strategies, clausal negation is not marked on any participant. The fourth strategy, negative concord, does mark clausal negation on a participant, which is ‘bad’, but what is ‘good’ is that there is also marking on the verb. One could thus say that the NC strategy is not as ‘bad’ as the NQ strategy, which is supported by the percentages in Table 4.2. Table 4.3 lists the languages that we claim to have the NQ strategy and it also shows how they distribute across six macro-areas. The list starts with the macro-areas with the highest number of languages showing NQ. As in Miestamo (2005), our sample includes one Creole, but since it is does not have the NQ strategy it is not included in Table 4.3 (for a study focussing on creoles see van der Auwera 2015). Table 4.3 does not confirm Haspelmath’s hypothesis that the NQ is typical for Western Europe. Of course, our sample hardly has any ‘Western European’ languages, for next to Icelandic (with NQ) it only has Albanian, Basque and Finnish (without NQ), none of them particularly ‘Western European’. The languages that Haspelmath (1997: 202) mentions are Germanic, Romance and Maltese and this pleads for a reformulation in terms of the more focused ‘Standard Average European’ notion (van der Auwera 2011). The point about Mesoamerica is fully confirmed though. Mesoamerica, which is included under North America, has 10 of the 12 North American languages.

114 table 4.3

van der auwera and van alsenoy The NQ strategy in our sample

Language

Family

Macro-area

n/m

%

Yuchi

Yuchi

North America

12/36 33.3%

Chinook

Chinook

Upper Nexaca Totonac

Totonacan

South America

4/39

10.2%

1/15

6.7%

Otomí, Chalcatongo Mixtec, Oto-Manguean Chocho, Tepetotutla Chinantec Purépecha

Tarascan

Mam

Mayan

Nevome, Nahuatl, Huichol

Uto-Aztecan

Epena Pedee

Choco

Paumarí

Arauan

Warao

Warao

Mapuche

Araucanian

Icelandic

Indo-European Eurasia

Tiwi, Kayardild

Australian

Australia & Papua 2/38 New Guinea

5.3%

Nicobarese

Austro-Asiatic

South-East Asia & 1/21 Oceania

4.8%

Egyptian Arabic

Afro-Asiatic

Africa

3.4%

n = the number of languages that have NQ m = the number of languages in the sample

1/19

more ado about nothing

4

115

Strict and Non-strict NQ

For NC it has become customary, since Giannakidou (1998: 186), to distinguish between strict and non-strict types. A non-strict type has already been illustrated with Spanish (4), repeated below. (4) Spanish (Haspelmath 1997: 201) a. Nadie vino nobody came ‘Nobody came’ b. No vi a nadie neg saw to nobody ‘I didn’t see anybody’ What determines negative concord is word order, more particularly, what has been called the ‘Negative First’ principle (‘Neg First’) (Horn 1989: 293), credited to Jespersen (1917: 5). The phrasing below is Jespersen’s: There is a natural tendency, also for the sake of clearness, to place the negative first, or at any rate as soon as possible, very often immediately before the particular word to be negatived (generally the verb). In (4a), the preverbal negative indefinite makes for an early negation ‘or at any rate’ (to use Jespersen’s hedge) to satisfy the Neg First principle there is no need to add a clausal negator, which will have to come before the verb but after the negative indefinite. In (4b), however, the negative indefinite does not come early, but the clausal negator does. Polish, however, has strict NC, which means that, for the relevant construction, negative concord is obligatory and thus independent of word order. This is illustrated in (8). (8) Polish (Haspelmath 1997: 201) a. Nikt nie przyszedł nobody neg came ‘Nobody came’ b. Nie widziałam nikogo neg saw nobody ‘I saw nobody’

116

van der auwera and van alsenoy

In the study of NQ constructions no distinction between strict and non-strict types has been made, in part, perhaps, because NQ has attracted far less attention than NC. We now propose that a similar distinction has to be made for NQ, and what is more, the crucial factor is again the Neg First principle and whether or not ‘something’ is preverbal or postverbal. There is a difference too. For NC this ‘something’ is the negative indefinite and what makes the difference between strict and non-strict NC is the presence or absence of a verbal negator. For NQ the ‘something’ is the indefinite, whether negative or positive, and what makes the difference between strict and non-strict is precisely whether or not the indefinite is negative. Consider (1b) and (1d) again, repeated below. (1) b. I saw nobody d. I didn’t see anybody In both (1b) and (1d) the indefinites are postverbal. The postverbal nobody is negative, thus the negation is postverbal, which is not particularly early, but it is no worse than that of the verbal negator n’t in (1d). In (9a), however, nobody is preverbal, and it cannot alternate with the combination of anybody and a verbal negator. (9) a. Nobody has seen me b. *Anybody hasn’t seen me Here the two versions do differ with respect to the earliness maxim, and only the version with the preverbal negation in nobody is allowed. The facts of English can thus be seen as non-strict. Nobody is restricted in use and, just like for non-strict NC, the determining factor is the position of the indefinite relative to the verb. When the indefinite follows the verb, NQ is not obligatory: nobody has competition from anybody.3 It will be useful for what follows to have a ‘formula’ for the various types of NQ that we will distinguish. The English type is symbolized in (10), with ‘I’ for ‘indefinite’, ‘N’ for ‘negative’, ‘V’ for ‘finite verb’, and ‘>’ for simple linear precedence.4 The left-hand box represents the preverbal domain and the right-hand box the postverbal one. 3 Note there is no claim that two (or more) constructions presented as alternatives are fully equivalent. There can be pragmatic or sociolinguistic differences. We leave those out of account. 4 These formulas resemble the ones used in Haspelmath (1997: 201), when he compares ‘NV-NI’, ‘V-NI’ and ‘(N)V-NI’ types of co-occurrence between a negative indefinite and verbal negation.

more ado about nothing

(10)

NI > V

117

NV > I or V > NI

Next to non-strict NQ there is also a strict version. Standard Dutch will serve to illustrate this: it has a counterpart to nobody, but not to anybody. (11) Dutch a. Ik heb niemand gezien I have nobody seen ‘I have seen nobody’ b. Niemand heeft mij gezien nobody has me seen ‘Nobody has seen me’ (12) represents the strict type.5 (12)

NI > V

V > NI

Our sample of 21 languages has clear cases of both strict and non-strict NQ. For some languages it is unclear. When a grammarian claims that some or other indefinite is inherently negative and illustrates this with examples—and sometimes just one example—without a verbal negator, (s)he seldom explicitly says that the meaning cannot be expressed with any other means. For example, in Paumarí ni-nahina in (13) ni is a clause-initial or verb-initial negator (Chapman & Derbyshire 1986: 220) and nahina functions as ‘somebody/something’ as well as ‘who/what’—we just gloss it as ‘what’ in (13) (Chapman & Derbyshire 1986: 205). (13) Paumarí (Chapman & Derbyshire 1986: 272) ni-nahina-ra o-nofi-ki neg-what-obj 1sg-want-nontheme ‘I don’t want anything’

But there are two differences. First, our conception of what counts as a negative indefinite is different (see section 3). Second, we offer two formulas, one for each domain, i.e., that of the preverbal indefinite and that of the postverbal indefinite. 5 Note that the indefinite precedes the lexical verb in (11a) and (11b), but what matters here is the position of the indefinite relative to the finite verb, which is an auxiliary here. See also the end of section 6.

118

van der auwera and van alsenoy

table 4.4

Strict and non-strict NQ

Strict Icelandic, Tiwi Non-strict Egyptian Arabic, Nicobarese, Mapuche, Upper Nexaca Totonac, Chalcatongo Mixtec, Chocho, Tepetotutla Chinantec, Mam, Nevome, Nahuatl, Huichol Both Otomí Unclear Epena Pedee, Paumarí, Warao, Yuchi, Chinook, Purépecha, Kayardild

2 11

1 7

We don’t know though whether the pattern illustrated in (13) is the only one. So while it is clear that Paumarí has the NQ strategy, it is not clear whether it is strict or non-strict. Table 4.4 is a classification of the sample languages in terms of strictness, with Paumarí in the row with the unclear cases. This classification is tentative, since even for the languages which we dared to classify as either strict or non-strict (or both), our judgment is often based on less than optimal information. It is probably too early to judge, but if we can trust the clear cases, then non-strict NQ is more frequent than strict NQ. The reason could be that strict NQ always shows the form/meaning mismatch and nonstrict NQ does not. Note that it might well be the case that the two zones of concentration, Western Europe and Mesoamerica, have different preferences, with Icelandic—but of course also Dutch and German—going for strict NQ, but Mesoamerica for non-strict NQ. A clear case of strict NQ is Icelandic. Negatives like enginn ‘nobody’ and ekkert ‘nothing’ behave just like Dutch niemand. (14) Icelandic (Neijmann 2001: 130) a. Enginn er eins nobody is alike ‘Nobody is alike’ b. Ég heyri ekkert I hear nothing ‘I hear nothing’ But Icelandic is nevertheless a little special, for it has two sets of negative indefinites. Next to the NQ set, there is also an NC set, illustrated in (15).6 6 The NC is non-strict, but different from the better known non-strict type, illustrated with Spanish (4). See section 5 and especially van der Auwera & Van Alsenoy (2016).

more ado about nothing

119

(15) Icelandic (Nejmann 2001: 130, Haspelmath 1997: 214) a. Ég vissi ekki sjá neitt I did neg see nothing ‘I didn’t see anything’ b. *Neinn sag (ekki) mig nobody saw neg me ‘Nobody saw me’ We have also listed Tiwi and Otomí and these are also special. Tiwi is an Australian language and (16) is an example with a preverbal negative indefinite. (16) Tiwi (Osborne 1974: 69) akarə-kuwani jinimatakupauli neg-who he.come.back ‘No one will come back’ Tiwi is SVO and Osborne (1974: 69) states about the postverbal domain that V > NI is possible, adding however that ‘[t]he Tiwi usually avoid negative pronouns as object’. For instance, they would not say ‘I saw no one’ but ‘I looked. No one!’ And, interestingly, in the one example in the grammar with a postverbal negative indefinite in the English translation, the Tiwi sentence does not have a verb. (17) Tiwi (Osborne 1974: 57) ŋia karəkamini I nothing ‘I have got nothing’ The other language is the SVO Otomanguean language Otomí, which is listed as manifesting both strict and non-strict NQ. The word for ‘nobody’ is ni’na and it consists of the ni ‘NEG’ (probably borrowed from Spanish; Hekking 1995: 65) and na ‘one’ This element is claimed to be possible both preverbally and postverbally (Priego Monfort 1989: 66; Hekking 1995: 65) and thus shows strict NQ.7 The fully native word hinte for ‘nothing’, however, composed of the verbal

7 The fact that Spanish negative indefinites do not carry restrictions as to their position with regard to the verb might explain why ni’na might occur postverbally. However, in a corpus put together by Hekking ni’na too occurs only preverbally (Hekking p.c.).

120

van der auwera and van alsenoy

negator of hin and the interrogative te ‘who’, does not exhibit strict NQ, for it can only occur clause-initially (Priego 1989: 66, Hekking 1995: 65). (18) Otomí (Priego Montfort 1989: 65, Hekking 1995: 33, 65) Hinte bí dini nothing pst.3 found ‘He found nothing’ The hinte pattern is non-strict NQ, but different, of course, from the one shown in English. We symbolize the Otomí non-strict NQ of hinte in (19). We use the asterisk in the box below to mean that there is no negative clause with a postverbal indefinite, either with a negative indefinite or with a positive one. (19)

NI > V

*

With Otomí we have already illustrated a sample language that has non-strict NQ. Here is one more illustration. Example (20a) shows that in the Uto-Aztecan SVO language Nahuatl the negative quantifier amaka ‘nobody’ occurs preverbally. This quantifier amaka consists of a prefixed variety of the clausal negator amo and the indefinite-interrogative aka ‘who, somebody’. Amaka cannot occur postverbally. In postverbal position, illustrated in (20b), the indefinite has to be positive, but then we get the verbal negator amo in front of the verb. (20) Nahuatl (MacSwan 1999: 121, 123) a. Juan amaka okitak Juan nobody pst-3.s-3o.sg-see ‘Juan saw nobody’ b. Juan amo okitak aka Juan neg pst-3s-3o.sg-see somebody ‘Juan didn’t see anybody’ This is non-strict NQ, but a type that is different from both Otomí and English. (21)

NI > V

NV > I

We have thus already seen three types of non-strict NQ. In the next section we try to sketch the full typology.

121

more ado about nothing

5

Types of Non-strict NQ

The first parameter of variation in the typology of non-strict NQ is whether indefinites, positive or negative, occur in both postverbal and preverbal domains. In English and Nahuatl, they do, but in Otomí the negative indefinite hinte does not. table 4.5

Non-strict NQ types

Indefinites allowed in both preverbal and postverbal domains + English

NI > V

V > NI or NV > I

Nahuatl

NI > V

NV > I

Otomí hinte

NI > V

*

We first turn to the constructions that allow indefinites in both the preverbal and the postverbal domains. 5.1

Non-strict NQ with Indefinites Allowed in Both Preverbal and Postverbal Domains In Table 4.5 we symbolized the English and Nahuatl types. In Table 4.6 we do it again, this time together with strict NQ. We use shading and lines to show how the three strategies differ. Thus Dutch and English differ from one another in one respect: in the postverbal domain English has an alternative to NQ, viz. NV > I, and it is the presence of this pattern in English and its absence in Dutch that is shaded. Similarly, Nahuatl differs from English in just one respect, viz. the absence vs. the presence of a V > NI pattern. This is the pattern that is shaded. Meanwhile, it is also clear that Nahuatl and Dutch differ in two respects. In the three NQ types of Table 4.6 the differences concern the postverbal indefinite. One can imagine a non-strict NQ type that differs from strict NQ with respect to the preverbal indefinite. Thus there is the hypothetical non-strict NQ type marked with the dark-lined box in Table 4.7. This hypothetical type is the mirror image of English style non-strict NQ. Instead of a choice for the postverbal indefinite, there would be a choice for the preverbal indefinite. We illustrate this type with pseudo-English. (22) a. Nothing surprises me b. Anything doesn’t surprise me

122

van der auwera and van alsenoy

table 4.6

Non-strict NQ types

table 4.7

Another non-strict NQ type?

c. I surprise nobody d. *I don’t surprise anybody This type is not attested and there may be a good reason. The strict NQ type flouts the Neg First principle in the postverbal domain; what we see in Englishstyle non-strict NQ is that the V > NI that flouts the Neg First has a NV > I alternative, which is in conformity with the Neg First principle. English-style non-strict NQ is thus better than strict NQ, at least from the point of view of the Neg First principle. The unattested non-strict NQ type, however, is worse. It keeps the ‘bad’ V > NI construction, and it gives the ‘good’ NI > V construction a ‘bad’ I > NV alternative, that is, ‘bad’ from the point of view of the Neg First principle. We conjecture that ceteris paribus the unattested type does not exist. Table 4.8 introduces another hypothetical type.

more ado about nothing table 4.8

123

Another non-strict NQ?

The newly entered type differs from English non-strict NQ in one respect, just like Nahuatl non-strict NQ. With Nahuatl the choice in the postverbal domain was eliminated, now we get a choice in the preverbal domain. In other words, we need a language with a construction in which one always has a choice, independently of the position of the indefinite vis-à-vis the verb, between marking the negation on the indefinite or on the verb—though not on both at the same time, for that would constitute NC. Hiberno-English, at least for some speakers, and probably due to an Irish substratum, is such a language (see Filppula 1991: 179–183). What makes Hiberno-English special, i.e., different from most other kinds of English, is the occurrence of anybody in front of a negative verb. Since anybody can also occur after a negative verb and since nobody is also an option, both before and behind a positive verb, all four strategies are acceptable. (23) Hiberno-English (20b based on Filppula 1999: 180, Filppula p.c.) a. Nobody will know where we went b. Anybody won’t know where we went c. I know nobody here d. I don’t know anybody here In our sample Mapuche seems to be such a language too. In (24) we give the examples listed in the grammar by Smeets (1989). (24) Mapuche (Smeets 1989: 244) a. Doy chem rumé nie-ke-la-fu-y-iñ more what ever have-cf-neg-ipd-ind-1ns-pl ‘We did not have anything more’

124

van der auwera and van alsenoy

b. Doy chem nu rumé nie-ke-fu-y-iñ more what neg ever have-cf-ipd-ind-1ns-pl ‘We had nothing more’ In (24a) the verb is negative, and in (24b) the indefinite is negative. Both examples have the indefinite in the preverbal domain. The grammar doesn’t have examples for the postverbal domain, but the word order in Mapuche is rather free (Smeets 1989: 20, 347–354), so it is possible that the choice is available independently of the position of the indefinite relative to the verb and this hunch is confirmed. (24) Mapuche (Malvestiti p.c.) c. Doy nie-ke-la-fu-y-iñ chem rumé d. Doy nie-ke-fu-y-iñ chem nu rumé Table 4.9 shows us yet another hypothetical type. Different from Nahuatl, this type would allow I in the preverbal domain, and different from Hiberno-English, only the preverbal domain would have two options. The type is illustrated with pseudo-English in (25). table 4.9

(25) a. b. c. d.

Another non-strict NQ?

Nothing surprises me Anything doesn’t surprise me *I surprise nobody I don’t surprise anybody

more ado about nothing

125

This type is not attested either, just like the one shown on the left-hand side of Table 4.7, i.e., the type with NI > V or I > NV in the preverbal domain and V > NI in the postverbal domain. But note that the latter type fares worse than the one introduced in Table 4.9, relative at least to the Neg First principle. Of all the types discussed, attested and not attested, the pattern introduced in Table 4.7, i.e., NI > V or I > NV preverbally and V > NI postverbally, is the only one that has 2 out of 3 strategies flouting the Neg First maxim. The preverbal NI > V or I > NV and postverbal NV > I type of Table 4.9 only flouts it in 1 out of 3 strategies, which is the same ‘score’ as English. Furthermore, its existence makes good sense from a diachronic point. To see this, we first have to remember that the world’s preferred strategies use negative verbs and no negative indefinites (Tables 4.1 and 4.2). Table 4.10 adds these at the top. table 4.10 From strict NQ to the top strategies

It is the neutral or negatively polar I construction, which figures in both the preverbal and postverbal majority strategies, that the NI must derive its origin from, due to the attraction of the verbal negator to the indefinite or to the progressive specialization of negative polar contexts to negative one (Haspelmath 1997: 203–210). These processes are more likely to start with the indefinite in the preverbal domain than in the postverbal domain, because the preverbal

126

van der auwera and van alsenoy

domain is the one that the verbal negative is attracted to, given the Neg First maxim. This gives us the type with preverbal NI > V or I > NV and postverbal NV > I. We thus predict that it will be attested or, more carefully, the chance that we will find it is greater than for the unattested type of Table 4.7. 5.2

Non-strict NQ with Indefinites Prohibited in Either the Preverbal or the Postverbal Domain We have seen that in Otomí a sentence about ‘nothing’ has to use the word hinte and that this has to occur preverbally. There is no alternative strategy with an indefinite, whether positive or negative, following the verb, again whether positive or negative. We symbolized this strategy in (19), repeated below. (19)

NI > V

*

In (19) there is only one preverbal pattern, but we have seen in the preceding section that languages can have a choice in the preverbal domain: the negative can go with the verb or with the indefinite. This is the case in Hiberno-English. The pattern symbolized in (26) is like Hiberno-English preverbally and like Otomí hinte postverbally. (26)

NI > V or I > NV

*

This pattern is attested in the Uto-Aztecan language Nevome. In (27a) we see the verbal negator pima, which appears as part of a negative indefinite in (27b). (27c) shows that the negative indefinite is not obligatory, for pima does not negate the indefinite there.8 No indefinite, however, whether positive or negative, but translationally equivalent with nobody and the like, would appear to be possible after the verb. (27) Nevome (Shaul 1986: 87, 87, 88) a. ti-stuodiga dios s-cocoma pima sushima igui our-lord god stat-sick neg undergo irr ‘Our lord God does not suffer sickness’ b. pim’urhoi si-n’-ohada, an’-upu pim-hucudoi s’-ohoda nobody stat-me-envy 1sg-also neg-someone stat-envy ‘Nobody envies me, just as I envy no one’ 8 Note that pima, though arguably a clausal negator, is still clause-initial and not adjacent to the verb. This is another manifestation of the Neg First maxim.

more ado about nothing

127

c. pim’-an’-igui haitu nuhida neg-1sg-irr something look ‘I saw nothing’ Nevome is an SOV language. This is not the same as saying that it is (strictly) verb-final. In fact, (27a) shows that Nevome is not strictly verb-final. Still, presumably the major constituents will occur preverbally. Since the verbal negator will tend to occur preverbally too—because of the Neg First principle—there could be occasions for the verbal negator to be adjacent to an indefinite with the negator in front of the indefinite. This could lead to univerbation and the creation of a negative indefinite. This probably happened in Nevome. Haspelmath (1997: 207–208) notices this and he goes as far as to claim that this Neg First driven univerbation is expected ‘to occur with all indefinites in verb-final languages’ (Haspelmath 1997: 206). He supports this claim somewhat indirectly: he does not really know whether the languages that he describes are really verb-final, but at least they are SOV (and they are all, just like Nevome, Uto-Aztecan, viz. Tümpisa Shoshone, Hopi and Yaqui; Haspelmath 1997: 207– 208). To the extent that one can judge, these three languages indeed have nonstrict NQ constructions, more particularly of the hinte type. It is not true, however, that all SOV languages will exhibit non-strict NQ. Consider the East Bougainville SOV language Nasioi. There is no evidence that the verbal negative can attach to the indefinite. (28) illustrates this (with an example that has already figured in the typological literature—Kahrel 1996: 36). (28) Nasioi (Rausch 1912: 134) a. Nanin nánu-i someone go-rpst ‘Someone went’ b. Nanin nánu-aru-i someone go-neg-rpst ‘Noone went’ Nasioi simply takes the neutral indefinite, which, it will be remembered, is found in half of our sample languages, 89 to be precise. Of these 89 languages at least 36 have SOV as their dominant order, judging by Dryer (2013), and they show no sign of a NI pattern.9

9 These languages are Ijo, Koyraboro Senni, Iraqw, Supyire, Maba, Nubian, Beja for Africa,

128

van der auwera and van alsenoy

The impossibility of postverbal indefinites in negative sentences is not restricted to SOV. It is found in VSO languages too. Upper Necaxa Totonac (Beck 2004: 92, 95) and Mam illustrate this situation. In Mam, at least in the Central Mam variety that Collins (1994) reports on, negative indefinites seem to combine the indefinite with a negative marker mix, which is not used in isolation (Collins 1994: 378). (29) Mam (Collins 1994: 378) —Ti’n tzaj t-q’o’n te’y? what dir.come 3sg.erg-give you? ‘What did he give you? —Mix ti’ neg thing ‘Nothing’ The example is not perfect, because it is elliptical, and Collins does not actually claim that the negative indefinite is obligatorily sentence-initial, but it probably is, at least if this variety patterns with another variety of Mam described by England (1983: 244): Negation is accomplished through the use of negative particles which are first in the sentence and followed immediately by the phrase or clause being negated. This automatically focusses negated nominals.10 Similar facts obtain in the VOS languages Chalcatongo Mixtec (Macaulay 1996: 119–125) and Chocho (Veerman-Leichsenring 2000: 32, 73) and probably also

10

Evenki and Hindi for Eurasia, Jingulu, Nasioi, Skou, Nabak, Kobon, Menya, Sentani, Kunimaipa, Daga, Amele, Usan, Arapesh for Australia and Papua New Guinea, Maricopa, Washo, Seri, Koasati, Takelma, Wappo, Slave, Haida, Huichol, Central Pomo for North America, and Araona, Pirahã, Rama, Sikuani, Hup, Tuyuca for South America. Observe that a language can be SOV and allow a verbal negator with neutral indefinites but also have negative indefinites, the simple reason being that a language may have more than two strategies. In the set of 89 languages there are two such languages, viz. Imbabura Quechua and Brahui. There are not mentioned in this paper, though, for their negative indefinites are NC indefinites. England does not illustrate this claim with any sentence containing a negative indefinite, but she does report on the existence of a negative indefinite mijuun, in which mi is a negator and juun must be related to juuɁn glossed as ‘each one’ (England 1983: 248–249).

more ado about nothing

129

Tepetotutla Chinantec (Westley 1991: 17, 30). Interestingly, in Chocho the negative indefinites are bimorphemic and the first part is a negative existential ‘is not’ (Veerman-Leichsenring 2000: 32); so in this respect Choco is still, at least etymologically, verb-initial. Note that the fact that both VOS and VSO have preverbal negatives and do not seem to allow postverbal ones, whether negative or positive (but in the latter case with a clausal negator) belies Haspelmath’s additional expectation that the Neg First driven creation of negative indefinites could ‘not at all [occur’] in verb-initial languages’ (Haspelmath 1997: 206). It does occur. What might not occur, or at least we have no evidence for it, is what is symbolized in (30), i.e., a pattern forbidding preverbal indefinites and forcing the indefinites to be in postverbal slots. (30) a.

*

V > NI

b.

*

V > NI or NV > I

The reason for why at least the pattern of (30a) might not exist must be the Neg First principle.

6

NQ and NC

We have argued that NQ has strict and non-strict types just like NC. In van der Auwera & Van Alsenoy (2016) we show that non-strict NC comes in different types, and the previous section shows that this is also the case for non-strict NQ. It does not follow, of course, that there is any interesting similarity between the non-strict NC and NQ subtypes. However, at least with respect to the types of non-strict NQ and NC that allow both preverbal and postverbal indefinites, the parallelism is very strong. In this section we will review the typology of nonstrict NQ of Table 4.10 and we will show that each type is paired with a type of non-strict NC. Let us start from the strict NQ system exemplified by standard Dutch (11). There are two attested systems that differ in only one respect, each introducing an alternative strategy in the postverbal domain, and in both the verb has its negator. In the strict NQ system of English, the indefinite is no longer negative, but in the non-strict NC system of Brabantic Belgian Dutch, illustrated in (31), or in Brazilian Portuguese (Agostini & Schwenter 2015), the indefinite remains negative.

130

van der auwera and van alsenoy

table 4.11 Non-strict NC and NQ types, very close to strict NQ

(31) Brabantic Belgian Dutch11 a. Niemand heeft mij (*nie) gezien noboby has me neg seen ‘Nobody has seen me’ b. Ik heb niemand (nie) gezien I have nobody neg seen ‘I have seen nobody’ Note that, as already adumbrated at the end of section 2.1, one could consider the Brabantic non-strict NC constellation to exemplify non-strict NQ as well. Considering it ‘non-strict NC’ highlights the fact that NI and NV do not always go together, whereas considering it ‘non-strict NQ’ would highlight that NI and V do not always go together either. In fact, all of the non-strict NC constellations, this one and the ones that follow, can be regarded from a NQ perspective. We will follow tradition, however, and take the ‘NC’ perspective. Starting from both non-strict systems, i.e., the non-strict NQ system of English and the non-strict NC system of Brabantic Belgian Dutch, there are two 11

Note that the negator that negates the finite verb follows it. What matters in this context is whether or not the negative indefinite precedes or follows the negative finite verb, independently of how the latter is turned negative. But the position of the verbal negator does constitute a further parameter of variation, explored with relevance to NC in van der Auwera & Van Alsenoy (2016). Yet another parameter of variation is whether the negative indefinites of the two postverbal patterns are the same. In Brabantic Belgian Dutch and Brazilian Portuguese they are, but not in Icelandic (see (14) and (15)).

more ado about nothing

131

further systems, each again differing in one respect, that of adding an alternative, but now, the differences can concern both the preverbal and the postverbal domain. In the preverbal domain change, the alternative gives the verb its negator and then either the indefinite loses its negator (NQ, as in HibernoEnglish) or not (NC, as in French, illustrated in (32)). table 4.12 More non-strict NC and NQ types

(32) French a. Jamais il (n’)a vu une chose pareille never he neg.have seen a thing similar ‘Never has he seen anything similar’ b. Il (n’)a vu personne he neg.have seen nobody ‘He hasn’t seen anybody’ In the change in the postverbal domain the strategy with a positive verb and a negative postverbal indefinite disappears, leaving the strategy with a negative verb and either a positive indefinite (NQ as in Nahuatl) or a negative one (NC, as in Spanish).

132

van der auwera and van alsenoy

table 4.13 More non-strict NC and NQ types

Then, from the two sets of two strategies (two for NQ and two for NC), we arrive at one more non-strict strategy, again differing in one respect from the four nonstrict strategies just discussed, either in the preverbal realm or in the postverbal realm. table 4.14 More non-strict NC and NQ types

more ado about nothing

133

The NQ type is unattested, though we expect it to show up (as discussed in the previous section), but for the NC type there is Catalan. (33) Catalan (de Swart 2013: 173) a. Ningú (no) ha vist Joan nobody neg has seen John ‘Nobody has seen John’ b. En Pere *(no) ha fet res the Peter neg has done nothing ‘Peter has done nothing’ And finally, from both the top and the bottom strategies of Table 4.14, one change in the preverbal domain, taking away the pattern with a preverbal negative indefinite and a positive verb, leaves us with two simple patterns, viz. the ‘Majority strategies’ pattern without any negative indefinites and the strict NC pattern, with both negative verbs and negative indefinites. Note that the top box has the plural ‘strategies’ and not just ‘strategy’. We are dealing both with the neutral something of John didn’t buy something and the negatively polar anything of John didn’t buy anything (Tables 4.1 and 4.2) and chances are these types have their own subtypes too. table 4.15 More types

134

van der auwera and van alsenoy

Table 4.16 puts together all the constellations. There are four non-strict NC types and also four non-strict NQ types, though of the latter one is unattested. In addition there is strict NC at the bottom and at the opposite end there are the ‘Majority strategies’. The latter will have different types too, but they are not of concern in this paper. The fact that there are at least 9 possible NC and NQ types might give one the impression that nearly everything is possible but this is not true. Combining all preverbal ‘values’ with all postverbal ‘values’ gives us 25 combinations, yet only 10 (the 9 NC and NQ types and the ‘Majority strategies’ type) are attested. Table 4.16 may already be fairly complicated, but it is not complete. Two dimensions have not yet been shown. One is that when a language has a choice, the choice need not be restricted to two options, as has been the case so far. Consider the two options in the postverbal domain of Standard English again, with (1b) and (1d). (1) b. I saw nobody d. I didn’t see anybody However, some speakers will have yet a third option, viz. the NC strategy of (1e) and for these speakers we have the system shown in (34). (1) e. I didn’t see nobody. (34)

NI > V

NV > I or V > NI or NV > NI

The only other example of a triple choice that we know of takes us back to Mapuche. In addition to the two choices already shown in (24a–b), with either a negative indefinite followed by a positive verb or a positive indefinite followed by a negative verb, there is yet a third pattern, at least for some speakers, that of a negative indefinite followed by a negative verb. (24) Mapuche (Malvestiti p.c.) e. Doy chem no rumé nie-ke-la-fu-y-iñ more what neg ever have-cf-neg-ipd-ind-1ns-pl ‘We did not have anything more’ (35)

NI > V or I > NV or NI > NV

V > NI or NV > I

more ado about nothing

135

table 4.16 NQ and NC types

The other additional dimension of variation concerns the fact that either the postverbal or the preverbal domain may forbid indefinites, whether positive or negative. For impossible postverbal indefinites we have already documented the NQ types symbolized in (19) and (26), put together in Table 4.17.

136

van der auwera and van alsenoy

table 4.17 More non-strict NQ types

If we go ‘up’ from Otomí we indeed arrive at Nevome, but we can also go ‘down’ and then we arrive at West Armenian (Khanjian 2010), illustrating a non-strict NC type. table 4.18 More non-strict NC and NQ types

(36) Western Armenian (Khanjian 2010:5) votʃ-intʃ (tʃi)-desa neg-what (neg)-saw ‘I didn’t see anything’ Going up and down one step takes us to two additional hypothetical types. The NC type, yielding a structure in accordance with Verb First, is attested in Tamil, but the NQ type, which does not comply with Verb First, is not attested.

more ado about nothing

137

table 4.19 More strategies

(37) Tamil (Pappuswamy p.c.) naan yaar-ai.y-um12-(ee) paar-ka-v.illai who-acc-neg-emph see-inf-neg I ‘I didn’t see anybody’ So much for the impossibility of the indefinite in the postverbal domain. When we switch to the impossibility of indefinites in the preverbal domain, there is no parallelism between NC and NQ at all, at least as far as we can tell. Thus we have found none of the hypothetical NQ systems represented in Table 4.20. Yet the corresponding NC systems are attested.

12

Etymologically, um is a scalar additive marker (‘even’, Gast & van der Auwera 2011), but, pace Lehmann (1989: 197–198) and Bhat (2004: 253), the combination with an interrogative pronoun yields an inherently negative pronoun, at least for some speakers (Pappuswamy, p.c.).

138

van der auwera and van alsenoy

table 4.20 No parallelism

(38) Welsh (de Swart 2010: 187, based on Borsley & Morris Jones 2005: 22, 29) a. Ni soniodd neb am y digwyddia neg mentioned nobody about the event ‘Nobody mentioned the event’ b. Welish i neb saw I nobody ‘I saw nobody’ (15) Icelandic (Nejmann 2001: 130; Haspelmath 1997: 214) a. Ég vissi ekki sjá neitt I did neg see nothing ‘I didn’t see anything’ b. *Neinn sag (ekki) mig nobody saw neg me ‘Nobody saw me’

more ado about nothing

139

As already argued at the end of section 5, the reason for the potential nonexistence of the system in the middle of Table 4.20, allowing only V > NI, should be the strength of the Neg First principle. By the same token, the four other systems should exist, both the NQ and the NC systems, as they all repair the ‘bad’ V > NI system. We do not know why we have only found the NC systems, and interestingly, the two that we did find are both special. Icelandic, as we have already seen, has two sets of negative indefinites, and the other system shows strict NQ (see (14)), while Welsh has an additional NC system, in which it is not the position of the indefinite vis-à-vis the finite verb that counts, but its position vis-à-vis the lexical verb (see van der Auwera & Van Alsenoy 2016). This points to a further possible difference between NQ and NC. For both, the main parameter of variation is the position of the indefinite vis-à-vis the finite verb. NC is more complicated, however: there are more parameters of variation. Of course, NQ has not, so far, attracted as much scholarly attention as NC.

7

Paradigmatic Variation

For our last point NQ is similar to NC again. It is tempting to speak about ‘NQ’ or ‘NC’ languages, but if one uses those terms, they should really only mean ‘languages showing NQ or NC’, and not ‘languages that only show NQ or NC’. NC and NQ are fundamentally properties of indefinite constructions and only in a derivative sense of languages (see Hansen 2012, 2014 on French). Thus Icelandic has both NQ and NC constructions, so the language itself is both NC and NQ. Similarly, Otomí has both strict and non-strict NQ, depending on the construction. At this stage of the research, it is very difficult to estimate to what extent a language is NQ for all of its constructions, i.e., for all of the members of a paradigm comprising the equivalents to nobody, nothing, never, nowhere, in no way way and no. The reason is that grammarians rarely give explicit information on all of the members of such paradigms. So, when a grammarian just lists one negative indefinite, as e.g. Romero-Figeroa (1997: 30) for Warao with the phrase “Finally, there is an indefinite pronoun that is inherently negative ekira ‘nothing, none’”, we are not sure whether there are others. (39) Warao (Romero-Figueroa 1997: 69) Ine ekira ha I nothing cop ‘I have nothing’

140

van der auwera and van alsenoy

In this particular language, it is clear that ekira functions not only as a pronoun, but also as the determiner ‘no’ (Romero-Figeroa 1997: 28). This points at yet another parameter of variation, valid for both NQ and NC, viz. the granularity of the distinctions in the indefiniteness paradigm.

8

Conclusion

In this paper we have addressed the frequency of negative quantification and we have based our generalizations on a variety sample of 179 languages. We have argued that negative quantification has strict and non-strict types, that these are very similar to the ones exhibited by negative concord, and that this similarity can be captured in an integrated account. We have attempted to explain the variation, particularly with the Neg First principle and with the idea that negative quantification constitutes a form/meaning mismatch.

Abbreviations 1 acc cf cop dir emph erg ind inf ipd irr ns obj pl pst rpst sg stat

‘first person’ ‘accusative’ ‘constant feature’ ‘copula’ ‘directional’ ‘emphatic’ ‘ergative’ ‘indicative’ ‘infinitive’ ‘impeditive’ ‘irrealis’ ‘non-singular’ ‘object’ ‘plural’ ‘past’ ‘recent past’ ‘singular’ ‘stative’

more ado about nothing

141

Appendix 1. The Sample For every macro-area, languages are listed alphabetically, first according to the family they belong to and then to the genus within that family. When the language constitutes the genus and the genus constitutes the family, we repeat the names. Africa: Beja (Beja, Afro-Asiatic), Tera (Biu-Mandara, Afro-Asiatic), Somali (Eastern Cushitic, Afro-Asiatic), Arabic (Semitic, Afro-Asiatic), Iraqw (Southern Cushitic, AfroAsiatic), Hausa (West-Chadic, Afro-Asiatic), Nama (Central Khoisan, Khoisan), Ju’huan (Northern Khoisan, Khoisan), Bagirmi (Bongo-Bagirmi, Nilo-Saharan), Kresh (Kresh, Nilo-Saharan), So (Kuliak, Nilo-Saharan), Kunama (Kunama), Ngiti (Lendu, Nilo-Saharan), Ma’di (Moru-Ma’di), Mana (Maban, Nilo-Saharan), Lango (Nilotic, Nilo-Saharan), Nubian (Nubian, Nilo-Saharan), Koyraboro Senni (Songhay, Nilo-Saharan), Majang (Surmic, Nilo-Saharan), Kanuri (Western-Saharan, Nilo-Saharan), Yoruba (Defoid, Niger-Congo), Degena (Edoid, Niger-Congo), Supyire (Gur, Niger-Congo), Igbo (Igboid, Niger-Congo), Ijo (Ijoid, Niger-Congo), Ewe (Kwa, Niger-Congo), Diola-Fogny (Northern Atlantic, Niger-Congo), Nupe (Nupoid, Niger-Congo), Gbeya-Bossangoa (Ubangi, Niger-Congo) Eurasia: Albanian (Albanian, Indo-European), Basque (Basque, Basque), Brahui (Northern Dravidian, Dravidian), Armenian (Eastern, Armenian, Indo-European), Hindi (Indic, Indo-European), Icelandic (Germanic, Indo-European), Japanese (Japanese, Japanese), Korean (Korean, Korean), Khalkha (Mongolic, Mongolian), Hunzib (AvarAndic-Tseciz, Nakh-Dagestanian), Lezgian (Lezgic, Nakh-Dagestanian), Nivkh (Nivkh, Nivkh), Evenki (Tungusic, Tungus), Finnish (Finnic, Uralic), Mansi (Ugric, Uralic) South-East Asia and Oceania: Khasi (Khasian, Austro-Asiatic), Khmu’ (PaluangKhmuic, Austro-Asiatic), Khmer (Mon-Khmer, Austro-Asiatic), Nicobarese (Nicobarese, Austro-Asiatic), Pacoh (Katuic, Austro-Asiatic), Vietnamese (Viet-Muong, AustroAsiatic), Seediq (Atayalic, Austronesian), Kambera (Central Malayo-Polynesian, Austronesian), Chamorro (Chamorroc, Austronesian), Gayo (Gayo, Austronesian), Tagalog (Meso-Philippine, Austronesian), Lewo (Oceanic, Austronesian), Biak (South-Halmahera-West New Guinea, Austronesian), Muna (Sulawesi, Austronesian), Rukai (Tsouic, Austronesian), Tibetan (Bodic, Sino-Tibetan), Burmese (Burmese-Lodo, Sino-Tibetan), Cantonese (Chinese, Sino-Tibetan), Lai (Kuki-Chin, Sino-Tibetan), Meithei (Meithei, Sino-Tibetan), Thai (Kam-Tai, Tai-Kadai) Papua New Guinea and Australia: Nasioi (East Bougainville, East Bougainville), Ambai (Lower Mamberamo, Lower Mamberamo), Mende (Middle Sepik, Sepik-Ramu), Abau (Nor-Pondo-Ramu, Sepik-Ramu), Yimas (Upper Sepik, Sepik-Ramu), Skou (Western

142

van der auwera and van alsenoy

Skou, Skou), Lavukaleve (Lavukaleve, Solomons East Papuan), Arapesh (Kombio-Arapesh, Torricelli), Makalero (Makasae-Fataluku-Oirata, Timor-Alor-Pantar), Usan (Adalbert Range, Trans-New-Guinea), Menya (Angan, Trans-New-Guinea), Kobon (East New Guinea, Trans-New-Guinea), Orokolo (Eleman, Trans-New-Guinea), Nabak (FinisterreHuon, Trans-New-Guinea), Kunimaipa (Goilalan, Trans-New-Guinea), Koiari (Kioarian, Trans-New-Guinea), Amele (Madang, Trans-New-Guinea), Mauwake (Madang, Trans-New-Guinea), Abun (North Central Bird’s Head, Trans-New-Guinea), Kilmeri (Northern Trans-New-Guinea, Trans-New-Guinea), Mian (Ok, Trans-New-Guinea), Sentani (Sentani, Trans-New-Guinea), Inanwatan (South Bird’s Head, Trans-New-Guinea), Gooniyandi (Bunuban, Australian), Gaagudju (Gaagudju, Australian), Garrwa (Garrwan, Australian), Binin-Gun Wok (Gunwinyguan, Australian), Wardaman (Gunwinyguan, Australian), Maung (Iwaidjan, Australian), Mara (Maran, Australian), Bardi (Nyulnyulan, Australian), Ngiyambaa (Pama-Nyungan, Australian), Kayardild (Tangkic, Australian), Tiwi (Tiwian, Australian), Djingili (West Barkly, Australian), Murrinya Patha (Western Daly, Australian), North America: Cheyenne (Algonquian, Algic), Wiyot (Wiyot, Algic), Slave (Athapaskan, Na-Dene), Haida (Haida, Na-Dene), Karok (Karok, Hokan), Eastern Pomo (Pomoan, Hokan), Seri (Seri, Hokan), Washo (Washo, Hokan), Maricopa (Yuman, Hokan), Huave (Huave, Huave), Oneida (Northern Iroquian, Iroquoian), Koasati (Muskogean, Muskogean), Chinook (Chinookan, Penutian), Klamath (Klamath-Modoc, Penutian), Chiapas Zoque (Mixe-Zoque, Mixe-Zoque), Tetepotutla Chinantec (Chinantecan, Oto-Manguean), Mam (Mayan, Mayan), Chalcatongo Mixtec (Mixtecan, Oto-Manguean), Otomí (Otomian, Oto-Manguean), Chocho (Popolocan, Oto-Manguean), Northern Sierra Miwok (Miwokan, Penutian), Nez Perce (Sahaptian, Penutian), Siuslaw (Siuslawan, Penutian), Takelma (Takelman, Penutian), Purépecha (Tarascan, Tarascan), Tshimsian (Tsimshianic, Penutian), Bella Coola (Bella Coola, Salishan), Squamish (Central Salishan, Salish), Lakhota (Siouan, Siouan), Upper Necaxa Totonac (Totonacan, Totonacan), Nahuatl (Aztecan, Uto-Aztecan), Nevome (Tepiman, Uto-Aztecan), Huichol (Corachol, Uto-Aztecan), Makah (Southern Wakashan, Wakashan), Wappo (Wappo, Wappo-Yukian), Yuchi (Yuchi, Yuchi) South America: Andoke (Andoke, Andoke), Paumarí (Arauan, Arauan), Mapuche (Araucanian, Araucanian), Baure (Arawakan, Arawakan), Jaqaru (Aymaran, Aymaran), Awa Pit (Barbacoan, Barbacoan), Chayahuita (Cahuapanan, Cahuapanan), Wai Wai (Cariban, Cariban), Wari; (Chapacura, Wanham, Chapacura-Wanhan), Damana (Aruak, Chibchan), Border Kuna (Kuna, Chibcan), Rama (Rama, Chibchan), Teribe (Talamanca, Chibchan), Tehuelche (Chon Proper, Chon), Epena Pedee (Choco, Choco), Sikuani (Guahiban, Guahiban), Pilagá (Guaicuruan, Guaicuruan), Aguaruna (Jivaroan, Jivaroan), Kwazá (Kwazá, Kwazá), Chiquitano (Chiquitano, Macro-Ge), Canela-Krahô

more ado about nothing

143

(Gei-Kaingang, Macro-Ge), Wichí (Matacoan, Matacoan), Mosetén (Mosetenan, Mosetenan), Pirahã (Mura, Mura), Hup (Nadahup, Nadahup), Páez (Paezan, Paezan), Shipibo-Konibo (Panoan, Panoan), Yagua (Pebe-Yaguan, Pebe-Yaguan), Puinave (Puinave, Puinave), Imbabura Quechua (Quechuan, Quechuan), Araona (Tacanan, Tacanan), Trumai (Trumai, Trumai), Tuyuca (Tucanoan, Tucanoan), Guaraní (Tupian, Tupi-Guarani), Urarina (Urarina, Urarina), Chipaya (Uru-Chipaya, Uru-Chipaya), Waorani (Waorani, Waorani), Warao (Warao, Warao), Yanomámi (Yanoman, Yanoman) Creoles: Haitian Creole

References Agostini, Tainara & Scott Schwenter (2015) ‘Variable Negative Concord in Brazilian Portuguese, acceptability and frequency’, Paper presented at the 15th Hispanic Linguistics Symposium. Beck, David (2004) Upper Necaxa Totonac. München: Lincom Europa. Bhat, D.N.S. (2004) Pronouns. Oxford: Oxford University Press. Borsley, Robert D. & Bob Morris Jones (2005) Welsh negation and grammatical theory. Cardiff: University of Wales Press. Chapman, Shirley & Desmond C. Derbyshire (1986) ‘Paumarí’, In Desmond C. Derbyshire & Geoffrey K. Pullum (eds.) Handbook of Amazonian languages. Volume 3. Berlin: Mouton de Gruyter, 161–352. Collins, Wes (1994) ‘Maya-Mam’, in Peter Kahrel & René van den Berg (eds.) Typological studies in negation. Amsterdam: Benjamins, 365–381. de Swart, Henriëtte (2010) Expression and interpretation of negation. An OT typology. Dordrecht: Springer. Dryer, Matthew S. (2000) ‘Counting genera vs. counting languages’, Linguistic Typology 4: 334–350. Dryer, Matthew S. (2013) ‘Order of subject, object and verb’, In Dryer & Haspelmath (eds.) Available at http://wals.info/chapter/71, Accessed on 2015-02-09. Dryer, Matthew S. & Martin Haspelmath (eds.) (2013) The world atlas of language structures online. Available at http://wals.info England, Nora C. (1983) A grammar of Mam, a Mayan language. Austin: University of Texas Press. Filppula, Markku (1999) The grammar of Irish English. Language in Hibernian style. London: Routledge. Gast, Volker & Johan van der Auwera (2011) ‘Scalar additive operators in the languages of Europe’, Language 87: 2–54. Giannakidou, Anastasia (1998) Polarity sensitivity as (non)veridical dependency. Amsterdam: John Benjamins.

144

van der auwera and van alsenoy

Giannakidou, Anastasia (2006) N-words and negative concord, In Martin Everaert & Henk van Riemsdijk (eds) The Blackwell companion to syntax. Volume III. Oxford: Blackwell, 327–391. Hansen, Maj-Britt Mosegaard (2012) ‘On the evolution of temporal n-words in Medieval French’, Language Sciences 34: 76–91. Hansen, Maj-Britt Mosegaard (2014) ‘The grammaticalization of negative indefinites. The case of the temporal/aspectual n-words plus and mais in Medieval French’, In Hansen, Maj-Britt Mosegaard & Jacqueline Visconti (eds.) The diachrony of negation. Amsterdam: Benjamins, 185–212. Haspelmath, Martin (1997) Indefinite pronouns. Oxford: Clarendon Press. Haspelmath, Martin (2005) ‘Negative indefinite pronouns and predicate negation’, In Martin Haspelmath, Matthew S. Dryer, David Gil & Bernard Comrie (eds.) The world atlas of language structures. Oxford: Oxford University Press, 466–469. Haspelmath, Martin (2013) ‘Negative indefinite pronouns and predicate negation’, in Dryer & Haspelmath (eds.) Available at http://wals.info/chapter/115, Accessed on 2015-03-01 Heidolph, Karl Erich (1970) ‘Zur Bedeutung negativer Sätze’, In Manfred Bierwisch & Karl Erich Heidolph (eds.) Progress in linguistics. A collection of papers. The Hague: Mouton, 86–101. Hekking, Ewald (1995) El Otomí de Santiago Mexquititlán: desplazamiento lingüistico, préstamos y cambios gramaticales. Doctoral dissertation University of Amsterdam. Horn, Laurence R. (1989) A natural history of negation. Chicago: The University of Chicago Press. Horn, Laurence R. (2011) Negation. Oxford Bibliographies, DOI: 10.1093/obo/ 9780199772810-0032 Howe, Darin (2005) ‘Negation in African American Vernacular English’, In Yoko Iyeiri (ed.) Aspects of English Negation. Amsterdam: Benjamins, 173–203. Jespersen, Otto (1917) Negation in English and other languages. København: A.F. Høst & Søn. Jespersen, Otto (1922) Language. Its nature, development and origin. London: George Allen & Unwin. Kahrel, Peter (1996) Aspects of negation. Doctoral dissertation, University of Amsterdam. Kefer, Michel (1974) ‘Die syntaktische Funktion der Negation im deutschen Satz I’, Revue des Langues Vivantes 40: 66–84. Khanjian, Hrayr (2010) ‘Negative Concord in Western Armenian’, Paper presented at Berkeley Linguistics Society (BLS 36), University of California, Berkeley, CA. Available at http://web.mit.edu/khanjian/www/Site/Research_files/2nd%20Generals% 20take%2023%20post‑d.pdf, Accessed on 2013-01-14

more ado about nothing

145

Lehmann, Thomas (1989) A grammar of modern Tamil. Pondicherry: Pondicherry Institute of Linguistics and Culture. Macauley, Monica (1996) A grammar of Chalcatongo Mixtec. Berkeley: University of California Press. MacSwan, Jeff (1999) A minimalist approach to intrasentential code switching. New York: Garland. Mathesius, Vilém (1937) ‘Double negation and grammatical concord’, in Mélanges de linguistique et philologie offerts à Jacq. van Ginneken à l’occasion du soixantième anniversaire de sa naissance (21 avril 1937). Paris: Klincksieck, 79–83. Miestamo, Matti (2005) Standard negation. The negation of declarative verbal main clauses in a typological perspective. Berlin: Mouton de Gruyter. Nejmann, Daisy L. (2001) Colloquial Icelandic. The complete course for beginners. London: Routledge. Osborne, C.R. (1974) The Tiwi language. Canberra: Australian Institute of Aboriginal Studies. Priego Montfort de Mostaghimi, Maria Eugenia (1989) Gramatica del Otomí (Hñähñu) del Mezquital, Mexico. Doctoral dissertation, University of Bielefeld. Rausch, P.J. (1912) ‘Die Sprache von Südost-Bougainville, Deutsche Salomonsinseln’, Anthropos 7: 105–134, 585–616, 964–994. Romero-Figeroa, Andrés (1997) Warao. München: Lincom. Shaul, David Leedom (1986) Topics in Nevome syntax. Berkeley: University of California Press. Smeets, Ineke (1989) A grammar of Mapuche. Berlin: Mouton de Gruyter. Van Alsenoy, Lauren (2014) A new typology of indefinite pronouns, with a focus on negative indefinites. Doctoral dissertation, University of Antwerp. van der Auwera, Johan (2011) ‘Standard Average European’, In Bernd Kortmann & Johan van der Auwera (eds.) The Languages and Linguistics of Europe: A comprehensive guide. Berlin: Mouton, 291–306. van der Auwera, Johan (2015) ‘Les indéfinis négatifs dans les langues créoles’, Nouveaux cahiers de linguistique française 32: 27–43. van der Auwera, Johan, Ludovic De Cuypere & Annemie Neuckermans (2006) ‘Negative indefinites—a typological and diachronic perspective on a Brabantic construction’, In Terttu Nevalainen, Juhani Klemola & Mikko Laitinen (eds.) Types of variation. Diachronic, dialectal and typological interfaces. Amsterdam: Benjmins, 305– 319. van der Auwera, Johan & Lauren Van Alsenoy (2016) ‘On the typology of negative concord’, Studies in Language 40: 473–512. Veerman-Leichsenring, Annette (2000) Gramática del chocho de Santa Catarina Ocotlán, Oaxaca. Leiden: University of Leiden. Werle, Adam (2002) ‘A typology of negative indefinites’, In Mary Andronis, Erin Deben-

146

van der auwera and van alsenoy

port, Anne Pycha & Keiko Yoshimura (eds.) Papers from the 38th Meeting of the Chicago Linguistic Society, vol. 2: The Panels. Chicago: Chicago Linguistic Society, 127– 143. Westley, David O. (1991) Tepetotutla Chinantec syntax. Dallas: Summer Institute of Linguistics and University of Texas at Arlington.

part 2 On Sense-Generality and the Semantics/Pragmatics Landscape

∵

chapter 5

Distinguishing Ambiguity from Underspecificity* Una Stojnić, Matthew Stone and Ernie Lepore

Throughout his career, Jay Atlas has been a key player addressing one of the most important and fundamental questions in semantics and pragmatics: how far does linguistic knowledge of meaning extend? The examples in (1), compiled by Borg (2004, p. 34), illustrate the diverse settings in which this question has arisen, and show what is at stake in answering it. Each sentence in (1) contains a parenthetical element that supplements its content. (1) a. b. c. d. e. f. g. h.

Paracetamol is better [than aspirin]. You won’t die [from that cut]. I’ve eaten [recently]. It’s raining [where the speaker is]. Everybody [who came to the party] had a great time. Smith weighs 120 pounds [weighed before breakfast and undressed]. The apple is green [on the outside]. Holland is flat [for a country].

In the right circumstances, a speaker can use the simple sentence, but make a point that is best captured by the supplemented rendition. How does this work? In particular, what is the role of knowledge of meaning in prompting, licensing, or determining these specific interpretations? In response to this problem, Jay Atlas has been one of the earliest and most consistent defenders of the position that knowledge of meaning plays no role at all in such interpretive phenomena—they are entirely due to general principles of interpretive inference.1 Atlas’s theory of interpretation is founded on his doctrine of the underspecificity of linguistic meaning. For Atlas, linguistic knowledge characterizes the semantics of the simple examples of (1a–h) in a way that is neither truth evalu-

* Preparation of this chapter was supported in part by NSF grant IIS 1526723 and a sabbatical leave from Rutgers University to Stone. We are grateful to the editors for their thoughtful comments and suggestions on an earlier draft. 1 See Atlas (1977, 1989, 2005).

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_006

150

stojnić, stone and lepore

able nor incomplete. For example, on his view, the sentence ‘It’s raining’ cannot be modeled as directly encoding propositional information—in part, of course, because rain can fall at some places but not at others. At the same time, on his view, linguistic semantics does not instruct the hearer to recover a suitable situation of evaluation; there is no hidden variable (Stanley, 2000, Stanley and Szabó, 2000) or other context dependency (Kaplan, 1989) that explicitly marks the location parameter that is eventually supplied in interpretation. Instead, Atlas sees the meaning of ‘It’s raining’ as a univocal but non-propositional trigger for a process of interpretive inference—inference to the best explanation. It is this inference, which makes sense of why the speaker has used the utterance—rather than any encoded meaning, on his view—that ultimately issues in the propositional content of a specific interpretation. This picture ramifies throughout the pragmatics literature, finding echoes in the works of Sperber and Wilson (1986), Bach (1994, 1999, 2001), Carston (2002) and Recanati (2004) among others. At the same time, however, the twin claims that linguistic meaning is not truth-evaluable and not incomplete distinguish Atlas’s approach from many prominent alternatives.2 For example, the context-sensitive explanations that Atlas rejects have defenders such as Stanley (2000), who would describe the data in (1) on the model of true demonstratives (in the sense of Kaplan, 1989). These approaches can represent the simple sentences in (1) as semantically incomplete, by including suitable free variables in logical form. Some sort of non-linguistic input is then required to provide a value for these variables and thereby specify the truth conditions of the utterance of the sentence in a context. In particular, on the Kaplanian account (1989), expressions like ‘that’ or ‘he’ resolve to whatever a speaker intends with their uses.3 Conversely, to understand them, hearers must entertain open-ended reasoning about the speaker’s communicative

2 We note that one might deny that the cases in (1) should be grouped together; for instance, one might hold that some of the examples are genuine instances of context sensitivity while others are not. Alternatively, one might advocate different accounts of context sensitivity for different examples in (1). Our arguments are compatible with such explanations, as we use (1) for illustration only, and any subset of the examples in (1) suffices for our purposes. Of course, the alternative theories sketched here are typically envisioned as general theories of context sensitivity, and are in fact typically understood, implicitly or explicitly, to cover most, if not all, of the cases. 3 Alternatively, one might argue that the supplementation comes from non-linguistic features of the utterance context (Wettstein, 1984). On either construal, the meaning is supplemented by some mechanism that goes beyond linguistic meaning. On either of these construals, the open-ended reasoning is required and crucial for interpretation.

distinguishing ambiguity from underspecificity

151

and/or referential intentions. It’s just the same, this theory goes, for the contextsensitive resolution of ‘is better’ in (1a), ‘won’t die’ in (1b), and so forth. Of course, Atlas agrees that open-ended abductive inference is involved in interpretation—but that is precisely what makes him skeptical of contextdependent accounts of this kind. For Atlas, pragmatic reasoning inevitably enriches meanings on the way to constructing specific interpretations: that is the essence of abductive reasoning. (A very similar conclusion can be drawn on Sperber and Wilson’s 1986 theory of explicature in pragmatics, or Recanati’s 1994 theory of free enrichment.) If that’s how pragmatic reasoning works, then postulating a hidden, context-dependent element in logical form, whose value explicitly needs to be reconstructed, seems to do no work in the theory at all. A starkly different reaction to the data in (1) is semantic minimalism (Borg, 2004, Cappelen and Lepore, 2005). Semantic minimalists suggest that linguistic knowledge does determine a complete and truth-evaluable meaning for the simple sentences in (1). In other words—in contrast to Atlas’s underspecificity account, and in contrast to demonstrative context-sensitive approaches—according to semantic minimalism no non-linguistic input is required to interpret the simple examples of (1) as expressing propositions. Put concretely, the claim is that every utterance of sentence (1d), for example, expresses the proposition that it is raining! This is a bold claim, because it involves positing minimal truth conditions that seem to fall short of the truth conditions intuitively expressed by such utterances. (We shall return to this apparent weakness in what follows.) Abductive approaches, like that of Atlas, seem to suggest that the enrichments required to fix truth conditions for utterances of sentences (1a–h) can be arbitrarily flexible and require arbitrary reasoning on the part of the hearer to recover them. Yet we find little evidence for such profligate interpretive variability among speakers or such extremely sympathetic mind reading among hearers. Rather, we find constrained and efficient strategies for getting our thoughts across with words.4 Semantic minimalism is a provocative starting point for explaining this coordination—but one that would be much more satisfying if we could see how the rules of language give us more than just a minimal proposition that a sentence encodes across all its occasions of use.5 In

4 A number of authors have tried to impose constraints on acceptable communicative intentions, in an attempt to streamline the reasoning involved in recovering them (e.g., King 2014); we are skeptical that this theoretical strategy can succeed. For some of our criticisms, see Stojnić, Stone and Lepore (2013, 2017), Stojnić (2016). 5 In fact, although they maintain that linguistic meaning fully specifies truth conditions, since these truth conditions are minimal, and not the ones intuitively expressed by most uses of

152

stojnić, stone and lepore

short, it is still an open problem to meet Jay Atlas’s challenge, and find any way of relating the enriched interpretations that our utterances intuitively have to some linguistic knowledge that underlies them. Our tribute to Atlas in this chapter is to attempt to meet this challenge in a novel way—by borrowing Kaplan’s (1989) distinction between true demonstratives and pure indexicals. As we have seen, true demonstratives are understood to be incomplete without non-linguistic supplementation. By contrast, pure indexical expressions are not: Their values are fixed as a matter of linguistic convention. For example, the first person pronoun ‘I’ is a pure indexical inasmuch as, regardless of whomever its user intends to reference, its uses always pick out the speaker. In this regard, the meaning of ‘I’ fully determines its semantic value in any context of use; it is not necessary to invoke pragmatic reasoning to infer or construct a semantic value for ‘I’ in context.6 We believe that there are completions of at least some of the sentences in (1a–h) that are linguistically governed in just this way.7 We will suggest that an indexical treatment of these interpretive effects is not prone to the free-for-all we believe comes with intentionalism and general abductive inference, but that it can nevertheless accommodate some of the interpretive intuitions that semantic minimalism abandons. In effect, once we give up mind reading as a path to resolution and replace it with conventionality, then much of the data that led to semantic minimalism in the first place can be explained without incurring

a given sentence, semantic minimalists have typically resorted to pragmatic mechanisms to explain how we normally manage to assert truth conditions richer then the minimal propositions. This is particularly evident in Cappelen and Lepore’s (2005) Speech Act Pluralism, for example. Such pragmatic theses about the intuitively asserted truth conditions come with all the profligate flexibility of abductive approaches. By contrast, in what follows we will argue that the linguistic meaning specifies intuitively expressed truth conditions. Thus we will claim the benefits of semantic minimalism, while avoiding the looseness of abductive approaches. 6 We note that, though there are so called descriptive uses of ‘I’, just as there are descriptive uses of ‘he’, we take this phenomenon to be orthogonal to the question of pure indexicality vs. true demonstrativeness (Nunberg, 1993). The character of ‘I’ needn’t necessarily be adjusted to allow for these uses any more (or less) then the character for ‘he’ has to be so adjusted. (We normally take the character of ‘he’, roughly, to be ‘the most salient, singular, third person, masculine individual’ which does not automatically capture the descriptive meaning of ‘he’ either.) 7 We shall not attempt to defend this position in its full generality in this chapter, but will provide enough evidence to fuel our argument against the underspecification account. For a defense of this position in its full generality, see Stojnić (2016).

distinguishing ambiguity from underspecificity

153

any of minimalism’s shortcomings, including positing hard to evaluate minimal propositions. This amounts to a third way couched in between minimalism and full-tilt intentionalism, and it’s the one we want to explore in what follows. A number of ideas must be combined together, however, to make the indexical view a compelling alternative to underspecificity. Our strategy in this chapter depends on showing how our indexical view is compatible with the manifest variety of attested interpretations of utterances in context. Our response is to appeal to covert ambiguity in the organization of discourse, as described by theories of discourse coherence (Hobbs, 1979, Kehler, 2002, Asher and Lascarides, 2003). Such ambiguities are extremely contentious—Atlas’s own theory works to undermine the postulation of such ambiguities both on empirical grounds and on methodological ones. So the bulk of this chapter is devoted to establishing the viability of our view. We argue that discourse exhibits qualitative ambiguities, on a par with the attachment ambiguities of sentence-level syntactic structure. Such ambiguities aren’t susceptible to Gricean ‘modified Occam’s razor’ style arguments because they follow from the most compact and economical statement of the relevant rules for composing linguistic structures. Second, we suggest that ambiguity resolution is a matter of constrained inference that plausibly escapes the open-ended abduction of general intention recognition. Thus, we can postulate covert ambiguities without threatening the explanations for the transparency of communication that we hope to inherit from semantic minimalism. Only in tandem with these arguments can an indexical view offer a meaningful alternative to Atlas’s underspecificity. We begin with some elaboration of Atlas on underspecificity.

1

Underspecificity

As early as 1977, Atlas affirmed that the lexical “entry for ‘not’ is such that ‘not’ is [semantically] non-specific. […] A reading or semantic representation of a [negative] sentence is not a bearer of truth-value; it is not a proposition or a logical form, though it determines what propositions can be literally meant by the sentence” (Atlas 1977, p. 322; see, also, Atlas 2005, p. 33). In other words, the sentence type corresponding to, e.g., (2), is not given a semantic representation that captures its truth conditions or encodes a proposition. (2) The king of France is not bald. As Atlas (2007, p. 218) writes:

154

stojnić, stone and lepore

[2 is] an unambiguous sentence, one without ambiguities of scope, or homonymy of lexical items, for ‘not’. In one context an utterance of the sentence would literally express the narrow-scope, predicate negation proposition (or truth-conditions), and in other contexts it would literally express the wide-scope, sentence negation proposition (or truthconditions). Neither proposition is a literal meaning of the sentence. The sentence-type is not ambiguous between these interpretations or contextual understandings. It is “neutral” between these meanings; it is, technically speaking, semantically non-specific between them.8 Atlas summarizes his view with a quote from Travis (1996, p. 451.): “What words mean plays a role in fixing when they would be true; but not an exhaustive one. Meaning leaves room for the variation in truth conditions from one speaking to another” (Atlas, 2007, p. 219). He finds underspecificity particularly damaging for semantic minimalists since he believes they have no recourse to account for the differing interpretations of utterances of (2). He believes the minimalist is committed to every one of its utterances being “true just in case [the king of France is not bald], and express[ing] the proposition that [the king of France is not bald]” (Atlas, 2007, p. 223).9 But, he emphasizes, this “is just not true of any utterance of (2). An utterance of the sentence might express the sentence (exclusion) negation; it might express the predicate (choice) negation” (Atlas, 2007, p. 223). Though Atlas denies that he is a Gricean, his view is Gricean in one key respect: the longstanding intellectual thread running throughout Grice’s work on implicature, neo- Gricean work on generalized conversational implicature, Relevance Theory’s explicature, theories of free enrichment in pragmatics, and Atlas’ own theory built on attributions of underspecificity is that language use is governed by general psychological principles that reinterpret grammatically specified meanings, and thereby, associate alternative speaker meanings to utterances in context. In his own words (Atlas, 2005, p. 34):

8 See also Atlas 2005, Ch. 1, for his discussion of the underspecified ‘boy’, as in “My cousin isn’t a boy any longer.” A discussion of this type of example can also be found in Stalnaker (1974/1999, pp. 53–54) and Langedoen (1971, p. 343). 9 We are unclear why a minimalist cannot argue there’s scope ambiguity. A minimalist denies there’s context sensitivity, but certainly they agree there’s syntactic scope. Analogously, provided that the idea is cashed out in terms of a difference in truth-conditions, no principles prevent a minimalist from endorsing the thesis that the difference between predicate negation and sentential negation derives from polysemy in the lexical item ‘not’. Again, minimalists have no objection to lexical ambiguity.

distinguishing ambiguity from underspecificity

155

semantical nonspecificity with respect to a semantical property—for instance, the nonspecificity with respect to [MALE] and [–MALE] of ‘neighbor’ by contrast with the specification of the lexical feature [–MALE] in ‘poetess’—requires inferential mechanisms to produce from the semantically nonspecific literal meanings of sentence-types the interpretations of speakers’ utterances of sentence-tokens. Atlas separates the attribution of underspecificity to (2) from an alleged scopal ambiguity. His argument is that attributing ambiguity is redundant (Atlas 2005, p. 38). “If pragmatic inferences were a necessary part of a theory of utterance-interpretation and the sentences were ambiguous, then inferences would be essential to disambiguation, to the selection of the appropriate sense in the context of utterance.” But Atlas (2005, p. 38) claims that the same inferential mechanisms must be invoked for precisifying semantically nonspecific interpretations—“one [mechanism] to select a reading, the other to construct a more specific or precise interpretation.” Atlas is also clear that by ‘underspecification’ he does not mean ‘context sensitivity’ either. Take ‘neighbor’ again: no one thinks that in some contexts it picks out a male and in others a female, as a function of its meaning and its context of use, as is the case with, say, the demonstrative pronouns ‘I’ and ‘he’. In other words, it is not a part of the character (in the sense of Kaplan (1989)) that ‘neighbor’ is an individual of a particular contextually specified gender. Instead, its meaning is neutral or underspecifies such considerations. Atlas would take the same approach for (1a–h). In defending underspecificity against context sensitivity and ambiguity, Atlas is essentially running an Occam’s razor style argument: his view is that pragmatic explanations that construct enrichments of univocal meanings are better than explanations positing ambiguities or Kaplanian contextual supplementation. There is no need to appeal to ambiguous or incomplete meanings to explain the pragmatics of the utterances, and in any case, resolving the intended meaning would require pragmatic explanations, anyway. So, since his account posits fewer and simpler meanings and yet relies on the same resources for fixing truth conditions that the theorist would have to appeal to anyway, it is to be preferred by reason of economy. In keeping with the Gricean tradition, Atlas frames this criticism in very general terms. We cannot respond to it at the same level of generality. However, when we explain our indexical view in detail, we will be able to show why its ambiguities arise and what they contribute to interpretation. We can then argue that ambiguity plays a principled and important role in our account: indeed, we can argue, contra Atlas, that what he deems underspecification really is a genuine case of disambiguation.

156 2

stojnić, stone and lepore

Developing the Indexical View

For Kaplan (1989), pure indexicals get their semantic values as a function of the context of use. For example, the meaning of the first person pronoun ‘I’ fully determines its semantic value in any context of use: the semantic value of ‘I’ is the speaker of the utterance. We believe—and have defended elsewhere (Stojnić, Stone, Lepore, 2013, 2017, Stojnić, 2016)—the view that the category of pure indexicals is in fact much broader than Kaplan realized. It even includes third person pronouns like ‘he’ and ‘him’! In fact, we think it covers many of the interpretive elaborations found in (1a–h) as well. A problem immediately arises for the suggestion that such items get their semantic values as a function of the context, however: cases of their use where multiple interpretations seem to be possible. Take example (3), from Smyth (1994), for example. (3) Phil tickled Stanley. Liz poked him. Two potential interpretations for (3) are probably obvious. One suggests that Liz’s poking was provoked by Phil’s action, maybe she’s retaliating for Stanley; on this construal, ‘him’ gets resolved to Phil. For the second interpretation, think of (3) as comparing two different but related actions. On this construal, ‘him’ gets resolved to Stanley. This ambiguity is no challenge, of course, for the prevailing view that ‘him’ in cases like (3) is a true demonstrative. On this view, the linguistic meaning of ‘him’ is something like ‘the most salient male’, which constrains but ultimately leaves open the semantic value it takes on an occasion of use.10 What does determine its semantic value, on the demonstrative view, is the speaker’s referential intention.11 Thus, according to this view, resolving ‘him’ involves recognizing the speaker’s referential intention, through reasoning about her mental 10

11

We take the linguistic meaning to be the Kaplanian character. Some authors prefer to think of the gender feature of a pronoun as a part of the presupposed content. (See Cooper, 1983, Heim and Kratzer, 1998, Roberts, 2002, Heim, 2008, Sauerland, 2008, inter alia). Though the issue is somewhat orthogonal, our preferred view treats the gender feature as a part of the character, characterized as a search function we associate with the pronoun (Stojnić, Stone and Lepore, 2013, 2014). As we argue in (Stojnić, Stone and Lepore, 2014), this is perfectly compatible with speakers sometimes managing to get a point across—i.e. to speaker refer, in the sense of Kripke (1977)—say, to a male individual with an utterance of ‘she’. We don’t pursue the issue here, in the interest of space. Or, as mentioned before, a set of contextually specified, but non-linguistic, epistemic cues (Wettstein, 1984).

distinguishing ambiguity from underspecificity

157

states, and by exploiting various epistemic cues that the context renders available in order to uncover these intentions. Indeed, we naturally expect that the prior discussion might render one resolution of ‘him’ particularly salient, or an accompanying gesture or other cue may show what resolution the speaker has in mind. However, such explanations are not open to the indexical view. Fundamentally, there is no reason to assume that the extralinguistic context in which an utterance of (3) is used necessarily determines a unique interpretation for ‘him’. Thus, on a naïve understanding of context, our examination of (3) shows one context with two possible semantic values for ‘him’. The conclusion seems inescapable that the semantic value of ‘him’ cannot be a function of the context. Our response is to develop a more refined account of context, according to which context evolves, clause by clause or even word by word, in a rulegoverned way, as a byproduct of the organization of discourse. Let us sketch an outline of our view. In brief, utterances of linguistic material such as (3) can harbor different implicit organizations at the level of discourse. These different possibilities affect how the context evolves from one sentence to the next. In particular, the different organizations of (3) can affect whether the context for its second sentence sets up Phil or Stanley as ‘him’ (or, as we might put it, whether Phil or Stanley is the center of the context, abusing the terminology of Grosz, Joshi and Weinstein’s 1995 centering theory, to coin a term that more clearly parallels the indexical descriptions ‘the speaker’ and ‘the addressee’). We motivate this view with the obvious observation that a discourse, no matter how simple, so, e.g., even (3), includes more than just its sequence of sentences. In addition, the sequence must be organized around familiar relationships. A famous illustration of this sort of organization is in Hobbs (1979): (4) John took the train from Paris to Istanbul. He has family there. (5) John took the train from Paris to Istanbul. He likes spinach. (4) is offering us a reason for why John went to Istanbul. In effect, a relation of Explanation links its two sentences (cf. Asher and Lascarides 2003, Kehler 2002). The explanatory relation is crucial to the interpretation of the pronominal expressions in the example: not only must ‘He’ pick out John, but ‘there’ must pick out Istanbul. Only the opportunities available at the destination of a trip can explain why someone makes it.12 12

For more on the correlation between discourse and anaphora, see Asher and Lascarides (2003) or Kehler at al. (2008).

158

stojnić, stone and lepore

More interestingly, (5) seems prima facie odd. Confronted by an utterance of it out of the blue we search for a connection. We wonder how the preference for spinach explains the train trip.13 If we fail to establish the connection, we fail to fully understand (5). Just the same, rhetorical relations are crucial to the interpretation of (3). In (3) we find two sentences connected either by a relation of Result or one of Parallel. Once we opt for one of these coherence relations over the other as the best fit in the context overall, notice that we automatically fix its pronoun resolution (to either Phil or Stanley). In short, once we settle on a coherence relation, say, Parallel or Result, the resolution of ‘him’ is automatically fixed. Atlas, of course, might take these considerations of coherence as evidence of an inferential relationship between a speaker’s intention in organizing the discourse and her referential intentions. Given his commitment to underspecificity, he would have to maintain that discourses like (3) underspecify pronoun resolution as well as which coherence relations tie the discourse in (3) together. Our view is that this pragmatic theory mistakes the relationship between how the discourse is organized and how its pronoun gets resolved. We are instead committed to a much tighter connection between coherence and pronoun resolution. We maintain, and argue as much elsewhere (Stojnić, Stone and Lepore, 2017, Stojnić, 2016), that not only is there a correlation between the implicit organization of the discourse and the resolution of a pronoun, but even stronger, the choice of the relation organizing a discourse determines pronoun resolutions as a matter of an underlying linguistic convention—that is, as a matter of English grammar. So construed, the reference of ‘him’ in (3) is not underspecified; on the contrary, rules of language fix its interpretation. Rather than appeal to linguistic underspecificity to explain the potential different readings a hearer might entertain when confronted by an utterance of (3), we appeal to straightforward and ordinary ambiguity. In particular, there is an ambiguity as to which coherence relation the discourse in (3) harbors in its

13

Surely, learning that Istanbul is famous for its spinach would help establish the Explanatory connection. However, note that interestingly, out of the blue, we tend to search for an Explanation relation, not just any old relation both in (4) and in (5). This is surprising since, of course a number of others may be available. We take this to suggest that the choice of the relation itself is primed by a linguistic cue, the issue we will return to below. Note that, likewise, if (4) or (5) are not uttered out of the blue, but are a part of a larger discourse, we might get a preference for other relations. As it will become clear below, the question of the choice of a relation, for us, is a question of disambiguation of the logical form of a discourse. (Cf. Stojnić, Stone and Lepore, 2017.)

distinguishing ambiguity from underspecificity

159

LF—if it is Parallel, the center of the context when we interpret the pronoun ‘him’ is Stanley, and so ‘him’ must pick out Stanley; if it is Result, the center of the context when we interpret the pronoun ‘him’ is Phil, and so ‘him’ must pick out Phil. This explains, on our view, how even third person singular pronouns like ‘him’ can be given indexical interpretations. Similar considerations, we predict, hold of other cases of alleged underspecificity, as, e.g., in (1).14 In short, contrary to what Atlas believes, we hold that meaning resolution in general is not underspecified. However, this leaves open that discourses can be ambiguous. If we are right, then in order to interpret an utterance, the hearer must first settle between these ambiguities. How interlocutors achieve this may involve identifying plausible inferential links conveyed by coherence relations. It may also include choosing the reading that yields the most plausible resolutions. This sort of reasoning obviously makes up a key part of disambiguation, but it does not contribute content to logical form. Rather, given the available logical forms for the string in question the sort of reasoning can privilege one ascription of logical form to the discourse over other candidates generated by the grammar. To borrow Kaplan’s apt locution, this is a part of the pre-semantic enterprise each speaker participates in for any discourse. Is this speech? If so, which language am I being presented with? Which sentence? Which words? What’s the speaker saying? Our claim is that once a coherence relation has been identified or justifiably ascribed, then the pronoun resolution is automatically fixed by the grammar. If this is right, and moreover, if (as we believe15), coherence affects the resolution in the case of (1a–h), then literal meaning does, contra Atlas, as a matter of fact, suffice to fix truth conditions. That is, the grammar leaves no room for linguistic underspecificity in the relevant cases.

3

Assessing the Alternatives

We now have all the ingredients needed to set up a debate between Atlas’s underspecification view and our own ambiguity one. In accounting for cases such as (1), our view postulates that linguistic knowledge determines the specification of semantic values for the relevant expressions in context. Atlas’s view, by contrast, holds that linguistic knowledge underspecifies their interpretations. On our view, interpretive variability is possible in such cases because dis-

14 15

See Stojnić (2016) for further discussion. See Stojnić (2016) for a defense.

160

stojnić, stone and lepore

courses can be disambiguated in different ways. For Atlas, by contrast, different readings are constructed through a pragmatic process of abductive inference that can entertain alternative hypotheses about the speaker’s communicative intention. To decide between the views, we need to deal with both points of disagreement. In particular, we assume that Altas would challenge our claim that the effects of discourse coherence are grammatically encoded. Thus, to defend our view, we need to marshal evidence that the effects of discourse coherence on features such as pronoun interpretation are not merely common sense but instead have a recognizable conventional component. In fact, we think there is an abundance of such evidence, which we outline in the first part of this section. The conventionality of the effect of coherence is only part of the story, however. In response to our evidence, we suspect that Atlas would still insist that hearers’ knowledge of coherence relations simply factors into the general abductive inference the hearer uses to construct a specific interpretation from an underspecified meaning. This is, after all, his general view about how we move from underspecified meanings of sentences to fixed truth conditions for their legitimate uses. So construed, even if a coherence relation participates in fixing the interpretation of a use of a pronoun, the choice between alternative coherence relations organizing a discourse is not a consequence of any linguistic ambiguity, but simply a consequence of pragmatic reasoning from an underspecified meaning. In short, the conventionality of the effect of coherence is not necessarily in itself an effective response to Grice’s—and Atlas’s—modified Occam’s razor arguments. Thus, to defend our view, we also need to give positive reasons for distinguishing the disambiguation we think is involved in interpretation from open-ended abductive reasoning about speaker intentions. We conclude this section with the considerations we think show the special status of disambiguation. 3.1 Coherence and Conventionality Our argument starts from the observation that the set of indicators interlocutors rely on in settling pronoun resolution is rather restricted. Indeed, it looks as if, contra Atlas, interlocutors favor certain linguistic cues over broader constraints created by background knowledge and rational inference. Consider, for example, (6) from Kehler (2002): (6) Margaret Thatcher admires Ronald Reagan and George W. Bush absolutely worships her.

distinguishing ambiguity from underspecificity

161

Kehler reports that (6) is generally judged anomalous. As we noted elsewhere (Stojnic, Stone and Lepore, 2017), this would be easily explained were we to understand the choice of a coherence relation to force a certain resolution of its pronoun as a matter of linguistic convention. In particular, following ‘admires’ with the stronger ‘absolutely worships’ indicates (6) is organized around a contrast between Thatcher’s and Bush’s comparable attitudes, signaling a Parallel coherence relation. But this relation requires, as a rule of English, that the pronoun in (6) gets resolved to an antecedent with the same grammatical role; here, the pronoun in the direct object position is searching for an antecedent in the direct object position of the previous clause. But this results in a gender clash, and hence, infelicity. As Kehler reports, (6) is understood as if the speaker mistakenly used ‘her’ instead of ‘him’. This explanation shows the active role that the attribution of a Parallel relation has in the interpretation of (6). In (6), we don’t simply see that the interpretation is being constructed holistically to give a result that makes sense, and that similarities among propositions are being exploited to find connections among sentences in discourse. In fact, the Parallel relation is imposing constraints that prevent us from making sense of the discourse in what otherwise appears to be the natural way. Thatcher has been evoked in the first clause, and she is a well-known object of Bush’s admiration, and a speaker could certainly be interested in highlighting the strengthening transatlantic relationship illustrated by these reciprocal attitudes. Moreover, considerations of charity would favor the plausible interpretation as well, since that would not assume the speaker has spoken in error. Thus, unless we adopt a view where the Parallel relation has a distinctive force in organizing the discourse, one that goes beyond mere rationality, we have to predict that it should be easy to refer to Thatcher with ‘her’. Yet, this is not what we find. Parallel in (6) indicates Reagan in a way that’s difficult to override for the sort of common-sense inference Atlas advocates. This is just what we would expect if Parallel signaled this interpretation as a matter of linguistic convention.16

16

Of course, as Kehler (2002) himself notes, and as we discuss elsewhere (Stojnić, Stone, Lepore, 2017), there are ways of making (6) felicitous, with ‘her’ picking out Thatcher. For instance, putting a focal stress on ‘her’ can have such an effect, as can embedding (6) in a larger discourse, or making a context such that a relation other than Parallel is preferred. This is perfectly consistent with our view: we have already seen that discourses can be ambiguous with respect to which relation they harbor—as witnessed by (3)—and our preferences for resolving ambiguities can change in the presence of additional linguistic cues, such as accentuation on ‘her’ in (6), or further discourse context that can give rise to new interpretive dependencies. Our point is simply that, holding the relation fixed, the

162

stojnić, stone and lepore

3.2 Ambiguity versus Abduction Our explanation of the effect of Parallel is that it shows that the logical forms of different discourses can harbor different coherence relations, which, in turn, fully determine resolution as a matter of grammar. Part and parcel of this view is that there are covert ambiguities in the organization of discourse. But why can’t the familiar Occam’s razor argument be run against these alleged ambiguities? That is, isn’t it the case that we need to employ general pragmatic reasoning in disambiguating different logical forms of a discourse, harboring different coherence relations? And if so, wouldn’t the same general pragmatic reasoning we need to employ again suffice to specify the meaning, and so isn’t positing ambiguity over and above the underspecification redundant? In other words, once confronted by our data and arguments, Atlas could question the very significance of rejecting reinterpretation, arguing that the theoretical machinery we are keen to banish will inevitably figure (and so, resurface) in an account of disambiguation. What are cases of disambiguation if not merely cases of cooperative intention recognition? Why, then, isn’t it correct to characterize cases of disambiguation as discarding grammatically underspecified meanings, and delivering specified interpretations based entirely on language-independent considerations? In short, why haven’t we, in accommodating the need for intention recognition in the role of disambiguation, completely surrendered our case to Atlas? Ambiguity in language is a fact of life. Grammar often delivers multiple candidate logical forms for a given discourse. To understand the speaker correctly, hearers have to recover the reading the speaker had in mind. It is impossible to put a priori limits on the kinds of considerations that provide relevant evidence or on the reasoning that a hearer should appeal to in the course of this inference. For example, (3), according to us, out of the blue, is ambiguous between a discourse containing the Result coherence relation and one featuring the Parallel coherence relation. General reasoning might very well be invoked in disambiguating between these two grammatically encoded readings, much as general reasoning might be involved in figuring out whether a speaker means the financial institution or the riverbank with a use of ‘bank’; or whether the negation in “The king of France is not bald” is a sentential or a predicate negation, or even whether the speaker’s idiolect of English agrees on the utterance with that of the hearer or whether it treats it differently in one of the many systematic and subtle ways English speakers diverge across the world.

interpretation of the pronoun is fixed. We discuss ambiguity resolution in our theory more directly below.

distinguishing ambiguity from underspecificity

163

To say that arbitrary considerations might matter for disambiguation, however, is not to say that they often do. Ambiguities can be—and, indeed, we think, typically are—resolved in very straightforward ways. In the usual case, hearers know the possible resolutions in advance: the candidates are given by the rules of language. So hearers don’t have to invoke explanatory reasoning to construct the possible interpretations at all. Moreover, in the usual case, there is simple and overwhelming evidence about the right resolution from superficial cues that are readily extracted from the linguistic and non-linguistic context.17 Thus, it’s part and parcel of the ambiguity view that ambiguities—like the right lexical sense in context for the word ‘bank’—should normally be quick and easy to resolve correctly. The same goes, on our view, for disambiguating coherence relations. When multiple options are possible for organizing a discourse, we think, hearers primarily rely on a narrow set of shallow linguistic cues that favor one disambiguation over another. We see this already in (5). Even though, out of the blue, the discourse is deemed infelicitous, interpreters attribute the coherence relation of Explanation to it. That’s why we ask ourselves—is Istanbul famous for good spinach? Is there some hidden connection between spinach and trains? But surely it isn’t common sense reasoning about the speaker’s intentions that settles this disambiguation: the point of the example is that we have no idea how a speaker could intend to explain John’s trip via spinach. Rather, it seems that we are exploiting linguistic cues that signal that the most likely relation to organize the discourse is Explanation, rather then any of the other possible relations.18 In our view, the flexibility and abstraction of discourse coherence has mistakenly led researchers to assume that anything goes. Take (7) and (8). (7) I knocked the glass off. It broke. (8) The glass broke. I knocked it off.

17

18

Even in such cases, our view holds that ambiguity resolution reflects efficient but heuristic probabilistic inference about the intended structure of a linguistic signal, and thus is neither a matter of applying linguistic rules nor a matter of interpreting context dependence. That certain linguistic cues favor particular coherence relations has been noted in the literature (see e.g. Kehler, 2002, Asher and Lascarides, 2003). For instance, the type of verb in the first clause, and the thematic roles played by its NPs often serve as cues for a particular relation.

164

stojnić, stone and lepore

While (7) is understood as an instance of a discourse organized by the Result relation, (8) is understood as organized by Explanation. Both relations are instances of a cause-effect organization of the discourse; thus, (7) and (8) could very well be used to describe exactly the same sequence of events in the world. It’s tempting to respond to these examples with the assumption that discourse is organized by whatever causal relationships make the most sense—a suggestion driven home by the data in (9), based on an example of Larry Horn’s (pc). (9) The glass broke. I knocked it off, and shards wound up all over the floor. This discourse, where ‘knocked it off’ offers an Explanation but ‘wound up’ offers a Result continuation, reminds us that the inferential connections in extended discourse can embrace multiple relations and need not necessarily align with sentence boundaries (a point developed further by Danlos, 2009, in her formal approach to discourse structure and meaning). What we want to emphasize here, however, is that the patterns in (7) and (8) are acutely sensitive to sentence semantics, in ways that crosscut common sense causation. For example, it matters that ‘break’ describes a single event solely in terms of its results, while leaving open the causal mechanism involved. Though continuous processes can stand in causal relations, sentences that report them don’t easily lead to the temporal reversal seen in (8), as the contrast of (10) and (11) shows. (10) Chris steered the car downhill. It sped up. (11) The car sped up. Chris steered it downhill. Conversely, although states can certainly result from events, stative descriptions tend to be interpreted as providing background to ongoing events, rather than as providing their results, as (12) suggests. (12) Chris steered the car downhill. It was speeding up. Thus, even in these cases, it seems like there is an element of convention at play that draws crucially on linguistic form, not just common-sense inference.19 A final piece of evidence for the disambiguation view is cross-linguistic variation. For instance, while one might expect that we understand the second

19

See Stone and Stojnić (2015) for further examples and discussion.

distinguishing ambiguity from underspecificity

165

sentence in (13) as elaborating on the first one, and hence, inferring an Elaboration coherence relation, by way of common sense reasoning, Asher and McCready (2006) point out that its Japanese correlate translation is infelicitous unless it contains an overt discourse marker that signals the appropriate relation. (13) A wolf might walk in. It would eat you first.20 Again, what we take these examples to suggest is that disambiguation cannot be dispensed with, or reduced to, general pragmatic reasoning.21 Our general point is that even if disambiguation were a process that sometimes required speaker intention recognition, we believe we have said enough to render plausible our departure from Atlas. After all, we have said enough to leave disambiguation in semantics on a par with disambiguation in syntax or morphology. We are far from having settled—or even spelled out—all the considerations involved in distinguishing among the theoretical possibilities. However, we think the view of disambiguation we have been defending is correct.

4

Conclusion

Like Atlas, we too want to ascribe as few truth-condition-determining meanings as we can; where we differ, of course, is that we deny that these ascribed meanings underspecify semantic value. On our account, the literal meaning of a discourse determines the resolution of all of many expressions (and, so we claim, not just pronouns, although we only sketched in any detail how our argument would go for pronouns alone here) without any invocation of mind reading.22 What we have sketched is, naturally, provisional in several respects. For one, we have remained silent about many ambiguities associated with discourses containing pronouns—both within English and across languages. And we have remained silent about how far we can take this move to all the expressions in (1a–h)—and so, it remains an open question whether our view can be extended to every case of alleged underspecificity endorsed by Atlas. However, we hope it is clear that the style of argument Atlas runs throughout his writings will not settle the debate between ambiguity and underspecification since this 20 21 22

The example (12) is a classic case of modal subordination, from Roberts (1989). Other reasons for rejecting the view that disambiguation is a pragmatic affair are on offer elsewhere (Lepore and Stone, 2015). We will not rehearse them here. See Stojnić (2016) for a full defense of this position.

166

stojnić, stone and lepore

decision, as we hope we have established, is not merely a matter of choice and parsimony. Atlas employs many different sorts of linguistic examples to argue for semantic underspecificity, as we have already seen with the examples in (1). The kind of underspecificity he has in mind is one where he sees linguistic meaning as freely selecting one interpretation from an open-ended array of candidates by invoking broadly pragmatic and open-ended mechanisms. If we are right, however, the intrusion of both context and background knowledge is not, and need not be, as powerful as Atlas has presumed. In interpreting even a straightforward case like (2), our discourse ultimately lacks a dependency on non-linguistic context. This hints at a larger project of extending the account we merely sketched here to the other sorts of cases of the apparent interpretive variability that Atlas has advanced. If this approach were to succeed in capturing the apparent interpretive variability of such cases while still positing uniform meanings and constrained variation, then Atlas would have to give up on his presumption that there is widespread underspecificity.

References Asher, Nicholas, and Alex Lascarides. 2003. Logics of Conversation. Cambridge: Cambridge University Press. Asher, Nicholas and Eric McCready. 2006. “Modal subordination in Japanese: Dynamics and evidentiality.” University of Pennsylvania Working Papers in Linguistics 12: 237– 249. Atlas, Jay D. 1977. Negation, ambiguity, and presupposition. Linguistics and Philosophy 1 (3):321–336 Atlas, Jay D. 1989. Philosophy without Ambiguity. Clarendon Press: Oxford. Atlas, Jay D. 2005. Logic, Meaning, and Conversation Semantical Underdeterminacy, Implicature, and Their Interface, Oxford University Press: Oxford. Atlas, Jay D. 2007. Meaning, propositions, context, and semantic underdeterminacy. In G. Preyer & G. Peter (eds.), Context-Sensitivity and Semantic Minimalism, 217–239. Oxford: Oxford U. Press. Bach, Kent. 1994. Conversational impliciture. Mind and Language 9: 124–162. Bach, Kent. 1999. The myth of conventional implicature. Linguistics and Philosophy, 22(4): 327–366. Bach, Kent. 2001. You don’t say? Synthese 127: 11–31. Bezuidenhout, Anne. 2001. Metaphor and what is said: a defense of a direct expression view of metaphor. Midwest Studies in Philosophy, 25: 156–186. Bittner, Maria. 2014. Temporality: Universals and Variation. Oxford: Wiley-Blackwell.

distinguishing ambiguity from underspecificity

167

Borg, Emma. 2004. Minimal Semantics. Oxford: Oxford University Press. Cappelen, Herman and Ernest Lepore. 2005. Insensitive Semantics, Wiley: Oxford. Carston, Robyn. 2002. Thoughts and Utterances: The Pragmatics of Explicit Communication. Oxford: Blackwell Cooper, Robin. 1983. Quantification and Semantic Theory. Dordrecht: Reidel. Danlos, Laurence. 2011. D-STAG: a formalism for discourse analysis based on SDRT and using synchronous TAG. In Philippe de Groote, Markus Egg, and Laura Kallmeyer (eds), 14th conference on Formal Grammar—FG 2009, 64–84. LNCS/LNAI Vol. 5591. Berlin: Springer. Grosz, Barbara J., Arivind Joshi, and Scott Weinstein. 1995. Centering: a framework for modeling the local coherence of discourse. Computational Linguistics, 21(2): 227– 253. Heim, Irene. 2008. Features on bound pronouns. In Daniel Harbour, David Adger & Susana Bejar (eds.), Phi Theory: Phi-Features Across Modules and Interfaces, 35–56. Oxford University Press. Heim, Irene & Angelika Kratzer. 1998. Semantics in Generative Grammar. Oxford: Blackwell. Hobbs, Jerry R. 1979. Coherence and coreference. Cognitive Science, 3(1): 67–90. Kaplan, David. 1989. Demonstratives. In Joesph Almog, John Perry, and Howard Wettstein (eds), Themes from Kaplan, 481–563. Oxford: Oxford University Press. Kehler, Andrew. 2001. Coherence, Reference and the Theory of Grammar, Stanford, CA: CSLI. Kehler, Andrew, Laura Kertz, Hanna Rohde and Jeffrey L. Elman. 2008. Coherence and coreference revisited. Journal of Semantics, 25, 1–44. Kameyama, Megumi. 1986. A property-sharing constraint in centering. Proceedings of the 24 Annual Meeting of the Association for Computational Linguistics, 200–206. New York. Kameyama, Megumi. 1999. Stressed and unstressed pronouns: Complementary preferences. In P. Bosch, & R. van der Sandt (eds.), Focus: Linguistic, Cognitive and Computational Perspectives, 306–321. Cambridge: Cambridge University Press. King, Jeffrey C. 2014. Speaker intentions in context. Noûs 48(2): 219–237. Kripke, Saul. 1977. Speaker’s reference and semantic reference. In Peter A. French, Theodore E. Uehling, and Howard Wettstein (eds.), Midwest Studies in Philosophy, volume II, 255–276. University of Minnesota Press. Langendoen, D. Terence. 1971. Presupposition and assertion in the semantic analysis of nouns and verbs in English. In Danny D. Steinberg & Leon A. Jakobovits (eds.), Semantics: An Interdisciplinary Reader in Philosophy, Linguistics, and Psychology, 341–344. Cambridge: Cambridge University Press. Lepore, Ernest and Matthew Stone. 2015. Imagination and Convention, Oxford University Press: Oxford.

168

stojnić, stone and lepore

Nunberg, Geoffrey. 1993. Indexicality and deixis. Linguistics and Philosophy. 16: 1–43. Recanati, François. 2004. Literal Meaning. Cambridge: Cambridge University Press. Roberts, Craige. 1989. Modal subordination and pronominal anaphora in discourse. Linguistics and Philosophy 12(6): 683–721. Roberts, Craige. 2002. Demonstratives as definites. In Kees van Deemter and Rodger Kibble (eds.), Information Sharing: Reference and Presupposition in Language Generation and Interpretation, 89–137. Stanford: CSLI Press. Sauerland, Uli. 2008. On the semantics markedness of phi-features. In Daniel Harbour, David Adger & Susana Béjar (eds.), Phi Theory: Phi-Features across Modules and Interfaces, 57–82. Oxford: Oxford University Press. Smyth, Ron. 1994. Grammatical determinants of ambiguous pronoun resolution. Journal of Psycholinguistic Research 23: 197–229. Sperber, Dan and Deirdre Wilson, 1986. Relevance: Communication and Cognition. Cambridge, MA: Harvard University Press. Stalnaker, Robert. 1974. Pragmatic presuppositions. Reprinted in Context and Content, 47–62. Oxford: Oxford University Press. Stanley, Jason, 2000. Context and logical form. Linguistics and Philosophy 23: 391–434. Stanley, Jason, and Zoltan Gendler Szabó. 2000. On quantifier domain restriction. Mind & Language. 15: 219–261. Stojnić, Una. 2016. Context Sensitivity in a Coherent Discourse, PhD Thesis, Rutgers University. Stojnić, Una, Matthew Stone, and Ernest Lepore. 2013. Deixis (even without pointing). Philosophical Perspectives, 26(1): 502–525 Stojnić, Una, Matthew Stone and Ernest Lepore. 2017. Discourse and logical form: Pronouns, attention and coherence, Linguistics and Philosophy 40(5): 519–547. Stone, Matthew and Una Stojnić. 2015. Meaning and demonstration. Review of Philosophy and Psychology, Special Issue on Pictorial and Spatial Representation, 2015: 69–97. Walker, Marilyn, Iida Masayo and Sharon Cote. 1994. Japanese discourse and the process of centering. Computational Linguistics 20(2), 193–232. Wettstein, Howard K. 1984. How to bridge the gap between meaning and reference. Synthese 58(1): 63–84.

chapter 6

Metaphor, Minimalism, and Semantic Generality: Seeing Things in Context Michiel Leezenberg*

Introduction Metaphor is alive and kicking as a topic of analytical-philosophical and linguistic inquiry. Having led a dormant existence for much of the twentieth century, it shot to prominence in the late 1970s, thanks to a number of seminal collections of papers like Ortony (1979) and Sacks (1979), and to books like Lakoff & Johnson’s Metaphors We Live By (1980). Since then, it has remained an object of intense scrutiny; but this increased attention has hardly led to any convergence of opinions. Although a number of long-held and influential doctrines, e.g., the belief that metaphor is an abbreviated simile, or that it is a marginal phenomenon that purely serves poetic embellishment, have by now largely been discredited, there is no consensus concerning even basic questions of how metaphor works. Already among scholars working in analytical philosophy of language and adjacent fields, let alone cognitive approaches and continental frameworks, there is wide disagreement as to what kind of processes are involved in metaphorical interpretation. I come to these questions with a metaphor-related history of my own. Long ago, I developed a semantic approach to metaphor that, on the one hand, describes how metaphors systematically depend on their context of utterance; on the other hand, it explores how metaphors change their context (Leezenberg 2001). I did so by, first, extending a Kaplan-style indexical semantics to accommodate the—independently motivated—context-dependence of property expressions, captured in the so-called thematic dimension of the context indicating the kind of property that is being talked about; and second, by investigating the exact articulation of asserted, presupposed, and implied information in the utterance of a metaphor. On such an account, in a context of, say,

* An earlier version of this paper was presented at the workshop Radical Pragmatics 40 Years On, Cambridge University, March 2017. I thank those present, in particular Jay Atlas, Kasia Jasczcolt, and Ruth Kempson, for their comments.

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_007

170

leezenberg

personality properties, a metaphor like John is a wolf expresses the assertion that John has a particular property, say, cruelty; it presupposes the thematic dimension of personality properties. I still believe this account to be substantially correct. Independently, and earlier, Josef Stern had developed a similar approach inspired by Kaplan’s logic of demonstratives.1 In this paper, I will contrast this account with two rival approaches that have gained prominence, or a renewed lease of life, in more recent years. First, I will discuss the thesis that metaphor involves semantic underdeterminacy rather than context-dependence, as defended by Jay David Atlas; second, I discuss the recent attempt by two advocates of so-called ‘semantic minimalism’ to rehabilitate Donald Davidson’s controversial account, which denies metaphor all meaning or cognitive content; third, I explore whether these accounts do justice to the context-dependence of metaphor, as argued for by authors like Stern and myself, and try to provide some principled argumentation as to exactly what kind of context-dependence is involved in metaphor.

1

Atlas: Two Dogmas of Literary Modernism

In ‘Metaphor, nonspecific meaning, and utterance interpretation: Two dogmas of literary modernism’, chapter 1.1. of his Logic, Meaning, and Conversation (2005), Jay David Atlas inveighs against two dogmas that, he claims, inform twentieth-century philosophy of language as much as studies of literary style.2 The first, and most important, dogma is the belief in an essential cleavage between literal (or ‘ordinary’) and figurative (or ‘literary’) language; the second dogma is the belief that metaphors are literally false or anomalous, because they involve a ‘category mistake’, or a ‘violation of selection restrictions,’ and that it is precisely the impossibility of determining an adequate literal meaning which triggers the hearer to construe an alternative, figurative interpretation. In fact, Atlas argues, this second dogma amounts to a reduction of metaphoricity to semantic anomaly (2005: 13). Closely related to this dogma, he adds, is the belief that such category mistakes or anomalies nevertheless have meaning (at what we would call the level of speaker’s meaning), and that these meanings 1 See, in particular, Stern (1985, 2000). For the canonical statement of Kaplan’s logic of demonstratives, see Kaplan (1989). 2 Although published only in 2005, this text was written in 1975; the original version is briefly discussed in Leezenberg (2001: 211–213); apart from a number of updated references and footnotes, there are no major divergences between the 1975 and the 2005 version.

metaphor, minimalism, and semantic generality

171

“may vary from person to person, are probably not recursively specifiable, and are highly sensitive to context” (2005: 12). Many of the arguments in favor of the literal-figurative distinction, he continues, are circular, in that they presuppose some version of that distinction. The more important question, however, is whether it does any explanatory work in linguistics; he argues that it does not (2005: 11). Thus, Atlas does not claim, with Relevance Theorists like Sperber & Wilson (1986), that metaphor and literal language differ in degree rather than kind; instead, he rejects the distinction between literal and metaphorical language altogether.3 On his account, metaphors are what he calls ‘semantically general’ or ‘semantically indeterminate’ between a literal and a metaphorical reading: by themselves, they do not express any proposition and do not have determinate truth conditions. Rejecting formalist or structuralist attempts to characterize metaphor in structural syntactic or semantic terms, he argues that metaphors are very much like other semantically general, but literal, sentences; hence, there is no distinction between literal and metaphorical language to be drawn at all; or put differently: there is no difference in logical type, whatever the differences in perlocutionary effect may be (2005: 15). Metaphor, Atlas concludes, is a perfectly ordinary aspect of ordinary language, which, like semantic underdeterminacy more generally, “provides us the economical alternative to vast elaborations of primitive vocabulary and conceptual distinctions” (2005: 16). On Atlas’s account, modernism’s main mistake lies in treating metaphor as something out of the ordinary. Human genius, he argues, lies not with the creation of poetic metaphors, but rather with the exact opposite: with the creation of a realm of ‘purely literal’ use of language, especially in mathematics and physics (2005: 16).4 Thus, he concludes, logical empiricism and literary modernism are guilty of the same mistake: they take the language of science as the norm rather than as the exception (2005: 16).

3 An account in terms of enrichment or loosening, as has become popular in RelevanceTheoretical circles (e.g. Sperber & Wilson 1986, Carston 2010, Bezuidenhout 2001), also seems too unconstrained to account for the relatively determinate content that metaphors acquire in particular contexts, but that is not yet a decisive argument against it: one might simply postulate a set of additional contextual constraints operating after the loosening that, as Relevance Theorists claim, generates a broader range of possible interpretations. 4 I am not sure that the language-ideological characterization of science and mathematics as domains of purely literal language would survive closer conceptual and historical scrutiny, but will not pursue this question here. Cf. Ludwig Wittgenstein: “there is no religious denomination in which so much sin has been committed through the misuse of metaphorical speculation as in mathematics” (1994: 3e).

172

leezenberg

In this context, Atlas—rightly, I believe—traces the doctrine or dogma that poetic language is emotive rather than cognitive or rational in nature to John Locke’s famous—or notorious—rejection of rhetorical eloquence, and by implication, all figurative language, as designed “for nothing else but to insinuate wrong ideas, move the passions, and thereby mislead the judgment;”5 but we could elaborate on the precise character of, and motives for, this doctrine as originally found in Locke and as elaborated in later authors. Recently, Bauman & Briggs (2003: ch. 2) have argued that Locke’s ‘purification’ of literal meaning was part and parcel, not just of a philosophical project of making ordinary language fit for epistemological purposes, but of a downright political project of restricting legitimate public debate to specific kinds of language (literal language that does not move passions seen as feminine) and to specific classes of speakers (gentlemen, i.e., upper-class males). Likewise, the objection that the logical positivists, in their legitimate critique of metaphysics, not only threw away poetic language with the anti-metaphysical bathwater, but reduced vast areas of intuitively significant everyday language to the status of meaningless pseudopropositions, has been raised before (see e.g. Levinson 1983: 227); but it has been observed less often that this criticism of metaphysics was at least in part motivated by political concerns (in particular the struggle against nationalism, antisemitism, and the continuing societal dominance of the Catholic church), and in turn implied a problematic ‘politics of language’ of its own. Such analyses open up the vast field of the political aspects of language and speech, a topic sadly neglected in analytical philosophy of language and its empirical offshoots. Atlas makes a number of further observations on metaphor in a more recent, and as of yet unpublished, book-length manuscript, Disfiguring the Figurative: Metaphor, Irony, Speech Acts and Robert Frost’s Philosophy of Language (2017). In this text, Atlas argues, among others, that the early Frost was something of a late Wittgensteinian, in seeing the literariness of literary texts as a public effect of public language, rather than a private mental phenomenon involving the imagination, or creating a poetic state of mind (2017: 21); that Frost, a bit like Frege, replaced the word with the sentence, or as he calls it ‘the vital sentence,’ as the basic unit of meaning (2017: 18); and that Frost defended something like a semantic generality account of metaphor that avoids both literary modernist dogmas.6 He then proceeds to give novel interpretations of a number of Frost 5 Locke, Essay, 3.10.34, quoted in Atlas (2005: 7). 6 Atlas construes Frost’s contrast between the grammatical sentence and the vital sentence as one between a grammatically well-formed string of words and the utterance of the grammatical sentence in a particular context (2017: 69); but Frost himself appears to reject the

metaphor, minimalism, and semantic generality

173

poems. Repeating his earlier claim that “metaphor is an ordinary feature of ordinary language,” and that the semantic mechanism involved in it “creates a new word with a more general, non-specific, literal [sic] meaning” (2017: 45); he also suggests that the (first) dogma of a fundamental literal-figurative opposition is derived from the (second) dogma that metaphors are literally false or anomalous. More importantly perhaps, he now sees his account of metaphor as part of a non-psychological characterization of literary language: “in literature, especially poetry, the sound and sense of a sentence or of a word are inseparable; in ordinary uses of language they are not” (2017: 16); as a result of this blending of sense and sound, metaphor, and poetic language more generally, cannot be paraphrased.7 Even with these later additions, however, Atlas’s argument is at best incomplete, as it gives no detailed account, or even an outline, of the actual process of metaphorical interpretation. Some of the main formal features of such an account are clear, however. First, it does not simply involve semantic ambiguity or context-dependence, implying that semantically, a metaphor does not express a determinate proposition even given a context of utterance; 0n this view, a metaphor in isolation expresses neither a proposition nor even a function from contexts to propositions (i.e., a character in Kaplan’s sense). Second, Atlas argues that the process of metaphorical interpretation involves a pragmatic ‘inference to the best interpretation,’ analogous to, but not identical with, Peirce’s abduction or Aristotle’s practical syllogisms (Atlas 2005: 68–79): it amounts to inferring the premise, and thus making explicit the deductive relationship in which it figures. The most important formal feature of such inferential patterns is that they are not deductively valid but defeasible, since they can be denied without any sense of logical contradiction. On Atlas’s account, then, metaphorical interpretations are much like classical (particularized) conversational implicatures as originally characterized by Grice. They differ from Gricean implicatures as usually construed, however, in that they involve a pre-semantic pragmatic inferencing (otherwise known as ‘pragmatic intrusion’) that operates prior to the determination of propositional content, rather than on the basis of a semantic interpretation seen as false, anomalous, or uninformative. former more radically, arguing against the very idea that the sentence is a ‘grammatical cluster of words;’ the [vital] sentence, he argues, is a ‘sound in itself apart from the word sounds,’ whereas the grammatical sentence, typically written than spoken, is ‘merely accessory to the other’ (Frost 1995: 681). 7 I think that this non-psychological characterization of literary language (2017: 16) and Atlas’s refusal to conflate semantic interpretation and psychological explanation point to a more radical project of de-psychologizing semantics; but I have no room to discuss these matters here.

174

leezenberg

Having myself both rejected the notion of literal meaning as a myth and criticized the assumption of falsehood or anomaly as a criterion for metaphor (Leezenberg 2001: 87–88; 301–304), I have considerable sympathy for Atlas’s identification, and rejection, of both dogmas of literary modernism. On several points of substance, however, notably his claim that metaphor involves semantic generality rather than context-dependence (and hence is interpreted using pragmatic principles rather than semantic rules), I am not convinced. I will not enter into a discussion of these claims straight away, however, but first present another view, which has become as influential as it is controversial.

2

Davidson and Rorty: Dogmas of Romantic Pragmatism

Atlas’s criticisms appear to apply with full force to one of the more notorious analytical-philosophical views on metaphor: Donald Davidson’s influential rejection of metaphorical meanings, and its offshoot in Richard Rorty’s assimilation of metaphor to unfamiliar natural phenomena, like pulsars and platypuses. Metaphors, Davidson famously argues in ‘What Metaphors Mean,’ “mean what the words, in their most literal interpretation, mean, and nothing more” (1984: 245). That is, whatever meaning metaphors express is just their literal, truth-conditional meaning; and whatever further effect they may have cannot be a meaning or cognitive content. He elaborates this claim by adding that metaphors do not express any determinate cognitive content, because they do not state that things are such and such, but make us see things as such and such and such. A metaphor, that is, does not describe a fact or express a proposition, but conveys an insight that is visual rather than propositional, since “seeing as is not seeing that” (1984: 263). Thus, Davidson appears to make three distinct, and not necessarily compatible, claims about metaphor: first, he argues that metaphors are not distinguished by any specifically metaphorical meaning, but by their use: “what distinguishes metaphor is not meaning but use—in this, it is like assertion, hinting, lying, promising, or criticizing” (1984: 259). This claim implies that metaphor is a distinct kind of speech act, a claim that has been challenged already by Max Black (1979). Second, however, Davidson consistently suggests that it is not speakers using metaphors but the metaphors themselves which produce or express such insights. Third, he suggests, but does not explicitly argue, that metaphors cause, rather than express or represent, beliefs. To see why it is impossible to hold all three of these claims simultaneously, let us look at each of them in more detail.

metaphor, minimalism, and semantic generality

175

First, Davidson’s repeated claims that “metaphor belongs exclusively to the domain of use” (e.g., p. 247, 259) suggest that whatever effect a metaphor has is achieved by speakers rather than sentences, and would lead one to expect a pragmatic account of metaphor in terms of speaker’s intentions and speaker’s meanings, for example, along the lines developed by Grice. Throughout his article, however, Davidson is keen to talk not of the speaker using a metaphor, but the metaphor itself, as ‘making us attend to some likeness’ (p. 247), as ‘nudging us into noting’ likenesses (p. 253), etc. This suggests, against Davidson’s first claim, that metaphors do have a semantic effect after all, in that here, it is the sentence rather than the utterance of that sentence by a speaker, which does this or that thing with the hearer. Yet, Davidson’s phrasing seems no coincidence or slip of the pen: by analyzing the effect of a metaphor as resulting from a speaker intending to communicate something, he would relapse into characterizing this effect as a speaker’s meaning or utterance occasion meaning (or what Grice would call ‘non-natural meaning’), and thus as a propositional content after all. Finally, Davidson consistently suggests that metaphors (or, again, speakers using metaphors) cause rather than represent or justify particular beliefs or insights: metaphors, he writes, ‘provoke’ or ‘invite’ thoughts (p. 261); they ‘make us notice aspects’ (p. 261); they ‘make us see one thing as another’ (p. 263), and finally, “metaphor can, like a picture or a bump n the head, make us appreciate some fact—but not by standing for, or expressing, the fact” (p. 262). Thus, he clearly implies that metaphor is a cause rather than an expression or representation of a cognitive content, a beliefs, or a thought; but oddly, he stops short of saying so explicitly. Only on p. 263 does he explicitly state “what we are caused to notice is generally not propositional in character” [emph. added]. Here, the use of the impersonal passive once again leaves implicit whether it is the metaphor or the speaker doing the causing. The reason for Davidson’s silence and ambivalence is presumably his hesitance to draw the the all-too-radical, and highly implausible, conclusion that metaphor falls outside the realm of communication altogether. This, however, is exactly the conclusion drawn by Richard Rorty: he explicitly claims that metaphors are causes rather than representations of beliefs, and as such fall outside the realm of rational and intentional communication: Davidson lets us see metaphors on the model of unfamiliar events in the natural world—causes of changing beliefs and desires—rather than on the model of representations of unfamiliar worlds. 1987: 284; emph. in original

176

leezenberg

Put differently, live metaphors play no role in justifying beliefs (p. 293), and do not communicate a non-natural meaning (p. 294); it is only when a metaphor dies, Rorty continues, that it crosses the ‘fuzzy and fluctuating line’ between natural and non-natural meaning, and starts to convey information, that is, to act as a justification of beliefs (295). In illustration of this claim, Rorty likens metaphor to birdsong: it may be a stimulus or cause to knowledge, he writes, but in itself, it falls outside the rational practice of giving and asking for reasons. I am not sure whether Rorty does full justice to Davidson’s ambiguous position here, nor do I know whether he is correct in assuming that metaphor-using poets and chirping birds do not engage in intentional communication;8 but his view of poets as resembling chirping birds rather than talking humans rests on an extremely narrow construal of what qualifies as ‘rational’ communication, and, by extension, of what qualifies as ‘human’—that is, intentional—action. However, I will not address these philosophical-anthropological questions, but only note that both Rorty’s and Davidson’s accounts imply that metaphors as such have no place in the inferential practice of giving and asking for reasons, that is, they are never and cannot be used as such in arguments,—and this claim is, in principle, testable. The crucial question then becomes whether or not one can argue about metaphors or by using metaphors; or, to put it differently, whether we can ever use metaphors in giving and asking for reasons. I will return to this question below; for now, the main point is that this radical doctrine is clearly driven by pragmatist philosophical considerations rather than by any detailed study of how metaphors actually function in linguistic communication. For Rorty, metaphors are on a par with anomalous natural phenomena: the Davidsonian view, he argues, “takes them out of the sphere of what Grice calls ‘non-natural meaning’ and reduces them to the level of mere stimuli” (1987: 291). Suspicions against thus treating metaphor, he continues, rest on the mistaken—and Kantian—presumption that there is an inviolable, ‘metaphysical’ break between the natural and the non-natural, or between scheme and content. Thus, Rorty’s—and, if we are to believe him, also Davidson’s—account of metaphor tunes in with, or is driven by, the pragmatist rejection of what Davidson (1984: 189–190) calls the ‘third dogma of empiricism,’ and what Sellars (1997 [1956]) calls the ‘myth of the given:’ the belief

8 On this moot question, cf. the highly instructive discussion between two characters in Gerard Reve’s novel, Het Boek van Violet en Dood (ch. VIII), about the question of whether birds sing in order to praise their creator, or instead produce melodies in order to delimit their territory and invite other birds to sexual intercourse.

metaphor, minimalism, and semantic generality

177

in a radical break between what is given to the mind and what is added by the mind, between conceptual scheme and content, or between language and experience. On Rorty’s account, this break is not metaphysical, but merely “the temporary and pragmatic break between stimuli whose occurrences are more or less predictable […] and stimuli which are not” (1987: 292). Elsewhere, he even describes this break as cultural rather than metaphysical: both scientists and poets, he believes, fall outside all language games, in that they are primarily engaged in presenting new metaphors, rather than developing better or more adequate descriptions of an antecedently given reality.9 One would not be exaggerating, I think, in calling this position romantic; the romanticism implicit in Rorty’s pragmatism becomes especially clear in his conclusion that asking ‘how metaphor works’ is as pointless as asking how genius works: if we could explain them, he writes, they would become superfluous, and “matters of amusement […] rather than indispensable instruments of moral and intellectual progress” (1987: 296). Temporary and pragmatic the break between causes and reasons, or stimuli and representations, may be, as Rorty claims; nonetheless, on his account, it is essential, and indeed unbridgeable at any one moment. If it were not, Rorty’s entire argument, which is predicated on precisely this distinction both here and in Philosophy and the Mirror of Nature, would collapse. For example, it is the irreducible difference between the space of reasons and the space of causes that allows Rorty to reject Quine’s proposal to reduce epistemology to developmental psychology (1979: ch. V). Hence, I would not call this distinction ‘metaphysical’, but rather epistemological or logical: after all, it is logically impossible, we have been told since Hume, to reduce normative states to purely causal or descriptive ones; thus, trying to reduce epistemic states of knowledge (which involve normative notions like truth, correctness, and validity), to purely causal terms involves a logical mistake, on a par, Sellars famously argues, ‘with the naturalist fallacy in ethics’ (1997: §5). Thus, Davidson’s and Rorty’s views on metaphor are crucially shaped not only by the Quinean rejection of reductionism and the analytic-synthetic distinction, but also by the Sellarsian rejection of the distinction between scheme and content. From Quine, both inherit the rejection of reified meanings; from Sellars, they inherit the rejection of conceptual schemes as distinct from neutral contents, and the concomitant attempt to separate the notions of truth and meaning (cf. Rorty 1979: 301). Despite this explicit disavowal of the three empiricist dogmas, however, Davidson and Rorty clearly maintain the two modernist

9 See e.g. Rorty (1989: 8).

178

leezenberg

dogmas identified by Atlas. Thus, Davidson clearly believes that there is a strict distinction between literal and figurative: he claims that, even if the boundaries between literal and metaphorical uses of words cannot be drawn clearly, as others have plausibly argued, “this fuzziness […] cannot erase the line between what a sentence means (given its context) and what it ‘draws our attention to’” (p. 260). He also adheres to Atlas’s second dogma, witness his remark that a sentence, even if it is not actually false, must be taken to be false for the hearer to “accept it as a metaphor and to hunt out the hidden implication” (pp. 257– 258). Much the same can be said of Rorty, who sees the distinction between literal and figurative as irreducible, witness his remark that “it is essential to Davidson’s view that dead metaphors are not metaphors” (p. 292n); his belief that metaphors are anomalous sentences is betrayed by his claim that they are like anomalous natural phenomena (p. 290). Rorty’s claim that metaphors are causes rather than representations of beliefs is thus a pragmatist restatement of the dogma that there is an essential difference between literal and figurative language. Perhaps, then, we may also think of the distinction between normative reasons and naturalist causes as another dogma—not of empiricism, as there seems to be nothing specifically empiricist about it, but rather of romantic pragmatism. This is not necessarily to say, of course, that there simply is no distinction between causes and reasons, or between facts and norms, or that naturalist fallacies as attacked by Hume, Moore and Sellars are not fallacies after all; it is merely to say that the spatial imagery of the ‘space of causes’ and the ‘space of reasons’ may obscure as much as it highlights. More specifically, this imagery downplays the irreducible temporality of linguistic practices and the historical variability of norms and normativity, and thus overlooks the fact that this distinction may be more radically contingent and historically variable than Sellars, Rorty, and others allow for. Whatever its philosophical motivation, the doctrine that metaphor does not involve metaphorical meanings runs into serious difficulties, and has met with substantial criticisms. In fact, in a later paper, ‘A Nice Derangement of Epitaphs’ (2005 [1986]), Davidson appears to withdraw or contradict virtually every substantial claim he had made on metaphor. Although this paper strictly speaking concerns malapropisms or ‘slips of the tongue’ rather than metaphors, its implications for his earlier views are obvious.10 The existence of such malapropisms,

10

For criticism of Davidson’s earlier views, see e.g. Black 1979; Stern 1991. For more detailed discussion of the relation between Davidson (1984) and (2005), see Leezenberg (2001), esp. pp. 124–134; cf. Camp 2013.

metaphor, minimalism, and semantic generality

179

Davidson argues, like saying ‘epitaphs’ when one means epithets, proves that speakers do not need shared sets of rules or conventions for being able to communicate. Instead, they communicate what he calls ‘first meanings’ rather than conventional meanings. These first meanings are what the speaker communicates on a particular occasion rather than what his words mean as a matter of convention, and thus come close to what Grice and others would call ‘utterance (or utterer’s) occasion meaning.’ Thus, Davidson, famously concludes, the existence of malapropisms proves that ‘there is no such thing as a language’ (2005: 107); that is, successful communication does not require shared conventions. Rorty, however, construes this rejection of shared rules as a more radical rejection of language tout court as an intermediate between our beliefs and the world: “[Davidson] would like us to stop thinking that there is something called ‘language’ which is a ‘scheme’ which can organize, or fit, or stand in some other noncausal relation to, a ‘content’ called ‘the world’” (1991: 60). Thus, by leaving out just the one letter a, he radically transforms Davidson’s assertion: on this reading, it amounts to the rejection not of shared languages in favor of idiolects or of shared conventions in explaining communication, but of the distinction between the world as given and language as a conceptual scheme to organize that world. True, Davidson on occasion appears to make statements similar to Rorty’s, witness his comment that “we have erased the boundary between knowing a language and knowing our way around in the world generally” (2005: 107); but his point in this particular paper is that we cannot, and need not, explain communication in terms of shared rules or conventions, reproducing a point about meaning made earlier by Grice (1957). Davidson appears to believe that a speaker using a metaphor also communicates a first meaning, witness his admission that that in ‘What Metaphors Mean,’ he was “foolishly stubborn about the word meaning when all I cared about was the primacy of first meaning” (2005: 173n). This comment, however, masks a far more radical shift: what he calls ‘first meaning’ here is not, as he suggests, word meaning, that is, literal or conventional meaning, but rather speaker’s meaning or meaningnn, which is determined by intentions rather than conventions. The belief expressed here, that literal meanings are speaker’s meanings (or, more precisely, utterer’s occasion meanings) is, of course, a far cry from the received view, which is also Davidson’s earlier view, that literal meanings are conventional word or sentence meanings. Further, in (2005), Davidson describes successful malapropisms as accepted or agreed on deviations from ordinary usages, and hence as what he calls ‘first meanings;’ but this implies that a metaphor that is accepted or appreciated by its audience may likewise qualify as a first meaning; that is, that as a metaphor, it does in fact communicate a meaning or cognitive content.

180

leezenberg

As an account of communication that does away with the need for shared rules or conventions, Davidson’s argument fails miserably. Malapropisms are not ‘sheer invention’ as he believes (2005: 141); rather, they clearly depend on, and exploit, the conventional phonetic resources of a language (thus, for example, replacing ‘epithet’ by the similar-sounding ‘epitaph’), as much as metaphors exploit its semantic resources. It also implies, against Davidson’s earlier argument, that metaphors, or speakers using metaphors, may communicate (first) meanings after all. In short, the later Davidson appears to withdraw much of what he had earlier said about metaphor; but these changes are not necessarily improvements.

3

A Dogma of Semantic Minimalism?

We thus have three competing, and incompatible, accounts of metaphor: the Davidsonian account that denies metaphor any meaning; the semantic-generality view that treats metaphorical interpretation as involving a pre-semantic inference to the best interpretation; and an account in terms of context-dependence, as developed by Stern 2000 and Leezenberg 2001.11 In the following, I would like to tease out the philosophical assumptions of these positions: minimalists, with Davidson, will probably—though, as we shall see, not necessarily—reject the idea of metaphor as having contextually varying truth conditions; semantic generalists see contextual variation in metaphor as pragmatically driven rather than semantically and compositionally generated; and Stern and I explain such variation in compositional semantic terms. First, let us explore minimalism and its Davidsonian roots. First, semantic minimalists tend to reject the postulation of various kinds of hidden contextdependence in everyday language as an explanatory strategy in semantics and pragmatics. Apart from a relatively small set of lexical items like I, this, here, now, etc., they argue, context plays little if any role in interpretation. Rather, they continue, whatever context-dependence we do encounter is overtly grammatically encoded; and every sentence expresses a (basically context-free) ‘minimal semantic content,’ which is shared by all utterances of that sentence.12

11

12

This distinction virtually coincides with what in the literature have come to be called, respectively, semantic minimalism, contextualism, and literalism (cf. e.g. Stern 2009); but I find this terminology needlessly confusing. In line with this general attitude, Cappelen & Dever (2013) argue that context-dependence

metaphor, minimalism, and semantic generality

181

Semantic minimalism is clearly, if not very emphatically, inspired by Davidson’s truth-conditional theory of interpretation, which aims at keeping the semantics simple by employing an extensional first-order predicate logic. This may make it tempting for minimalists to take recourse to Davidson’s account of metaphor as well; but is that strategy really forced upon them? Put differently: is it essential to semantic minimalism that metaphor not only does not involve any hidden or overt context-dependence, but also falls outside the realm of meaning altogether? Josef Stern (2009) clearly thinks it is not. He argues that metaphor in fact passes two tests proposed by Cappelen & Lepore for determining whether a given expression is semantically context-dependent, viz., the ‘inter-contextual disquotational indirect report’ test and the ‘blocking of collective descriptions’ test.13 On the basis of this, he then argues that an account of metaphor as involving semantic context-dependence (which he, as noted, calls ‘literalist’) is in fact compatible with semantic minimalism, and that minimalists “can agree that the only semantic content of a metaphor is its metaphorical interpretation, without conceding that claim for every non-literal interpretation” (2009: 284). This would, of course, require minimalists like Cappelen & Lepore to agree that metaphor involves a kind of non-overt contextdependence similar to that found in indexicals, which—given their general dislike of the appeal to context-dependence as an explanatory strategy—seems unlikely. Thus, a Davidsonian account of metaphor may not be necessitated by semantic minimalism; but it certainly fits in well with it. This need not cause any surprise, of course, given the latter’s roots in Davidson’s truth-theoretical theory of meaning. A Davidsonian inspiration is also quite visible in Lepore & Stone’s (2010) minimalist account of metaphor. Metaphor, they claim, does not have any specifically metaphorical meaning or discourse effect; or, as they put it, it does not add anything to the ‘conversational scoreboard’ (cf. Lewis 1979). Unlike Rorty, however, they do see metaphor as belonging to the realm of intentional communication, but not as communicating anything in virtue of the hearer recognizing the speaker’s intention, as Grice claims is the case with non-natural meaning. This leads to puzzling results: Lepore & Stone’s attempt to preserve a Davidsonian-style account of metaphor as not involving meaning or cognitive con-

13

is rather less widespread and rather less philosophically interesting than has been thought in the wake of Kaplan’s seminal writings. Stern (2009); cf. Camp (2007).

182

leezenberg

tent, while at the same time maintaining that it should be treated within a theory of action formulated in Gricean terms of intention recognition, is as baffling as Davidson’s about-face concerning metaphorical meaning. A bit like Davidson’s first meanings, the results of such communication closely resemble what Grice would call ‘non-natural meaning’ or ‘meaningnn;’ but Lepore & Stone are adamant that they not be called ‘meanings’. Although a speaker using a metaphor may intend to draw a hearer’s attention to similarities, this is not, they claim, an intention to convey a propositional content p, i.e., he does not “elicit an effect given by p in his audience by means of recognizing the speaker’s intention to elicit this effect” (2010: 170). Unlike or against Davidson (1984) and Rorty (1987), they unequivocally argue that a speaker uttering a metaphor is engaged in intentional communication, or, as they call it, ‘cooperative interaction;’ but like them, they reject the idea that she is communicating a meaning, i.e., a content that “must be identified either with what a speaker intends her audience to recognize as trying to communicate; or as what must be added to the conversational record in consequence of coordination between her and her audience” (2010: 170). Metaphors, they argue, are like jokes or hints: the goal of their being uttered is for appreciation to occur rather than for information to be exchanged by the recognition of intentions (2010: 171). This argument puts excessively strong a priori constraints on what one can and cannot do with metaphors; one can, of course, intend to do all kinds of things in using a metaphor, whether to ‘put things before our eyes,’ to invite or to suggest, to praise or to insult, to appeal to the audience’s expectations or to play on its emotions. There seems to be no good reason why all, and only, metaphors should involve the particular intention to non-propositionally suggest likenesses. In claiming, with Lepore & Stone, that they do, one not only risks reducing them to particular kinds of speech acts, but also reproducing the dogmatic belief in a strict and radical distinction between literal and metaphorical language. It is one thing to argue that the effects of metaphor need not be (exclusively) propositional in nature; few would dispute that metaphors can help us visualize things in unexpected or vivid ways, or, to speak with Aristotle, ‘to put things before our eyes’ (pros ommatôn poiein, Rhet. 1411b6 ff.). It is another thing altogether to claim that these effects cannot be propositional. Both in Davidson and in Lepore & Stone, this more radical claim appears to be motivated less by an analysis of the actual behavior of metaphors in their wider settings than by a priori considerations about what a theory of meaning should look like, and in particular by the desire for a restricted semantic theory that minimizes appeals to context-dependence and does away with intensional notions. This desire becomes visible in their claim that progress in the realm of meaning “depends on developing and applying clear standards that demarcate

metaphor, minimalism, and semantic generality

183

semantic phenomena narrowly. If we can locate metaphor elsewhere, it is good news for meaning” (2010: 179). This drive to keep semantics simple and clearly delimited may sound appealing enough; but in this paper at least, Lepore and Stone do not develop any clear standards of demarcation to decide whether or not metaphor should be treated as a semantic phenomenon; rather, their treatment is driven by general philosophical considerations. Particularly revealing in this context is their comment that “our intuitions about literal meaning place heavy constraints on key semantic notions: truth, reference, context” (ibid.). This suggests that Lepore & Stone are as much beholden to a dogmatic concept of literal meaning as Davidson (1984) and Rorty (1987). It also suggests that, with respect to metaphor at least, the notion of a ‘minimal proposition,’ too, may reproduce a problematic—if not dogmatic—notion of literal meaning rather than providing a novel explanatory account.14

4

Dogmas Aside

Despite the obvious differences between Davidson’s and Atlas’s accounts of metaphor, there are some intriguing convergences between semantic minimalism and semantic generalism. In his discussion of semantic minimalism, Atlas (2007) claims that, unlike Cappelen & Lepore, he is not a ‘propositionalist’ about literal meaning, let alone a ‘minimal propositionalist;’ but like them, he rejects the postulating of widespread ambiguity or context-dependence in natural language as an explanatory strategy. Every sentence type, he elaborates, has a literal (sic) meaning; but not all sentences necessarily express propositions that can be evaluated truth-conditionally. Thus, Atlas identifies a large class of sentences (or sentence types, to be precise) that have a literal (or, more precisely, a semantically indeterminate or semantically general) meaning, but neither express a determinate proposition nor have any determinate truth conditions in isolation from a particular context of utterance. This contextdependence, however, is qualitatively different both from semantic ambiguity and from indexicality, in that it is guided by pragmatic principles for inferring speaker’s intentions rather than semantic rules.

14

Stern (2009) argues that the minimal proposition expressed by a metaphor is in fact its metaphorical interpretation; but on his view, metaphor crucially involves a covert contextdependence of the kind rejected by Cappelen & Lepore, in particular through the operator Mthat, which he sees as present in deep structure or logical form, but generally not overtly realized.

184

leezenberg

There are several recent accounts of metaphor that resemble Atlas’s, most importantly in claiming that metaphor involves Gricean inferencing even if it does not rest on a defective or anomalous literal meaning. Thus, the ‘direct expression view’ proposed by Bezuidenhout (2001) rejects the view of metaphor as a case of classical (particularized) conversational implicature triggered by an apparently inappropriate, uninformative, or trivial literal interpretation; but the interpretive mechanism she suggests is clearly of a Gricean character, as it involves a defeasible process of inferring or ‘calculating’ the speaker’s intentions on the basis of assumptions—held to be universal—about the speaker’s rationality. Likewise, Recanati’s self-proclaimed ‘quasi-contextualist’ position resembles Atlas’s in its rejection of the notion of a minimal proposition as ‘theoretically useless’ (2004: 86), and in its conviction that large areas of language interpretation, including metaphor, involve pragmatic principles rather than semantic rules. On Recanati’s account, semantic interpretation only yields a ‘semantic schema’ or a ‘propositional radical’ rather than a full-blown proposition.15 In his conclusion that most context-dependent expressions are ‘semantically indeterminate rather than indexical in the strict sense’ (2001: 85), Recanati comes close to Atlas’s position that metaphors are semantically underdeterminate and that their interpretation involves a defeasible inference to the best interpretation. Given the open-ended and indeterminate quality often ascribed to metaphorical interpretation, it certainly seems tempting—not to say obvious—to conclude that metaphor involves what Recanati calls “a more radical form of [pragmatic] context-dependence” based on the indeterminate and defeasible calculation of speaker’s intentions rather than the ‘very straightforward,’ i.e., rule-governed and deductive, semantic context-dependence typical of indexicals like I or here (2001: 85). Tempting as this conclusion is, however, it should be put to the test: one would like to see whether we can actually prove, rather than merely claim, that metaphorical interpretations involve Gricean inferencing rather than semantic-context-dependence. Recanati argues that from a psychological point of view, it is impossible to separate semantic interpretation based on rules and a more pragmatic, optional process of enrichment or loosening representations; both, he suggests, are “indissociable, mutually dependent aspects

15

Recanati identifies this propositional radical with what Russell called a ‘propositional function’; but this may be misleading, as one might construe this with what authors like Dana Scott call propositional functions, i.e., functions from contexts to propositions, or what Kaplan (1989) called ‘characters’, which is precisely not what Recanati intends.

metaphor, minimalism, and semantic generality

185

of a single process of pragmatic interpretation” (2001: 88). If we follow this argument to its logical conclusion, we may well end up claiming that literal meaning, or sentence meaning, is an illusion; but Recanati stops short of this claim. Thus, Bezuidenhout and Recanati are not quite as bold as Atlas in rejecting the notion of literal meaning as an unfounded dogma. But dogmas aside, how can we adjudicate between the minimalist approach, the semantic generality view, and a ‘literalist’, or, as I would prefer to call it, a direct contextual interpretation account of metaphor? It would be nice to have a diagnostic for determining whether metaphor involves no meaning at all, pragmatic intrusion, or semantic, rule-based context-dependence. Now, in fact there is such a diagnostic for distinguishing entailment, presupposition, and implicature: they display rather different inferential behavior. For example, unlike entailments, implicatures are cancelable; unlike both implicature and entailment, presupposition is preserved under negation and modals, and so on. I am not sure we can come up with a similar diagnostic for metaphor as such; but by exploring whether its interpretation displays the inferential properties of semantic entailment, implicature, or presupposition, we may be able to test these positions at least indirectly.16 Let us begin with the Davidsonian/Rortyan claim that metaphor falls outside rational communication. A different way of putting this is that there can be no inferential relations involving metaphorically interpreted sentences; instead, only the literal meaning may show relations of entailment, contradiction and the like; but since this generally (though not necessarily) involves a falsehood, obviously, anything will follow from it. More generally, this claim implies that metaphors as such do not establish discourse relations with the sentences or utterances surrounding them, for example, by setting up referents for anaphoric pronouns, by expressing temporal or causal relations between different sentences or phrases, etc.; or that metaphors make no contribution to the conversational scoreboard.17 Now this is, in principle, a testable claim: if we can establish that there are metaphors that do show logical, or inferential, relations of e.g., entailment or contradiction with other sentences, or if succeeding anaphorical pronouns do pick up referents established by metaphorically

16 17

I have discussed the diagnostics for conversational implicature and their applicability to metaphor in Leezenberg (2001: 102–108). Lepore & Stone (2010), following Brandom (1994) and Lewis (1979). It should be noted, incidentally, that Brandom’s influential formulation of an inferentialist semantics has little to say about the inferential particularities of conversational implicature, and nothing at all about the place of metaphor and other figures of speech in communication.

186

leezenberg

interpreted noun phrases, it may be considered refuted.18 Here, the odds seem heavily against Davidsonian accounts: it seems perfectly possible to reason or argue with, and about, metaphors, as in (1) John is not a wolf but a weasel Prima facie, one would think that one metaphorically expressed property, say, cruelty, is denied and another (say, cowardice) is affirmed of John. That is, the negation does not cancel or undo the metaphorical interpretation altogether, but merely rejects one specific metaphorically expressed property in favor of another. To judge from such examples, metaphors can indeed be part of a practice of asserting and denying, and thus of giving and asking for reasons; here, the burden of proof would seem to lie with those who claim otherwise. Metaphors also appear able to set up discourse referents, and thus to make contributions to the conversational scoreboard after all. When John Maynard Keynes wrote about Wittgenstein’s arrival in Cambridge: (2) Well, God has arrived. I met him on the 5.00 PM train He clearly intended his metaphorically used God in referring to Wittgenstein, among others, to establish an antecedent for the anaphoric pronoun him in the following sentence. In other words, whatever other effects (2) may have, it does make a contribution to the conversational scoreboard; otherwise the referent of him would remain a mystery, to be solved by other means. Another case that seems difficult if not impossible to account for in Davidsonian terms is presented by metaphors involving if-then clauses: (3) If art is the tip of the iceberg, then I’m the part that’s sinking below (Lou Reed) Here, again, the metaphorical antecedent phrase would seem to set up a set of discourse referents to be picked up in the consequent. Both examples suggest that metaphors, as metaphors, do establish discourse relations with the clauses, phrases, sentences and/or utterances surrounding them. When seen in their wider linguistic contexts, that is, metaphors appear to express contents after all. In fact, whatever plausibility Davidson’s view has results in part from

18

This aspect of a Davidsonian account has been noticed, and criticized, before: see e.g. Bergmann 1982, Bezuidenhout 2001, Leezenberg (2001: 233–238).

metaphor, minimalism, and semantic generality

187

his limiting his attention to decontextualized sentences of the simple A is a B type; it progressively decreases if we look at the behavior of actual metaphors in their wider linguistic setting and nonlinguistic contexts.19 All of the above seems to militate, rather decisively, against approaches that deny metaphors all cognitive content, inferential role, or place in intentional communication. Next, if one accepts that metaphors may maintain inferential or discourse relations with their linguistic environment, one would like to know whether those relations primarily involve semantic or pragmatic principles. The fact that metaphor can be used in any kind of reasoning or inferential process, whether or not defeasible, militates against Davidson’s and Rorty’s view; but by exploring exactly which kind of inferential processes are involved, we can perhaps make a reasoned decision, as opposed to a possibly dogmatic claim, as to whether metaphor involves semantic generality or indeterminacy, or rather some kind of context-dependence. We can pose this question in testable terms by asking whether metaphorical interpretations have the inferential properties of conversational implicatures, rather than those of asserted or presupposed semantic information. Defeasibility is one relatively familiar diagnostic to establish whether a particular inference is semantic or pragmatic in character. Grice argues that, unlike deductively valid semantic entailments, conversational implicatures are defeasible in that they can be denied without a sense of logical contradiction (1989: 39). The question then becomes whether metaphorical interpretations are defeasible in anything like this sense. If a metaphorical interpretation is indeed a Gricean inference or something like it, it must be cancellable; that is, the metaphorical content can be denied without any sense of logical contradiction. But when Dutch novelist W.F. Hermans denied his own earlier characterization of his former publisher, Geert van Oorschot, as a ‘vulture’ (aasgier) for allegedly cheating him out of part of his royalties, he clearly did not exactly cancel his metaphor: (4) Van Oorschot is geen aasgier. Aasgieren zijn tenslotte hoogvliegers. Van Oorschot is not a vulture. After all, vultures are high-flyers. Here, he did not cancel the metaphorical interpretation altogether, but merely one way of construing or explicating it: obviously, his sarcastic elaboration does not amount to a denial of his earlier metaphorical claim that his publisher is a scavenger, picks the bones of his victims, etc. Defenders of a pragmatic

19

It should be noted, however, that the discourse effects of metaphor remain a relatively underexplored topic.

188

leezenberg

inference view have claimed that what is defeated or canceled in metaphorical denials like (4), a particular metaphorical interpretation can in fact be cancelled in favor of another metaphorical interpretation (cf., e.g., Bezuidenhout 2001: 183); but such claims appear to rest on an erroneous, or at least tendentious, reading of Grice’s original characterization of defeasibility: on Grice’s own account, implicatures can be canceled simpliciter, not merely canceled in favor of some other implicature. Another diagnostic that seems of particular relevance to metaphor is the claimed universal character of principles of rational action, as contrasted with language-or culture-specific semantic rules. Semantic interpretation, it has been argued, involves a knowledge of language-specific rules or conventions; pragmatic inferencing, by contrast, involves not language-or culture specific conventions, but universal principles of human action (Levinson 1983: 120–121; cf. Recanati 2001a: 82). On such an account, the same metaphors would lead to similar—if defeasible—inferences about the speaker’s intentions in different languages, based on the assumption that the latter is behaving rationally. Again, treating metaphor as involving a universal inferencing of this kind rather than a more convention-based form of interpretation would lead to the rather implausible suggestion that metaphorical interpretations are determined largely independently from linguistic or cultural inventions. But, for example, a term like sweet, as used in (5) John’s words are sweet may receive rather different metaphorical interpretations in English, where it means or communicates that John’s words are pleasant; but in Chinese, sweet would mean that his words are specious.20 Likewise, in English, (6a). (6) a. John is a fox b. John-wa kitsune desu communicates that John is sly and cunning; but its Japanese translation (6b) conveys that John is a malicious spirit; in fact, the latter may well be taken as literally stating that John is a fox spirit by many a Japanese speaker. Thus, against a semantic-generality account, there seem to be good reasons not to see metaphor as involving a Gricean kind of inferencing; but exactly what kind of inferential process is involved? The fact, indicated by examples

20

Cf. Stern (1985: 685), Leezenberg (2001: 216).

metaphor, minimalism, and semantic generality

189

(1) and (4) above, that metaphorical interpretations survive negation suggests that they share a number of relevant structural features with presupposition rather than implicature. Hence, it now remains to get clearer about exactly what these features are, and thus to explicate exactly what is presupposed and what is asserted in the utterance of a metaphor. Much like presuppositions, metaphorical interpretations appear to survive negation, modals, and embedding in if-then clauses: (7) a. b. c. d.

John is a wolf John is not a wolf Perhaps John is a wolf If John is a wolf, we’d better keep an eye on him.

This suggests that at least part of what is communicated in metaphor is presupposed information; and this presupposed part is what I call the thematic dimension, that is, the kind of property involved (say, biological taxonomic properties as opposed to moral personality properties). The existence of thematic dimensions as contextual factors should not be in doubt: it is independently motivated by the distinct inferential behavior of so-called predicatelimiting adverbials (Leezenberg 2001: 168ff.; cf. Bartsch 1987) like healthwise or financially: (8) Healthwise/Financially, John is doing well The adverbial explicates the thematic dimension here, leading to two different properties being ascribed to John in each case: respectively, being healthy, and being wealthy. Further, it has been argued on independent grounds that thematic dimensions involve presupposed rather than asserted information, and that presuppositions survive exactly the kind of embedding shown in (7a–d)21 One cannot conclude from this, however, that the property metaphorical ascribed to John in (7a) comes from a set of ‘presupposed properties’ as Stern (1985, 2000) would have it.22 Rather, upon further exploration, it emerges that metaphor

21 22

Cf. Bartsch (1987: 5); Levinson (1983: 177–199). Stern’s view of presuppositions in terms of sets of properties rather than propositions does not clearly distinguish between what is asserted and what is presupposed in a metaphor, leaving him unable to account for negation and the kinds of embedding noted in (7a–d); cf. Leezenberg (2001: 202–208).

190

leezenberg

crucially involves the so-called ‘thematic dimension,’ i.e., the kind of properties talked about, as the relevant contextual variable. In a thematic dimension of taxonomic properties, (7a) is simply literally false; but in a dimension of personality properties, it is true if John is cruel, lonely in old age, etc., and false if, for example, he is cowardly rather than cruel. On such an account, which one may label one of direct contextual interpretation, metaphor involves the interpretation of a predicate in a novel or unexpected thematic dimension. This account, moreover, suggests that what is presupposed in metaphor is the thematic dimension (e.g., that of personality properties in (7a–d)), and what is asserted is that John has the property of cruel, lonely, etc., given that dimension. Such an account goes a long way in explaining the actual inferential behavior of metaphors, including its embedding under negation, modals, and so on. Thus, the case for metaphor as not involving meanings, contents or contributions to the conversational record appears to collapse when we explore the effects of specific metaphors in their wider discursive context; likewise, the case for metaphor as an example of semantic underdeterminacy or generality, and as involving pragmatic intrusion (that is, pre-semantic Gricean inferencing) appears at best unproven and at worst implausible. Rather, the inferential evidence discussed above would seem to suggest that metaphor involves a particular—and independently motivated—kind of context-dependence of property expressions, and a particular articulation of presupposed and asserted information.

Conclusion: Just One More Dogma Metaphor, in short, appears to involve semantic context-dependence rather than pragmatic inferencing or entirely non-linguistic (cognitive) processes. Against semantic minimalists, one may reply that metaphors do express specifically metaphorical propositions or contents, and involve a kind of contextdependence, the reality of which can be established on independent grounds; the onus of proof would seem to lie with those who deny, against all appearances, that metaphors make any contribution to the conversational scoreboard. Against semantic generalists, one may argue that metaphors appear to involve context-dependence rather than semantic generality, and hence are interpreted by resorting to semantic rules rather than pragmatic principles. There may be no way of telling directly whether a metaphor in isolation does or does not express a proposition; but one may give some reasonably plausible indirect arguments. Most importantly, the inferential behavior

metaphor, minimalism, and semantic generality

191

of metaphors appears not to conform to the patterns predicted or demanded either by a semantic-generality account that denies them any determinate propositional content at sentence level and relies on Gricean inferencing, or by a semantic-minimalism account that denies metaphor meaning altogether. This inferential behavior provides suggestive, if partly indirect, evidence that, pace Davidson, metaphors can be used to make assertions, or more generally, contributions to the conversational scoreboard; and, pace Atlas, that metaphor involves a semantic form of context-dependence rather than a pragmatic—and defeasible—inference to the best interpretation. Davidson’s and Rorty’s accounts of metaphor presuppose both dogmas of literary modernism; as such, they not only rest on a very restrictive notion of literal meaning, but also presuppose an extremely narrow construal of what constitutes rational communication. Davidsonians, and a good many others who see communication as, first and foremost, a practice of giving and asking for reasons, give an unwarranted priority to assertive language and the exchange of information; Griceans, construing rational linguistic behavior as cooperative, take a culturally and historically very specific ideal of polite conversation over tea as embodying a universal notion of rationality. Both positions, then, appear to rest on folk-theoretical assumptions about rationality, cooperativeness, and speakers’ agency and intentions, that have too unthinkingly, and for far too long, been taken for granted as primitive, self-evident, and universal explanatory notions. One may ask if the (ultimately Kantian) concept of rationality underlying the pragmatic-inference approach to metaphor is in the end not as narrow and ethnocentric as the Davidsonian (and ultimately Lockean) notion of literal language. If it is, we may well have isolated another dogma that has strongly—if largely silently—dominated the analytical philosophy of language. My argument here is not so much that metaphor, or poetic language in general, is more ‘rational’ than authors from Locke to Rorty would have it; rather, it is that the linguistic forms and practices considered ‘rational’ and ‘literal’ in analytic philosophy, and in semantics and pragmatics, rest on a prior exclusion of other kinds and usages of language, which reduces such uses and functions to a secondary, marginal, or even illegitimate status. The literary modernist beliefs in literal meaning and in the semantic anomaly of metaphors really are dogmas, that is, tacit and tenacious assumptions held to be true as a matter of course rather than explicit doctrines put up for challenges in a rational debate. Likewise, the third dogma discussed above, i.e., the pragmatist belief in a strict break between the factual and the normative, or between causes and reasons, is generally assumed rather than argued for. We may now have uncovered a final dogma of philosophical pragmatics: the— ultimately Kantian—belief that linguistic communication normally involves

192

leezenberg

the rational and cooperative action of fully conscious and intentionally acting (that is, autonomous) actors. This initially plausible picture, however, is in fact a highly culturally specific one (cf. Leezenberg 2010). Put differently, to speak with Wittgenstein, these dogmas are among the misleading assumptions that ‘bewitch’ our thought; or, to speak with Michael Silverstein, they are ‘linguistic ideologies’ that rationalize or legitimize certain linguistic practices. But such ideologies, even if dubious or debatable when made explicit, can still have a productive or positive role to play, in that they may shape our actual linguistic behavior. This leaves us with the challenge of, first, identifying and distinguishing specific language-ideological beliefs and concepts of literal meaning, and, second, of tracing how these have guided, constrained, and regimented actual linguistic practices; but I leave the exploration of this—vast—task for another occasion. If, however, what analytical philosophers have called rational communication, i.e., the practice of giving and asking for reasons, is as particular, variable and historically contingent a way of regimenting language as the concept of literal language is, then analytical philosophy, semantics and pragmatics have an entire, and largely still unexplored, field of study lying in front of them.

References Atlas, Jay David 2005 Logic, Meaning, and Conversation. Oxford University Press 2005. Atlas, Jay David 2007 ‘Meanings, Propositions, Context, and Semantic Indeterminacy’. In Preyer & Peter (eds.), pp. 217–239 Atlas, Jay David 2017 Disfiguring the Figurative: Metaphor, Irony, Speech Acts and Robert Frost’s Philosophy of Language. MS, draft 4B.1, February 2017 Bartsch, Renate Bartsch, ‘Context-Dependent interpretations of lexical items’. In J. Groenendijk a.o. (eds.) Foundations of Pragmatics and Lexical Semantics. Dordrecht: Foris, pp. 1–26 Bauman, Richard & Briggs, Charles Voices of Modernity: Language Ideologies and the Politics of Inequality. Cambridge University Press 2003 Bergmann, Merrie ‘Metaphorical Assertions.’ Philosophical Review 91 (1982): 229–245 Bezuidenhout, Anne ‘Metaphor and What Is Said: A Defense of a Direct Expression View of Metaphor’. Midwest Studies in Philosophy XXV (2001): 156–186 Black, Max ‘How metaphors work: A reply to Donald Davidson’. In Sacks (ed.) 1979, pp. 181–192 Borg, Emma Minimal Semantics. Oxford University Press 2004 Camp, Elisabeth 2007 ‘Prudent semantics meets wanton speech act pluralism’. In Preyer & Peter (eds.), pp. 194–215

metaphor, minimalism, and semantic generality

193

Camp, Elisabeth 2013 ‘Metaphor and varieties of meaning’. In E. Lepore & K. Ludwig (eds.) A Companion to Davidson (Oxford: Blackwell), pp. 361–378 Cappelen, Herman & Lepore, Ernie 2005 Insensitive Semantics. Oxford University Press Cappelen, Herman & Dever, Josh 2013 The Inessential Indexical: On the Philosophical Insignificance of Perspective and the First Person. Oxford University Press Carston, Robyn 2010 ‘Metaphor: Ad hoc concepts, literal meaning, and mental images’. Proceedings of the Aristotelian Society (New Series) 110: 295–321 Davidson, Donald 1984 ‘What Metaphors Mean’. In Inquiries into Truth and Interpretation (Oxford University Press), pp. 245–264 Davidson, Donald 2005 ‘A Nice Derangement of Epitaphs’. In Truth, Language, and History (Oxford University Press), pp. 89–109 Frost, Robert 1995 Collected Poems, Prose, & Plays. New York: The Library of America Grice, H.P. 1957 ‘Meaning’. Philosophical Review 64: 377–388 Kaplan, David 1989 ‘Demonstratives.’ In J. Almog a.o. (eds.) Themes From Kaplan. Oxford: Oxford University Press, pp. 481–563 Lakoff, George & Johnson, Mark 1980 Metaphors We Live By. Chicago: University of Chicago Press Leezenberg, Michiel 2001 Contexts of Metaphor. Elsevier Science Leezenberg, Michiel 2010 Gricean and Confucian Pragmatics: A Contrastive Approach. In D.F. Shu & K.P. Turner (eds.) Contrasting Meaning in Languages of the East and West. Berlin: Peter Lang, pp. 3–32 Lepore, Ernie & Stone, Matthew 2010 ‘Against Metaphorical Meaning’. Topoi 29: 165– 180 Levinson, Stephen C. 1983 Pragmatics. Cambridge University Press Lewis, David 1979 ‘Scorekeeping in a Language Game.’ Journal of Philosophical Logic 8: 339–359 Ortony, Andrew (ed.) 1979 Metaphor & Thought. Cambridge University Press; second, expanded edition 1993 Preyer, G. & Peter, G. (eds.) 2007 Context-Sensitivity and Semantic Minimalism. Oxford University Press Recanati, François 2001 ‘What is said’. Synthese 128: 75–91 Recanati, François 2004 Literal Meaning. Cambridge University Press Rorty, Richard 1979 Philosophy and the Mirror of Nature. Princeton University Press. Rorty, Richard 1987 ‘Unfamiliar noises: Hesse and Davidson on Metaphor’. The Aristotelian Society, Supplementary Vol. LXI: 283–296 Sacks, Sheldon (ed.) 1979 On Metaphor. University of Chicago Press Sellars, Wilfrid 1997 [1956] Empiricism and the Philosophy of Mind. Cambridge MA: Harvard University Press Stern, Josef 1985 ‘Metaphor as Demonstrative.’ Journal of Philosophy LXXXII: 677– 710

194

leezenberg

Stern, Josef 1991 ‘What Metaphors Do Not Mean.’ Midwest Studies in Philosophy XVI: 13– 52 Stern, Josef 2000 Metaphor in Context. Cambridge MA: MIT Press. Stern, Josef 2009 ‘Metaphor and Minimalism.’ Philosophical Studies 153: 273–298 Sperber, Dan & Wilson, Deirdre 1986 ‘Loose Talk’. Proceedings of the Aristotelian Society, New Series, LXXXVI: 153–171. Wittgenstein, Ludwig 1994 Culture and Value (Revised edition). Oxford: Blackwell

chapter 7

A Radically Pragmatic Account of Number Words and the Reversibility of Scales Jerrold Sadock

1

Radical and Reactionary Theories of the Number Words

At the tag end of the period when radical pragmatics was in vogue,1 I considered an extremely radical analysis of number word expressions in natural languages according to which they do not provide sentences with either upper bounded semantics, lower bounded semantics, or exact semantics, but are general with respect to such readings. I failed to provide a precise idea of what the meaning of number expressions would be on that theory, but concluded that it would not provide enough meaning to support mathematics. (See Sadock 1984.) In its place I offered a reactionary solution according to which the number words lend expressions exact meanings; a sentence like I have two grandchildren would be true if I have exactly two grandchildren, and false otherwise. That would require both the lower bounded (at least two) and upper bounded interpretations (at most two) to be pragmatic weakenings of a strong sense. Jay Atlas (Atlas 2005: 186–190) criticized the reactionary view on the grounds that nobody has yet produced anything like a full-fledged theory of pragmatic weakening. Atlas preferred an analysis according to which both the “at least”, “at most” and “exactly” uses were to be explained as pragmatically induced precisifications of a noncommittal sense. Carston (1988) has also endorsed the underspecification theory, and on the other hand, Koenig 1991, Horn 1992, and Geurts 1998 accept, or are at least not uncomfortable with, the bilateral, upward and downward semantically bounded meanings for cardinal number expressions (see also Geurts 2007). My present thinking agrees with both Carston and with Atlas. But if the meaning of number words is non-committal with regard to semantic boundedness, what meaning do numerical adjectives have that provides a basis for arithmetic? Jay’s solution was to say that every number expression is ambiguous between ordi-

1 Larry Horn (2006) has called this period the GAPP, the Golden Age of Pure Pragmatics. I prefer “The Golden Age of Radical Pragmatics: GARP.”

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_008

196

sadock

nary English words, which he disambiguates graphically by spelling the words out, and cardinal numbers, for which he uses Arabic numerals. Those numbers written as alphabetic words are underspecified, lacking upper and lower semantic bounds; those numbers written with numeric characters are precise, and therefore suitable for mathematical purposes. I don’t find this a very satisfying solution. I would prefer a univocal account of all numerals and will offer one below. There are languages with two number systems, one of which is usually inherited and the other borrowed. Japanese, Korean, and West Greenlandic Inuit are examples. In the last mentioned language, the native system is rarely used for numbers above twelve, and the borrowed system, based on the Danish number system, is the one that almost exclusively is used in mathematical calculations of various kinds. The difference in usage is similar, but not identical to the two poles of the Atlasian ambiguity. But there is a very big difference between the single English set of number words and the bifurcated Greenlandic vocabulary for them. There are separate terms for each system, the first seven of which are shown in (1). (1) Native and borrowed number words in West Greenlandic Inuit: Native

Borrowed

ataasey marluk pingasut sisamat tallimat arfineq arfineq marluk …

en to tre fire fem seks syv

There are some areas of overlap in the use of these two systems, but suppose there weren’t. Suppose the native language system were used exactly in those cases where Atlas would say the underspecified, ordinary language forms would be used, and the borrowed system was employed exactly where his exact numbers would be used. Would that be an argument supporting the postulation of an ambiguity in the English system and others where there is terminology for all uses? Not at all. By the same reasoning we could argue spuriously that the English word aunt is ambiguous between “mother’s sister” and “father’s

a radically pragmatic account of number words and scales

197

sister” because Danish has moster and faster respectively. In a language like English, where there is but one set of number terms, the invocation of ambiguity is especially suspect. It would be a remarkable coincidence that the endless collection of number expressions, whether the underspecified ones or the precise ones, are always pronounced identically. One of the principal goals of radical pragmatics is to avoid falling back on ambiguity to account for different uses of the same expression. Ambiguity explains nothing. But there is the possibility of explanation if it can be shown that various uses of an expression can be worked out on the basis of pragmatic principles and the speaker’s and hearer’s goals in the circumstances in which the number expressions occur. Furthermore, if we could show by structural arguments, for example, that number expressions of English were in fact ambiguous. Then some method would have to be assumed to choose the likely reading on the basis of context. As far as I can see, that method would make use of the very same contextual principles that a pragmatic system would use to determine the contextualized construal use of an underspecified meaning. That would make the ambiguity redundant. In the absence of direct evidence of ambiguity per se, the postulation of double meaning would thus be otiose. I therefore reject Atlas’s method of dealing with the ‘problem’ of mathematics as ad hoc and profligate. (See Zwicky and Sadock 1975 for a discussion of arguments for and against ambiguity.)

2

The Meaning of Numbers

Here in a nutshell is what I want to say about the meaning of number words in natural languages: (3) The meaning of a cardinal number word n+1 is “SUCCESSOR (n)”, where n is a cardinal number word. The meaning of seventeen in English is “SUCCESSOR(sixteen)”.2 Knowing English includes knowing the analogous meaning of eighteen, which is “SUCCES2 A few years after delivering this talk at Pomona College, I discovered that I was not the first to adopt this ultra-pragmatic conception of number words. In the linguistic context, that would be my illustrious predecessor at the University of Chicago, Leonard Bloomfield. Here is how Bloomfield (1935, 509) put it: “In every community children are taught to recite certain words in a fixed order; these words have no meaning, outside of the fact that there exists this conventional way of recit-

198

sadock

SOR(seventeen)”, and so on, and knowing the recursive scheme that allows successors to be produced without limit. We can then derive a further meaning property of seventeen, viz., “PREDECESSOR(eighteen)”, according to the following definition: (4) (SUCCESSOR(n) = n+1) ⇔ (PREDECESSOR(n+1) = n) Children know their numbers when they can reel off the first twenty in their proper order, and later when they know the words for the decades in their order, the words hundred, thousand, million, etc., and the principles of recursion that allow one to produce the name of the successor of any arbitrarily large n. The number words of English, then, do not mean any of “n at least”, “n at most”, or “exactly n”. Their meaning is nothing more than a specification of their position in the well-ordered sequence of number words. They derive their power to enumerate the elements of a set from their strict order, and to convey set cardinality owing to the fact that once a set is counted, the cardinality of ing them. These are the cardinal numerals. The tenuity of their meaning makes them ideal for scientific discourse. Nevertheless, this meaning involves certain features of man and his world, such as the linear character of our movement in time. We cannot escape some undefined terms of the character of before (or predecessor or the like). The cardinal numerals have a variety of uses in ordinary speech. The simplest is that of putting sets of objects into oneto-one correspondence without actually confronting them. Since in this operation human beings show great uniformity of response, it plays an important part in science; probably every science that is to speak in numerical terms will have to presuppose the existence of distinct objects and our trick of assigning objects to classes, often in arbitrary ways, a long chapter of psychology and anthropology. In some languages the cardinal numerals are said to be very few, e.g. three or four. In English and many other languages they are systematized, by features of morphology and syntax, in such a manner that, given any numeral word or phrase, a hearer can utter a numeral word or phrase in a stated before-and-after relation to the given numeral; for instance, we can (uniformly and predictably) name the immediate successor of any cardinal numeral. There is nothing mysterious about this, since all that is required is a verbal habit: the phrase immediate successor of … would suffice.” I wrote what follows in happy ignorance of Bloomfield’s prior discovery, unaware that both his theory and his explanation of the pragmatics were more-or-less identical to what I would say seventy years later. I deserve a great deal of criticism for my slipshod scholarship, but I give myself a little credit for independently hitting upon something that one of the most influential linguists of the previous century had to say. I should also mention that neither Bloomfield nor I acknowledged fact that Peano (Kennedy 1973) and Dedekind 1888 showed that the properties of arithmetic follow from the postulate that the SUCCESSOR function defines the natural numbers.

a radically pragmatic account of number words and scales

199

the set has been determined. The properties of the number words in natural languages derive not entirely from their meager meanings, but to a large extent from the uses to which those very spare meanings are conventionally put. Here are some simple, intuitive, didactic facts that I believe support the view that the use of number expressions to count and to recognize cardinality are serially derivative of a simple, non-mathematical meaning. My expertise in these matters stems entirely from my experiences as the grandfather of two bright boys now eight and three. When children learn numbers, they learn them as no more than a sequence of articulations. A young child who says, “I can count to ten,” means “I can recite a series of syllables in an order that adults approve of, the last of which is [tεn]”. They learn to recite nursery rhymes, prayers in a foreign language, the words to the Pledge of Allegiance and so on, without necessarily understanding much or even any of what they are saying. Notice that when the learner first learns to count, she is not counting things; she’s just counting, just reciting a list. The intransitive verb count and the transitive verb count do not describe the same activity. At this point the child may be taught to associate graphemes with the units of the syllable sequence. If she has not already figured it out, this exercise will let her know that the phonic stretch [sεvǝnèjt] consists of two units and that they are [sεvǝn] and [ejt]. The next phase in the child’s education is the demonstration of the applicability of the numbers—the segments of the counting recitation—to the task of counting things. You don’t teach a child his numbers individually by, say, pronouncing “one” and pointing to a single block, then “two”, pointing to a group of two blocks, and so on. The counting-things lesson is conducted by obtaining a set of similar things and associating the members of the set with the number words, starting with one and proceeding through the ordered list until the last element of the set has been associated and its sequential number pronounced with list-final intonation. Initially the results of the counting classes may not seem promising. The child may take counting things to be a game that consists of reciting the numbers in order, and with each one pointing to a member of the collection, missing some, pointing to some twice, and then, with an intonation that signals triumph, the child names the next number in the sequence, and stops. Eventually the child will figure out that exhaustivity and non-redundancy are important aspects of counting things: the idea of set cardinality is now emerging, and when it is fully developed, the child has one of the central concepts needed for arithmetic.

200 3

sadock

Independence of Meaning and Use

The separation of meaning and use is not confined to the cardinal numbers. Here are other linguistic facts that demonstrate the autonomy of semantics and pragmatics: There are elements of language that are useful, but meaningless: Every natural language, I believe, contains expressions that don’t seem to have any meaning in the sense of determining truth or verifiability, but nevertheless have uses. Obvious examples are expletives like hi, ouch, or darn, that are used, respectively, to greet, to express a genuine or feigned feeling of physical pain, and to convey mental annoyance. If someone says “Ouch!” when I accidentally step on his toe, I might say, “C’mon, now; that didn’t hurt.”, but “False,” would be inappropriate. A similar argument can be made on the basis of onomatopoeic forms like kaBANG, and kaPOW and chugga-chugga-chugga.3 An example similar to, but independent of the cardinal numbers is the alphabet, which will be discussed more thoroughly below. The sequence “a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t, u, v, w, x, y, (and,) z” has no meaning, but it has a range of very important functions.

4

Expressions with Two Meanings and One Use

There are also expressions with two meanings and only one use. Some trademarked names of products have an ordinary meaning, like the Procter and 3 An esteemed colleague at Yale has brought to my attention David Kaplan’s (2004) video lecture on ouch and oops. Kaplan tries to convince his audience that “Ouch!” and “I am in pain” contain the same semantic information, that in some sense they have the same meaning. I disagree. There are substantive differences. To give just one counter argument, “I am in pain” refers to the speaker; “Ouch” does not. There is a vicarious “Ouch!” which you can utter when you see someone walk into a post, but in that circumstance you cannot use “I am in pain” to achieve the same effect. I believe that Kaplan’s attempt to assimilate whatever is communicated by an utterance to semantic information is mostly equivocation, but I do not have the space to argue that here. His lecture covers some of the same ground that Jerry Morgan’s important, and to my mind more cogent 1978 publication does. I should also mention that in Sadock 1994 I propose that we recognize three distinct dimensions of content: a representational dimension for expressing propositions, an affective dimension in which attitudes are presented (but not represented), and an effective dimension in which social effects are achieved. These dimensions correspond intuitively to Kaplan’s three varieties of “semantic information”, the descriptive, the expressive, and the performative, this last mentioned only in passing in Kaplan’s lecture.

a radically pragmatic account of number words and scales

201

Gamble’s brands Cheer®, Bounty®, and Downy®. The product names have nontrademarked meanings that suggest pleasant things alongside their legally protected reference. We know what it means to be of good cheer, to enjoy bounty, and to be soft and downy. If these meanings were irrelevant, then Procter and Gamble could just as well have named their wares Gloom, Dearth, and Scratchy. Frozen metaphors and certain idioms belong in this category as well. Metaphors with an established use, like prince in “Harry’s a real prince,” which is used to indicate that the subject has a noble character not because, or just because, princes are supposed be of noble character, but also because of a convention of usage (Morgan 1978) that goes directly from the form to its metaphorical use without necessarily using the literal meaning as a stepping stone. Still, the user of such frozen metaphors understands the words themselves, and they are important for learning and remembering.4 A problem of long standing in speech act theory is the proper account of the mechanism by which performative utterances come to do what they do when they’re used. How does saying “I now pronounce you husband and wife,” (or something like that) when uttered by the right person under the right circumstances as part of a complete and correct ceremony, actually manage to accomplish the uniting of a couple in marriage? In an unpublished talk that anticipated his classic 1978 paper, Jerry Morgan gave a surprising, but intriguing answer: He said, in essence, that performatives work by a kind of social magic. Morgan asserted that it doesn’t really matter what is said as long as there exists a ritual in which the utterance of particular words—they might even be words of a foreign language—is taken as the effective utterance or a necessary part of the accepted procedure. Why, then, do we say “I christen …” as part of the act of christening, “I baptize …” in a baptism, and “I now pronounce you husband and wife” to join a couple in matrimony? Because, said Morgan, the christener, baptizer, or marrier would feel silly saying, if it happened to be the accepted formula, “Yes, we have no bananas.” The same consideration explains why we say, “Wake up!” when waking someone up, rather than “Don’t wake up!” The point is that the meaning of utterances in these cases, while sensible, is in fact

4 Larry Horn correctly observes that compositional expressions that are not fully compositional range from those where the literal meaning is recognizable to most, if not all speakers and plays at least a metaphorical role in their meaning, to those that most speakers do not see the connection between the meaning of the individual words and the conventional sense. For example, I like sailing and therefore know the origin of the phrase three sheets to the wind, but I assume that there are many other speakers who directly connect that sequence of words with a meaning like “quite drunk”.

202

sadock

irrelevant to their efficacy. Children have genuine magic formulas like olly olly oxen free and dibs. Children apparently have a higher tolerance for silliness than adults.

5

Quantity Implicature and Scale Reversal

Horn (1972, 1989) recognized the importance of implicational scales in the generation of pragmatic inferences. One way to define implicational scales is this: (5) a. A minimal implicational scale is a pair of lexical items W1, W2, and an expression function E(Wn) such that: If E(W2) is true, then E(W1) is true, And it is not the case that if E(W1) is true, then E(W2) is true b. More generally, an ordered set of lexical items ⟨W1, W2, W3, … Wn⟩ is an implicational scale if (5a) holds of all ordered pairs ⟨Wj, Wj+1⟩. Horn’s idea of how the pragmatic effects depend on implicational scales is this: where an implicational scale exists, a speaker who uses an expression with a value that is not the highest on that scale will generally convey the implicature that he or she could not have cooperatively used a stronger member of the scale, either because she or he lacks evidence for the stronger statement, or because she or he believes it would be false. For example, the ordered set ⟨acceptable, good, very good, excellent⟩ (Becker 2002)5 is an implicational scale with respect to the expression function “The meal was W”, since “The meal was excellent,” implies that the meal was at least very good, good, and acceptable. Therefore, if I were to tell you that the meal was acceptable, I implicate, but do not say, that it wasn’t good, or very good, or even excellent.

6

The Generality of Number Expressions

As linguists know, a phrase like “Jay’s picture” has a number of reasonable, contextually suggested paraphrases: “the picture of Jay”, “the picture Jay took”, “the 5 Here I follow the modern convention of presenting Horn scales as going from weak on the left to strong on the right, rather than the other way around as in Becker’s (2002) ⟨ excellent, very good, good, acceptable ⟩. Becker follows Laurence Horn’s original style in, for example, Horn 1972. I thank Professor Horn for pointing this out.

a radically pragmatic account of number words and scales

203

picture Jay voted for in the Kutest Kitty Kontest,” and so on. I do not think that these are specific meanings of the phrase; I do not think that the phrase is literally multiguous among these various interpretations. Instead I suggest, following Atlas’s lead, that it is general with respect to all these more particular interpretations. So what does Jay’s picture literally mean? It’s got to be something very, very unspecific, like: “the picture relating to Jay”, with the specific relationship to be filled in pragmatically.6 In an analogous fashion I’d like to claim that three apples literally means something quite non-specific, something on the order of: “group of apples relating to the successor of two and the predecessor of four”. That’s too spare a meaning to be useful and that’s where the pragmatics comes in: to help you figure out just how, in the context of a particular utterance, this phrase can be used and interpreted as “group of at least three apples”, “group of exactly three apples”, “group of at most three apples”. The fact that someone has chosen three as the member of the sequence of numbers rather than any of the infinitely many others indicates that the speaker intends that number word to play a part in determining the contextual significance of the phrase. As with all pragmatic reasoning, it depends on the ostensible purposes of the utterance, as Grice 1975 originally suggested in the formulation of his cooperative principle, viz., to “Make your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange.” In the context of an accounting class, say, the general meaning of a noun phrase like three hundred dollars is likely to have the bilateral interpretation since, among other things, addition and subtraction figure prominently in that discipline. Addition sort of works on the neo-Gricean treatment: “At least three hundred dollars plus at least two hundred dollars” comes out semantically as “at least five hundred dollars” which is (at least) true. Subtraction, on the other hand is a disaster; “At least three hundred dollars minus at least two hundred dollars” is any amount whatsoever.7

6 Recanati (2004, 7) writes: “Genitives provide another well-known example: an utterance including the phrase ‘John’s book’ does not express a complete proposition unless a particular relation has been identified as holding between the book and John.” That point of view differs from the one expressed here only in that I would allow John’s book to have an evaluable meaning, but one so weak as to demand contextual particularization in order to constitute a contribution to an ordinary discourse. The difference is either too philosophically intricate to receive adequate treatment here, or too little to worry about. 7 The subtraction would yield $ 300 minus $ 200 plus n dollars minus m dollars, i.e., any number of dollars.

204 7

sadock

Scale Reversal and the Theory of Radically Underspecified Number Words

And finally, there is the phenomenon of scale reversal (Horn 1972, Sadock 1984, Carston 1998). Carston 1998 provides the following examples of situational reversal of the cardinal number scale: (6) a. She can have 2000 calories without putting on weight. b. The council houses are big enough for families with three kids. c. You may attend six courses (and must attend three). In the last example, we understand the first clause as indicating a maximum number of courses. It is equivalent to “at most six courses.” The upper bound is not an implicature since “You may attend at least six courses (and must attend three)” is incoherent. My examples in 1984 compared golf scores with bowling scores and are now a little trite, so I’ll bring in a different sport: 501 darts. In this game each player starts a leg (a single game) with a score of 501 Points indicated by where the darts land on the board are subtracted from the latest score. The object is to reach zero exactly. So if Anderson is ahead of Davis in a dart game, he has fewer points than Davis; in most other contests, the player who is ahead has more points than the opponent. In my 1984 article I rejected the theory according to which the number terms do not impose either upper or lower bounds on their values. I did that without considering what, on the radically pragmatic account, the meaning of the number words should actually be. Now I have a meaning: Three means “the successor of two and the predecessor of four”. Three and every other number refer by definition to a position in an infinite sequence defined by the SUCCESSOR relation. That sequence can be usefully employed in the enumeration of sets, and the result of an accurate enumeration is the cardinality of the set. Though I rejected the radically pragmatic treatment of numbers in Sadock 1984, I still found that it displayed at least one favorable property: Surprisingly, there is actually an advantage to the outrageously pragmatic theory of number words in that it treats them differently from the inexact quantifiers like some, many, most, and all. With these we find exactly the same upward dual understanding as with the numerical quantifiers (some being variably ‘some only’ or ‘at least some’), but we do not find any scale reversal. It does not seem possible to use some, for example, in such a way as to implicate ‘at most some’. Sadock 1984, 143

a radically pragmatic account of number words and scales

205

This is an important point that deserves elaboration. Here is a more detailed example. Bruce Snowdon, aka Harold Huge, the last sideshow fat man, passed away in 2009 at the age of 64 weighing 607 pounds.8 He was, by anyone’s estimation, extremely high on the scale of human heftiness. I assume that the scale has at least the positions overweight, fat, obese, and extremely obese. Suppose, contrary to fact, that Mr. Snowdon had decided to diet and exercise.9 Under the supervision of a doctor, he loses a lot of weight and at his next check up the doctor says, “Congratulations! You’re obese.” Even under these circumstances, it would be hard to construe the doctor’s pronouncement as conveying that Mr. Snowdon is no longer extremely obese.10 In Sadock 1984 I referred to the difference between number scales like the score in a football game, and non-numerical scales such as the degree of acceptability of linguistic example sentences as exact and inexact scales, respectively. But I no longer believe that that is the crucial difference between the reversible and non-reversible scales. Exact definitions can be given to the various terms of an inexact scale, and the National Institutes of Health have in fact quantized the terms overweight, obese, and extremely obese in terms of body mass index (BMI) as show in (7).11

(7) BMI 18.5–24.9 25.0–29.9 30.0–39.9 40.0 and above

8 9 10

11

Classification Normal weight Overweight Obese Extreme obesity [sic]

Max lbs for 6’ height 184 220 294 –

I can think of no better place to discover the properties of scales than the measurement of weight. Bruce Snowden earned his money being a fat man. He worked hard at keeping his weight up. Larry Horn (p.c.) adds the following interesting and useful addendum to this discussion: “Another illustration of this is the point I made (Horn 1989: 240) inspired by my then 6 year old son David who found the bathwater too hot and insisted that I make it warmer: Thus warmer and hotter are both read as ‘higher on the < …, hot, warm, … > scale’, cooler and colder as ‘higher on the ⟨cold, cool, …⟩ scale’. It is only for children who have not yet acquired the subtleties of full scalar competence that warmer can denote ‘less hot’, ‘closer to (exactly) warm’.” http://www.nhlbi.nih.gov/health/health‑topics/topics/obe/diagnosis, accessed 11/28/ 2015, and http://www.nhlbi.nih.gov/health/educational/lose_wt/BMI/bmicalc.htm, accessed 3/7/2016.

206

sadock

Providing such numerical values does not, however, turn the implicational scale into a non-implicational scale that can be situationally reversed. Suppose Mr. Snowdon worked really hard, getting down to 282 pounds, giving him a body mass index of 39.9, just within the obesity category. Snowdon’s doctor would still not be able to say, “Congratulations: you’re obese!” But she could say, “Congratulations: You weigh 294 pounds.” The scale of avoirdupois is reversible; the scale of heftiness is not. Here is another difference between the cases where reversal is possible and where it is not. For non-numerical scales, the following are good discourses: (8) Is Noam smart? OKI’ll say! OKOh, my God, yes! OKVery! (9) Does Noam have an IQ of 120? ?I’ll say! ??Oh, my God, yes! ??Very much so! The reason for this difference is that the non-numerical scales are natural Horn scales—implicational scales in virtue of the meaning of the elements. If someone admits to being obese, he can hardly deny being fat. But my eight-year-old grandson will staunchly deny that he is five years old. Obesity, no matter how quantified, entails fatness; being eight years old does not entail being five years old. All that Max is eight years old entails is that he is of an age that has something to do with the successor of seven. We can rephrase the statement such that it is a Horn scale. “Max has lived for eight years” entails that he has lived for five years, and I think Max would agree.

8

Reversible Non-numerical Scales

There are, as a matter of fact, non-numerical sequences that allow scale reversal for the same reason that the sequence of numbers does. Like the natural numbers, they are pure sequences with no semantic implicational consequences other than those that follow from the sequence. The two examples I will present also lack the mathematical applications of the sequence of numbers, but both serve special purposes for which, in fact they were invented.

a radically pragmatic account of number words and scales

207

The first such sequence is the alphabet. There is no natural order to the sequence of letters of the alphabet, and therefore the opposite order is possible and is frequently used. For example, co-authors are often named in alphabetical order, but in an effort to be fairer to people like Arnold Zwicky, the authors are sometimes listed in reverse alphabetical order. As with number words, we learn the alphabet words in particular order and abstractly. Children learn “their letters” in much the same way as they learn “their numbers”. First they are taught a jingle of syllables which might be divided into letter-names, but might also include polysyllabic words like ellemennow. Later kids learn that the pieces of the jingle have graphic representations. Relatives will wax ecstatic when the child demonstrates her prowess by recognizing and correctly naming random letters, which at this point is the only reward the child receives. The reversibility of the alphabetic sequence is just a reflection of the arbitrary, accepted order in which the letters are learned. A is the first letter; B = SUCCESSOR(A); C = SUCCESSOR(B), and so on. Therefore we define the PREDECESSOR relation as before: (10) (SUCCESSOR(x) = y) ⇔ (PREDECESSOR( y) = x) Later the function of the letters as the elements of the written language is learned, but while important, it is wholly irrelevant to the reversibility of the alphabetic succession.

9

The Highs and Lows of Poker

The second example of a reversible sequence is the ordered list of winning poker hands, where “>” symbolizes the transitive relation “beats a”: (11) straight flush > four of a kind > full house > flush > straight > three of a kind > two pair > one pair > ace high > king high > … six high12 In five card poker with no wild cards the ordering of the winning hands reflects the likelihood of getting a particular kind of hand. In other common varieties of poker, deuces wild, for example, the chances change and the order based on probability of having a particular type of hand changes as well. The likelihood

12

Six high is the lowest hand because five high would be 5, 4, 3, 2, A, which is a straight.

208

sadock

in percentages for five card straight poker and five card deuces wild is given in (12):13 The columns are arranged in increasing likelihood of the various combinations. (12) Probability of hands in A. straight poker and B. deuces wild Straight poker

%

0.000 straight flush .00015 four of a kind .0204 full house .1441 straight .3925 three of a kind 2.113 two pair 4.754 pair 42.26 no pair 50.12

Deuces wild

%

five of a kind .0258 straight flush .0982 full house .4876 flush .5568 four of a kind 1.214 straight 2.395 two pair 3.657 no pair 30.77 pair 47.13

We see that in deuces wild, a flush is less likely than four of a kind, and no pair is less likely than one pair. Despite the change in probabilities, the “beats a” relation remains what we see in (11) showing that (11) is actually a sequence, and not an implicational scale. A simpler demonstration of the sequential, non scalar nature of the ordered list of winning hands comes from the role that the ranking of the cards plays. If, say, two players each have a pair, the player with the higher ranking pair wins, where rank is determined by the sequence (13) from which the implicational “beats a” scale (14) is derived. (13) ⟨A, K, Q, J, 10, 9, 8, 7, 6, 5, 4, 3, 2⟩ (14) A > K > Q > J > 10> 9 > 8 > 7 > 6 > 5 > 4 > 3 > 2 It might not look it, but the sequence in (13) is, in fact, entirely arbitrary. The probability of getting a pair of deuces is exactly the same as the likelihood of getting a pair of queens: 42.26%. Therefore the pragmatically derived scale

13

This table is a version of Cactus Kev’s Poker Table #1, http://www.suffe.cool/poker/table1 .html. Accessed 02/11/2016.

a radically pragmatic account of number words and scales

209

ought to be reversible, and indeed, it is. There are “lowball” games in which the opposite order of hands determines the winner. Ordinarily in such a game, straights and flushes do not count and aces are the lowest cards. Thus the 5 high—5, 4, 3, 2, A—is the best lowball hand. (15) 5 high > 6 high > … > Q high > K high > pair > two pair > three of a kind > full house > four of a kind

10

Non-quantitative Uses of Number Expressions

Certain uses of number words themselves are remote from their application to arithmetic. If you ask me for my phone number, you are not asking me for a quantificational expression at all. The sum of your phone number and my phone number is meaningless unless you are a numerologist. This is why, when I’m asked for my phone number, I might say something like, “five seven two (pause) three six two five”, and not “five million, seven hundred and twenty three thousand, six hundred and twenty five.” The number words are handy here, since each one can be made unique within a certain group, but it’s hardly the only solution. The alphabet could be used as well. I could also answer, using the alphabetic characters on the telephone keys: “J, S, A, D, O, C, K.”

11

Conclusions

I hereby withdraw my own objections to the idea that, as a matter of definition, number words have no upper limit or lower limit. What I did wrong in Sadock 1984 was, first of all, to fail to be specific as to what the meaning of the number words actually is. Adopting the ideas of Giuseppe Peano and Leonard Bloomfield (which I was unaware that I was doing) solves the problem. The number words are defined by their position in the infinite sequence of number words. That meager meaning enables the number words to be used in counting, to specify quantity, and to ground mathematics. Those are uses to which the number words are put, not separate meanings. We do not need to grasp at ambiguity as an account of the use of a number word n to indicate at least n, at most n, or exactly n; a single definition of the number words can handle math, darts, and telephone numbers. The idea that a number’s position in the sequence of number words is its genuine meaning allows us to explain why pragmatically derived scales allow context to determine which end of the scale is up. Intrinsic scales, however, are

210

sadock

based on inbuilt entailments between words, and therefore resist reversal. The difference is that the numbers, in virtue of their definitions, form a sequence and not an implicational scale. That something excellent is also good, but not vice versa, follows from what the words mean, so the only way that the scale may be inverted is through negation or any operator that reverses semantic entailment. (See Jespersen 1924, Horn 1972, and Fauconnier 1975.) In this particular case of radical underspecification, then, I now see that Jay Atlas was right all along.

References Atlas, Jay David. 2005. Logic, Meaning, and Conversation: Semantical Underdeterminacy, Implicature, and their Interface. New York: Oxford U.P. Becker, Thomas. 2002. Autohyponymy: Implicature in Lexical Semantics, Word Formation, and Grammar. Journal of Germanic Linguistics, 105–136. Bloomfield, Leonard. 1935. Linguistic Aspects of Science. Philosophy of Science, Vol. 2, 499–517. Carston, Robyn. 1988. Implicature, Explicature, and Truth-Conditional Semantics. In Ruth Kempson, ed., Mental Representations: The Interface Between Language and Reality. 155–181. Cambridge: Cambridge University Press. Carston, Robyn. 1998. Informativeness, Relevance, and Scalar Implicature. In R. Carston and S. Uchica, eds. Relevance Theory: Applications and Implications, 179–236. Amsterdam: John Benjamins. Dedekind, Richard. 1888. Was sind und was sollen die Zahlen? Braunschweig: Vieweg. Fauconnier, Gilles. 1975. Pragmatic Scales and Logical Structure. Linguistic Inquiry, Vol. 6, 353–375. Grice, H.P. 1975. Logic and Conversation. In Syntax and Semantics, Volume 3: Speech Acts, edited by P. Cole and J. Morgan. New York: Academic Press, 41–58. Reprinted in H.P. Grice. 1989. Studies in the Way of Words. Cambridge MA: Harvard University Press, 22–40. Geurts, Bart. 1998. Scalars. In Petra Ludewig and Bart Geurts, eds. Lexikalische Semantik aus kognitiver Sicht, 95–117. Tübingen: Gunter Narr. Geurts, Bart. 2007. Are the Number Words Learnable? Unpublished pdf: https://www .researchgate.net/publication/255626972_Are_the_number_words_learnable. Accessed Feb. 28, 2016. Horn, Laurence R. 1972. On the Semantic Properties of Logical Operators in English. UCLA dissertation. Distributed by Indiana University Linguistics Club. 1976. Horn, Laurence R. 1989. A Natural History of Negation. Chicago: University of Chicago Press. Reissued, Stanford: CSLI, 2001.

a radically pragmatic account of number words and scales

211

Horn, Laurence R. 1992. The Said and the Unsaid. In Chris Barker and David Dowty, eds. SALT II: Proceedings of the Second Conference on Semantics and Linguistic Theory. Columbus OH: The Ohio State University Department of Linguistics. Horn, Laurence R. 2006. The Border Wars: a Neo-Gricean Perspective. In Klaus von Heusinger and Ken Turner, eds. Where Semantics meets Pragmatics, Volume 16 (Current Research in the Semantics/Pragmatics Interface). Amsterdam: Elsevier, 21–48. Jespersen, Otto. 1924. The Philosophy of Grammar. London: Allen Unwin. Reprinted 1965. New York: Norton. Kaplan, David. 2004. The Meaning of “Ouch” and “Oops”. Howison Lecture. Berkeley: University of California at Berkeley. https://www.youtube.com/watch?v= iaGRLlgPl6w. Accessed February 24, 2016. Kennedy, Hubert C. 1973. Selected Works of Giuseppe Peano. Translated from the Italian and edited, with a biographical sketch and bibliography. Toronto: University of Toronto Press. Koenig, Jean-Pierre 1991. Scalar predicates and negation: Punctual semantics and interval interpretations. CLS 27, 140–155. Morgan, Jerry L. 1978. Two Types of Convention in Indirect Speech Acts. In P. Cole, ed., Syntax and Semantics, Vol. 9: Pragmatics, 261–280. New York: Academic Press. Reprinted in Steven Davis, ed. 1991. Pragmatics: A Reader. New York: Oxford University Press, 242–253. Recanati, François. 2004. Literal Meaning. Cambridge UK: Cambridge University Press. Sadock, Jerrold M. 1984. Whither Radical Pragmatics? In Deborah Shiffrin, ed. Meaning, Form and Use in Context: Linguistic Applications. Georgetown University Round Table on Languages and Linguistics, 139–149. Washington: Georgetown University Press. Sadock, Jerrold M. 1994. Toward a Grammatically Realistic Theory of Speech Acts. In Savas Tsohatzidis, ed. Foundations of Speech Act Theory: Philosophical and Linguistic Perspectives, 393–406. London: Routledge. Sadock, Jerrold M. 2012. The Modular Architecture of Grammar. New York: Cambridge University Press. Zwicky, Arnold M., and Jerrold M. Sadock. 1975. Ambiguity Tests and How to Fail Them. In J.P. Kimball, ed., Syntax and Semantics Volume 4, 1–36. New York: Academic Press.

chapter 8

Utterances and Expressions in Semantics and Logic* David Braun

“I love you,” says Chris to Kim. “I love you,” says Kim in return. How sweet! And how linguistically interesting! Though Chris and Kim utter the same sentence, Chris says that Chris loves Kim, whereas Kim says that Kim loves Chris. The difference in what they say should have some consequences for linguistic theory. Perhaps it shows that distinct utterances of ‘I love you’ can semantically express distinct propositions. If that is right, then semantic theory should pay close attention to utterances. A semantic theory should, for instance, entail that distinct utterances of ‘I’ can refer to distinct people, and distinct utterances of ‘I love you’ can semantically express distinct propositions. Yet many semantic theories never mention utterances. They instead ascribe semantic properties to linguistic expressions. Even when semantic theories deal with indexicals, such as ‘I’ and ‘you’, many merely relativize reference and content to contexts, without mentioning utterances. Is this a mistake? Should semantics deal with utterances? Utterances must, of course, be mentioned somewhere within a comprehensive theory of language. But the traditional distinction between semantics and pragmatics suggests that utterances belong in pragmatics. Semantics, according to this traditional distinction, describes the semantic properties of linguistic expressions. Pragmatics describes what speakers do with those expressions when they utter them. So, the theory of speech acts is a part of pragmatics. Utterances are speech acts. So, utterances should appear in pragmatics, not semantics, on this traditional conception. But perhaps the traditional division of duties between semantics and pragmatics breaks down when it comes to indexicals. It’s entirely reasonable to

* I am grateful for the opportunity to contribute to a volume in honor of Jay Atlas. I was fortunate to have Jay as an undergraduate teacher and as a supervisor for my undergraduate thesis. He has been my mentor ever since. I thank him for his help and for numerous conversations, over many years, about semantics, pragmatics, philosophy of language, and philosophy in general, including conversations about utterances and expressions in semantics.

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_009

utterances and expressions in semantics and logic

213

think that the meaning (or content) of an indexical depends on the circumstances of its utterance. So, it may appear that semantics needs to refer to utterances. Below, I consider various arguments for and against including utterances in semantics. Some of these arguments may be familiar; others may not. In the end, I conclude that utterance-semantics is viable, but must incorporate a rather surprising view of linguistic meaning in order to avoid serious objections. The most serious objections concern (a) expressions that are metaphysically impossible to utter and (b) synonymous simple expressions.1 The view of linguistic meaning needed by utterance-semantics brings it closer to expression-semantics than one might expect.

1

Linguistic Expressions, Utterances, Contexts, and Contents

Utterances are acts in which an agent utters a linguistic expression. So utterances are events of a certain sort. Utterances, like other events, occur at times and locations. Linguistic expressions, by contrast, are neither actions nor events. They either have no times and locations (if they are completely abstract entities) or they have times and locations that are distinct from those of their utterances.2 For example, some utterances of ‘I’ occurred entirely during the 4th of July 2015 in Cleveland, but the word ‘I’ itself does not occur, or exist, entirely within that time and location. Thus, each utterance of a linguistic expression is distinct from that linguistic expression. (Some philosophers seemingly take utterances to be tokens of expressions, rather than actions or events. I discuss the contrast between utterances and tokens in section 6.2. But notice that tokens of expressions are also distinct from expressions.) I will initially take a semantic theory to be an utterance theory iff, roughly, (a) it attributes paradigmatic semantic properties, such as referring, having a content, and being true, to utterances and (b) it takes linguistic meanings of expressions to be functions from utterances to contents, or relations between

1 Zimmermann (1997) discovered the apparent problem with synonymy for what he calls ‘possible utterances’. I independently discovered the apparent problem shortly after Zimmermann’s paper was published, but before I read the paper. I attempt to clarify and generalize the apparent problem in section 9 below. Bach (2005), Salmon (2004b), and Neale (forthcoming) also discuss expression-semantics vs. utterance-semantics, and come down against utterance-semantics. 2 Kaplan (1990) argues that words are contingent objects created by speakers.

214

braun

utterances and contents.3 I will later loosen these conditions, as I attempt to solve apparent problems with utterance-semantics and compare utterancesemantics with expression-semantics. (See sections 5.2, 5.4, and 9.6.) John Perry argues in favor of utterance-semantics in this sense. (See Perry 2006, p. 382, and Perry 2001, chapters 3 and 5. What I call ‘content’ above, Perry calls ‘referential content’ or ‘official content’.) On Perry’s view, different utterances of ‘I’ can have different contents. Nevertheless, the expression ‘I’ has a constant linguistic meaning. (I use ‘linguistic meaning’ where Perry uses simply ‘meaning’.) The linguistic meaning of ‘I’ is a certain relation that holds between u and A iff u is an utterance of ‘I’ and A is the agent of u. The linguistic meaning of ‘I’ determines a content given a particular utterance: for example, the content of a particular utterance u0 of ‘I’ is the agent of u0. ‘Smoke’ also has a linguistic meaning that determines a content for each utterance of it, but the same content for each utterance, namely the property of smoking. The linguistic meaning of the sentence ‘I smoke’ is determined by the linguistic meanings of the words in ‘I smoke’ and the sentence’s syntactic structure. This linguistic meaning determines a content for every utterance of ‘I smoke’. The content of each utterance of ‘I smoke’ is a structured proposition, whose constituents are the contents of the relevant utterance of ‘I’ and the content of the relevant utterance of ‘smoke’. So, the content of an utterance of ‘I smoke’ is the structured proposition whose constituents are the agent of u and smoking. We can represent it with the ordered pair ⟨the agent of u, smoking⟩. There are strong similarities between Perry’s theory and David Kaplan’s famous theory of indexicals (Kaplan 1989a). But Kaplan’s theory does not ascribe contents to utterances. His theory instead ascribes contents to linguistic expressions with respect to contexts. Kaplan tells us that each context c determines an agent of c, a location of c, a time of c, and a possible world of c. The content of ‘I’, with respect to any context c, is the agent of c. Agents vary from context to context, so the content of ‘I’ varies from context to context. The contents of non-indexical expressions do not vary from context to context. For example, the content of ‘smoke’ at every context is the property of smoking. The content of ‘I smoke’ at a context c is the structured proposition 3 Kent Bach (personal communication) says that it is odd to claim that utterances refer, for utterances are acts, and it is odd to say that acts refer. I agree with Bach, but I believe that utterance-semanticists need not take this seeming clash between their theory and ordinary judgment very seriously. Utterance-semanticists are attempting to describe a relation between utterances and objects that they think is important for semantics. This relation resembles reference in certain ways, and so they use ‘refer’ in a technical sense to label that relation. See Perry 2007, p. 514, in response to Bach.

utterances and expressions in semantics and logic

215

whose constituents are the agent of c and the property of smoking, which we can represent with the ordered pair ⟨the agent of c, smoking⟩. Every expression has a stable kind of meaning that is fixed by linguistic convention. The word ‘I’, for instance, has a kind of meaning that is fixed by linguistic convention, and which (seemingly) determines that it has different contents with respect to different contexts. Kaplan calls this kind of meaning character. Kaplan represents characters with functions from contexts to contents. For instance, the character of ‘I’ is a function on contexts whose value at any context is the agent of the context. Kaplan identifies the linguistic meaning of an expression with its character. Kaplan calls his contexts ‘contexts of utterance’, but that is misleading. On his view, expressions can have contents with respect to contexts whose agents utter nothing (at the time and world of the context). For example, the sentence ‘I utter nothing’ has a content in all contexts, including those in which the agent of the context utters nothing (at the time and world of the context). That sentence is true at those contexts. Utterances make no substantial appearance in Kaplan’s semantic theory. Why does Kaplan abstract away from utterances so severely? Kaplan usually justifies doing so by appealing to the needs of logic (Kaplan 1989a, p. 522; 1989b, pp. 584–585). We will consider Kaplan’s reasons later (section 8). More importantly, we will also consider whether the needs of semantics require such severe abstraction away from utterances. I will take Kaplan and Perry to be my representative advocates of expressionsemantics and utterance-semantics, respectively. But the issues I will discuss arise for all semantic theories that have the two characteristics of utterancetheories that I mentioned above, even those on which semantic contents are quite different from Kaplan’s and Perry’s. For example, Jay Atlas (2005, pp. 40– 41) attributes highly nonspecific meanings to sentence-types, which are incapable of having truth-values. He holds that utterance-tokens of a sentence, in a context, have semantic interpretations, which can have truth-values. It would be natural, on his view, to take the linguistic meaning of a sentence to be a function from utterances (or pairs of utterances and contexts) to semantic interpretations. Such a view faces many of the issues I discuss below.

2

Two Initial Arguments for Utterances in Semantics, and Replies

In this section, I critically evaluate two arguments in favor of including utterances in semantics. Both arguments appeal to foundational issues in semantics. (I was inspired to formulate the following arguments by Barwise and Perry 1983;

216

braun

Perry 2001; Schiffer 2005; and Korta and Perry, 2011. But none of them presents exactly the arguments below.) 2.1

First Argument for Utterance-Semantics: Utterances Determine Meanings of Expressions Here is the first argument in favor of utterance-semantics: According to most theories of meaning, utterances are crucial to determining the semantic properties of linguistic expressions in a community. For example: (a) Causal theories of proper names (Donnellan 1970, Kripke 1980) say that the causal history of utterances of a proper name determines its reference, in a community’s language. (b) Intentionalist theories of meaning, such as those of Grice (1957), Schiffer (1972), and Lewis (1975), claim that expressions acquire meanings in a community because of regularities in the thoughts and intentions of those who utter those expressions. (c) Soames (2002, chapter 3) says that the semantic content of a sentence, in a community, is determined by the propositions that speakers in the community assert when they utter those sentences. Therefore, semantic theories should ascribe semantic properties to utterances. Reply: Expression semantics attributes semantic properties primarily to expressions, but that is completely consistent with acknowledging that regularities in utterances determine the semantic properties of expressions. We should distinguish, as Kaplan does (1989b, p. 573), between semantics and meta-semantics. Semantics describes the semantic properties of the expressions (and perhaps utterances) in a language. Meta-semantics describes what makes it the case that a community uses a particular language, or how the meanings of the expressions in those languages are fixed or determined by other facts, such as usage. These enterprises are distinct. Two semanticists may agree on the semantic properties of ‘yell’; they may agree, for instance, that its content, in all contexts (or in all of its utterances), is the property of yelling. Yet they may disagree about why it has that meaning. One of them may think that ‘yell’ has that meaning, in that community, because of certain regularities in its utterances. The other may think it has that meaning in virtue of primitive, irreducible laws of intentionality, or in virtue of God’s commands. Such semanticists would agree on the semantics of ‘yell’, but disagree about its meta-semantics. Meta-semantic theories almost certainly must pay attention to utterances. But this does not show that semantic theories must. Expressionsemantics is consistent with all of the above meta-semantic theories that mention utterances.

utterances and expressions in semantics and logic

2.2

217

Second Argument for Utterance-Semantics: Utterances are Crucial to the Psychology and Epistemology of Language Here is a second argument for utterances in semantics: Theories of language comprehension and production must mention utterances. Hearers’ sensations of utterances, and their beliefs about utterances, are an important part of their evidence for their beliefs about what speakers mean when they utter expressions. Speakers choose to utter (produce) one expression, rather than another, in light of the preceding facts. Semantic theories that do not mention utterances neglect these facts about language. Reply: Once again, we need a distinction, this time between semantics, on the one hand, and psycholinguistics and linguistic epistemology, on the other. Psycholinguistics describes the complex psychological processes by which speakers come to understand utterances, have beliefs about what people say and what words mean, decide to utter expressions, and produce particular utterances. Linguistic epistemology describes the conditions under which beliefs about speakers, hearers, language, and utterances are justified and constitute knowledge. Semantics does none of this. Semantics merely describes semantic properties and semantic values. Thus expression semantics is entirely consistent with the preceding claims in psycholinguistics and linguistic epistemology. Counter-reply to the preceding reply to the second argument: Granted, semantics is not psycholinguistics or epistemology. But semantics must be consistent with psycholinguistics and epistemology. So, psycholinguistics and epistemology place constraints on semantics. Semantic theories that do not mention utterances cannot describe the facts about utterances that constrain the semantic properties that expressions can have. Counter-reply to the preceding counter-reply: How psychology and epistemology constrain semantics is, again, a meta-semantic matter. (Two theorists can agree on semantics, and yet disagree on the meta-semantic matter of how semantic content relates to utterances, psychology, and epistemology.) So, semantics does not describe the constraints that psycholinguistic and epistemological facts about utterances place on semantic theory. But these constraints can be described in auxiliary hypotheses or bridging principles. For example, here are two hypotheses that describe some relations between Kaplan-style expression-semantics and speech acts, and therefore the relations between the thoughts and intentions of speakers who utter expressions and their semantic contents; these hypotheses might play a role in describing relations between semantics, on the one hand, and psycholinguistics, and epistemology, on the other. (a) An utterance by a speaker, at a time and world, determines a Kaplanian context. (b) Speakers who assertively and

218

braun

literally utter sentences typically assert the contents of those sentences, with respect to the contexts determined by their utterances. If, for instance, Alice utters ‘I smoke’ in a world, then her utterance has an agent at that world, a time at which it occurs in that world, and a location at which it occurs in that world. This determines a Kaplanian context C. If she utters that sentence assertively and literally, then she probably asserts the semantic content of ‘I smoke’ with respect to C. If she does not assert it, then that is prima facie evidence against Kaplan’s theory. The preceding distinctions between semantics, on the one hand, and other subject matters, such as meta-semantics and psycholinguistics, on the other, seem principled and natural. I conclude that neither of the preceding arguments in favor of utterance-semantics is compelling.

3

An Argument for Utterance-Semantics from True Demonstratives

I turn now to an argument in favor of utterance-semantics that largely consists in an argument against expression-semantics. We begin with Kaplan’s distinction between true demonstratives and pure indexicals. (See Kaplan 1989a, pp. 489–492. Perry, in his 1997, 2001, and 2006, makes a closely related distinction between automatic and discretionary indexicals.) The pure indexicals include the expressions ‘I’, ‘today’, and ‘tomorrow’. The true demonstratives include ‘that’, ‘you’, and ‘there’. The reference and content of a pure indexical seems, roughly speaking, not to depend on the intentions of an agent, whereas the reference and content of a true demonstrative does. More precisely, a speaker who utters ‘I’ always refers to herself, and a speaker who utters ‘tomorrow’ always refers to the day after she speaks, as long as she intends to use these expressions literally with their usual (English) meanings. (Thus Perry calls these indexicals ‘automatic’.) This is why Kaplan says that, in absolutely every context, the content of ‘I’ is the agent of the context, and the content of ‘tomorrow’ is the day after the day of the context. By contrast, a speaker who utters a true demonstrative must have intentions beyond the intention to speak English literally, in order to use these expressions to refer to objects. (Thus Perry calls these indexicals ‘discretionary’.) She should, roughly speaking, have some intention to refer to some object using the demonstrative. Let’s call these intentions demonstrative intentions. True demonstratives raise a prima facie difficulty for expression-semantics, such as Kaplan’s theory (as Kaplan 1989b, pp. 585–587, points out; see also Kaplan 1999, p. xxii). Kaplan’s theory is designed to assign a content to every expression at every context. But consider a Kaplan-style context in which the

utterances and expressions in semantics and logic

219

agent is not uttering ‘that’, and has no intention to refer demonstratively to any object. It seems artificial to say that the expression ‘that’ has a content or referent with respect to such a context. The problem may generalize. Many philosophers and linguists have claimed that context-sensitivity extends far beyond Kaplan’s list, and includes such words as ‘rich’, ‘ready’, ‘might’, ‘know’, and many others. (See Bach 2012 and Braun 2015b for surveys.) If these expressions are indexicals, then they are clearly discretionary, in Perry’s sense, and so they pose the same sort of problem for expression-semantic theories that true demonstratives do. The artificiality of assigning a content to a true demonstrative at every Kaplan-style context is not an issue for utterance theories, for utterance-theories assign contents to utterances, and speakers who utter ‘that’ do have demonstrative intentions. (I will mention some qualifications to this claim in section 6.2 below.) So, we now have an argument against expression-semantics and in favor of utterance-semantics. Some expression-oriented semantic theorists might respond to this argument by replying that true demonstratives do not have semantic contents, or semantic references, with respect to contexts. They might say that the linguistic meanings of true demonstratives allow them to be used by speakers to speakerrefer to objects, but the words themselves never semantically refer to objects, or semantically express contents, with respect to contexts. (See Schiffer 2005, p. 1141; Bach 2006, pp. 543–549; Heck 2014; Soames 2015, pp. 109–110; Bach forthcoming; and Neale forthcoming.) This response is worth considering at length, but I won’t do so here, because the response rejects content and reference for both expressions (demonstratives) and their utterances, in favor of mere speaker-reference and speaker-meaning. I will instead concentrate on defending a Kaplan-style theory that does allow ‘that’ to have a content and reference in a context. Reply: An expression-theorist can approach this problem in stages. At the first stage, she can add a demonstratum (demonstrated object) to at least some contexts, and hold that the content of ‘that’ with respect to a context is the demonstratum of the context, if it has one. (See Kaplan 1989b, pp. 585–587.) If a context has no demonstratum, then ‘that’ does not have a content with respect to it. This modification entails that some expressions fail to have contents at some contexts, so expression-theorists then must deal with the consequences of content-failure and reference-failure. I think that the best way to do so is to add a theory of gappy propositions, but I shall not pursue this further here. (See Braun 2005 and 2008 for some discussion.) However reference-failure is handled, the modified theory is still an expressions-at-contexts theory, rather than an utterance theory.

220

braun

At the second stage, an expression-theorist must decide how the demonstrative intentions (if any) of a context’s agent, at the world and time of the context, are related to the demonstratum (if any) of that context. An expression-theorist has two different issues to resolve: (a) Is the agent’s having a demonstrative intention that determines an object sufficient for the context’s having that object as its demonstratum? (b) Is it necessary? Let’s take up necessity first. Suppose an expression-theorist denies necessity: she allows her semantic theory to mention some contexts that have a demonstratum, though the agents of those contexts do not have demonstrative intentions that determine those objects (at the world and time of the context). There are two types of context to consider. In the first type, the agent has a demonstrative intention that determines some object other than the (official) demonstratum of the context. In the second type, the agent has no demonstrative intentions at all. It may seem odd for a semantic theory to assign demonstrata to contexts in which the agents lack demonstrative intentions, or in which the agent’s demonstrative intentions determine some object other than the (official) demonstratum. But perhaps the strangeness can be mitigated by noting that such contexts cannot, in a certain sense, be realized or determined by a metaphysically possible utterance of ‘that’. If an agent utters a demonstrative at a (metaphysically) possible world, then she has demonstrative intentions at that world. So, no metaphysically possible utterance of ‘that’ realizes a context in which the agent has no demonstrative intentions.4 Moreover, if an agent has a demonstrative intention at a time and world, then the object to which she refers (if any) is the object determined by her demonstrative intention at that time and world. But (plausibly) if a context has a demonstratum, then an utterance of ‘that’, in a world, realizes that context only if the utterance’s agent is the agent of that context, and the demonstratum of that context is the object determined by the demonstrative intentions of the utterance’s agent at that time and world. Thus, the oddball contexts contemplated above would not be realized by any metaphysically possible utterance. The preceding denial of necessity is defensible. I am, however, inclined to accept necessity. (I here agree with Kaplan 1989b, pp. 585–587.) As the previous paragraph makes clear, it is metaphysically necessary that an object is an agent’s demonstratum (the object to which an agent refers with a demonstrative) at a time and world iff that object is determined by that agent’s demon-

4 I am writing here of utterances in which ‘that’ is used in a locutionary manner, and not merely mentioned. I say more on these matters in section 6.2

utterances and expressions in semantics and logic

221

strative intentions at that time and world. A semantic theory that includes “contexts” that violate this metaphysical constraint on demonstratives seems poorly motivated. Such a theory is comparable to a biological theory whose taxonomy includes a category of whales that are fish, but which also entails that it is metaphysically impossible for a whale to be a fish. So, I believe that expression-theorists should accept the necessity condition on contexts that I mentioned two paragraphs back.5 Moreover, I argue below that accepting this necessary condition has no bad consequences for expression-semantics. If we follow my suggestion, then the following sentence will be untrue in all contexts, on the modified theory: ‘That exists but I do not intend to refer to it’.6 That consequence might worry some. One might think that if ‘That exists but I do not intend to refer to it’ is untrue in all contexts, then its negation is true in all contexts, and one might worry that this seeming-consequence is unacceptable. But the seeming-consequence may not be a consequence. Whether it is a consequence depends on complicated issues about truth-values and negations. Is ‘That exists and I do not intend to refer to it’ false, or neither true nor false, in contexts in which ‘that’ fails to refer? Is the negation of a sentence that is neither true nor false in a context also neither true nor false in that context? Or is it true? I will discuss some of these issues in section 8.3 below. (See also Braun 2005.) Here’s another objection: If ‘That exists but I do not intend to refer to it’ is untrue in all contexts, then ‘If that exists, then I intend to refer to it’ is true in all contexts, and so is logically true. But the latter is not logically true. So, we should reject necessity. (For an objection roughly like this, see Salmon 2002, n. 29.) Reply in two parts: First, as I pointed out in the last paragraph, there are complicated interacting issues about reference-failure, the truth-values of simple sentences, and the truth-values of complex sentences (such as negations and conditionals) containing such simple sentences. So, the conditional sentence may fail to be true in all contexts, even though the first sentence is untrue in all contexts. (See section 8.3.) Second, suppose that the conditional sentence is true in all contexts. This does not entail that it is logically true, contrary to the objection. Logical truth (or validity), on a Kaplanian conception, is 5 I am not saying that every context that includes a demonstratum must have an agent who utters ‘that’. That would lead to issues with synonymy like those that I describe in section 9. 6 Read this sentence in the following way: ‘That is such that I intend to refer to it but it does not exist’. The alleged consequence does not entail that, in all contexts, the sentence expresses a proposition that is necessarily false. Consider a context in which ‘that’ does refer, and consider the proposition the sentence expresses at that context, then evaluate that proposition at a world in which the agent of the original context fails to exist.

222

braun

not truth in all contexts, but rather truth in all “contexts” in all structures. See section 7 below for discussion of validity.7 Let us turn to sufficiency. Suppose that the agent of a context has a demonstrative intention that determines an object (at the time and world of the context). Must an expression-theorist say that the context has that object as its demonstratum? If the theorist were to say ‘no’, then she would be allowing that, in some contexts of this sort, there is no demonstratum, or the demonstratum is some object other than the object determined by the agent’s demonstrative intention. This again seems strange. Such a theorist might again try to mitigate its strangeness by arguing that such contexts are not realized by any metaphysically possible utterances. Again, this seems to me to be a defensible option. But again, I believe there is little point to it. It is more natural for a semantic theory to discuss only contexts that satisfy the metaphysical constraint that I mentioned earlier. Furthermore, accepting sufficiency does not have negative consequences for validity. I discuss this point in sections 7 and 8.8 Summarizing: The objection from true demonstratives against expressionsemantics is, at most, inconclusive. Expression-theorists have various reasonable, and even attractive, options for dealing with the problem that this objection raises.

4

An Argument for Utterance-Semantics from Multiple Occurrences of Demonstratives

Let’s turn to another argument in favor of utterance-semantics. This argument, like the last one, consists mostly of an objection to expression-semantics for 7 Kaplan and Salmon disagree about whether to impose restrictions on contexts such as the one I have accepted. Kaplan (1989b, pp. 585–586) favors such restrictions. For example, he proposes to mark contexts in which an agent does not intend to address someone as inappropriate for ‘you’, and he proposes to define validity as truth-in-all-appropriate-possible-contexts. (Kaplan evidently means truth-in-all-appropriate-contexts-in-all-structures.) Salmon (2002, n. 29) rejects an analogous restriction to appropriate contexts for ‘that’ because (he thinks) it would have the consequence that ‘That object (assuming it exists) is now being demonstrated’ is logically true. I believe Kaplan is right to accept such restrictions on contexts in the intended structure. But Salmon is right to reject such restrictions on “contexts” in unintended structures, and therefore ‘That object (assuming it exists) is now being demonstrated’ is not logically true. See section 8.3 below. 8 We might want to consider a further break from Kaplan at this point, by requiring that the speaker utter something in every context in the intended structure. I discuss this in section 9.6. Thanks to Alex Radulescu for discussion.

utterances and expressions in semantics and logic

223

true demonstratives. But it focuses on multiple occurrences of demonstratives. I think it is the strongest argument in favor of including utterances in semantics.9 To begin, consider sentence (1). (1) That is larger than that. Speakers can use (1) to assert a truth, by demonstrating two different objects. That gives us some reason to think that a correct expression-based semantic theory of (1), in the Kaplan style, should entail that (1) is true with respect to some contexts. But the word ‘that’ is unambiguous. So, it has a single linguistic meaning. If linguistic meaning determines reference in a context, in the way that Kaplanian characters do, then the expression ‘that’ has exactly one reference in every context. Now suppose the extension of a complex expression, in a context, is compositionally determined by the extensions of its constituent expressions, in that same context. Then (1) is true in a context c iff the referent of ‘that’ in c is larger than the referent of ‘that’ in c. But then (1) is true in a context iff some object is larger than itself. But that’s impossible. So, there is no context in which (1) is true. So, a Kaplan-style theory does not tell us why (1) can be used to assert a truth. Kaplan has responded to this problem by, roughly speaking, ambiguating the word ‘that’. First, he subscripts occurrences of ‘that’.10 Thus he can represent (1), or uses of (1), with (2). (2) That1 is larger than that2. Second, he introduces sequences of multiple demonstrata to contexts. Third, he stipulates that the referent of ‘thatn’ in a context is the n-th demonstratum of the context. Doing this allows (2) to be true in some contexts. But it also has the consequence that the expressions ‘that1’ and ‘that2’ have different characters. The word ‘that’, however, has a single linguistic meaning. So, we cannot identify

9

10

The negative part of the following argument against Kaplan’s theory appears in Braun 1996. Wettstein (1986, p. 196 n. 17) presents nearly the same negative argument, but without going into detail about Kaplan’s use of subscripts. Kaplan 1989b, pp. 585–590, discusses issues raised by multiple occurrences of demonstratives. Barwise and Perry 1983, pp. 133– 136, and Crimmins 1995 discuss the use of subscripted pronouns in semantics and what they wrongly suggest about the linguistic meanings of demonstratives. See also notes 10 and 11 below. Kaplan’s treatment of demonstratives in his 1989a uses dthat-terms. I am ignoring that view here. See Braun 1996 and Salmon 2005a for discussion of that theory.

224

braun

the linguistic meaning of ‘that’ with the character of any of these subscripted ‘that’s. So, Kaplan’s response to the problem creates a different problem, namely failing to provide a plausible candidate for the linguistic meaning of ‘that’. That’s the objection to expression-semantics. Here is the consequent argument in favor of utterance-semantics: Every utterance of (1) contains two different utterances of ‘that’. An utterance-theory can assign distinct references and contents to distinct utterances of ‘that’, and so can allow an utterance of (1) to be true. So, in principle, an utterance-semantics can deal with (1) easily. There is no adequate analysis of (1) within expression-semantics. So, all expression-semantic theories of true demonstratives are false, and some utterance-semantic theory of true demonstratives is true.11 Reply: Sentences such as (1) show that expression-theories must deal not only in the semantics of expressions, but also in the semantics of occurrences of expressions. (Occurrences are not utterances. Some never-uttered sentences contain many occurrences of ‘that’.) A revised expression-semantics must entail that the expression ‘that’ has a single linguistic meaning (which may or may not be a Kaplan-style character), but also allow distinct occurrences of ‘that’ in a complex expression, such as (1), to have different contents with respect to appropriate parameters, such as a context. I proposed two such theories in a previous paper (Braun 1996).12 According to the first theory, the expression ‘that’ has a single content with respect to any given context. So, it is possible to use a function from contexts to contents to represent the single linguistic meaning of ‘that’. But the semantics allows context-shifts in mid-sentence, and so allows distinct occurrences of ‘that’ in a sentence to be evaluated for reference and content with respect to different contexts. As a result, the content of sentence (1) at some contexts is a proposition that says that x is not identical with y, where x and y really are distinct. That proposition is true. This theory denies that the content of a (complex) expression at a context c is fully determined by the contents of its constituent expressions at that same context.

11

12

The positive argument here in favor of utterance-theory appears (more or less) in Braun 1996. Garcia-Carpintero (1998) takes multiple occurrences of demonstratives as a reason to favor semantic theories that assign contents to tokens. Kaplan (1999, p. xxii) also mentions multiple occurrences of demonstratives as a motivation for a semantics that assigns contents either to tokens, or to expressions with respect to contexts in which tokens of true demonstratives are deployed. For other attempts to deal with the problem of multiple occurrences of demonstratives within an expression-semantics, see Gauker 2015, Georgi 2015, and Pickel, Rabern, and Dever, 2015.

utterances and expressions in semantics and logic

225

The second theory presented in that paper distinguishes between the linguistic meaning of ‘that’, on the one hand, and the Kaplanian characters that occurrences of ‘that’ have with respect to sequences of demonstrative intentions (or demonstrations), on the other. The word ‘that’ has a single linguistic meaning, but distinct occurrences may have distinct Kaplanian characters with respect to an association of demonstrative intentions with occurrences of demonstratives. This second theory uses a technique similar to the preceding one to assign distinct demonstrative intentions and characters to distinct occurrences of ‘that’ within a single sentence.13 I tentatively conclude that a modified version of Kaplan’s theory can deal well with sentences that contain multiple occurrences of demonstratives. I do concede, however, that the problem of multiple occurrences of demonstratives is a reasonable motivation for seeking an utterance-semantic theory. That is one of my motivations for thinking about utterance-semantics.

5

An Objection to Utterance-Semantics from Un-uttered Expressions and Too Many Synonymous Expressions, and Some Replies

I turn now to objections to utterance-semantics. The first objection I shall consider does not succeed, but considering it will motivate certain important features of utterance theories. The following is a typical claim of an utterance theory: For every u, if u is an utterance of ‘I’, and A is the agent of u, then the content of u is A. This claim specifies the content of every utterance of ‘I’. The linguistic meaning of ‘I’ (the expression) should be something that determines the contents of its utterances. So, it seems that in an utterance-based semantic theory, the linguistic meaning of ‘I’ is, or can be represented by, a simple extensional function, which is equivalent to a set of ordered pairs. The function’s domain is the set of utterances of ‘I’. Its range is the set of agents of those utterances. For each utterance in the domain, the function’s value is the agent of that utterance. Parallel points hold for all other simple expressions. The linguistic meaning of a complex expression is also a function, determined by the linguistic meanings of its parts and its syntactic structure. (See section 1.) So, the linguistic meaning of the sentence ‘I

13

Pickel, Rabern, and Dever (2015) point out a technical error in the first theory, and by implication, in the second theory as well. They propose a fix for the first theory. I have proposed a related fix to the first theory in Braun 2015a, which can also be extended to the second theory. For details see Braun 1996; Pickel, Rabern, and Dever 2015; and Braun 2015a.

226

braun

love you’ is a function whose domain is all utterances of that sentence. Its value for any utterance of the sentence is the proposition whose constituents are the agent of the sub-utterance of ‘I’, the relation of loving, and the addressee of the sub-utterance of ‘you’. But un-uttered complex expressions raise an apparent problem for any view on which linguistic meanings are functions from utterances to contents. Consider the sentence ‘Houdini heaved harpsichords’. No one has uttered this sentence. (I have merely mentioned it. I will say more in section 6 about the sorts of utterances that utterance-semantics must deal with.) The above theory entails that the linguistic meaning of this sentence is the function whose domain is the set of utterances of it, and whose value, for each such utterance, is the proposition that Houdini heaved harpsichords. But there are no utterances of this sentence. So, the domain of this function is empty. So, its range is also empty. Therefore, this function is the empty function, which is equivalent to the empty set of ordered pairs.14 A similar argument leads to the conclusion that the (alleged) functional linguistic meaning of ‘Obama obliterated obelisks’ is the empty function. So, if linguistic meanings of these two sentences are functions on their utterances, then they have the same linguistic meaning— they are synonymous. A similar point holds for all un-uttered sentences. But obviously some pairs of un-uttered sentences differ in linguistic meaning. The above two sentences are good examples. So, utterance-semantics is false. Let’s call this the Too Much Synonymy Objection. I will consider four attempts at to deal with the Too Much Synonymy Objection. Each theory will improve on its predecessor. The first two attempts will be inadequate. The third may be adequate, given a somewhat controversial metaphysical assumption. The fourth is adequate, but requires a slight departure from standard utterance-semantics. 5.1 Reply #1: Appeal to Merely Possible Utterances The first solution allows utterance-semantics to deal with merely possible utterances. No one has actually uttered ‘Houdini heaved harpsichords’, but it’s pos-

14

Some technical explanation: If we take functions to be primitive, but to have extensions, then we should take functions that have empty domains to have empty ranges, and take the extensions of such functions to be the empty set (of ordered pairs). If we take functions to be identical with sets of ordered pairs, then we could define the Houdini-function as follows: the set of all ordered pairs ⟨u, P⟩ such that u is an actual utterance of ‘Houdini heaved harpsichords’ and P is the proposition that Houdini heaved harpsichords. Since there are no actual utterances of this sentence, there are no such ordered pairs, and so the definite description ‘the set of all ordered pairs ⟨u, P⟩ such that …’ refers to the empty set.

utterances and expressions in semantics and logic

227

sible for someone to do so, and so there are possible utterances of that sentence. Let the linguistic meaning of any word be a function whose domain is all possible utterances of it, and whose value is the content of that possible utterance. These functions, together with the sentence’s syntax, determine a function from possible utterances of the sentence to the proposition that Houdini heaved harpsichords. There is a similar function determined, in a parallel way, for ‘Obama obliterated obelisks’. The domain of the first functional linguistic meaning is the set of possible utterances of ‘Houdini heaved harpsichords’, whereas the domain of the second is the set of possible utterances of ‘Obama obliterated obelisks’. These domains are distinct. The values of these functions are different propositions. So, the two sentences have distinct linguistic meanings. So far, so good. But there is a complication. Plausibly, the same possible utterance can occur in two worlds that differ in ways that are semantically relevant to the uttered expression. For example, an actual utterance of ‘today’ that occurs near midnight might have occurred shortly after midnight instead. So, that utterance might have had a different content, and therefore does have a different content in some possible world. So, as Salmon (1987, section V) and Kaplan (1999, p. xxi) point out, there is no such thing as the content of that possible utterance. Many other possible utterances similarly vary in content from world to world. 5.2 Reply #2: Appeal to Pairs of Possible Utterances and Contexts Our second theory deals with the variation in content of possible utterances from world to world by bringing in something very like a Kaplanian context into utterance-semantics. (Kaplan [1999, p. xxi] suggests such an addition for token semantics.) On this proposal, the linguistic meaning of an expression is identified with a function whose domain is a set of pairs of possible utterances of that expression and contexts in which they are uttered, and whose value for each such pair is the content of that utterance in that context. We might initially identify these contexts with the possible worlds in which the utterances occur, because a possible utterance of an expression and a world together should determine a single content. Further developments of the theory may force us to add items to contexts other than worlds. I pause for an observation: Our second utterance-theory assigns contents to pairs of possible utterances and contexts, and these contexts include at least worlds. Moreover, we earlier assumed that each possible utterance at a world has a unique speaker, and occurs at a unique time and location. Now compare this theory with Kaplan’s: his theory assigns contents to expressions with respect to contexts that include agents, times, and worlds. So, the two

228

braun

theories use similar parameters (contexts, agents, worlds, times, locations). To bring them into closer alignment, we could formulate an utterance-semantics that resembles Kaplan’s theory in format, but which contains an extra, unKaplanian, condition: the linguistic meaning of an expression is a function from contexts in which the context’s agent utters the expression at the time and world of the context to contents. (See Kaplan 1999, p. xxi, for a parallel suggestion for token semantics.) There seems little reason to choose between the former conceptualization of utterance-semantics (as functions from pairs of possible utterances and contexts) and the latter, slightly more Kaplanian conceptualization of utterance-semantics—except that the latter helps facilitate comparisons with Kaplan’s theory. We will see more of this re-conceptualized version of the second theory below in sections 7 and 9.6. But for now, I return to the un-re-conceptualized version that assigns contents to pairs of possible utterances and contexts. Appealing to possible utterances and contexts may seem to deal adequately with the Too Much Synonymy Objection. But unfortunately, there is a remaining issue with possible utterances and un-uttered sentences. Perhaps there are sentences that have no metaphysically possible utterances. Consider a sentence formed by conjoining ‘Obama obliterated obelisks’ with itself ten trillion times. Perhaps it is metaphysically impossible for an agent to utter this sentence. (If this is not a good example, conjure up another.) Call such (alleged) sentences metaphysically un-utterable sentences. On the above proposal, the linguistic meaning of this metaphysically un-utterable sentence is a function whose domain is the set of pairs of metaphysically possible utterances of that expression and the contexts in which those possible utterances occur. But this domain is empty. So, its range is empty and its linguistic meaning is the empty function. Similarly, the (seemingly) metaphysically un-utterable sentence that consists of the sentence ‘Houdini heaved harpsichords’ conjoined with itself ten trillion times has the same linguistic meaning on the above proposal. So, these two sentences have the same linguistic meaning on this proposal. So do all other metaphysically un-utterable sentences. All of them are synonymous. But clearly not all metaphysically un-utterable sentences have the same linguistic meaning. An advocate of a possible-utterance theory could maintain that there are no metaphysically un-utterable sentences. Some utterance-theorists might be happy to make this assumption. In my opinion, it would be risky to rest one’s defense of a semantic theory on such a strong metaphysical claim. Alternatively, an utterance-theorist could insist that her theory quantifies over all logically possible utterances, which includes many metaphysically impossible ones. But idealizing to that degree risks changing the subject matter of the theory,

utterances and expressions in semantics and logic

229

for some merely logically possible utterances are not metaphysically possible utterances, and so (plausibly) not utterances at all. This seems contrary to the spirit of utterance-semantics.15 5.3 Reply #3: Appealing to Genuine Relations There is a third way for utterance-theorists to defend their view from the problem of un-uttered, and (allegedly) metaphysically un-utterable, sentences. They can, like Perry, identify linguistic meanings with genuine relations rather than extensional functions. (See Perry 1997, 2001, 2006. Braun 1995 identifies Kaplanian characters with genuine relations, for related reasons.) On this view, the linguistic meaning of ‘I’ is a certain relation that holds between u and A iff u is an utterance of ‘I’ and A is the agent of u.16 The linguistic meaning of ‘harpsichord’ is a (certain) relation that holds between any utterance u and the property of being a harpsichord iff u is an utterance of ‘harpsichord’. The relational linguistic meaning of the sentence ‘Houdini heaved harpsichords’ is determined by the (relational) linguistic meanings of its words and its syntax. The linguistic meaning of that sentence is a (certain) relation that holds between u and P iff u is an (actual) utterance of the sentence and P is the proposition that Houdini heaved harpsichords. Similarly, the linguistic meaning of ‘Obama obliterated obelisks’ is a (certain) relation that holds between u and P iff u is an (actual) utterance of the sentence and P is the proposition that Obama obliterated obelisks. Each of these relations has an (actually) empty extension. But they are distinct relations. We know that they are distinct, because they could (metaphysically) differ in extension. That is, it is possible for there to

15

16

I presented similar objections to utterance-semantics from metaphysically impossible utterances in Braun 1996. Garcia-Carpintero (1998) responded that (roughly speaking) token-semantics should quantify over all logically possible tokens. Below I will attempt to show that an utterance theorist would be better off saying that a semantic theory should quantify over metaphysically possible utterances and contexts in the intended structure, whereas a logic should quantify over all “contexts” in all structures, including unintended structures. The “contexts” of the unintended structures may be metaphysically impossible. See sections 7 and 8. I argued in section 5.2 that if linguistic meanings are extensional functions, then their domains should be pairs of possible utterances and contexts. The use of genuine relations eliminates the need to mention contexts. Let the linguistic meaning of ‘today’ be a certain relation T that holds between u and d iff d is the day on which u occurs. This relation varies in extension from world to world. Suppose u is an utterance of ‘today’ that occurs just before midnight in world w1 and just after midnight in w2. T may hold between u and day d1 in w1, but hold instead between u and day d2 in w2.

230

braun

be utterances of both sentences, so there are possible worlds in which these relations differ in extension. So, the linguistic meanings of these sentences are distinct. Let’s now extend this use of relations to the metaphysically un-utterable conjunctive sentences that we are considering. The linguistic meaning of the sentence consisting of ‘Houdini heaved harpsichords’ conjoined with itself ten trillion times will be a certain relation that holds between u and P iff u is an utterance of this sentence and P is the conjunctive-Houdini-proposition. A parallel point holds for the conjunctive ‘Obama’ sentence and the conjunctiveObama-proposition. If these relations are distinct, then the sentences have distinct linguistic meanings. Unfortunately, it is not altogether clear that the two relations are distinct. We are imagining that it is metaphysically impossible for there to be utterances of either conjunctive sentence. So, necessarily, the extension of each relation is the empty set. So, the relations are necessarily co-extensive. So, if we think that these relations are distinct, then we must endorse the view that there can be relations (and properties) that are distinct, yet necessarily co-extensive. Some philosophers are willing to accept the existence of necessarily co-extensive, but distinct, relations, but others are not. (See Orilia and Swoyer 2016, especially section 6.) If we endorse this view of relations, then we are relying on a contentious view in metaphysics to solve a problem in semantic theory. 5.4 Reply #4: Structured Linguistic Meanings There is a fourth way that utterance-semanticists might defend their view from the Too Much Synonymy Objection. This fourth way avoids the previous controversial metaphysical assumption about relations. It instead uses structured linguistic meanings. (Richard 1983 and Braun 1994 use structured linguistic meanings for unrelated reasons.) Let’s continue to suppose that the meaning of a simple expression is a function from utterances to contents, or a genuine relation that can hold between utterances and contents. But suppose that the linguistic meaning of a complex expression is a structured entity whose ultimate constituents are the linguistic meanings of the simple expressions that appear in the complex expression. For instance, the linguistic meaning of ‘Houdini heaved harpsichords’ is a structured entity whose ultimate constituents include the linguistic meaning of ‘Houdini’, the linguistic meaning of ‘heaved’, and the linguistic meaning of ‘harpsichords’. The linguistic meaning of ‘Obama obliterated obelisks’ contains the linguistic meanings of ‘Obama’, ‘obliterated’, and ‘obelisks’. So the structured linguistic meanings of the two sentences are distinct. If we identity the linguistic meaning of a complex expression with its structured linguistic meaning,

utterances and expressions in semantics and logic

231

then these sentences differ in linguistic meaning. And so do the very long, metaphysically un-utterable conjunctive sentences that we’ve been discussing. The main drawback of this reply for an utterance-theorist is that she could no longer say that all linguistic meanings are functions or relations. She must allow that some linguistic meanings are entities with constituent structures, whose ultimate constituents are functional or relational linguistic meanings. But that seems a rather low price to pay. I conclude that utterance-semanticists can deal adequately with problems of both un-uttered, and metaphysically un-utterable, linguistic expressions, if they are willing either (a) to assume that some necessarily co-extensive relations are distinct or (b) to assume that the linguistic meanings of complex expressions are structured entities whose ultimate constituents are functional or relational linguistic meanings of simple expressions. However, for the sake of simplicity I will frequently write below as though utterance theorists (should) take linguistic meanings to be simple functions from actual utterances to contents.

6

An Argument against Utterances-Semantics from Too Many Contents

Let us now turn to another serious argument against utterance-semantics. The objection is inspired by the works of Bach (2005) and Salmon (2004b), though neither presents the argument in the exact form I give it below. I will then consider a reply to that argument, replies to that reply, and so on. 6.1 The “Too Many Contents” Objection to Utterances in Semantics The “too many contents” objection to utterance-semantics points out that a speaker usually performs many different speech acts when she utters a sentence. The propositional objects of these speech acts are often distinct. For example, someone who utters ‘John voted for Obama and Mary voted for Romney’ typically asserts the proposition that the sentence semantically expresses, namely that John voted for Obama and Mary voted for Romney. But such a speaker also typically asserts at least two other propositions, namely the proposition that John voted for Obama and the proposition that Mary voted for Romney. Neither of the latter propositions is the semantic content of the sentence. Furthermore, a speaker who utters this sentence may also conversationally implicate several other propositions, such as the propositions that John is a Democrat and the proposition that Mary is a Republican. Here are some more (perhaps controversial) examples of multiple speech acts performed in a sin-

232

braun

gle utterance. Someone who utters the sentence ‘The current President of the USA smokes’ in 2015 may thereby assert both the (descriptive) proposition that the President of the USA smokes and the (singular) proposition that Obama smokes (Salmon 2004a). Someone who utters ‘Joe is an honest man’ might, if she is being ironic, thereby assert that Joe is not an honest man. In all of the preceding cases, a speaker who utters a single sentence performs several speech acts with distinct contents. All of the contents of these speech acts are, plausibly, contents of the relevant utterance, in some sense. So, there is no such thing as the content of an utterance. Moreover, most of these contents are not plausible candidates for the semantic content of the utterance. So, the objection concludes, a semantic theory should not identify the semantic content of an utterance with the content of the utterance. 6.2

Reply to the Objection from Too Many Contents: Find a Distinctive Speech Act An advocate of utterance-semantics might reply that there is a principled way to select only one of the speech acts performed in an utterance as the speech act primarily relevant to semantics. The advocate might then select the content of that speech act as the semantic content of the utterance, and so the content that is relevant to utterance-semantics. To select the correct content, we need to think more carefully about utterances and speech acts than we have thus far. (See Korta and Perry, 2011.) Up till now, I have mostly ignored the fact that the term ‘utterance’ is sometimes used for two different types of entities, namely actions (events) of a certain sort and products of such actions. When a speaker utters an expression, there is an event of intentionally producing the sound, on the one hand, and the sound produced, on the other. The former is an action. The latter is a sonic event. From here on, I shall use the term ‘utterance’ for actions and the term ‘token’ for the products of such actions. (I am here following Perry’s terminological lead: see Perry 1997, 2001, 2006. Kaplan 1999 makes the same distinction.) Tokens will include sonic events, and their analogs when it comes to writing and signlanguage. I will use ‘token’ so that tokens need not have a unique agent, time, or location.17 Utterances are certain sorts of actions of producing a sound, or a written mark, or a sign-language gesture.

17

I am being rather stipulatory about the nature of tokens here, but only so as to maintain a clear distinction between tokens and utterances. Many tokens are events. If the producers of these tokens were essential to them, then they would be hard to distinguish from actions, such as utterances. Thanks to Alex Radulescu for discussion.

utterances and expressions in semantics and logic

233

Utterances in the preceding sense are speech acts. J.L. Austin (1975) distinguished among three importantly different types of speech act that are performed in utterances, including locutionary acts, illocutionary acts, and perlocutionary acts. A speaker performs a locutionary act when she (simultaneously) performs three acts: (i) she produces some sounds (a phonetic act); (ii) she produces some sounds as expressions in a particular language (a phatic act); and (iii) she produces some sounds as expressions in a particular language, with a certain sense and reference (a rhetic act). For example, Alice may produce the sounds ‘He smokes’, while producing those sounds as a sentence of English, and while also using ‘he’ to refer to Obama and ‘smoke’ to express the property of smoking. If Alice performs all three acts when she performs the preceding phonetic act, then she performs a locutionary act. Alice may also thereby perform an illocutionary act of asserting that Obama smokes; she performs this act if she (roughly speaking) commits herself to the truth of the proposition that Obama smokes, when she locutes that Obama smokes. Finally, Alice may perform the perlocutionary act of surprising her auditor. The term ‘utterance’ might reasonably be applied to mere phonetic acts, mere phatic acts, mere locutionary acts, or mere illocutionary acts. Should utterance-semantics assign contents to tokens or to speech acts? If speech acts, then which speech acts? Garcia-Carpintero (1998) favors ascribing contents to tokens. Reichenbach (1947) seems to favor tokens, but often seems to have speech acts in mind when he speaks of tokens. Korta and Perry (2011) favor speech acts. There are good reasons to favor speech acts. Consider a typical claim of utterance-semantics: Every utterance of ‘I’ refers to the agent of the utterance. An utterance-semantics that includes such a claim assumes that utterances have unique agents. It also assumes that utterances have associated times and locations. Speech acts have unique agents, times, and locations. But not all tokens do. There are tokens of ‘I love you’ that are produced by the wind. There are scattered tokens of ‘I love you’ whose parts are produced by several different people: consider a token of the sentence whose token of ‘I’ is produced by Alice, whose token of ‘love’ is produced by Betty, and whose token of ‘you’ is produced by Carla. If utterance-semantics should ascribe semantic contents to speech acts, then to which speech acts should it ascribe semantic contents? Consider another typical claim made by an utterance theory: for every utterance u, if u is an utterance of ‘you’ in English, then u refers to the person addressed by the agent of u at the time of u. An agent who performs merely the phonetic act of uttering the sounds ‘I love you’ need not refer to a person when doing so. An agent who performs the phatic act of uttering the sentence as a sentence of English need

234

braun

not do so, either. But if an agent utters ‘You smoke’ and thereby performs a locutionary act, she will utter that sentence as a sentence of English with a certain “sense” and reference. So, she will refer to the person she is addressing, and the content of the locutionary act will be that A (the addressee) smokes. So, we have now found a type of speech act, locutionary acts, whose contents can plausibly be identified with semantic contents. So, we can reasonably conclude that utterance-semantics should ascribe semantic contents to locutionary acts. (See Korta and Perry, 2011. To generalize this point to sub-sentential expressions, we must count utterances of them as locutionary acts. For instance, we must count acts of referring to people with ‘you’ as locutionary acts.) Thus, utterance theories can avoid the problem of “too many contents.” Now that we have distinguished among types of speech acts, and decided that utterance-semantics should deal with locutionary acts, we should start using different terminology. We should speak of locutionary acts rather than utterances, when describing “utterance” semantics, and we should also speak of uttering a sentence as a sentence in language L in a locutionary manner, rather than simply speaking of uttering. We should also introduce a new term for the latter kind of relation; a term like luttering might serve well. But I will not be so careful below. I shall immediately lapse back into speaking simply of utterances and uttering. I shall use the longer, but more precise, terminology only when necessary. I conclude that utterance-semanticists have a principled reply to the “too many contents” objection.

7

An Argument against Utterance-Semantics from Logic, and a Reply

7.1 Utterance-Semantics and Utterance-Logic David Kaplan often emphasizes the difference between expressions and utterances, and argues that logical properties, such as validity, are properties of expressions (Kaplan 1989a, pp. 522–523; 1989b, pp. 584–585). He argues that a logic for utterances would have strongly unintuitive consequences. Can his remarks about logic be used to argue against utterance-semantics? The answer is not obvious. Semantics deals with semantic properties, such as semantic contents and linguistic meanings. Logic does not, or need not, deal with all of the semantic properties that semantics describes. Logic need not, for example, describe the linguistic meanings of expressions. Logic must, admittedly, deal with truth, and with extensions more generally, in order to analyze validity or logical truth. More precisely, if standard theories of validity are correct, then logic must employ notions of truth and extension that are relativized

utterances and expressions in semantics and logic

235

to appropriate parameters, such as models or structures. But to do this, logic need only recognize the types of extensions that expressions have. For example, a logic for English must recognize that the extensions of unary predicates in English are sets of individuals. But it need not recognize that the intended extension of ‘blue’ in English is the set of blue things. And it certainly need not recognize other semantic properties of ‘blue’. But perhaps logic’s need to deal with extensions establishes enough of a connection between logic and semantics to argue against utterance-semantics. Utterance-semantics entails that utterances are the bearers of content, and so are bearers of truth. If utterances are bearers of truth, then the bearers of logical truth (validity) should be utterances as well. But if utterances are not bearers of logical truth (as Kaplan maintains), then we should conclude that utterances are not bearers of truth, and so utterance-semantics is misguided. That’s a first stab at an argument from logic against utterance-semantics. Let’s look more closely. 7.2 A Short Review of Kaplan’s Logic for Indexicals Let us first remind ourselves of some highlights of David Kaplan’s logic for indexicals (as opposed to his semantics), and the respects in which it is not a logic of utterances. (For a more detailed introduction, see Braun 2015b. For the full theory, see Kaplan 1989a.) Logical truth, or validity, is usually analyzed as truth in all models or structures of an appropriate sort. Kaplan’s logic for indexicals uses LD structures. (‘LD’ stands for ‘Logic of Demonstratives’.) Each LD structure is a sequence containing a domain of individuals, a set of “contexts,” a set of “times,” a set of “worlds,” a set of “positions”, a set of “individuals”, and an “interpretation” function that assigns, to each predicate and functor of the language, a function from “time-world” pairs in the structure to extensions of appropriate types within the structure. Each “context” in a structure has an associated “agent,” “time,” “world,” and “position” in that structure. Kaplan defines the truth-value of every formula with respect to each quadruple consisting of a “context,” a “time,” and a “world,” in a given structure. A formula φ is true with respect to “context” c in structure S iff φ is true with respect to “context” c, the “time” of c, and the “world” of c in S. A formula φ is valid iff φ is true with respect to every “context” in every structure. Let’s pause to make a few observations. First, in Kaplan’s theory, expressions (formulas) are both (a) the bearers of truth in a “context” (in a structure) and (b) the bearers of validity. Utterances do not bear either of these properties, on his theory. Second, LD structures are sequences of (a) sets of arbitrary things and (b) an “interpretation” function, which assigns, to each predicate and functor,

236

braun

a function from “times” and “worlds” in the structure to members or subsets of the previous sets. But in many LD structures the sets of “contexts”, “times”, “worlds”, and “positions” are merely sets of numbers. In others, they are sets of dogs. In yet others, they are sets of pineapples. So, the sets in many structures are not sets of contexts, times, worlds, and positions. That is why I used scare quotes when I wrote above about “contexts,” “worlds,” and so on, in structures. Third, among the LD structures is the intended structure. In the intended structure, the set of “contexts” is the real set of contexts, the set of “worlds” is the real set of possible worlds, the “agents” of the “contexts” are real agents of those contexts, and so on.18 Fourth, Kaplan does not, strictly speaking, define validity as truth in all contexts. Strictly speaking, he defines validity as truth in all “contexts” in all structures. Many of these “contexts,” in many of these structures, are not really contexts. Fifth, one major reason to define validity in terms of structures, including unintended structures, is to avoid conflating validity with necessity. All worlds in the intended structure are metaphysically possible.19 So, the sentence ‘No gnu is a number’ is true in all contexts in the intended structure. But that sentence is not logically valid. And it is not valid on Kaplan’s analysis, because it is false in some “context” in some unintended structure. Presumably an utterance-oriented logic would also need to use unintended structures to avoid conflating validity with necessity. Sixth, the sentences ‘I exist’ and ‘I am here now’ are valid, according to Kaplan’s theory. But to obtain this result, Kaplan restricts the class of LD structures to include only those structures in which, for instance, at every “context” the extension of ‘I’ is a member of the

18

19

Here are some more semi-technical details that some may wish to skip: The intended structure for an interpreted language has an interpretation function that assigns the intended intensions to the predicates and functors of that language. For instance, the interpretation function of the intended structure for English assigns to the predicates and functors of English the intensions that they have in English. Moreover, the intended structure determines a function that represents the intended character of each expression, a function from contexts to an intended intension. (This function can represent the intended character insofar as intensions can represent the contents of expressions at contexts.) It is controversial what possible worlds and contexts really are, and hence controversial what the sets of worlds and contexts in intended structures for interpreted languages are. Perhaps the set of possible worlds that appears in all intended structures is a set of properties that the (actual) universe metaphysically might have had. Perhaps the members of the set of contexts are possible events, each of which essentially occurs at a certain time, world, and location, and essentially has a certain agent standing in certain relations to other individuals (such as demonstrata); but in many the agent does not speak. See note 15.

utterances and expressions in semantics and logic

237

extension of ‘exists’ at the “time” and “world” of the “context”. Presumably, an utterance-oriented logic of indexicals would impose analogous restrictions on structures, and perhaps other restrictions as well. 7.3 An Utterance-Oriented Logic for Indexicals Familiar analyses of validity attribute validity to expressions. Is it possible to formulate a theory that attributes validity primarily to utterances? By ‘utterance’, I mean real (possible) utterances, not the “utterances” of unintended structures (which may be numbers, dogs, pineapples, and so on). If so, it seems that an utterance-theorist should prefer such a logic. After all, as the previous objection to utterance-semantics said, logical truth seems to be a kind of truth, and utterance-theorists attribute truth to utterances, not expressions. Here is a poor first attempt at a theory of utterance-validity: an utterance is valid iff it is true. This analysis is obviously incorrect: My recent utterance of ‘Obama was President of the USA in 2014’ is true, but there is no reasonable sense in which it is logically valid. A parallel view of validity that appeals to possible utterances, with respect to contexts, is equally poor: A possible utterance, with respect to a context, is valid iff it is true with respect to the world of that context. That analysis fails, because my actual utterance of ‘Obama was President of the USA in 2014’ was true with respect to the world of its context, but is not valid. Perhaps we need to use both contexts and worlds. Consider the content of a possible utterance, relative to a context. That content is either true or false at every world. So, we can doubly-relativize truth of possible utterances, and try the following analysis: A possible utterance, with respect to a context, is valid iff it is true with respect to that context and the world of that context, and it is true with respect to that context and every world whatsoever. Unfortunately, this analysis conflates validity with necessity. A possible utterance of ‘No gnu is a number’, with respect to any given context, is true with respect to all worlds whatsoever. But we should not count that utterance as valid. To avoid conflating validity with necessity, we seemingly need to evaluate possible utterances for truth at alternative structures. So, let us consider structures much like Kaplan’s LD-structures, but whose “interpretation” functions are defined on possible utterances rather than expressions. The possible utterances in the domains of these “interpretation” functions are the real possible utterances, coming from the intended structure. A possible utterance whose content, at a context in the intended structure, is true at all times and worlds in the intended structure, may fail to be true at all “times” and “worlds” in an alternative structure. However, a structure’s “interpretation” function must be able to recognize when two possible utterances are utterances of the same expression. Consider

238

braun

two possible utterances of ‘red’ (and assume ‘red’ is context-insensitive). An “interpretation” function in an alternative structure should assign the same function (intension) from “time-world” pairs to extensions to each of these possible utterances. So, it seems that the domain of a structure’s “interpretation” function should be all pairs ⟨u, e⟩, where u is a possible utterance and e is the expression of which u is a possible utterance. There remain some puzzling questions. Alternative structures should assign extensions to ⟨u, e⟩ pairs, relative to parameters in the alternative structures. But should the possible utterances of the intended structure “exist” in alternative structures? Standard logics assign extensions to expressions in structures, but the expressions themselves need not appear in the unintended structures’ domains of individuals. Should we allow possible utterances from the intended structure to be similarly absent from unintended structures? If possible utterances of the intended structure should be in the domains of all alternatives structures, then should each possible utterance be something that is “uttered” by some “agent” in some “context” in each alternative structure? Formulating an analysis of validity for (possible) utterances raises difficult questions. Doing so may require some cheating—we have already noted that “interpretation” functions of unintended structures “need to know” the expressions uttered in possible utterances. Fortunately, there is a simpler theory that is consistent with the spirit of utterance-semantics and utterance-logic, and which raises fewer difficult questions. We saw part of it in section 5.2 (see Kaplan 1999). We can define truth for a formula (an expression) with respect to a “context,” “time,” and “world,” in a structure, as Kaplan does, but add a decidedly un-Kaplanian restriction: a formula φ is true with respect to c, t, w, and S only if the “agent” of c “utters” φ at the “time” of c and the “world” of c, in S. More technically: only if ⟨the “agent” of c, the extension of φ in S⟩ is a member of the extension of ‘utter’ with respect to “context” c, the “time” of c, and the “world” of c, in S. (Notice that this proposal requires that each structure S assign an extension to the quotation-name of every expression of the language, or that it do something similar. I will not pause to consider how this should be done. But note that the extension of a quotation name at an alternative structure need not be the quoted expression itself.) This theory is fairly Kaplanian in format, but is much friendlier to utterances than Kaplan’s. It also allows us to dodge some of the difficult questions about “interpretation” functions that assign extensions to possible utterances from the intended structure. With a definition of truth at a “context,” “time,” “world,” and structure in hand, we can define various utterance-friendly restrictions and logical notions in appropriate ways, as follows.

utterances and expressions in semantics and logic

239

(3) φ is true with respect to c, t, and w (in S) only if: the “agent” of c “utters” φ at the “time” of c, and the “world” of c, in S. (That is, only if ⟨the “agent” of c, the extension of φ in S⟩ is a member of the extension of ‘utter’ with respect to c, the “time” of c, and the “world” of c, in S.) (4) φ is true with respect to c (in S) iff: φ is true with respect to c, the “time” of c and the “world” of c, in S. (5) φ is valid iff: For every structure S, if c is a “context” in S and the “agent” of c “utters” φ at the “time” of c and the “world” of c, then φ is true in S. (See above for a precise version of ‘the “agent” of c “utters” φ’.)20 The following definition of logical consequence is consistent with the previous definition of validity for formulas and also consistent with the spirit of utterance-logic in general. (6) φ is a logical consequence of a set of sentences Γ iff: for every structure S, if c is a “context” in S, and the “agent” of c “utters” φ and every member of Γ at the “time” and “world” of c, and every member of Γ is true in c in S, then φ is true in c in S. (From here on, let’s say that the pair ⟨ Γ, φ⟩ is an argument whose premise-set is Γ and whose conclusion is φ.) We can also follow Kaplan’s stipulations regarding ‘I’, ‘now’, and ‘here’. We can, for instance, stipulate that in all structures, the extension of ‘I’ with respect to any “context,” in any structure, is the “agent” of the “context”, in that structure. The above proposal does not ascribe validity, or logical truth, to utterances. It instead ascribes validity, and logical truth, to expressions. Nevertheless, this proposal is much more utterance-oriented than ordinary analyses of validity and logical truth, because it restricts truth-in-a-context to expressions that are uttered in that context. Should utterance-theorists accept this theory? It seems no worse for them than accepting that linguistic meanings are properties

20

The clause ‘the “agent” of c “utters” φ’ in (5) is needed, even though the definition of truth in a “context” restricts truth to “contexts” in which the “agent” “utters” the sentence. Consider a revised definition of validity that drops that clause, and consider the sentence ‘All dogs are dogs’, and a “context” in a structure in which the “agent” does not “utter” this sentence. The definition of truth in a context entails that the sentence is false in this context. So, on the revised definition of validity, ‘All dogs are dogs’ is not valid.

240

braun

of expressions rather than utterances. Moreover, once an utterance-theorist accepts this analysis of expression-validity, she can propose a derivative notion of validity for utterances: a particular utterance is valid iff it is an utterance of a valid expression. If an utterance-theorist does accept the above analyses, then she should reject the following premise of the anti-utterance-semantics argument I presented in section 7.1 above: “If utterances are the bearers of truth, then utterances are also the bearers of logical truth (validity)”.

8

Some Arguments against Utterance-Logic, and Some Replies

The above utterance-friendly analysis of validity has its own potential problems. The problems are interesting in their own right, whether or not they raise serious issues for utterance-semantics. (As I indicated above, I doubt that the alleged problems with utterance-logic show that there are problems with utterance-semantics.) I discuss some of these problems below, before I return to criticizing utterance-semantics. Many of these problems are mentioned by Kaplan (1989a, 1989b). I think at least some can be avoided by distinguishing carefully between intended and unintended structures. 8.1 Objection #1: ‘I utter something’ Consider the sentence ‘I utter something’. This sentence is true with respect to all “contexts” in all structures in which the “agent” of the “context” “utters” that sentence. So, on the analysis of validity in (5), it is valid. But it isn’t valid. So, the analysis is incorrect. (See Kaplan 1989a, p. 522, and Kaplan 1989b, pp. 584–585.) Reply: The above analysis of validity does entail that ‘I utter something’ is valid. There is no plausible way for an utterance-logician to design a logic to avoid this rather strange consequence. But perhaps an utterance-logician could reasonably maintain that she is trying to explicate a notion of validity that is rather different from classical notions and Kaplan’s notion. The sentence ‘I utter something’ is true whenever uttered by a speaker in a locutionary manner, and in virtue of her uttering it with locutionary intentions. That is a property of a sentence worth noting, and one might reasonably say that this property is a kind of validity, namely utterance-validity. 8.2 Objection #2: ‘I am speaking English’ Consider the sentence ‘I am speaking English’. This sentence is true with respect to all contexts in which the agent of the context utters that sentence. So, the above theory says that it is valid. But it isn’t. So, the analysis is wrong. (Kaplan 1989b, p. 585, note 40, mentions related issues.)

utterances and expressions in semantics and logic

241

Reply: ‘I am speaking English’ is a sentence of English, but it is also a sentence of many other languages. Consider, for instance, Schmenglish, a language like English except that ‘cat’ means dog and ‘dog’ means cat and ‘English’ refers to Schmenglish. A typical Schmenglish speaker who utters ‘I am speaking English’ is not speaking English. However, someone who utters that sentence as a sentence of English is speaking English. Now as we noted earlier, “utterance”-semantics should ascribe contents to locutionary acts. And the relevant uttering-relation for an utterance-semantics of English is that of uttering in English in a locutionary manner. It is a necessary truth that someone who utters ‘I am speaking English’ as a sentence of English in a locutionary manner utters a sentence that is true (in English), with respect to the speaker’s context. So, it is a necessary truth that if an agent of a context utters ‘I am speaking English’ in a locutionary manner in English, then the sentence ‘I am speaking English’ is true (in English) with respect to her context, the time of her context, and the world of her context. Finally, the above analysis of validity was meant to be friendly to utterance-semantics, which ascribes contents to locutionary acts. So, when I used ‘utterance’ and ‘utter’ in the analysis, I really meant locutionary act and uttering a sentence in English in locutionary manner. Put all this together, and we get the following: The analysis entails that ‘I am speaking English’ is true (in English) at every context in the intended structure in which the agent utters it as a sentence of English in a locutionary manner. But none of this entails that ‘I am speaking English’ is valid on the analysis of validity in (5). The proposed utterance-friendly analysis of validity involves quantification over all structures, including unintended structures. One of the points of using unintended structures is to allow there to be sentences that are necessary truths that are invalid. The sentence ‘I am speaking English’ is not true in every “context” of every unintended structure in which the “agent” of the “context” stands in the extension of ‘utter’ at the “time” and “world” of the context. (One reason this is not the case is that in some unintended structures, ‘English’ refers, at some contexts and worlds, to something whose extension does not include the sentence ‘I am speaking English’.) So, it is invalid. An utterance-theorist could, if she wishes, change the above analysis of validity by restricting the class of structures to ones in which ‘I am speaking English’ is true with respect to every “context” in which the “agent” “utters” ‘I am speaking English’ at the “time” and “world” of the “context.” She could do so by requiring that every structure satisfy the following restriction: If c is a “context” and the pair ⟨the “agent” of c, the extension of ‘I am speaking English’ in S⟩ is in the extension of ‘utter’ at the “time” and “world” of c in S, then ⟨the “agent” of c, the extension of ‘English’ in S⟩ is in the extension of ‘speak’ at the “time” and

242

braun

“world” of c in S. But ascribing validity to the sentence is not forced on those who favor an utterance-friendly logic of English. 8.3 Objection #3: ‘If that exists, then I intend to refer to it’ Consider the sentence ‘If that exists, then I intend to refer to it’.21 It is valid according to the analysis in (5), for in every context in which the agent utters it, the agent intends to refer to the context’s demonstratum. So, the above analysis entails that it is valid. But it is not. Reply: This objection raises many of the same issues as the previous one, plus a few more. Let us suppose that when the objector says ‘utter’, he means utter as a sentence of English in a locutionary manner, or e-lutter, for short. It is true that, for every context in the intended structure, if the agent e-lutters ‘If that exists, then I intend to refer to it’, then the agent has a demonstrative intention to refer. If the context has a demonstratum, then the antecedent and consequent are both true in the context. However, we want to allow for reference failure in some contexts. In reference failure contexts, the antecedent is either false or neither-true-nor-false. The consequent is true, assuming that the infinitival phrase semantically functions more or less in the way that ‘that I refer to it’ does. (If it does, then the consequent “says” that the agent stands in the intending relation to a gappy proposition.) If the antecedent of the conditional is false, then the whole conditional is true. If the antecedent is neither true nor false, then the truth-value of the whole is less clear. Perhaps the whole sentence is true (as the strong Kleene table for material conditionals says) or perhaps it is neither true nor false (as the weak Kleene table says). Suppose (following the strong Kleene table) that the conditional is true. Then the sentence ‘If that exists, then I intend to refer to it’ is true in all contexts in the intended structure. Nevertheless, it is false with respect to some “context” in some unintended structure, and so it is not valid on the above analysis. A theorist of utterance-logic might want to restrict the set of unintended structures in some way so that the sentence is true in all “contexts” of the restricted set of (intended and unintended) structures, and analyze validity as truth in all “contexts” in the restricted class. But that is merely an option. 8.4 Objection #4: Very Long Sentences Plausibly, some sentences are so long that it is metaphysically impossible for any agent to utter them. (See section 5.) An example might be the sentence

21

Read this sentence as follows: ‘That is such that: if it exists, then I intend to refer to it’. See Kaplan 1989b and Salmon 2002, and note 7 above, for discussion of related issues.

utterances and expressions in semantics and logic

243

consisting of ‘David Kaplan is a plumber’ conjoined with itself 101,000,000,000 times. Let φ be such a sentence, and assume that it is (actually) false. It is vacuously true that φ is true in all contexts in which the agent of the context utters it, for there are no such contexts. So, the analysis in (5) entails that φ is valid. But φ is not valid. So, the analysis is wrong. (See Kaplan 1989a, p. 522, and 1999, p. xxii, n. 35.) Reply: Let’s grant that there are metaphysically un-utterable sentences. It is vacuously true that such a sentence is true in every context in the intended structure in which the agent utters it. But nothing yet follows about the unintended structures. The “worlds” of unintended structures need not be possible. So, there are “contexts” in some unintended structures whose “worlds” are not possible worlds. So, there is a “context” c in some unintended structure such that the pair ⟨the “agent” of c, the extension of φ in S⟩ is a member of the extension of ‘utters’ at the “time” and “world” of c in S. So, given the analysis, the sentence is not valid. Perhaps a theorist of utterance-logic would wish to restrict the structures in such a way that there are no “contexts” in which metaphysically un-utterable sentences are “uttered.” But it is difficult to see what such a restriction would be, short of restricting the structures to the intended structure. That restriction would be undesirable, for it would conflate validity with necessity. In short, it is easier to design a reasonable utterance-logic on which such a sentence is invalid than it is to design a reasonable utterance-logic on which it is valid. 8.5 Objection #5: Logical Consequence and Infinite Premise Sets Consider the earlier analysis of “φ is a logical consequence of Γ”. Consider the infinite premise set {‘1 is a number’, ‘2 is a number’, ‘3 is a number’, … } and the conclusion sentence ‘David Kaplan is a plumber’. There is no context in which the agent utters every member of this infinite premise set. So it is vacuously true that, for every context in which the agent utters the conclusion and every member of the premise set, if every member of the premise set is true in that context, then so is the conclusion. So, the above analysis of logical consequence in (6) entails that ‘David Kaplan is a plumber’ is a logical consequence of the premise set. But it isn’t. So, the analysis is incorrect. Reply: The reply to this objection strongly parallels the previous reply to the previous objection. There are no contexts in the intended structure in which the agent of the context utters an infinite number of sentences at the time and world of the context. But there are “contexts” in unintended structures in which the “agent” “utters” the conclusion and all of the sentences in the infinite premise set, and in which all of the premise-sentences are true but the conclusion is false. So, the earlier analysis of logical consequence does not

244

braun

entail that the conclusion is a logical consequence of the premise set, and the objection fails. 8.6 Objection #6: Too Few Valid Sentences Each of the following sentences has the form ˹If φ, then φ˺ and so is valid. (7) If today is Tuesday, then today is Tuesday. (8) If I am sitting, then I am sitting. (9) If Fred is sitting, then Fred is sitting. Yet it is possible to utter each of these sentences falsely. Consider an utterance of (7) that begins a moment before midnight on Tuesday, but whose consequent is uttered just after midnight on Wednesday. Consider an utterance of (8) by Fred, who utters the antecedent while he is seated, utters ‘then’ while he is rising, and utters the consequent after he is standing. Consider an utterance of (9) by someone who is observing Fred, and who produces the antecedent and consequent of (9) at the same moments that Fred produces the antecedent and consequent of (8). For each of these sentences, there is a context in which the agent utters the sentence, and that sentence is false with respect to the context, time of the context, and world of the context. So, none is valid according to the proposed analysis of validity. So, that analysis is incorrect. (See Kaplan 1989a, pp. 522; 1989b, pp. 584–585.) Reply: The above objection raises difficult issues for all logical theories, whether utterance-oriented or expression-oriented. The sentences appear to be valid. But it is possible for a speaker to utter them and say something that is false. How can this apparent clash be resolved? One natural response is to claim that these sentences are uttered falsely only when there’s a shift in context in the middle of the utterances. Utterance theorists could say that all utterances of sentence (7)–(9) that occur within a single context are true. The utterance-friendly analysis of validity speaks of truth of a sentence in a context, and truth of a sentence in a context requires that the agent utter that sentence at the time and world of the context. That last restriction should be understood to mean that the agent utters the entire sentence within the time of the context. A parallel point holds for all “contexts” in all unintended structures. So, sentence (7) is true with respect to any “context” in which the “agent” “utters” the (entire) sentence at (within) the “time” of c, and in the “world” of c. The same holds for the other sentences. So, the utterancefriendly analysis of validity entails that these sentences are valid after all.

utterances and expressions in semantics and logic

245

This reply may rescue the utterance-friendly analysis of validity from the objection. But it raises questions about how to formulate an utterance-semantics for utterances of (7)–(9) in which the agent speaks falsely. The utterancetheorist now seems committed to assigning contents to utterances in which there has been an important shift in context in mid-utterance. Perhaps the utterance theorist should speak of the context of the entire utterance of the sentence, and break this context down into smaller sub-contexts, and distinguish between those super-contexts in which there has been a shift in time between its sub-contexts and those in which there has been no such shift. Such a change in utterance-semantics suggests that a similar change should be made in utterance-logic. The utterance-semantics described above might remind readers of the first expression-semantics that I proposed earlier for multiple occurrences of demonstratives (section 4). That version of expression-semantics allowed shifts in context in mid-sentence. Multiple occurrences of tensed verbs, such as those in (7)–(9) above, raise issues for semantics much like those of multiple occurrences of demonstratives. Such sentences raise issues for both utteranceoriented and expression-oriented semantics, and for both utterance-oriented and expression-oriented logic. It is not at all clear that one style of theory handles the semantic and logical issues raised by such sentences better than the other. In short, some sentences whose forms suggest that they are logical truths can be uttered falsely. This creates problems for all semanticists and logicians. It would be hasty to conclude that sentences (7)–(9) create more problems for utterance-semantics than for expression-semantics. 8.7 Summary Any utterance-oriented logic would differ significantly from classical logic and Kaplan’s logic of indexicals. Some of the objections to utterance-oriented logic that we considered might be met by distinguishing carefully between the intended structure and unintended structures. But not all can be met in this way, and perhaps that is a reason to favor expression-logic. However, none of these objections to utterance-logic tells us much about the acceptability of utterance-semantics.

9

Utterance-Semantics and the No Synonyms Objection

I now leave aside utterance-logic, and return to utterance-semantics. My final objection to utterance-semantics is the most serious that I will consider in this

246

braun

paper. Until recently, I thought it was fatal. But I now believe that utterancesemanticists can answer it, if they are willing to reformulate their view of linguistic meaning in a simple, but rather surprising, way. 9.1 The No Synonyms Objection The words ‘yell’ and ‘shout’ are synonyms in English. So are ‘flammable’ and ‘inflammable’, ‘flapjack’ and ‘hotcake’, ‘pail’ and ‘bucket’, ‘woodchuck’ and ‘groundhog’, ‘wealthy’ and ‘affluent’, and ‘baffle’ and ‘bewilder’.22 These words are synonymous in the strictest sense: they have the same linguistic meaning. But it appears that utterance-semantics would assign them different linguistic meanings, simply because they are different words. Consider, for the sake of simplicity, the version of utterance-semantics that takes the linguistic meaning of a word to be a function from its actual utterances to the contents of those utterances. On this view, the linguistic meaning of ‘flammable’ is a function whose domain is the set of all actual utterances of ‘flammable’; its value, for each utterance in the domain, is the property of being flammable, which is the same as the property of being inflammable. (From here on I will use ‘(in)flammable’ to express this property.) The linguistic meaning of ‘inflammable’ is a function whose domain is the set of all utterances of ‘inflammable’; its value is the property of being (in)flammable. But no utterance of ‘flammable’ is an utterance of ‘inflammable’. (More cautiously: not every utterance of ‘flammable’ is also an utterance of ‘inflammable’.) Therefore, these two functions have different domains, and so are distinct functions. So, on the view that linguistic meanings are functions from actual utterances to contents, the words ‘flammable’ and ‘inflammable’ have different linguistic meanings. But these expressions are synonyms.23 A parallel point holds for sentences that contain these synonyms. ‘Gasoline is flammable’ and ‘Gasoline is inflammable’ are synonymous, but their linguistic meanings would be distinct on the above view.24

22

23

24

An advocate of very-fine-grained contents might hold that ‘flammable’ and ‘inflammable’ differ in their prefixes, so differ in syntactic structure, and so differ in content. I will ignore this view, and continue to assume that the two words are strictly synonymous. Other pairs of synonyms would serve my purposes just as well. A parallel problem arises for theories that identify linguistic meanings with partially defined functions, all of which have the same domain, the class of all utterances. On such a view, the linguistic meanings of ‘flammable’ and ‘inflammable’ are undefined on different utterances, and so are different functions. As I mentioned in note 1 above, I discovered this apparent problem with utterancesemantics in 1998. I afterward discovered that Zimmermann (1997) had found a similar

utterances and expressions in semantics and logic

247

The same problem arises for versions of utterance-semantics that take linguistic meanings to be functions defined on possible utterances, or on pairs of possible utterances and contexts. Not every possible utterance of ‘flammable’ is also a possible utterance of ‘inflammable’. So, the functional linguistic meanings of these expressions are distinct on this view. A parallel problem also arises for versions of utterance-semantics that take linguistic meanings to be (genuine) relations between utterances and contents. The relational linguistic meaning of ‘flammable’ is a certain relation that holds between u and F iff u is an utterance of ‘flammable’ and F is the property of being (in)flammable. The relational linguistic meaning of ‘inflammable’ is a certain relation that holds between u and F iff u is an utterance of ‘inflammable’ and F is the property of being (in)flammable. The former relation holds between utterances of ‘flammable’ and the property of being (in)flammable. The latter relation does not. So, they are distinct relations. Inter-linguistic synonymy raises a closely related problem. The linguistic meaning of ‘dog’ in English is the same as linguistic meaning of ‘Hund’ in German. But not every (possible) utterance of ‘dog’ in English is also an utterance of ‘Hund’ in German. So, the versions of utterance-semantics that we considered above ascribe a different functional or relational linguistic meaning to ‘dog’ in English than they do to ‘Hund’ in German. But their linguistic meanings are the same, in their respective languages. That’s the objection. Kaplan-style expression-oriented theories of linguistic meaning avoid this problem. On those theories, the linguistic meaning of ‘flammable’ is a function whose domain is the set of all contexts, no matter what words the agents of these contexts utter (if any). The function’s value for each such context is the property of being (in)flammable. The functional linguistic meaning of ‘inflammable’ has the same domain of contexts, and the same value for each context in the domain. So, according to these theories, the words have the same linguistic meanings. Let’s call the above objection the No Synonyms Objection, and consider some replies on behalf of utterance-semantics. 9.2 First Reply, and a Response First reply: The No Synonyms Objection mis-describes the relation of having the same linguistic meaning. The expressions ‘flammable’ and ‘inflammable’ problem with synonymy for semantic theories that deal only with what he calls “possible utterances.” I have attempted to clarify the problem, and generalize it so that (a) it applies to other utterance theories, such as those I mention in the paragraph that follows immediately below, and (b) it applies to inter-linguistic synonymy.

248

braun

share the following property: being an expression all of whose utterances have as their semantic contents the property of being (in)flammable. This is sufficient for them to have the same linguistic meaning. Similarly, ‘yell’ and ‘shout’ share the following property: being an expression all of whose utterances have as their semantic contents the property of yelling (which is identical with the property of shouting). Generalizing, we get the following analysis of having the same linguistic meaning: (10) Expression E has the same linguistic meaning as expression Eʹ iff: There is a function F that is the linguistic meaning of E, and a function Fʹ that is the linguistic meaning of Eʹ, and a semantic content C such that: for all utterances u of E, F(u)=C and for all utterances of uʹ of Eʹ, Fʹ(uʹ)=C. The generalization can be easily modified if linguistic meaning is a function on possible utterances, or on pairs of possible utterances and contexts, or is instead a genuine relation. Response to the First Reply: Analysis (10) of the relation having the same linguistic meaning is unacceptable. Suppose two expressions have linguistic meanings. Then they have the same linguistic meaning if and only if the linguistic meaning of the first is identical with the linguistic meaning of the second. But (10) allows two expressions to have the same linguistic meaning, though their linguistic meanings are not identical. That is bizarre. In any case, the proposed analysis in (10) does not generalize well to pairs of indexicals that have the same linguistic meaning. Consider a language, Schmenglish, that is just like English except that it contains an additional indexical, ‘Schmi’, that is (seemingly) synonymous with ‘I’. In particular, the content of every utterance of ‘I’ in this language is the agent of that utterance, and the content of every utterance of ‘Schmi’ is the agent of that utterance. These expressions apparently have the same linguistic meaning. But utterances of ‘I’ vary in content, as do utterances of ‘Schmi’. So, there is no content that all utterances of ‘I’ have, and no content that all utterances of ‘Schmi’ have. So, the analysis in (10) entails that they do not have the same linguistic meaning. 9.3 Second Reply, and a Response Second reply (a revised version of the First Reply): The No Synonyms Objection mis-describes the relation of having the same meaning as. Consider ‘I’ and ‘Schmi’ in Schmenglish. They have different (functional) linguistic meanings in Schmenglish. But there is a single relation, rule, or function that is distinct from the linguistic meanings of ‘I’ and ‘Schmi’, but which yields the content of an utterance of either expression. For example, the relevant function for ‘I’

utterances and expressions in semantics and logic

249

and ‘Schmi’ takes both utterances of ‘I’ and utterances of ‘Schmi’, and yields, for any utterance of either expression, the agent of that utterance. The existence of such a function is sufficient for the two words to bear the having the same linguistic meaning relation to each other in Schmenglish. Generalizing, we get the following analysis. (11) Expression E has the same linguistic meaning as expression Eʹ iff: (a) there is a function F, defined on all and only utterances of E, that is the linguistic meaning of E and (b) there is a function Fʹ, defined on all and only utterances of Eʹ, that is the linguistic meaning of Eʹ, and (c) there is a function G, defined on all and only utterances of either E and Eʹ, such that, for all utterances u of E, G(u)=F(u), and for all utterances uʹ of Eʹ, G(uʹ)=Fʹ(uʹ). The analysis can be modified to accommodate the view that linguistic meanings are functions on possible utterances, or functions on pairs of possible utterances and contexts, or genuine relations between utterances and contents. Response to Second Reply: The analysis of having the same linguistic meaning proposed in (11) preserves the conspicuously bizarre aspect of (10) that I pointed out earlier, namely that it allows two expressions to have the same linguistic meaning, though the linguistic meaning of one is not identical with the linguistic meaning of the other. Besides that, the analysis of having the same linguistic meaning in (11) is far too liberal. ‘Cat’ and ‘dog’ do not have the same linguistic meaning. Yet there is a single function G whose domain is the set of all utterances of either ‘cat’ or ‘dog’, and whose value for any utterance of ‘cat’ yields the property of being a cat, and whose value for any utterance of ‘dog’ is the property of being a dog. So, according to the preceding analysis, ‘cat’ and ‘dog’ have the same linguistic meaning. (A parallel problem holds for versions of (11) on which linguistic meanings are functions on possible utterances, or pairs of possible utterances and contexts, or are genuine relations.) 9.4 Third Reply, and a Response Third reply: The No Synonyms Objection mis-describes the linguistic meanings of ‘flammable’ and ‘inflammable’. The linguistic meaning of both is the function whose domain is the set of all utterances of either ‘flammable’ or ‘inflammable’, and whose value, given an utterance of either word, is the property of being (in)flammable. So, the correct utterance-semantic theory entails that the linguistic meanings of the two words are identical. Another example: the linguistic meaning of both ‘I’ and ‘Schmi’ in Schmenglish is the function whose domain is the set of all utterances of either expression and whose value,

250

braun

for each such utterance, is the agent of the utterance. More generally, if two expressions have the same linguistic meaning, then the linguistic meaning of each is a function whose domain is the set of all utterances of either expression, and whose value for any utterance in the domain is a content appropriate to that utterance. (The view can easily be modified to accommodate the view that the linguistic meaning of a word is a function from possible utterances to contents, or the view that it is a function from pairs of possible utterances contexts, or the view that it is a relation that holds between utterances of a word and contents.) Response to the Third Reply: The Third Reply has a number of strange, unacceptable consequences. A strange epistemic consequence: An utterance-semanticist who wishes to discover the linguistic meaning of the word ‘flammable’ in English cannot do so by discovering the contents of utterances of ‘flammable’ alone. She must also discover the contents of utterances of ‘inflammable’ in English. If she is unaware of the existence of the word ‘inflammable’, then she is bound to overlook the (alleged) fact that the domain of the linguistic meaning of ‘flammable’ includes utterances of the word ‘inflammable’, and so she is bound to mis-describe the linguistic meaning of ‘flammable’. But that consequence seems preposterous. A strange consequence for linguistic change: Suppose that English speakers coin a new term for expressing the property of being (in)flammable, say ‘enfireable’ (ignore the complex morphology). Once current English speakers add this item to their language, the linguistic meanings of ‘flammable’ and ‘inflammable’ in the enhanced language must be different from their meanings in the old language, for the domains of their functional linguistic meanings in the new language now include utterances of ‘enfireable’. This is bizarre. A strange epistemic consequence for cross-linguistic synonymy: ‘Dog’ has the same linguistic meaning (in English) as ‘Hund’ (in German), ‘chien’ (in French), ‘perro’ (in Spanish), and so on for many languages. So, the Third Reply seemingly entails that the linguistic meaning of ‘dog’ (in English) is a function whose domain includes certain utterances of ‘Hund’, certain utterances of ‘chien’, certain utterances of ‘perro’, and so on, for utterances of all other expressions with which ‘dog’ is cross-linguistically synonymous. Hence, to specify the meaning of ‘dog’ in English, one must also know of the existence of synonymous words in all other languages. This makes the description of the linguistic meaning of a word within a single language unachievable. The Third Reply tries to answer the No Synonyms Objection by widening the domain of linguistic meanings to include utterances of more than one expression. It fails. But perhaps it fails because it does not widen the domains of the functions enough.

utterances and expressions in semantics and logic

251

9.5 Fourth Reply, an Objection, and a Response Fourth reply: The No Synonyms Objection mis-describes the linguistic meanings of ‘flammable’ and ‘inflammable’. The linguistic meaning of ‘flammable’ in English is the function whose domain is the set of all (locutionary) utterances of all expressions in all languages. The function’s domain includes not only all (locutionary) utterances of ‘flammable’ and ‘inflammable’ in English, but also all (locutionary) utterances of ‘blue’, ‘apple’, ‘the’, ‘Hund’, ‘perro’ and so on, made in any language whatsoever. This function’s value for all of these utterances is the property of being (in)flammable. This function is also the linguistic meaning of ‘inflammable’ in English. So, the two expressions are synonyms of English, according to the correct utterance-semantic view of their linguistic meanings. The preceding view of linguistic meaning is invulnerable to the earlier objections to the Third Reply. (Recall that the Third Reply says that the domain of the linguistic meaning of ‘flammable’ is merely the set of all utterances of either ‘flammable’ or ‘inflammable’). An investigator could fully describe the linguistic meaning of ‘flammable’ in English without knowing whether English includes the word ‘inflammable’. She could do so simply by specifying that the domain of the functional linguistic meaning of ‘flammable’ is the set of all utterances of all expressions in all languages. If the language of the community that she is investigating acquired a new synonymous word, such as ‘enfireable’, then the linguistic meaning of ‘flammable’ of that community’s language would remain unchanged, according to her theory: the linguistic meaning of ‘flammable’ would still be the function whose domain contains all utterances of all words in all languages, and whose value is always the property of being (in)flammable. Cross-linguistic synonymy presents no problem in specifying the meaning of an expression in English. A person describing the linguistic meaning of ‘dog’ in English could specify the domain of this function without knowing whether there are, or could be, utterances of words in other languages whose contents are also the property of being a dog, as long as she specifies that the linguistic meaning is a function on all utterances of all expressions in all languages. This proposal is also invulnerable to the objections to the First and Second Reply. On the new view, ‘I’ and ‘Schmi’ have the same linguistic meaning in Schmenglish, because the linguistic meaning of ‘I’ in Schmenglish is identical with the linguistic meaning of ‘Schmi’: it is the function whose domain is all utterances of all expressions in all languages, and whose value for any such utterance is the agent of the utterance. ‘Cat’ and ‘dog’ do not have the same linguistic meaning in English under this latest proposal. The linguistic meaning of ‘cat’ in English is a function that yields the property of being a cat when

252

braun

applied to any utterance, whereas the linguistic meaning of ‘dog’ in English is a function that yields the property of being a dog when applied to any utterance. The above view of linguistic meaning could be modified to accommodate the other utterance-theoretic views of linguistic meaning mentioned in section 5. The linguistic meaning of ‘flammable’ in English could be (a) the function whose domain is the set of all possible utterances of any expression in any language or (b) the function whose domain is the set of all pairs of (i) all possible utterances of all expressions, in all languages, and (ii) contexts. The value of these functions would always be the property of being inflammable. Alternatively, the linguistic meaning could be (c) a genuine relation that holds between (and only between) any utterance of any expression in any language and the property of being (in)flammable. On all of these views, ‘flammable’ and ‘inflammable’ have the same linguistic meaning in English. For the sake of simplicity, I shall continue to concentrate below on the simplest functional theory of linguistic meaning. Objection: The domain of an expression’s linguistic meaning should not include utterances of all expressions in all languages. Consider the abovehypothesized linguistic meaning of ‘flammable’ in English. The domain of this function includes utterances of ‘blue’ in English. Applying this function to utterances of ‘blue’ yields the property of being (in)flammable. But the content of an utterance of ‘blue’ in English is not the property of being (in)flammable. So, the Fourth Reply’s view of linguistic meaning is incorrect.25 Reply to the Objection: To obtain the content of an utterance of ‘blue’ in English you should not apply the linguistic meaning of ‘flammable’ in English. Rather, you should apply the linguistic meaning of ‘blue’ in English. (Obviously.) This latter function yields the correct content of any utterance of ‘blue’ in English. More generally, the content in English of an utterance of expression E in English is the value of the linguistic meaning of E in English when applied to that utterance. Even more generally, if one wants to compute the content of an utterance u in L of expression E in L, one should apply the linguistic meaning of E in L to that utterance. (12) states this fact a bit more formally. (12) C is the content of u in L iff: There is an E such that: E is an expression of L, M is the linguistic meaning of E in L, u is a locutionary utterance of E in L, and M(u)=C. 25

This objection is equally relevant to the version of the preceding view that uses genuine relations, for we can define the domain of a (genuine) binary relation in the following rather standard way: x is a member of the domain of (genuine) binary relation R iff there is a y such that x bears R to y.

utterances and expressions in semantics and logic

253

All linguistic meanings, of all expressions in all languages, have the same domain. But only one of these functions is guaranteed to yield the correct content of an utterance of an expression E in language L, and that function is (unsurprisingly) the linguistic meaning of E in L. 9.6 Reflections on the Preceding The Fourth Reply answers the No Synonyms Objection by widening the domains of all linguistic meanings to include all utterances of all expressions in all languages. I tentatively believe that this reply to the No Synonyms Objection is adequate. It might be surprising that such a strange, but simple, device would work. But perhaps it shouldn’t be surprising. Utterance-semantics had trouble with synonymy because its linguistic meanings paid too much attention to the expressions uttered in utterances. The new linguistic meanings pay no attention to the expressions uttered in utterances, for each has in its domain all utterances of all expressions in all languages. Kaplan’s theory is invulnerable to the Synonymy Objection for an analogous reason: all of his linguistic meanings (characters) are defined on the same domain of contexts and pay no attention to the expressions (if any) that the agents of the contexts utter. If Kaplan’s theory evades problems with synonymy, then it seems likely that the new linguistic meanings of the Fourth Reply do as well. The new linguistic meanings are, admittedly, rather strange. Their domains include an amazingly disparate variety of actions. For example, in some languages, an act of throwing a cabbage at a king is an utterance of an expression, and an act of placing a shoe on top of some sealing wax is an utterance of a different expression. According to the Fourth Reply, all such utterances are included in the domain of every linguistic meaning in every language. Perhaps the heterogeneity of this domain will worry some. But they should note that the set of utterances on which these linguistic meanings are defined is no more varied than the set of contexts on which Kaplanian characters are defined. In fact, the new linguistic meanings are equivalent, in a certain sense, to Kaplanian-characters that are restricted to certain sorts of Kaplan-style contexts. (Thanks to Alex Radulescu for pointing this out to me.) In section 5.1, I pointed out that a single possible utterance can vary in content from world to world. So, in section 5.2, I considered an utterance-semantics that took linguistic meanings to be functions from pairs of utterances and contexts to contents. The new linguistic meanings described in this section are functions from all utterances of all expressions in all languages to contents. But if this view is to avoid the problem of variation in a single utterance’s content from world to world, then the new linguistic meanings must really take the domains of lin-

254

braun

guistic meanings to be pairs of (a) utterances in which an agent utters some expression of some language and (b) contexts that include at least worlds. But each utterance determines a unique speaker, time, and location at each world. So, we can “re-conceptualize” this utterance-semantics as a Kaplan-style expression-semantics, but with an un-Kaplanian constraint (which I italicize): the linguistic meaning of an expression is a function from Kaplan-style contexts in which the agent utters some expression in some language (at the time and world of the context) to contents. So, the newest view of linguistic meanings is much closer to Kaplan’s theory than perhaps some utterance-theorists would like. There are some remaining differences. One is that Kaplan’s characters for pure indexicals are total functions on his contexts. For example, the character of ‘I’ is a function that yields the agent of c for all contexts. The new view denies this: the linguistic meaning of ‘I’ yields a value at a (Kaplan-style) context only if the agent utters something. (The uttered expression need not be ‘I’, or any word with the same linguistic meaning, but something must be uttered.) So, from the Kaplanian perspective, the linguistic meaning of the new “reconceptualized” utterance-view are partial functions. There are perhaps more significant differences regarding truthin-all-contexts (in the intended structure). Kaplan’s view entails that ‘I exist’ is true at all contexts whatsoever, whether or not the agent utters something at the time and world of the context. The new utterance-oriented theory entails that ‘I exist’ is true at all contexts in which the agent utters something. (Again, the uttered expression need not be ‘I exist’.) On the new re-conceptualized utterance-semantics, the linguistic meaning of ‘I utter something’ yields a true proposition at every context (in the intended structure) on which it is defined. But on Kaplan’s view, there are plenty of contexts on which the character of ‘I utter something’ is defined for which it yields a false proposition. How significant are these remaining differences between the pure Kaplanian expression-semantics and the newest utterance-semantics? I am unsure. But I find it difficult to justify the new utterance-theory’s particular restrictions on the domains of linguistic meanings. According to the new utterancesemantics, the linguistic meaning of ‘I exist’ yields a content when applied to contexts in which the agent utters the Japanese translation of ‘All dogs bark’. And yet that same linguistic meaning cannot be applied to a context in which an English-speaking agent utters nothing at the time of the context, even if the agent uttered that sentence at a time just slightly before the time of the context and will again utter that sentence at a slightly later time. I suspect that this way of discriminating among contexts does not yield any theoretically interesting results.

utterances and expressions in semantics and logic

10

255

Conclusion

The most serious potential problems for utterance-semantics are the nonsynonymy of pairs of metaphysically un-utterable sentences and the synonymy of distinct simple expressions. But utterance-semanticists can avoid the first problem by taking linguistic meanings to be genuine relations between utterances and contents, or by taking the linguistic meanings of complex expressions to be structured linguistic meanings. They can avoid the second problem by taking the linguistic meanings of simple expressions to be functions or relations that hold between all utterances of all expressions of all languages (and contexts), on the one hand, and appropriate contents, on the other hand. The defensible versions of utterance-semantics more closely resemble expression-semantics than one might expect. When the defensible versions of utterance-semantics are reformulated in a more Kaplanian format, the most obvious difference is that Kaplan’s linguistic meanings apply to contexts in which the agent utters nothing, whereas the linguistic meanings of utterancesemantics require that the agent utter something (though not necessarily the expression whose content we seek). But as far as I can see, there are no compelling theoretical reasons to restrict the domains of linguistic meanings in the way that (the Kaplanian version of) utterance-semantics does. Moreover, utterance-semantics is inconsistent with the traditional distinction between semantics and pragmatics that I described at the beginning of this paper. That traditional distinction does seem wellmotivated and fruitful. I tentatively conclude that we still have some theoretical reason to prefer expression-semantics over utterance-semantics.26

26

This paper has an unusual history. I wrote the first version of it in 1998–1999, and presented parts of it at a semantics workshop at Cornell University in 1999. I set it aside until 2015, when I substantially revised it in response to an invitation from Ken Turner and Larry Horn to contribute to the present volume. I thank Jason Stanley and Zoltán Szabó for inviting me to the 1999 Cornell workshop. Sally McConnell-Ginet was my commentator at that event and I thank her for her useful and entertaining comments. Thanks also for questions and comments at Cornell from Carl Ginet, Harold Hodes, David Kaplan, Jeffrey King, Michael Glanzberg, Craige Roberts, Jason Stanley, and Zoltán Szabó. Thanks to Kent Bach for encouraging me to revise and publish the 1999 draft, and for comments on a 2015 draft. I presented an abbreviated version of this paper at the Kline Workshop on ContextSensitivity at the University of Missouri in November 2016. Thanks to Marina Folescu, Michael Glanzberg, Claire Horisk, Matthew McGrath, Gillian Russell, and Peter Vallentyne for their questions and comments. Thanks especially to Alexandru Radulescu for inviting me to the workshop, for his spoken and written comments, and for his extensive conversa-

256

braun

References Atlas, Jay David. 2005. Logic, Meaning, and Conversation. Oxford: Oxford University Press. Austin, J.L. 1975. How to Do Things with Words, Second Edition (J.O. Urmson and Marina Sbisà, Eds.). Cambridge, MA: Harvard University Press. Bach, Kent. 2005. “Context Ex Machina.” In Zoltán Szabó (ed.), Semantics vs. Pragmatics, pp. 15–44. New York: Oxford University Press. Bach, Kent. 2006. “What Does It Take to Refer?” In Ernest Lepore and Barry C. Smith (Eds.), The Oxford Handbook of Philosophy of Language, pp. 515–554. New York: Oxford University Press. Bach, Kent. 2012. “Context Dependence.” In Manuel Garcia-Carpintero and Max Kölbel (Eds.), The Continuum Companion to the Philosophy of Language, pp. 153–184. New York, NY: Continuum International Publishing Group. Bach, Kent. Forthcoming. “Reference, Intentions, and Context: Do Demonstratives Really Refer?” In Korta and Maria de Ponte Azkarate (Eds.), Reference and Representation in Thought and Language. Barwise, Jon, and Perry, John. 1983. Situations and Attitudes. Cambridge, MA: MIT Press. Braun, David. 1995. “What Is Character?” Journal of Philosophical Logic 24, pp. 227–240. Braun, David. 1994. “Structured Characters and Complex Demonstratives.” Philosophical Studies 74, pp. 193–219. Braun, David. 1996. “Demonstratives and Their Linguistic Meanings.” Noûs 30, pp. 145– 173. Braun, David. 2005. “Empty Names, Fictional Names, Mythical Names.” Noûs 39, pp. 596–631. Braun, David. 2008. “Complex Demonstratives and Their Singular Contents.”Linguistics and Philosophy 31, pp. 57–99. Braun, David. 2015a. “Comments on Pickel, Rabern, and Dever’s ‘A Solution to the Problem of Recurring Demonstratives’.” Talk presented at the Pacific division meeting of the American Philosophical Association in Vancouver, BC on April 1, 2015. Braun, David. 2015b. “Indexicals.” Stanford Encyclopedia of Philosophy. http://plato .stanford.edu/entries/indexicals/. Crimmins, Mark. 1995. “Contextuality, Iteration, Reflexivity, Logic.” Philosophical Perspectives 9, pp. 381–399. Donnellan, Keith. 1970. “Proper Names and Identifying Descriptions.” Synthese 21, pp. 335–358.

tions with me about this paper and related topics. Finally, thanks to Ken Turner and Larry Horn for inviting me to contribute to this volume.

utterances and expressions in semantics and logic

257

Garcia-Carpintero, Manuel. 1998. “Indexicals as Token Reflexives.” Mind 107, pp. 529– 563. Gauker, Christopher. 2015. “How Many Bare Demonstratives are there in English?” Linguistics and Philosophy 37, pp. 291–314 Georgi, Geoff. 2015. “Logic for Languages Containing Referentially Promiscuous Expressions.” Journal of Philosophical Logic 44, pp. 429–451 Grice, H.P. 1957. “Meaning.” Philosophical Review 66, pp. 377–388. Heck, Richard. 2014. “Semantics and Context-Dependence: Towards a Strawsonian Account.” In Alexis Burgess and Brett Sherman (Eds.), Metasemantics: New Essays on the Foundations of Meaning, pp. 327–364. New York: Oxford University Press. Kaplan, David. 1989a. “Demonstratives.” In Joseph Almog, John Perry, and Howard Wettstein (Eds.), Themes from Kaplan, pp. 481–563. Oxford: Oxford University Press. Kaplan, David. 1989b. “Afterthoughts.” In Joseph Almog, John Perry, and Howard Wettstein (Eds.), Themes from Kaplan, pp. 565–613. Oxford: Oxford University Press. Kaplan, David. 1990. “Words.” Proceedings of the Aristotelian Society 54, pp. 93–119. Kaplan, David. 1999. “Reichenbachs Grundzüge der Symbolischen Logik,” translated from English by A. Kamlah. In Hans Reichenbach (A. Kamlah and Maria Reichenbach, Eds.), Gesammelte Werke in 9 Bänden, Band 6: Grundzüge der Symbolischen Logik (German translation of Reichenbach’s Elements of Symbolic Logic), pp. xi–xxii. Vienna: Springer-Verlag Wien. Korta, Kepa and John Perry. 2011. Critical Pragmatics: An Inquiry into Reference and Communication. Cambridge: Cambridge University Press. Kripke, Saul. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press. Lewis, David. 1975. “Languages and Language.” In Keith Gunderson (ed.), Minnesota Studies in the Philosophy of Science, Volume VII, pp. 513–543. Minneapolis: University of Minnesota Press. Reprinted in David Lewis, 1983, Philosophical Papers, Volume I, pp. 163–188, Oxford: Oxford University Press. Neale, Stephen. Forthcoming. “Silent Reference.” G. Ostertag (Ed.), Meanings and Other Things: Essays in Honor of Stephen Schiffer. Oxford University Press. Orilia, Francesco and Swoyer, Chris. 2016. “Properties.” In Edward N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Winter 2016 Edition). URL = http://plato .stanford.edu/archives/win2016/entries/properties/. Perry, John. 1997. “Indexicals and Demonstratives.” In Bob Hale and Crispin Wright (Eds.), A Companion to Philosophy of Language, pp. 586–612. Oxford: Blackwell. Perry, John. 2001. Reference and Reflexivity. Stanford, CA: CSLI Publications. Perry, John. 2006. “Using Indexicals.” In Michael Devitt and Richard Hanley (Eds.), The Blackwell Guide to the Philosophy of Language, pp. 314–344. Oxford: Blackwell. Perry, John. 2007. “Situating Semantics: A Response.” In Michael O’Rourke and Corey Washington (Eds.), Situating Semantics: Essays on the Philosophy of John Perry, pp. 507–575. Cambridge, MA: MIT Press.

258

braun

Pickel, Bryan, Brian Rabern, and Josh Dever. 2015. “A Solution to the Problem of Recurring Demonstratives.” Talk presented at the Pacific division meeting of the American Philosophical Association in Vancouver, BC on April 1, 2015. Reichenbach, Hans. 1947. Elements of Symbolic Logic. New York: Macmillan. Richard, Mark. 1983. “Direct Reference and Ascriptions of Belief.” Journal of Philosophical Logic 12, pp. 425–452. Salmon, Nathan. 1987. “Existence.” Philosophical Perspectives 1, pp. 49–108. Reprinted in Salmon 2005b, pp. 9–49. Salmon, Nathan. 2002. “Demonstrating and Necessity.”Philosophical Review 111, pp. 497– 537. Salmon, Nathan. 2004a. “The Good, the Bad, and the Ugly.” In A. Bezuidenout and M. Reimer (eds.), Descriptions and Beyond, pp. 230–260. New York: Oxford University Press. Reprinted in Salmon 2007, pp. 309–339. Salmon, Nathan. 2004b. “Two Conceptions of Semantics.” In Zoltan Szabó (ed.), Semantics and Pragmatics, pp. 317–328. New York: Oxford University Press. Reprinted in Salmon 2007, pp. 340–349. Salmon, Nathan. 2005a. “Demonstrating and Necessity.” Philosophical Review 111, pp. 497–537. Reprinted in Salmon 2007, pp. 67–99. Salmon, Nathan. 2005b. Metaphysics, Mathematics, and Meaning: Philosophical Papers Volume 1. New York: Oxford University Press. Salmon, Nathan. 2007. Content, Cognition, and Communication: Philosophical Papers, Volume 2. New York: Oxford University Press. Schiffer, Stephen. 1972. Meaning. Oxford: Oxford University Press. Schiffer, Stephen. 2005. “Russell’s Theory of Descriptions.” Mind 114, pp. 1135–1183 Soames, Scott. 2002. Beyond Rigidity. New York: Oxford University Press. Soames, Scott. 2015. Rethinking Language, Mind, and Meaning. Princeton, NJ: Princeton University Press. Wettstein, Howard. 1986. “Has Semantics Rested on a Mistake?” Journal of Philosophy 83, pp. 185–209. Zimmermann, Thomas. 1997. “The Addressing Puzzle.” In Wolfgang Künne, Albert Newen, and Martin Anduschus (Eds.), Direct Reference, Indexicality, and Propositional Attitudes, pp. 133–153. Stanford: CSLI Publications.

part 3 On Grammar, Inference, and Truth

∵

chapter 9

Grammar as Procedures: Language, Interaction, and the Predictive Turn Ruth Kempson and Ronnie Cann

1

Grammar, Processing and Interaction

Building on the trend initiated in part by the insight of Atlas 1977 that linguistic content had to be seen in terms more abstract than is expressible in denotational terms, at least for negation, this chapter argues for a new perspective on language in which language is characterised in procedural “knowing how” terms and emerges as a natural subpart of an overall cognitive system, bringing to it the added dimension of cross-party interaction. This is achieved by bringing together two independent strands of research. On the one hand, we sketch the Dynamic Syntax view of language (Dynamic Syntax: DS Cann et al. 2005), in which the concept of procedure is central: languages are defined to be sets of conditional actions incrementally inducing the building up of representations of content relative to emergent and shifting contexts to yield effects of linearisation (production) and interpretation (parsing). Furthermore, because these procedures constitute a grammar, producer and hearer are expected to use the same predictive procedures for string-content processing, both incrementally building up structured representations relative to context and some anticipated goal; so interaction between conversational agents is directly anticipated. In then turning to a broader cognitive science background, we find confirming evidence of this perception-as-action perspective in the way this dynamic is shared by the brain itself. The brain is claimed to be a predictive engine using immediate and encyclopaedic context constructively at every step to predict the structure/shape of the incoming sensory array (The Predictive Processing (PP) model of cognition: A. Clark 2013, 2016). Within this model, as in the DS account of language, all activity is seen as process-based and embodied: action and perception alike involve anticipatory projections as to what the input stimulus might be providing, with that stimulus acting as a filter, confirming or correcting those anticipations. Given the consonance of these two models, and adding to the PP framework the dimension of interaction which the combined assumptions of DS and PP yield, language emerges as one of a number of cog-

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_010

262

kempson and cann

nitive processes which lead to manifest joint interaction through the embodiment of procedures for interacting with the external world (others being music making and dance: Cross 2012, Pezzulo 2011). With this move, the competence-performance divide is left behind in favour of a model in which the very “knowledge of language” itself is defined procedurally as involving mechanisms whose effects are directly witnessed in actual processing behaviours (parsing and production). And from this interdisciplinary marriage, we have the promise of a wholly new perspective on language evolution following the pattern of first-language acquisition as grounded in the effectiveness of language as an evolved set of mechanisms for human interaction. In order to be able to appreciate the significance of this interdisciplinary union, there are two important sub-tasks to address: (i) why the DS model of language is a grammar formalism in its own right and not just a model of some aspect of information growth, parasitic on some static grammar formalism; (ii) what evidence there is for claiming that language is grounded in actions yielding interaction between participants through such growth. 1.1 Dialogue Interactivity The starting point of our argument is the widespread recognition of endemic context-relativity of both natural language interpretation and language production, so in modelling language processing, both perception and production must be seen as involving not merely the concept of evolving content but also its twin concept of perpetually evolving context (Pickering and Garrod 2004, 2013, and for a useful survey see Clifton et al 2012). Some of the most striking evidence that modelling language processing involves the articulation of evolving representations of content comes from the data of conversational dialogue. In informal conversational exchanges, notably the only data to which a small child is exposed, the emergent conversations are littered with so-called fragments where speakers and hearers simply assume that in some sense to be made explicit, the context provides all that is needed to see their apparently fragmentary contribution as providing a coherent and consistent add-on to what has just preceded. We can join one another’s conversations, becoming a new speaker while the previous speaker shifts into being a hearer, a shifting between speaker/hearer roles taking place repeatedly during the course of the exchange. Indeed, this shift of roles is the sine qua non of informal conversations, monologues being the exception rather than the norm: (1) A: I need a a B: mattock. For breaking up clods of earth

[BNC]

grammar as procedures

(2) Jack: I just returned Kathy: from Jack: Finland

263

[Lerner 2004]

The take over may take the form of an overt attempt to be helpful to the other party, but this is by no means their only function. One may be simply seeking to extend what the other has said: (3) A: B: C: B: C: A: C: B: C:

We’re going to Marlborough Marlborough? to see Granny with the dogs? if you can keep them under control in the garden? unless it rains which it always does

Conversations such as these, where there may be covert competition as to who can have the last word, are the bedrock of informal conversation, by no means constrained by any over-arching intentionally held aim of achieving coordination. And, of course, extensions such as these may be uttered by a single interlocutor refining and elaborating an initial sentence, as in (4) where we have a first extension providing the time of the event given in the initial utterance, then a source of the event, then a modifier of the subject: (4) Mary’s back. Late last night. From the US. Tired and frustrated at the delays. There is an open-ended flexibility in how such interactive construction of dialogues can be achieved. The adding on of a fragment in context may be used for novel effect at the level of the speech act, when apparently dual speech acts can be achieved within a single emergent utterance: (5) A: Are you left or B: right-handed Such interaction is effective even when the parties may be in sharp dispute: (6) A: It’s obvious from what he says (that) B: (that) you are wrong

264

kempson and cann

Partial fragments may also be being used to direct one’s interlocutor to construct something novel, an ad-hoc context-specific concept that depends on knowledge of the individuals in question: (7) Bird-expert: I watched birds yesterday all afternoon, in the sunshine Friend: with your grand-son? Bird-expert: It was great fun. He did really well … for 20 minutes. Metaphorical utterances can be taken up and extended: (8) A: How’s work? B: Oh, I shuffled along A: without getting much done, I imagine. And even very young language-acquiring children can join in on such interactions, either following up on some provided frame as in (9)–(11) or in initiating a process of frame construction as in one-word utterances, with no preceding linguistic context, as in (12): (9) A: B: A: B:

Old MacDonald had a farm. A-I-E-I-O. And on that farm he had a cow. And the cow goes Moo.

(10) A (carer to each child in turn in the nursery-group): And your name is … B (child): Mary. (11) Experimenter: This is (not) a … Child: comb! [reconstructed from De Villiers and Tager Flusberg, 1975] (12) Eliot (2 year old on mum’s bike waving at empty mooring on the other side of the canal): Daddy. Mother: That’s right dear, you were here yesterday with Daddy clearing out the boat. [direct observation] The general dynamic is that each such contribution adds unproblematically to whatever string of words has been set out so far, irrespective of whether or not what precedes it is a full sentence, and whether or not the add-on contributes to what the other party is trying to get across. The effect is one of a rich potential for interactivity between dialogue participants, available to all speakers of a language from very early stages of language development.

grammar as procedures

265

1.2 Implications for Grammar The existence of interactive constructing of utterances, and the fact that utterances by a single individual may be only a partial contribution to some gradual build up of information, pose significant problems for all conventional grammatical frameworks, in every sub-area of the grammar. From a syntactic perspective, we need to ask: how are split utterances possible; why are they so common in conversational dialogue; and how is it that the shift from one interlocutor to another can occur at any point in a sentence-construction process? Questions such as these might be, and often are, set aside as performance considerations only. But there is one consideration, overwhelmingly, which resists any attempt to dismiss the data as peripheral—the seamless fluency with which individuals take on or hand over utterance responsibility, apparently distributing syntactic and semantic dependencies across more than one participant. The problem for conventional grammars is the commitment to model an individual speaker’s linguistic competence as involving knowledge of a set of rules (however defined) that licenses all and only the strings of the language which that speaker judges to be well-formed; and, moreover, only those strings that are defined as sentences with representations rooted in the largest constituent recognised in the theory: S, CP, VP[+SBJ], type t, or whatever. Even if the goal of a grammar were weakened to characterise only a subset of the total phenomena displayed by a language, the fact that utterances can apparently split apart every single dependency within a language leads to the result that every generalisation articulated within a standard grammar will be incomplete. According to even the most modest evaluation metric, the grammar will fail. This is because, in all cases, there will be split utterance data which, despite being wholly acceptable to individuals in some speech situation, will not be given any characterisation by that grammar in virtue of their apparently fragmentary nature, not only non-sentential but possibly also not even constituting any phrasal type, as in the first utterance in (1), (3), or the second in (2) and (11). In order, then, to retain the grammar as an explanatory tool, such expressions uttered by the individual speaker, independent of any interaction with other speakers, will not be characterised as falling within the remit of the grammar. They will accordingly have to be deemed to be irrelevant, characterisable by some other set of principles. But any such dismissal would solely be on grounds that the distribution of such fragments is inexpressible within the grammar methodology adopted, and so not deemed part of any sort of competence on the part of a speaker (Chomsky 1965). But such fragments are not, by any stretch of the imagination, dysfluent uses of language: to the contrary, they are achieved with total fluency, demonstrating a fine level of sensitivity to the evolving development of the discourse.

266

kempson and cann

The ever-extendability of a structure across more than one speaker may at first sight seem to be a mere idiosyncracy of informal conversational exchange, where people may allow a single composite sentence to unfold apparently indefinitely as in (3), with each providing their own add-on, one after the other. There are two issues here. First, there is the issue of how fully determined the concept of a sentence boundary is. It is commonplace to observe of languages with which we are not familiar that the sentences “seem to go on for ever” and it is hard to find out where one begins and the other ends. But what data such as (3) illustrate is that this is an attribute of spoken conversation. The assumption that this doesn’t occur in well-documented languages is no more than a parochial folk-prescriptive attitude to “correct grammar” as essentially that of written language (Miller and Weinert 1998). It might then seem to be but a short step to see that in conversation, one contributor to the conversation is merely taking over responsibility for some emergent sentential sequence from someone else, with the activity being in some sense shared, an explicitly joint activity. So what seems to require building up is not the construction of sentences at all, but rather some structured representation of content. In some cases, indeed, the subparts of the sentence-string which appear to be omitted would seem to be either internally inconsistent or even not well-formed, considered as a single sequence of words: (13) Eleni: Is this yours or Yo: yours.

[natural data]

(14) {A emerging from a smoking kitchen} A: I’ve burnt the kitchen rather badly. B: Have you burnt A: Myself? No. In (14), the relevant sub-parts, put back together, yield an ungrammatical string due to the conflict in person between the binder and the reflexive pronoun: *Have you burnt myself. Yet the exchange is perfectly well-formed, the reflexive pronoun being interpreted locally as the subject of the predicate just constructed from B’s utterance, but with the switch of speaker having to be commensurate with attributes of the new speaker, now A. Intuitively, in some sense, these two parts have to go together to determine a whole syntactic structure, even though the result string-wise is not grammatical. What such data indicate is that these exchanges are not about putting word sequences together to form some string of words constituting a sentence or

grammar as procedures

267

even a string of sentences in a discourse (see also (3) and (13) above which, when put together into a single discourse are incoherent, despite the interaction being perfectly acceptable). Instead, what we appear to need to characterise such dialogue is a grammar that licenses, not sentences as traditionally construed, but sub-sentential units that put together may, or may not, yield a well-formed sentence as defined within a language’s grammatical tradition depending on speech act and other contextual factors. By implication, this yields a view of grammar as providing mechanisms for successful linguistic interaction, rather than a set of rules that define the set of all and only the grammatical sentences taken by speaker-hearers to constitute a language. However, even though it may be that representations of meanings are what is constructed rather than representations of putative structure defined over strings of words, such structures are nevertheless not wholly insensitive to the morpho-syntactic constraints imposed by the words of the language, as the required reflexive form in (14) indicates. This problem cannot be dismissed as a language-particular idiosyncracy. In all case-rich languages, fragments have to adopt a morphological form appropriate for what would be expected in an overt clausal sequence, commonly replicating the pattern provided by the immediately antecedent context. Thus in (15), the form of the fragment pronoun has to be the nominative form ego, not the accusative emena: (15) A:

Maria to egrapse to grama? B: Oxi, I the.Nom Maria it.Acc write.3.sg.past the letter No (*emena) ego I.Nom (*me.Acc) A: ‘Did Maria write the letter? B: No, I did (* Me did)’

Furthermore such fragments are required to take a definitive morphological form whether or not there is some overt antecedent in a previous clause to provide the appropriate pattern, again as in Modern Greek (Gregoromichelaki 2012): (16) [Context: A is contemplating the space under the mirror while re-arranging the furniture and B brings her a chair] A to B: tin (*i) karekla tis mamas? Ise the.ACC (*the.NOM) chair the.GEN mum.GEN be.2.sg.pres treli? crazy ‘Mum’s chair? Are you crazy?’

268

kempson and cann

Thus, first and foremost, relative to any sentence-based grammar, the split utterance data seem to fall irretrievably outside its remit. But even if the remit of grammar is somehow extendable to include these data, the structured representations which have to be induced appear to have both to be representations of content and yet also sensitive to what morphological constraints the particular grammar may impose on what representations can legitimately be constructed. Moreover such constraints must be expressible as lexical properties of the given words, despite possibly lacking the environment these need for successful completion. There are thus major hurdles for any conventional syntactic framework in addressing these data, if the framework is to have the expressivity needed to incorporate them within its remit. The problems posed by naturalistic data are barely less for semantics. First is the issue of how it is that almost every word can have its interpretation modified in context with compositionality to be defined over such modifications. The meaning attributable to our words reflects the context to which it is intended to contribute, shifting so seamlessly from one understanding to another that it is often hard to notice even that there has been such a shift—in (17) three different concepts of ‘burn’ emerge within a single exchange. (17) A: I’m afraid I’ve burned the kitchen ceiling. The paper’s blackened and part of it has come away. B: Did you burn A: myself? No fortunately not. Well, only my hair. And in (18), an expression is placed in the conversational arena, so to speak, and immediately modified, or at least justified by one particular attribute of stage-managing: (18) M: He’s stage manager [pause] He’s actually first assistant but- he’s calling the show.

[Hough 2015]

There is then the question of hand-on from one speaker to another. If meanings evolve over the course of an exchange, with shifts which we are barely aware of, how is it that propositional content can nevertheless be seen to emerge incrementally across a group of speakers? Things are no better for pragmatics. Probably the most basic questions needing an answer are: how can speaking and hearing be so seamlessly interwoven; and why is there such systematic cross-speaker use of non-sentential utterances in dialogue? The reason these questions are fundamental is that the answers to them impact on two of the most common assumptions made

grammar as procedures

269

in pragmatic theory: that pragmatic inference operates over propositions as expressed by sentences; and that successful communication is only achieved if a hearer can identify the proposition or thought that a speaker intends (or could have intended) to convey (Grice 1975, Sperber and Wilson 1995, Bach 1994). The former assumption is seriously called into question by examples like (3). Here, the conditional if you can keep them under control scopes, not over the proposition purportedly expressed by the preceding utterances, We’re going to Marlborough to see Granny with the dogs but only over the non-sentential prepositional phrase with the dogs: going to Marlborough to see Granny is not contingent on C’s being able to control the dogs. If inference is defined solely over complete propositions, then any such theory will have to be powerful enough to reconstruct some proposition from the prepositional phrase and further to explain why the scope of the conditional is not over the proposition expressed by the previous utterances taken together. Additionally, the construction of a speech act, usually taken to be something that only sentences are the vehicle for conveying (Austin 1962, although see Stainton 2006 for a discussion of speech-acts conveyed by sub-sentential utterances), is nevertheless able to be constructed as in (13) above and (19) below in which the first utterances are incomplete interrogatives implying a question, but whose interactive completion yields a statement: (19) Lawyer: Will you choose your son as your attorney or Client: my wife. With respect to the question of intention recognition, there is plenty of evidence in dialogue that, while building on some estimation of the other person’s mind-set is commonly made use of both in parsing and in production, it is not a prerequisite for successful communication (Gregoromichelaki et al. 2011). For example, the clarification request in (20) is made before B could possibly have recognised any intended meaning by A: (20) A: They X-rayed me, and took a urine sample, took a blood sample. Er, the doctor B: Chorlton? A: Chorlton, mhm, he examined me, erm, he, he said now they were on about a slight [shadow] on my heart. [BNC: KPY 1005–1008] In non-co-operative examples like that in (6) above and (21) below the second speaker completes the previous utterance without any necessary consideration of the intentions of the first speaker. In (21), the son is certainly not waiting for

270

kempson and cann

the third of the commands even if he has bothered to process the second, but nevertheless he has responded wholly appropriately. (21) Mother: Son:

This afternoon first you’ll do your homework, then wash the dishes and then you’ll give me 10 pounds?

And, most strikingly of all, such split-utterances are employed in interactive exchanges with very young children, even when a compound speech-act effect is achieved as in (10) parallel to (19). Indeed they are one means for detecting complex conceptual abilities like negation as in (11). This strongly suggests that recognition of the content of other people’s intentions is not a necessary condition for acts of communication to be successful and, more generally that acquisition need not make reference to higher order reasoning (Gregoromichelaki et al 2011). Taking a step back, what appears to be needed to model these data is a framework in which strings, content, context and even speech-act determination can all be seen as evolving during the course of a conversational interaction. This involves a radical shift of assumptions. Instead of presuming on a methodology which isolates the capacity for language from all attributes of its use, the concept of the process of building up structured representations needs to be taken to be central. Instead of such a system being assumed to be articulated in a vocabulary wholly encapsulated from general cognitive considerations, the emergent representations will need to be conceptual representations, so in principle allowing free interface between such grammar-internal specifications and general context considerations. Any such level of representation will nonetheless have to be sensitive to whatever morpho-syntactic constraints the forms of the language may impose, so the articulation of morphosyntactic constraints will have to be in terms of constraints on the mappings from the morphological forms onto such conceptual representations. In short, we are turning to the exploration of a view of language which breaks out from the stranglehold of the competence-performance division in two ways, putting in its place a methodology for defining language as a set of actions for coordination between interlocutors. The concept of sentence will no longer be central to determining the remit of grammar; and the structural properties of any natural language will be defined in terms of conditional actions mapping words and word sequences onto representations of content. So the system will no longer be an encapsulated set of properties defined over the strings of words and lacking any interface with general cognitive restrictions (as in conventional frameworks), but rather will be defined in a domain-general

grammar as procedures

271

vocabulary and able to function as a proper subpart of the general cognitive architecture. Principles underpinning structured strings are thus transformed into procedures for incrementally building structural representations of content, a stance which enables language acquisition to be seen as a continuous developmental process rather than a sequence of different grammar formalisms with the child having to hypothesise the chain of grammar improvement. With this transformation, the concept of mapping expressions onto emergent contents and contexts as these become available lies at the centre of what has to be expressed, so the concept of procedure becomes central, together with the twinned concepts of initial underspecification and subsequent update.

2

Modelling Incrementality and Context-Dependent Interpretations

The framework we adopt to model this phenomenon, is Dynamic Syntax (DS), which defines grammar as mechanisms for incremental growth of interpretation: the concept of underspecification plus update is central to the grammar (Kempson et al 2001, Cann et al 2005, Kempson et al 2015, 2016). DS is a representationalist model of interpretation of which the core structural notion is interpretation growth relative to context, so a process-based model. What has gone in this approach to language modelling is any concept of static, purely syntactic structure, inhabited by strings so a fixed pairing. What is retained is a DRT-like concept of building representations of content relative to context (Kamp and Reyle 1993). The formal devices are defined to model how to pair emergent interpretations with words being uttered in sequence. The system defines actions that give rise to expectations of further actions, all involving incremental updates towards some overall goal. On this view, the setting out of such dynamics constitute the grammar: syntax just is a set of principles for inducing growth of such structures. The syntactic mechanisms are procedures that define how parts of representations of content can be introduced and updated, all such growth being relative to context. Moreover, context is as dynamic and structural as the concept of content with which it is twinned, a record of the emergent structures that represent the unfolding of interpretational content plus the actions used to develop this incremental process (Purver et al. 2006). So the overall picture is a shifting one: shifting goals, shifting contents, and shifting contexts, as each established content becomes part of the context for the next step in the process. These goal-driven sequences of actions are taken to involve building partial trees, with, as output, a tree whose nodes reflect the content of some utterance,

272

kempson and cann

as in (22) which represents the content of an utterance of Who hugged Mary? in binary tree form.

(22)

The display in (22) indicates only the output representation of the utterance— a static structure. What is not shown is the way the process giving rise to this output is initiated and developed. It is thus the process of tree growth and not the structure of the output trees (mere reflections of functor-argument structure) that provides the explanatory core of the framework. The point of departure for the analysis of Who hugged Mary? is some partial tree whose only decoration gives a skeletal indication of the goal to be achieved, a propositional formula, otherwise unspecified, i.e. a single node with the single decoration a goal requirement to derive a formula of the specified type, ?Ty(t):1 (23) Tn(0), ?Ty(t), ◊

What has to be built to meet this goal requirement is a formula matching this type, a result which is achieved on the basis of what the context already provides, what general strategies for progressive growth of structure are available, and information provided by words used. In any complete tree, each node has a content formula, shown to the left of the colon “:” and a type specification to its right (e for entity, es for event, e → (es → t) for predicates, etc.). The trees are binary, with, by convention, the functor node on the right, and the argument on the left. Individual terms are invariably of type e, presuming on an account of quantification, which we will sketch later. There is a particularised type es 1 The label Tn(0) gives the binary address of the node in the unfolding tree, here address 0, the rootnode. ◊ is the pointer indicating the node currently under development.

grammar as procedures

273

for event terms, only schematically represented here (Gregoromichelaki 2006, Cann 2011). The dynamics of growth is partly top down—as general strategies open up possibilities—and partly bottom up—as actions induced by the words provide terminal decorations which provide the input to a bottom-up process yielding the effect of compositionality in the resultant tree. To flesh out this concept of growth, we need first a vocabulary for defining trees, and then a vocabulary for defining how such trees can be seen to grow. The modal logic of trees of (Blackburn and Meyer-Viol 1994) provides the language of tree description, with two basic modalities. ⟨↓⟩: ‘⟨↓⟩α holds at a node if α holds at its daughter (one node down)’ for which there are variants ⟨↓0 ⟩ and ⟨↓1 ⟩ for argument and functor daughter relations respectively. And the inverse ⟨↑⟩: ‘⟨↑⟩α holds at a mother node, one node up, if α holds at its mother’, equally with argument and functor variants. There are also Kleene star operators which yield concepts of underspecified dominate and be dominated by: ⟨↓∗ ⟩Tn(n) holds at a node when a node Tn(n) is somewhere below it (along an arbitrary sequence of daughter relations), ⟨↑∗ ⟩Tn(n) holds at a node when a node Tn(n) is somewhere above it.2 There are three principal kinds of underspecification that require updates. First, there are partially specified trees as some nodes or formulae remain yet to be developed. Secondly, there may be underspecified content formulae which are assigned a type but as yet with only a place-holding metavariable as formula, itself needing update to some specified formula value. Of these, anaphoric devices are the familiar cases of content underspecification, encoded as projecting such a metavariable, possibly with constraints on substitutions for them: for example, UFemale′ : e for she/her or U3Plural′ : e for they/them. Thirdly, the concept of underspecification is extended also to structural underspecification, in which it is a tree-relation that may be incompletely specified, so that a node introduced into a tree may lack a fixed site, hence referred to as an “unfixed node”, defined by the Kleene* operator ⟨↑∗ ⟩ and its variants. This is the mechanism which underpins long-distance dependencies and local scrambling. Underspecification of any parameter (type, formula or position within a tree) is invariably twinned with an associated requirement, symbolised as ?X, for some annotation X. So we have type requirements ?Ty(α) to find a formula of some type α; ?∃x. Fo(x) for a requirement to find a value for a metavariable; and ?∃x. Tn(x) to find a fixed treenode address for an unfixed node. Require2 Various types of locality can be defined over dominance relations, inducing different domains in which unfixed nodes can be fixed. See Cann et al (2005) for details.

274

kempson and cann

ments may be modal as well as non-modal, with modal requirements imposing constraints on how a structure may be developed. For example ?⟨↑0 ⟩Ty(es → t) is a requirement expressing an output constraint imposed by nominative casemarking that a node so decorated be the highest individual argument of a predicate, still needing an event term incorporating temporal specification with which to combine. Requirements are essential for reflecting the incrementality involved in progressively building trees, providing goals to be achieved that may be satisfied immediately or not until substantially later in a derivation. 2.1 Actions for Tree Growth The concept of underspecification is itself a static notion, a property of a given tree configuration. To define the process of tree-growth, we adopt a language for formulating actions that determine the (monotonic) transitions between partial trees. There are four primitive actions: make(X) for constructing a new node; go(X) that moves the pointer to a node; and put(Y) that annotates a node with some information. There is also an Abort action, which terminates an action sequence. These are used to define a range of action sequences. Computational Actions constitute generally available strategies for treegrowth, either inducing the unfolding of an emergent tree on a top-down basis, or inducing bottom-up processes which, once appropriate terminal nodes are decorated, lead to the decorations for all non-terminal nodes. For example, as one of a very restricted number of initial steps in a derivation starting from a propositional requirement, ?Ty(t) as in (22), a computational rule licenses the construction of a node characterised solely as being unfixed within a given tree domain, as expressed by decoration ⟨↑∗ ⟩Tn(a) which indicates that “somewhere above me is the Treenode a”.3 Rules are given in standard ⟨IF…, THEN…, ELSE…⟩ statements. The example in (24) shows the actions and output tree that define the construction of such an unfixed node.4

3 The tree displays diagrammatically the unfixed node, using a dashed line. Fixed tree relations are shown with a solid line. Note that such tree diagrams record only the effect of the rule, and not conditions on its application. 4 There is a condition on this action that it can only take place if no other node from this point of departure already exists within this newly emergent tree. This is the second condition preceding the sequence of actions, aborting the action if there are already constructed nodes within this emergent tree. There is a distinct local variant of this action inducing underspecified substructure within individual proposition-templates, but for simplicity we omit this here.

grammar as procedures

275

(24)

Words also invoke actions of the same sort: making and annotating nodes and moving the pointer around a tree. The simplest of such macros of actions is when nodes are merely annotated with formulae as in a parse of the name Mary as in the entry in (25):5 (25) Verbs induce more structure than names, triggering conditional actions that induce all the nodes that define their predicate-argument structure, including an event argument where lexical aspect information may be added and tense is specified by finite verb forms. Languages vary as to what options for the argument node decorations they license. In English, being not a “prodrop” language, verbs specify argument nodes as open type requirements of the appropriate sort (?Ty(e) for individual term arguments, ?Ty(t) for propositional complements, etc.) which ensures that there has to be some further step of linguistic-input to satisfy the type requirement. Languages also vary in word order, with English finite main verbs unable to appear sentence initially and necessarily appearing after the subject and before a direct object (modulo discontinuous dependencies). These constraints are guaranteed by the set of actions in (26) where the second condition requires some structure to have already been built, while the first two actions ensure that this is identified as the logical subject (for non-passive forms). The remaining actions build an event argument which is annotated with lexical aspectual information and tense (for finite forms) and then construct the remaining predicate-argument structure, leaving the pointer on the open logical object node, thus ensuring that some object term be identified next. This complex of conditions and update actions are illustrated in the lexical specification of the verb hugged in (26):6 5 This is an oversimplified characterisation omitting all discussion of how or whether names are associated with a uniqueness constraint or what form that constraint should take. 6 These actions have been slightly simplified for expository reasons. See Cann (2011) for details and discussion.

276

kempson and cann

(26)

Relative to the prefixed condition, the subsequent sequence specifies step by step how to build the necessary tree structure, for each relation between nodes in such a structure, building that relation, going along it to the new node so constructed; decorating it, coming back to some more central point, and repeating this sequence of actions as many times as there are argument and functor nodes. The result is a skeletal propositional template associated with the predicate Hug′ , together with a partially specified event term (only sketched here), notably with only that predicate node having some content specification. 2.2 Combining Actions in Sequence With this minimal sketch in place, we can see how structure is built up for (22). The first step listed is to induce an unfixed node from the underspecified tree in (23) by the actions specified in (24) yielding the tree specified there. This provides the context (an open term requirement) for the parse of who which annotates the node with a particularised metavariable, WH and a type specification. With the pointer returned to the topnode, the trigger for the processing of the verb hugged is satisfied, inducing the sequence of actions given in (26). These steps are shown in (27):

(27)

grammar as procedures

277

This update leaves the pointer at the logical object node, ensuring that the syntactic object, Mary, can be parsed next to decorate the relevant node with type and formula information as in (25). One of the effects of the actions induced by the verb is that the unfixed node is required to be the logical subject given the position of the wh expression immediately before the verb, and this is achieved by a process of unifying the unfixed node with the open term node to yield (28) satisfying all outstanding requirements on the unified nodes. From this interim result, a simple sequence of applications of the computational action of modalised function-application is all that is needed to successively satisfy all nonterminal node requirements up the functor spine of the tree, giving (22) above:

(28)

Of course, there is more to be said, in particular because these actions build relative to context, with both hearer and speaker following the same sort of activity. The only difference between the dynamics of parsing and production is that, in parsing, there is only the relatively weak goal of trying to build some propositional formula relative to the utterance and context, while, in production, there is some more particular goal to be achieved, some (possibly partial) tree representing something to be communicated. So, for the speaker, every update associated with a putative word must be extendable towards that goal tree to allow that word to be produced. The only difference between being a speaker or a hearer is thus the checking by the speaker of the subsumption relation between the tree under construction and some entertained (partial) goal tree.7 So from this perspective the shift from hearer to speaker and back again

7 The hearer may well also have expectations as to what is to follow in the form of some hypothesised extension of what they are currently parsing, but such anticipatory steps are not an essential accompaniment.

278

kempson and cann

is seamless and unproblematic: both interlocutors entertain a partial tree representing the content of the utterance so far. To start generating an utterance all a hearer has to do is extend the current parse state to act as a new goal tree and then start speaking, again checking output parse with the goal tree (Gregoromichelaki et al 2013, Gregoromichelaki and Kempson 2015). And this may go on as long and as often as the interlocutors want to. In the same way, a hearer does not have to recognise the speaker’s intended communication before starting to speak: he merely has to decide to extend what has already been said into something he wants to say, whether or not he has anticipated what the speaker was intending to utter. In consequence, the model is entirely flexible with respect to our intentions to interact with each other and how big a role such intentions might play.

3

Cross-Turn Syntactic Licensing: Split Utterances

It is this free switching of rules which constitutes the split utterance phenomena, the most striking aspect of which is how the shift of roles from speaker to hearer and conversely can take place apparently splitting dependencies of all sorts. By way of illustration why such phenomena can, and must, be licensed cross-turn and distributed across speakers, we take morphosyntactic dependency between a reflexive anaphor and some locally available antecedent: (29) Mary: Did you burn Bob: myself? No. (30) displays the partial tree upon Mary’s utterance having been processed. At this point, with burn as well as you processed, Mary’s utterance of the indexical pronoun you will have served to annotate from context the subject node with a conceptual representation of the current addressee, Bob. Given this as a context, Bob is now licensed to complete the utterance using the first-person reflexive, because this expression, by definition, induces a copy of the formula residing in a local co-argument node just in case that formula satisfies the conditions set by the person and number features of the uttered reflexive, i.e., the current speaker:8

8 Bob’s utterance here serves simultaneously as a completion and a clarification request reinforcing the DS claim that all such phenomena be treated uniformly by elaborating subpropo-

grammar as procedures

279

(30)

The fact that the structure may have been induced in part by someone else is irrelevant. This phenomenon, as here modelled, is specific to the construal of the reflexive pronoun, said to be subject to a Principle A constraint (Chomsky 1986); and the associated “Principle B” constraint imposed on a nonreflexive pronoun debarring it from picking up an antecedent from this domain is equally simple to express in the same terms (Cann et al 2005; Gregoromichelaki 2006, 2013). In both cases, the morphosyntactic restrictions are expressible as constraints circumscribing the goal-directed search for achieving interpretation of linguistically-underspecified elements. Though this seamlessly achieved splitting of utterances can apparently separate off any expression from the syntactic environment it needs for its wellformedness, this effect is achieved because speakers and hearers both use the defined tree-growth mechanisms to construct a representation of what is being said taking the immediate context as input. The only difference between speaker and hearer is the additional requirement on speakers that the construction process has to be commensurate with some more richly annotated (possibly incomplete) structure corresponding to what they have in mind to say. This dynamic predicts that switching from parsing to production, and conversely, will be seamless, yielding the effect of in-tandem construction without needing to invoke higher levels of sitional types, for more details see Kempson et al (2007); Gargett et al (2008); Purver et al (2010); Eshghi et al (2015); Gregoromichelaki and Kempson (2015), Gregoromichelaki (2016).

280

kempson and cann

inference (Poesio & Rieser 2010) or superimposed duplication of the one type of activity upon the other (Pickering & Garrod 2013): each individual will simply be constructing the emergent structure relative to the context which they themselves have just constructed in their other capacity (Gregoromichelaki et al. 2011, 2013), so there is in effect no splitting of the dependency environment. Given the bidirectionality of the grammar, parser-generator switches are seamless and the resulting interlocutor coordination is secured by the subpersonal processing mechanisms of the grammar directly without needing to invoke high-order inference. Such shifts of roles from speaker to hearer are displayed even in supposed hard-core cases of syntactic dependencies, such as the classic “movement” cases of long-distance dependency: (31) A: Mary, John … B: trusts? A: Well, she clearly likes him. In this case what A and B severally as speaker and hearer have built at the point of switch is a partial structure. Following the switch, whether as speaker or as hearer, the particular form of the verb, here the finite form trusts, will fix the interpretation of John′ as logical subject (in virtue of the property of English finite verbs determining that the most local unfixed node already in the structure under construction must be construed as logical subject: see Kempson et al 2016 for details). The form of the verb will also fix the interpretation of Mary′ as object in virtue of the step of unification immediately after the processing of the verb trusts. Such restrictions will be applied by both speaker and hearer within the partial trees they have individually constructed within their respective activities whether as speaker or as hearer.

4

Extending Content: Building Relative to Context

There is of course much more to add to have even a minimal indication of why DS constitutes a grammar formalism. In particular, the sketch so far has placed relatively little emphasis on the context. The exegesis in the main here presumes some propositional requirement leading to some output of type t. But, as these split utterances vividly demonstrate, speakers and hearers can start these activities from partial structures of whatever sort, so the concept of growth itself needs to be defined in these terms, for both content and context. Generalising then to partial trees as input or output from any task, we need to model how speakers may start out with only a partial thought in mind

grammar as procedures

281

(with a partial tree as goal), may intervene with some partial contribution to the emergent structure (with partial trees both in context and as goal) and, equally, hearers may start with some partial tree as context, updating it as appropriate driven by the utterance of incoming words relative to that context. The minimal context for both parsing and production thus must include partial trees under construction,9 the sequence of words so far uttered, and the sequence of actions so far used.10 The status of this composite context as a record of tree growth within the grammar is confirmed by the fact that languages all contain explicit devices for indicating links with some constructed context, more familiarly known as the head-adjunct relation. Examples include (32) John, a lovely guy, is becoming Chairman of the company. (33) John, who I admire, is becoming Chairman of the company. The process of adjunction, by definition, involves the adoption of some term as a context from which to articulate some ancillary add-on—the adjunct—so that the result is the extension of some partial tree by additions which in some sense preserve the context provided by that simple primary tree, nonetheless adding additional “adjunct”-provided information to it. As (32)–(33) show, such adjuncts can be of different types, here a noun phrase and a relative clause. DS models this directly as a process of building so-called linked trees in which from any node once its requirement is satisfied, a LINK transition onto a newly emergent tree is licensed with a requirement that the result is a node sharing a term with the node at point of departure—the context relative to which this transition is licensed. The canonical case of these is relative clauses such as (33) which can be added on at any point in the overall tree being developed: (34) I walked up to John, who was ignoring me, with a gift for his wife. (35) John, and not Mary, is about to be voted in as Chairman, who I warned about this some weeks back.

9 10

The context will characteristically also include a wider set of structures, by assumption also modellable in tree-theoretic form. In Eshghi et al 2011, this is modelled via a mapping onto a Type Theory with Records formulation (TTR), but we suppress these details here: see Sato 2011; Hough and Purver 2014; Eshghi et al 2013; Hough 2015.

282

kempson and cann

(36) I approached a man who had been annoying people at reception, with a request that he should leave. (37) Someone came in to the lecturehall, who had been annoying people at reception. On this analysis, such structures, which can occur anywhere in a string in English as long as relatable to that head, are modelled preserving Jespersen’s insight of relative pronoun being an encoded quasi-anaphoric device (Jespersen 1949); on the DS view these add-ons are projected as independent structures through the addition of a LINK transition onto this newly initiated structure (Cann et al 2005), the relative pronoun in effect projecting a constraint on a transition across from one partial structure to another, which its actions induce. Though as with relatives, the new structure is propositional in type, this is by no means a general constraint on adjunct structures, as witness (32) and (38), for which the requirement of a shared term at the content level of DS tree-construction is preserved, even though notably not observable from the sequence of words themselves: (38) A1: B: A2:

Mary’s just returned. From Germany? Very tired.

We omit all details here (but see Cann and Kempson 2017), but merely note that the requirement of shared content is met by B’s utterance having to be construed as the predicate ‘returned from Germany’ and A’s follow-up utterance providing a new adjunction structure modifying the subject ‘Mary’. An important further reflection is what this direct modelling of adjuncts allows. Because these loosely connected structures are not structures defining hierarchical and linear dependencies between words, but rather expressing links between structured concepts, they provide a vehicle for modelling a whole range of possibly covert inferences—ranging over propositional attitudes, politeness indications about either speaker/hearer or individuals described, and more generally the whole class of implications broadly characterised as implicatures (see Cann and Kempson 2017). The overall effect thus is that structure content and context all evolve, step by step. “Content” associated with words involves very much more than a mere itemising of conceptual content: to the contrary these macroed-sequences called up by a word characteristically project local structural update but yet invariably less than a fixed denotational content within a fixed structure.

grammar as procedures

283

The evolving nature of both context and content is modelled directly in DS through defining a composite evolving structure represented as a directed acyclic graph (DAG: Eshghi et al 2011, 2013, 2015; Hough 2015). This keeps track of choices available and selects amongst them as soon as possible, a device which, depicting transitions between processing states Ti graphically as in Figure 9.1, shows how burgeoning choices may yet serve to narrow down derivational choices as these become available. We take only the simplest sequence of steps here, that involved in processing the sequence of an auxiliary plus you construed as subject of the transitive verb burn, but even in such a short three-word sequence, additional options can be seen to open up but be swiftly closed off. In the case of split utterances then, as with all other utterances, it is the emergent DAG which serves as recording for speaker and hearer alike the shifting triples of context, words and content.

figure 9.1 DS context plus content as a graph

A further bonus of incorporating the DAG within the formalism itself rather than as some external “performance-only” add-on, is that it provides a basis for self-correction, allowing local backtracking in the face of inconsistency to some first compatible point in the path, displaying precisely the fine-grained incrementality needed to model the flexibility of on-line production and parsing both in what are manifest corrections ((39),(42)), and others which may get identified as “apposition” structures, rather than the online revisions which they often serve as ((40)–(41)): (39) The yell-uh purple square (40) The yellow square, the one with a black dot on it, … (41) The proper conclusion, the conclusion to be drawn from these arguments, is …

284

kempson and cann

We thus now have the basis for anticipating encoded lexical specifications for expressions such as or rather in (42) which are defined to make reference to local back-tracking: (42) The yellow or rather orange square Standing back from the details, what we now see is an action-based grammar, and a non-representationalist view of “syntax”, syntax being transformed into a set of conditional update actions that induce or develop partial representations of content relative to context. These are prediction-driven strategies, inducing structure-developing options that get progressively developed or locally corrected, with both local and universal grammar-defined procedures that have the effect of inducing interaction and coordination effects. Furthermore, though there appears to be rich interaction between minds with clear capacity for inference and, at least, in principle sensitivity for other people’s needs, successful communication does not need the capacity of either mindreading or recognition of the other party’s intentions before successful communication can be said to have taken place (contra Breheny 2006). To the contrary, the effect of interaction takes place simply because parsing and production use the same processes, as definitive of the DS formalism: no separate parsing and processing modules have to be co-defined explicitly for such interactive effect. And with grammars defined as mechanisms for information/action coordination, languages can be seen as tools for interaction.

5

Language as Actions

In drawing such linguistics-internal arguments together, we note ongoing research in cognitive psychology, where embodied representations have been explicitly argued for as a means of explaining the tight physical coordination that exists between individuals engaging in manifestly interactive behaviour, where there is striking parallelism between the coordination demonstrably sustained in split utterance exchanges and these more general actions. In studying activity such as ballroom dancing, for example, Pezzulo and colleagues (Pezzulo 2011) have argued against invocation of higher order reasoning as the basis for such cases of rapid cross-modal coordination: to the contrary, they provide evidence that both action and perception involve low-level procedural mechanisms which determine agents’ actions, being both predictive and simulating. The general framework for this embodied view is put forward by Andy Clark (Clark 2013, 2016), who argues that all cognitive behaviour involves

grammar as procedures

285

constructive actions, so that perception and action both involve anticipations constructed relative to immediate and more long-term context in advance of the designated activity itself, the input stimuli merely acting as a filter on such anticipatory guesses, rebutting or confirming the expectations with which we greet the world. With even half an eye on this neighbouring research activity, the parallelism with the language case is striking: taking heed of the world of cognition as espoused by Pezzulo, Clark and colleagues has become an imperative. 5.1 Extending the Predictive Processing Model In the so-called Predictive Processing (PP) model of Clark 2016, the bottom up flow of information of the input system model advocated by Fodor (1983) and Marr (1982) is thus replaced in total by a generative probabilistic model of cognition, in which cognitive assumptions are created on the basis of the context, anticipations which constitute 80 percent of the explanatory load. Unlike the Fodor/Marr model in which the input stimulus is the point of departure, on the PP model, the incoming stimulus functions as a filter that checks for prediction errors to which that anticipatory input might give rise, contributing a mere 20 per cent load of the explanation. According to Clark, this is the pattern for the whole cognitive system. Perception, action (and imagination too) are all said to construct guesses about incoming signal: what they differ in is the response to any recognised inconsistencies. Perception involves a filter weeding out impossible and unlikely construals. Action involves physical action to make true some hypothesis, in order to avoid what would otherwise be filtered out on the appropriate anticipatory hypothesis as transparently not matched by the input stimulus. And imagination involves a suppression of the inconsistency filter itself that would otherwise involve either rejection or action. In such a system, as in DS, context is everything. Brains are not the passive modular input systems of Marr (1982). To the contrary, they are predictive engines using their own immediate and encyclopedic context at every step to guess the structure/shape of the incoming sensory array, whether for perception or action. The status of the incoming signal, which changes moment by moment, is on this view substantially reduced from that attributed to it in the static multilevel hypothesis of Marr. Instead, there is an affordance competition hypothesis whereby the brain specifies in parallel several potential actions which compete against each other for further processing relative to probabilistic weightings (these determined by iterative experiences, cognitively or normatively culturally imposed), until, standardly, all but one of these is winnowed out. While the model Clark depicts is richly articulated, being a claim about cognition in general, his view is strikingly a generalisation of the DS stance: language, like other

286

kempson and cann

cognitive activities, on this view involves the active construction of assumptions in real-time as the grounding for all language-based activity. In the light of this parallelism, the relatively weak characterisaton of language in Clark (2016) comes as a surprise. He draws current research on language processing together to obtain a conclusion that is both too strong and too weak. It is too strong because Clark argues against the necessity of invoking a so-called efference-copy view of perception and its counterpart action. Nonetheless, Clark cites with approval the Pickering and Garrod (2013) efference-copy view of the shared activity involved in split utterance processing.11 Furthermore, despite rejecting inferentialist views of perception/action in favour of his own PP model in which embodied mechanisms carry the explanatory load of how cognitive activity takes place, he nevertheless suggests that language enables top-level “script sharing” and “brain to brain coupling”, along lines of the “shared brain” view of Frith 2007 (though see the modified view of Friston and Frith 2015);12 but these variants are all solidly inferentialist, being grounded in Gricean assumptions of meaning, with the additional twist for the associated explanation of language acquisition of requiring that the capacity for such higher-order mind-reading capacity be in place prior to the very earliest stage of language acquisition, as a pre-requisite for language learning to take place at all, a view which both Tomasello (2008) and Christiansen and Chater (2016) embrace. So the view of language which Clark promulgates has all the richness which his own PP model signally does not endorse. On the other hand the characterisation of language processing in Clark 2016 is too weak, in that it fails to capture the coordinated social interaction endemic to language use because the language perspective he presumes upon preserves the view of language as a statically available body of knowledge. This means that he misses the parallel between language and cognitive architectures, missing the potential for replacing top-down inferential accounts with emergent systemic effects. The answer to this conundrum is straightforward: what the Predicting Processing perspective lacks and needs is the rounding out of the PP model to incorporate a model of language as evolved for interaction, such as DS.

11

12

It is notable that their account fails to provide more than a partial account of the phenomenon, capturing as it does only the “helpful completion” type of data where the interrupting participant is actively seeking to construct the completion their collaborator seems to need help with. Friston and Frith (2015) retract from the Friston “social brain” view to one in which coordinated behaviour involves an abductive step of self-other identity.

grammar as procedures

287

5.2 The DS/PP Model Bringing the DS characterisation of language into the PP landscape, the unitary characterisation of language as a sub-part of a Predictive Process cognitive system that is specialised for its interaction effects emerges wholly naturally, and the anomalies of the Clark 2016 discussion of language are resolved. Both the DS model of language and the PP model of cognition are fundamentally action-oriented, involving predictive moment by moment processing, evolving context dependency at every level with continuous intermingling of production, thought, encyclopedic and immediate context, with choices made relative to the individual’s own context. In this system, choices between competing hypotheses are made with respect to an individual’s own context. There is no cross-individual replication or “mutual knowledge” constraint: the dynamic is hypothesis-construction relative to context plus error-correction narrowing down possible alternatives, a dynamic by claim underpinning all cognitive activity. Despite close parallelism between complementary activities such as perception and action, no “efference-copy” mechanism is required to duplicate the one constructive process relative to the other, nor, as Clark explicitly notes, is any higher-order inference required by either party conjecturing another’s mental states. What DS brings to the Predictive Processing perspective is that the very same PP dynamic extends to cross-individual exchanges simply in virtue of the construal of language as a mechanism for information growth, whether in parsing (perception) or production(action). The hearer and the producer in any utterance exchange are both engaged in the action of structure building and it is for this reason that at any point in the utterance sequence, they can exchange roles—in both cases, the process of structural growth will simply be proceeding from where they had got to. Following the Clark parallelism between perception and action, parsing can now be seen as hypothesis formation constrained by an associated acoustic filter eliminating impossible interpretations; and production can be seen as hypothesis formation in a richer, plan-driven, enterprise, involving transmission to the realisation of bodily actions to make true a correspondence between some assumption and its outcome. So though the two counterparts may be seen as feeding into distinct physiological realisations, in other frameworks taken to constitute self-evident grounds for the disjointness of these two inverse activities (Hendriks 2014), on Clark’s view they are both driven by the same pressure to make true the pairing of structure and output, and so essentially integrated through a common mechanism. Hence, on this joint DS/PP view, syntax is an embodied skill consisting of coupled interlocutor actions for incremental predictive processing in context.13 13

The existence of “prosodic words” which are signally not noticed by either producer or

288

kempson and cann

The effects of interaction so transparently displayed in conversational dialogue can thus be seen as emergent from the system with no essential high level inference or mind reading, nor any risk of implicitly invoking an invisible agent, this the well-known homunculus charge of Putnam (1967). There is a further commonality. The specifically probabilistic generative model, with shifting probabilities at every level as filtered by the carrying forward of putative prediction errors, which might seem to distinguish the PP model from DS, is to the contrary, reflected in the most recent DS implementation (Eshghi et al 2013, 2015; Hough and Purver 2014; Hough 2015) by the addition of probabilistic weightings into the language system as part of the macro sequences of actions stored in the lexicon, completing the parallelism. As Clark’s perspective would require, the DS system has thus been developed to incorporate a built-in self- and other-correction mechanism which yields stepwise weeding out of options that are probabilistically implausible as well as categorically impossible (see Bickhard 2009 who claims this is diagnostic of an interactive system). Such weightings derive from the accumulated weight of iteration, personal and even social experience through exposure to cultural norms that dictate perspectives for understanding phenomena in context.14 Finally there is the constraint imposed by the cognitive system itself, driven by economy for some given effect, a property often invoked by pragmatists as determining communicative success, but here seen as the natural emergence of the way the brain has evolved to determine success in its task of providing a filter on the wealth of incoming information to determine that its task in real time is manageable (what Clark refers to as the lazy brain hypothesis, citing Herbert Simon 1956 for the original idea). No longer do linguists have to presume that the performance factors involved in communication of

14

hearer, suggest that the mechanisms for realisation are a different level of abstraction than the dynamic core of language itself. In claiming that ‘she’s a genius’ by uttering the prosodic word /ʃiːzə/ the speaker will be building a structure commensurate with the thought that Mary, let us say, is a genius, without any hiccough caused by the fact that his articulation is at best apparently nonoptimal. On the other hand, the hearer, processing the speaker’s utterance only has to be able to retrieve the appropriate interpretation from the very high probability, given its very common occurrence, that this word comprises the sequence of ‘she’ plus ‘is’ plus ‘a’, so both parties will be successfully building up the thought that Mary is a genius without the lack of the supposedly indicated lexical entry causing any impediment to the success of the exchange. See the neuro-science perspective as reported in Rose and Abi-Rached (2013) in which the brain is demonstrated to be responsive to iterative cultural processes and so subject to both short- and long-term context-relative change.

grammar as procedures

289

necessity involve steps of a higher-order inference constrained by assumptions of normativity (the Gricean perspective: Grice 1975) or by cognitive constraints imposing limits on such inferences as stored in some pragmatics module (the relevance-theoretic perspective: Sperber and Wilson 1995, Sperber 2002). On the new view such constraints are the constraints of the cognitive system in general, imposed on all cognitive activity, being implementations of systeminternal interdependencies (Clark 2016 among others). On the combined DS/PP perspective, the phenomenon of a language and its processing, at any one time are, indeed, not essentially distinct: both are ephemeral ever-changing constructs. What is relatively constant is the capacity for language, and the cognitive system more generally, this being the predictive device which constructs hypotheses for action/perception, systematically able to correct any such putative actions/judgements relative to its inbuilt correction device, hence enabling learning, ongoing interaction and modification.

6

Evolution: Surfing Uncertainty Together?

With language now seen as a tool for interaction, the perspective from which to develop the view of natural language as an evolved adaptive system shifts into a less inferentialist perspective while remaining broadly functionalist. Natural languages are perhaps the quintessential case of adaptivity, and the view that language is an adaptive system was first suggested by Darwin. The adaptationist claim is disputed by some (Fitch, Hauser and Chomsky 2005, Lightfoot 1999) and strongly advocated by others (Christiansen and Chater 2016 among others). According to the non-adaptationist view, the architecture of the mind/brain determines the form of natural language in virtue of domain-specific attributes of a language-specific module nested in the brain determining the limits shared by all natural languages, which could not be subject to any gradualist account. Pinker and Bloom (1990), despite defending an encapsulated language module nonetheless defend an adaptationist stance. Others claim to the contrary that language is a product of cultural evolution; and that languages have developed in the light of pressure from cultural and pragmatic (inference-driven) pressures to match the needs of our brain, e.g. evolving in the direction of progressive lowering of cognitive costs in order to meet the pressures for fast processing which are imposed by a system evolved to make very rapid choices within a very narrow and endlessly shifting window of time (Christiansen and Chater together with Hawkins (2004), Bybee (2010) and other construction theorists). These latter proposals all pursue an essentially functionalist form of explanation—language as shaped by the brain.

290

kempson and cann

There are many intermediate positions in this debate; yet these varying stances characteristically leave essential questions unaddressed. Though they address (and disagree about) the nature of human language in the current evolutionary window of time, abducing from that to some pre-language state, by and large they do not address the “how questions”: how the properties particular to language such as its notable signal-content correlations came into being; how these attributes evolved alongside the emergent general human potential for context-relative inference, before, after, or in tandem; and so on. Yet, as Kirby et al (2008) rightly stress, essential to any characterisation of language as an adaptive system is a specification of the nature of its interdependence with other adaptive systems with which it interacts, notably the human biological system, the human rich cultural capacity, and the human capacity for learning which gives rise to the adaptivity of language through use. As Scott-Phillips et al (2009) have emphasised, studies of language evolution have to address such questions as how signalhood itself evolved, as they put it “How does a signal signal its own signalhood”. To address these questions, they develop a significant methodology for experimental testing, via both computational and cognitivepsychological test methodologies. Despite often marked differences between these various stand-points, two assumptions nevertheless have remained common to all: (i) that establishing signalhood, i.e. string content correspondences has to involve assigning structure to the natural language strings themselves with compositionality to be defined over those strings; (ii) that successful communication requires establishing some fixed signal-content correspondence as intended by the speaker to be recovered by the hearer.15 The new dimension that DS brings to the debate is the claim that it is interaction which is critical to successful communicative exchanges rather than agreement of the necessity to recover some priorly fixed signal-content correspondence which the speaker of the utterance might have in mind, and the attendant claim that language is not so much a set of string-structure pairings which have to be subject to interpretation but a set of processes for building up structure-content correspondences relative to each participant’s own (structured) context. The starting point for exploring this shift in perspective is the very familiar observation that all utterance exchanges will, on DS assumptions, necessarily and visibly involve moment by moment interaction between participants in

15

Kirby et al modify this assumption in providing evidence that in cross-generational acquisition, positing of polysemies enhances rather than disadvantages language learning/ change (Kirby et al 2008).

grammar as procedures

291

the communicative activity as they severally build up structured representations of content. But we can go further than this by extrapolating from first language acquisition exchanges between carers and infants the evidence for activities that can be said by abstraction also to be available to human agents at an evolutionary stage immediately previous to language emergence. Before any of the emergence of language, higher primates were displaying some potential for interactive behaviour, often but not exclusively associated with females with the nurturing of offspring to maturity, grooming, etc (as witness the many tv programmes we are currently inundated with). We can thus legitimately presume that the capacity for interaction itself was in place at the time of emergence of humans as a species. What was missing and had to be secured, was the recognition by the parties of the manifestness of such interactivity. Without seeking to trivialise the leap involved in attribution of content to signals (and from there to sign and symbol recognition), nonetheless, we can glean a clue from the peekaboo games which pre-linguistic children so much enjoy, where there is no attribution of content. These illustrate the fluency with which pre-language children can join in on interactional exchanges despite there being no goal of estabishing shared content, the sole goal being the pleasure of the interaction as the much-enjoyed effect manifestly displayed by both participants.16 We can thus relatively safely assume that interaction and manifest pleasure in such interaction were in place before the emergence of language. There is another clue from acquisition, for, in the next phase, building on the joint manifest activity and the pleasure in doing so continues, despite asymmetry between participants. As reported by Hilbrink et al (2015) and Clark and Cassillas (2016), in the pre-language stage of the infant, the carer characteristically uses her language fluently, even though possibly self-consciously simplified, invariably interpreting the non-languagebased verbalising behaviour of the child within her own conception of what this activity leads to, or if the babbling infant is the instigator of some coactivity by providing some take-up which she as carer can construe as some opening request/query to which she provides a response (Clark and Cassillas 2016 report rich exchanges taking such forms between a Korean carer and her child). The invariable result is an interactive effect between the child and carer even in the absence of any expectation of the child’s understanding of the 16

An anecdotal report of one such exchange witnessed by RK confirms the speech-act efficacy of such exchanges. A pre-language infant in her push-chair was persisting with great pleasure in responding in a peekaboo exchange with an older language-controlling child. After several such exchanges, the child broke off saying “Stop bullying me” and rushed back to his mother.

292

kempson and cann

exchange—the sole reward of this joint activity for both being the rich bonding achieved by this interactive behaviour.17 Out of this interactive behaviour, the emergent language speaking capacity can be seen to emerge, not upon the necessity of having already in place an established rich theory of mind, as supposedly needed to secure agreed recognition of some intended content (Tomasello 2008 and elsewhere), but from the reiteration of these exchanges over the course of the days in which these semiritualised shared activities of feeding, nappy- changing etc take place. Events for a pre-language-speaking child follow a regular pattern, reiterated at regularly recurring intervals each day; and for each such case, both child and adult build their own private representation of what is offered them and what they themselves engage in as overt actions. For the child this will mean either providing some completion to a schematic representation just provided by the adult’s utterance, or if if the child is initiating some exchange with such a fragment it is on the expectation that the adult will then develop it (as in (5)–(6) above). In the earliest occurrences of language, both producing and processing such a construction would, following the anticipatory Clarkian dynamic, involve constructing assumptions or putative actions filtered by the input stimulus. In recurrent occurrences of such scenarios, all these actions will thus invariably be buttressed by the fact that the embodied language capacity necessarily coordinates and in this sense duplicates speaker and hearing actions. So no matter how disjoint or asymmetric the construals of these events by the several participants may be in these exchanges, the duplicative effect of participant actions as either parser or producer will determine regular reiteration by the individual of the pattern of activity, thereby familiarising the language-acquiring child with the language input of the carer, with the invariable reward of achieving the emotionally bonding interactive effect. And, by the very same pattern, each further addition of a word plus the update it encodes will multiply the possible cross-individual interactions. In all these cases, at each stage, high-level inferences or other-self abductions do not need to be invoked (contra Sperber and Wilson 1995; Frith 2007; Friston and Frith 2015). Learning can moreover still take place without clarification requests or corrections (though see Saxton (2010) for their efficacy in the learning progression). Acquisition and evolution alike can thus be seen to emerge on a gradualist approach without recourse to either rich innate language structure or an a priori rich mind-reading capacity. 17

Communication in language exchanges between adults in highly asymmetric nativespeaker and struggling second-language participants displays a similar pattern, as does the exchange of fluent speakers in speaking with people with language difficulties caused by brain trauma.

grammar as procedures

293

Furthermore, with the parse DAG presumed to be integral to the DS system, the language system is defined as having an inbuilt correction mechanism, which, as Bickhard 2009 notes, is quintessential to learning. We use language throughout our day to day activities, so iteration of such interactions is ensured as needed for consolidation. Such iterations of language continue month after month, and, with successive reiterations, routinised sequences may then become stored as a macro unit sequence of actions. The triggers for these sequences of action will often be stored economically in reduced phonetic form, and cued by intonation/consonant clusters and other prosodic devices for ease of retrieval.18 With all such cross-party reiterations, group effects can be seen as emerging without having to invoke the normativity which is implicit in Gricean accounts of communication and overtly postulated in Bickhard (2009) as an external normatively imposed pressure.19 It is important to stress however, that none of these exchanges preclude putative mind-reading activities. To the contrary, such tightly coordinated activity between pairs/groups of people undoubtedly buttresses the manifest group-consolidating effect of all language, for it consolidates groups of individuals who give evidence of sharing a system sufficiently closely aligned that the lack of perfect correlation may well go unnoticed, while nevertheless excluding from such groupings all those who fail to command such a set of procedures. But, importantly, what the communicative effectiveness of small children and the split utterance data demonstrate is that such mind-reading is not integral to the core of the communicative ability which the language device so signally successfully drives. In consequence, the evolutionary system notably does not need to have in place any theory of mind prior to the emergence of language (contra Tomasello 2008). To the contrary, the very activity of language processing, because it involves an intrinsically interactive exchange, will guarantee such group effects without any prior aspiration of altruistic group-

18 19

In this, the diachronic emergence of clitic clusters is illustrative—see Bouzouita 2008, Chatzikyriakidis and Kempson 2011. In this paper, we leave on one side the arguments of Sober and Sloan Wilson 1998, Sloan Wilson 2002 in which they argue that groups form adaptive units whose status competes with adaptivity at the individual level, with optimal evolution involving a trade-off between group level and individual-level adaptivity, successful groups evolving to determine control over individual behaviours in order to ensure optimal group-level adaptivity. However we note in passing that human language constitutes a striking counter-example to the need to temper the normal “selfish” pattern for survival at the individual level with intra-group mechanisms of control, as, both at individual and group levels, capacity for language is strikingly advantageous.

294

kempson and cann

consolidating coordination. All that is required to trigger the process is the recognition of the behaviour as a manifest signal of interaction. The only shift that needs to have emerged is the means of making manifest such desire for interaction (unlike Scott-Phillips et al 2009).20 In this, our account broadly follows Christiansen and Chater 2016 in seeing language as an evolving device emerging along the same pathway as language change. However, unlike Christiansen and Chater, there is no commitment to higher-order inference mechanisms underpinning communication having to be in place prior to language development, a stance notably confirmed by the research reported here: the need to invoke shared agreed content either as the result or as the grounding for successful communication has been bypassed.21 And as Kirby et al (2008) demonstrate, the emergence of structure and compositionality strikingly emerge without any intentional recognition of its efficacy, emerging simply through re-iteration across generations. So finally, language enables us to “surf uncertainty together”, to adopt Clark’s revealing metaphor, even though characteristically building representations of content on the basis of our own structured context. On this view, we have left behind the necessity of invoking a rich set of unexplained innate properties. To the contrary, innateness is reduced to an inbuilt tendency towards interaction, a minimal Baldwin effect confirmed by the manifest success of group-forming activities to consolidate their evolutionary status. No hidden homunculus threatens, or any rich in-built structural-design feature. We have also left behind the need for a sudden-switch explanation of how language evolved, for language evolution is seen in the same terms as acquisition, an essentially interactive process (Clark, E. 2009; Clark E. and Estigarribia 2011). And there is no recourse to a problematic attendant “social brain” hypothesis as invoked in passing by Wilson 2002. Finally, we have now no need to invoke language evolution as being in principle different in kind from that of histor-

20

21

The experiment in Scott-Phillips et al (2009) presumes that successful interaction determines some agreed content, conflating the determination of interaction between contributing parties with the determination of attributable content. It is notable that discussion of binding principles aside, for which Christiansen and Chater give a pragmatic account following Reinhart 1983 (which in being Gricean would be counter-intuitively underlyingly normative), their exegesis of their own Darwinian account of evolution of language as having adapted to our brains makes no further reference to pragmatic considerations, despite the accumulated evidence they provide of how processing-cost constraints have affected language over a long time-scale, a null return which itself suggests that the specifically Gricean form of communicative constraints underpinning their own account plays little or no role in the explanation. And, with

grammar as procedures

295

ical change: all three can be analysed in the same gradualist terms. The result is that we can now see language itself as an evolving subsystem of the general cognitive system, from whose activities group effects can emerge without ratified common-ground or any concept of a shared world view. The explanation we have provided is, it must be declared, manifestly functionalist; but it is not circular (the charge of Sperber 1996). Indeed, we urge that it is now high time to set aside the dismissal of functionalist arguments out of hand that have been common-place in linguistics (Berwick and Chomsky 2016, Newmeyer 1988), as in many other disciplines (Sober and Sloan-Wilson 1998, Sloan-Wilson 2002). The closing hope is that in now seeing language as intrinsically a tool for interacting in real time (Kempson et al 2016, 2017a, 2017b), we have the chance to explain why and how the language capacity might have developed inexorably.

References Atlas, Jay. 1977. Negation, ambiguity, and presupposition. Linguistics and Philosophy 1.3 21–36. Austin, J.L. 1962. How to do things with words. London, Oxford University Press. Bach, Kent. 1994. Conversational impliciture. Mind & Language 9(2), 124–162. Berwick, Robert & Noam Chomsky, 2011 Why Only Us? Language and Evolution MIT Press. Bickhard, Mark. 2009. The interactivist model. Synthese 166. 547–591 Blackburn, Patrick & Wilfried Meyer-Viol. 1994. Linguistics, logic and finite trees. Logic Journal of the Interest Group of Pure and Applied Logics, 2(1):3–29. Bouzouita, Miriam. 2008. The Diachronic Development of Spanish Clitic Placement. Ph.D King’s College London. Breheny, Richard. 2006. Communication and folk psychology. Language and Mind 21. 74–107. Bybee, Joan. 2010. Language, Usage and Cognition. Cambridge: Cambridge University Press. Cann, Ronnie. 2011. Towards an account of the English auxiliary system. In Kempson, R. Gregoromichelaki, E. & Howes, C. (eds.) The Dynamics of Lexical Interfaces. CSLI Publications: 279–317. respect to the anaphoric binding principles, an alternative account might take the form of emergent encodings of local boundary demarcations, themselves having the functionality of determining the phrasal chunks which participants need to construct during the utterance process, an account fully commensurate with the lazy brain hypothesis (see Cann et al 2005, Kempson et al 2016 for a DS formulation of reflexive and pronominal anaphora).

296

kempson and cann

Cann, Ronnie & Ruth Kempson. 2017. What do words do for us? Diachronica 71: 425– 460. Cann, Ronnie, Ruth Kempson & Lutz Marten. 2005. The Dynamics of Language. Oxford, Elsevier. Chatzikyriakidis, Stergios & Ruth Kempson. 2011. Standard Modern and Pontic Greek Person Restrictions: a feature-free dynamic account. Journal of Greek Linguistics 11(2): 127–166. Chomsky, Noam. 1965. Aspects of the Theory of Syntax, Cambridge, MA, MIT Press. Chomsky, Noam. 1986. Knowledge of Language: Its Nature, Origin and Use. Westport, CT, Praeger. Christiansen, Morton & Chater, Nick. 2016. Creating Language: Integrating Evolution, Acquisition and Processing. Cambridge, Mass: MIT Press. Clark, Andy. 2013. Predictive brains, situated agents, and the future of cognitive science. Brain and Behavioral Science 36: 181–204. Clark, Andy. 2016. Surfing Uncertainty: Prediction, Action and the Embodied Mind. Oxford: Oxford University Press. Clark, Eve. 2009. Language Acquisition Cambridge: Cambridge University Press. Clark, Eve & Marisa Casillas. 2016. First language acquisition. In Keith Allan (ed.) The Routledge Handbook of Linguistics pp. 311–329. New York: Routledge. Clark, Eve & Estigarribia, Bruno. 2011. Using speech and gesture to introduce new objects to young children. Gesture, 11, 1–23. Clifton, Charles, Antje Meyer, Lee Wurm & Rebecca Treiman. 2012. Language comprehension and production. In A.F. Healy & R.W. Proctor (Eds.), Vol. 4 Experimental Psychology. Volume 4 in I.B. Weiner (Editor-in-Chief), Handbook of Psychology, New York, Wiley. Cross, I. 2012. Music as a social and cognitive process. In P. Rebuschat, M. Rorhrmeier, J.A. Hawkins & I. Cross (eds.), Language and music as cognitive systems (pp. 313–328). Oxford: Oxford University Press. de Villiers, J.& H. Tager Flusberg. 1975. Some facts one can’t deny. Journal of Child Language 2. 279–286. Eshghi, Arash, Matthew Purver & Julian Hough. 2011. DyLan: parser for Dynamic Syntax. Technical Report, Queen Mary, University of London. Eshghi, Arash, Matthew Purver, Julian Hough. 2013. Probabilistic Induction for an Incremental Semantic Grammar. In Proceedings of the 10th International Conference on Computational Semantics (IWCS), Potsdam, March 2013. Eshghi, Arash, Christine Howes, Eleni Gregoromichelaki, Julian Hough & Matthew Purver. 2015. Feedback in conversation as incremental semantic update. Proceedings of the 11th International Conference on Computational Semantics (IWCS), London. Fitch, W. Tecumseh, Marc Hauser, & Noam Chomsky. 2005. The evolution of the language faculty: clarifications and implication. Cognition 97. 197–210. Fodor, Jerry A. 1983. The Modularity of the Mind. Cambridge, MA: MIT Press.

grammar as procedures

297

Friston, Karl & Frith, Christopher. 2015. A duet for one Consciousness and Cognition 36. 390–405. Frith, Christopher. 2007. Making up the Mind: How the Brain Creates our Mental World. Oxford: Wiley-Blackwell. Gargett, A., Gregoromichelaki, E., Howes, C. & Sato, Y. 2008. Dialogue-grammar correspondence in Dynamic Syntax. In Proceedings of the 12th Workshop on the Semantics and Pragmatics of Dialogue. (LONDIAL’08). Gregoromichelaki, Eleni. 2006. Conditionals in Dynamic Syntax. Ph.D King’s College London. Gregoromichelaki, Eleni. 2012. Review of J. Ginzburg (2012) The Interactive Stance. Folia Linguistica 47:293–316. Gregoromichelaki, Eleni. 2013. Clitic left dislocation and clitic doubling: A dynamic perspective on left-right asymmetries in Greek. In Gert Webelhuth, Manfred Sailer, Heike Walker (eds.) Rightward Movement in a Comparative Perspective. Berlin: John Benjamins. Gregoromichelaki, Eleni. 2016. Quotation in dialogue. In Paul Saka & Michael Johnson, M. (eds.) The Pragmatics of Quotation. Springer. Gregoromichelaki, Eleni & Ruth Kempson. 2015. Joint utterances and the (split-)turn taking puzzle. In Alessandro Capone & Jakob L. Mey. Interdisciplinary Studies in Pragmatics, Language and Culture. pp. 703–744. Dordrecht: Springer. Gregoromichelaki, Eleni, Ruth Kempson, Matthew Purver, Gregory J. Mills, Ronnie Cann, Wilfried Meyer-Viol & Patrick G.T. Healey. 2011. Incrementality and intentionrecognition in utterance processing. Dialogue and Discourse, 2(1): 199–233. Gregoromichelaki, Eleni, Ruth Kempson, Christine Howes & Arash Eshghi. 2013. On making syntax dynamic: the challenge of compound utterances and the architecture of the grammar. In: I. Wachsmuth, J. de Ruiter, P. Jaecks, and S. Kopp (eds.) Alignment in Communication: Towards a New Theory of Communication. John Benjamins. Grice, Paul. 1975. Logic and Conversation. In P. Cole and J. Morgan (eds.), Syntax and Semantics, 3, pp. 41–58. New York: Academic Press. Hawkins, John. 2004. Efficiency and Complexity in Grammars. Oxford: Oxford University Press. Hendriks, Petra. 2014. Asymmetries between Language Production and Comprehension. Dordrecht: Springer. Hilbrink, Elma, Herideth Gattis & Stephen Levinson. 2015. Early developmental changes in the timing of turn-taking: a longitudinal study of mother-infant interaction. Frontiers in Psychology 6, 1492, doi: 10.3389/fpsyg.2015.01492. Hough, Julian. 2015. Modelling Incremental Self-Repair Processing in Dialogue. Ph.D Queen Mary University of London. Hough, Julian & Matthew Purver. 2014. Probabilistic type theory for incremental dia-

298

kempson and cann

logue processing. In Proceedings of the EACL 2014 Workshop on Type Theory and Natural Language Semantics, pp. 80–88. ACL 2016, Gothenburg Sweden Jespersen, Otto. 1949. A Modern English Grammar on Historical Principles, Part III Syntax. New York, London: Routledge. Kamp, Hans & Uwe Reyle. 1993. From Discourse To Logic. Dordrecht, Kluwer Academic Publishers. Kempson, Ruth, Wilfried Meyer-Viol & Dov Gabbay. 2001. Dynamic Syntax: The Flow of Language Understanding. Oxford, Blackwell. Kempson, R., Gargett, A. & Gregoromichelaki, E. 2007. Clarification requests: an incremental account. In Proceedings of the 11th Workshop on the Semantics and Pragmatics of Dialogue (Decalog 2007). Kempson, Ruth, Ronnie Cann, Arash Eshghi, Eleni Gregoromichelaki & Matthew Purver. 2015. Ellipsis. In Lappin, S., Fox, C. (eds.) Handbook of Contemporary Semantics 2nd editn. 114–140. New York, London: Wiley-Blackwell. Kempson, Ruth, Ronnie Cann, Eleni Gregoromichelaki & Stergios Chatzikyriakidis. 2016. Language as mechanisms for interaction Theoretical Linguistics 42(3–4), 203– 276. Kempson, Ruth, Ronnie Cann, Eleni Gregoromichelaki & Stergios Chatzikyriakidis. 2017a. Grammar-based action. Theoretical Linguistics 43(1–2). Kempson, Ruth, Ronnie Cann, Christine Howe, & Stergios Chatzikyriakidis. 2017b. Cognitive science, language as a tool for interaction, and a new look at language evolution. Proceedings for Formal approaches to the dynamics of linguistic interaction ESSLLI 29, Toulouse. Kirby, Simon, Kenny Smith & Hannah Cornish. 2008. Language, learning and cultural evolution: how linguistic transmission leads to cumulative adaptation. In Robin Cooper & Ruth Kempson (eds.) Language in Flux; dialogue coordination, language variation, change and evolution. 81–108. London: College Publications. Lerner, Gene. 2004. On the place of linguistic resources in the organization of talk-ininteraction: Grammar as action in prompting a speaker to elaborate. Research on Language and Social Interaction, 37(2) 151–184. Lightfoot, David. 1999. The Development of Language: Acquisition, Change and Evolution. Oxford: Blackwell. Marr, David. 1982. Vision. MIT Press. Miller, Jim & Regina Weinert. 1998. Spontaneous Spoken Language. Oxford, Clarendon Press. Newmeyer, Fritz. 1998. Language Form and Language Function. Oxford: Oxford University Press. Pezzulo, Giovanni. 2011. Shared representations as coordination tools for interaction. Review of Philosophy and Psychology, 2(2):303–333. Pickering, Martin & Simon Garrod. 2004. Toward a mechanistic psychology of dialogue. Behavioral and Brain Sciences 27(2):169–226.

grammar as procedures

299

Pickering, Martin & Simon Garrod. 2013. An integrated theory of language production and comprehension Behavioral and Brain Sciences, 36(4):329–347. Pinker, Steven & Paul Bloom. 1990. Natural language and natural selection. Brain and Behavioral Sciences 13. 707–784. Poesio, Massimo & Hannes Rieser. 2010. Completions, co-ordination and alignment in dialogue. Dialogue and Discourse 1: 1–89. Purver, Matthew, Ronnie Cann & Ruth Kempson. 2006. Grammars as parsers: meeting the dialogue challenge. Research on Language and Computation 4, 289–326. Purver, Matthew, Eleni Gregoromichelaki, Wilfried Meyer-Viol & Ronnie Cann. 2010. Splitting the I’s and Crossing the You’s: context, speech acts and grammar. In Proceedings of the Semantics and Pragmatics of Dialogue (POZDIAL). Poznan, Poland. June 2010. Putnam, Hilary. 1967. The nature of mental states. (Originally published under the title ‘Psychological Predicates’) In Mind, Language and Reality: Philosophical Papers, Volume 2. Cambridge: Cambridge University Press. Reinhart, Tanya. 1983. Coreference and bound anaphora. Linguistics and Philosophy 6. 47–88. Rose, Nikolas & Abi-Rached, Joelle. 2013. Neuro: The New Brain Sciences and the Management of the Mind. Princeton: Harvard University Press. Sato, Yo. 2011. Local ambiguity, search strategies and parsing in DS. In R. Kempson, E. Gregoromichelaki & C. Howes (eds.) Dynamics of Lexical Interfaces, Chicago: CSLI Press, 201–228. Saxton, Matthew. 2010. Child Language: Acquisition and Development. Sage Publications. Scott-Phillips, Thomas, Kirby Simon, & Ritchie Graham. 2009. Signalling signalhood and the emergence of comunication. Cognition 113(2):226–233. Simon, Herbert. 1956. Rational choice and the structure of the environment. Psychological Review 63(2), 129–138. Sober, Elliott, & Wilson, D.S. 1998. Unto Others. The Evolution and Psychology of Unselfish Behavior. Boston Mass: Harvard University Press. Sperber, Dan. 1996. Explaining Culture: A Naturalistic Approach. Oxford: Blackwell. Sperber, Dan. 2002. In defense of massive modularity. In E. Dupoux (ed.) Language, Brain and Cognitive Development: Essays in Honor of Jacques Mehler. 47–57. Cambridge MA: MIT Press. Sperber, Dan & Deirdre Wilson. 1995. Relevance: Communication and Cognition (2nd edition). Oxford, Blackwell. Stainton, Robert. 2006. Words and Thoughts Oxford, Oxford University Press. Tomasello, M. 2008. The Origins of Human Communication. Cambridge Mass: MIT Press. Wilson, David Sloan. 2002. Darwin’s Cathedral: Evolution, Religion and the Nature of Society. Chicago: University of Chicago Press.

chapter 10

Illusory Inferences in a Question-Based Theory of Reasoning Philipp Koralus and Salvador Mascarenhas*

1

Introduction

The capacity for reasoning is central to modern human endeavors. Philosophers and linguists have often focused on idealized forms of reasoning as part of an account of utterance interpretation. Systematic consideration of reasoning as a cognitive capacity has largely been left to psychologists. The most widely discussed approaches to our reasoning capacity are mental logic (Rips, 1994), Bayesianism (Oaksford and Chater, 2007), and mental model theory (JohnsonLaird, 1983). Each of these approaches has important advantages. Mental logic approaches incorporate the insight that a fully satisfactory theory of reasoning should be formally precise in the sense that it is possible to calculate predictions of the theory for an unbounded number of reasoning problems from a clearly specified set of axioms. Bayesian approaches have the advantage that they seek to account for our reasoning performance in ways that would flow from a simple core idea of what our reasoning capacity is aiming at, e.g. rationally updating a probability distribution in light of evidence. Mental model theory has the advantage of making central the nature of the representations we build as we interpret premises and how those representations can be less detailed than the premise statements in terms of their information content. We have argued for a new theory of reasoning that we believe unites those advantages, the erotetic theory of reasoning (ETR) (Koralus and Mascarenhas, 2013), taking cues from a related account of utterance interpretation (Koralus, 2012) but moving beyond interpretation to reasoning proper. The core idea behind ETR is the notion that we reason by raising questions and seeking to answer them as directly as possible. This intuitive notion has been given a formally rigorous description (Koralus and Mascarenhas, 2013) elsewhere. In this paper, we focus on presenting three new experiments on reasoning documenting novel

* The research reported here was supported by a research grant from the Laces Trust. We thank Sunny Khemlani and Phil Johnson-Laird for helpful discussions.

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_011

illusory inferences in a question-based theory of reasoning

301

systematic fallacies predicted by the erotetic theory. We thus broadly follow a vision according to which tools primarily used in the study of language can yield insights about cognition more broadly (Atlas, 2005).1 Inferences from disjunctive statements involving ‘or’ are among the simplest non-trivial cases of reasoning. For example, if we accept that either there is an apple on the table or else an orange, and we further accept that there isn’t an orange on the table, we may straightforwardly conclude that there is an apple. Now, we are subject to systematic fallacies of reasoning with premises that are similarly simple. Walsh and Johnson-Laird (2004) presented participants in an experiment with the following problem: (1) (P1) Either Jane is kneeling by the fire and she is looking at the TV or otherwise Mark is standing at the window and he is peering into the garden. (P2) Jane is kneeling by the fire. Does it follow that she is looking at the TV? Remarkably, only about 10% of participants gave correct answers to problems of this form, prompting the authors to call them “illusory inferences.”2 Most participants say that it follows from the two premises that Jane is looking at the TV. However, on reflection, we can see that this is a fallacy. The truth of the premises is compatible with a situation in which Jane is kneeling by the fire but not looking at the TV, while Mark is standing by the window and is peering into the garden. The general pattern of these types of fallacies is not readily explained by recourse to pragmatic interpretive factors. For example, the conclusion that Jane is looking at the TV still does not follow under an exclusive-or interpretation of ‘or’. We will say more about more sophisticated pragmatic accounts toward the end of the paper after discussing our experiments and looking at the relevant patterns in greater generality. To date, the only systematic accounts of illusory inferences of this kind are offered by mental model theory and by the erotetic theory of reasoning. Both

1 We also follow the great Atlas tradition of philosophico-linguistic collaboration. Moreover, the procedural take on how linguistic meaning gets processed, as well as the idea that we can take both logic and cognition seriously at the same time would have been recognizable to Jay even back in the early 1970s when such notions would have been seen as science fiction by most. He might well say, “what took you so long?” Koralus remains deeply indebted to him as an intellectual example and model of intellectual fearlessness. 2 A correct inference from the premises in (1) would be, for example, any restatement of the premises, as well as a statement that “nothing follows.”

302

koralus and mascarenhas

theories account not only for such inferences with disjunctions and conjunctions as in (1) (mental models theory: Walsh and Johnson-Laird, 2004; erotetic theory: Koralus and Mascarenhas, 2013), but also for illusory inferences with conditionals (MMT: Johnson-Laird and Savary, 1999; ETR: Koralus and Mascarenhas, 2013) and with quantifiers (MMT: Khemlani and Johnson-Laird, 2012; ETR: Mascarenhas and Koralus, 2015). The mental-model based explanation Walsh and Johnson-Laird propose for this illusory inference is based on the idea that we build mental models of only some of the alternative possibilities compatible with the first premise, notably of the alternatives in which Jane is kneeling and looking, the alternative in which Mark is standing and peering, and the alternative which combines both of the former. Walsh and Johnson-Laird suggest that when we encounter the second premise we “match” that premise to the alternatives from the first premise that are partly co-referential with it. This match is then treated as definitively establishing that one of these alternatives holds. In this case, all envisaged alternatives that “match” Jane kneeling by the fire also include that Jane is looking at the TV, yielding the illusory inference. ETR proposes a rather different explanation of illusory inferences. The core idea is that a disjunctive premise statement raises the question of which of the disjuncts is the case. Effectively, (P1) is interpreted as akin to the question, “am I in a kneeling and looking situation or in a standing and peering situation?” (P2) is then taken as akin to “you are in a kneeling situation!”, interpreted as a maximally strong answer. Taking P2 as a maximally strong answer to P1 would lead us to conclude that we are in a kneeling and looking situation. This question/answer-based explanation of the illusory inference pattern does not require co-reference between expressions in the premise statements. Moreover, it predicts an order effect and other cases of “illusory” inferences, which we address in experiments reported further below. Our results show that the pattern of “illusory” inferences is much more general and systematic than has previously been reported. We argue that existing theories do not account for these data and use this to motivate the notion of reasoning as questionanswering. We end the paper with a description of how this idea yields a new general account of propositional reasoning that can make sense of both fallacies and of the possibility of valid reasoning.

2

Experiment 1 Illusory Inferences from Disjunctions without Co-reference

Our first concern was to establish whether the illusory inference pattern is in fact due to co-reference as Walsh and Johnson-Laird (2004) suggest, or whether

illusory inferences in a question-based theory of reasoning

303

it applies more broadly to premise pairs of the form “(P1) A and B or B and C. (P2) A.” To do this, we relied on indefinite expressions. Successive uses of indefinites like ‘a’ and ‘an’ are not typically interpreted as referring to the same individual (Heim, 1982), so their use would not provide for interpretations on which successive premises are linked by co-reference, unlike expressions like proper names and indexical expression like ‘it’, which were used by Walsh and Johnson-Laird (2004). In experiment 1, we examined whether illusory inferences from disjunctions are driven by processes that are specific to building mental models with coreference across premises. We examined four illusory inference problems and four control problems that were not hypothesized to yield illusory inferences. Both types of problems involved two premises, where the first premise consisted of a disjunction and the second premise consisted of an atomic proposition or a negated atomic proposition. The target and control problems were variants of the following two examples: (2) Sample Target Problem. There is an ace and a queen, or else a king and a ten. There is a king. What if anything follows? (3) Sample Control Problem. There is an ace and a king, or else a queen and a jack. There isn’t an ace. What if anything follows? We predicted that, in the target problem, participants should systematically draw the illusory inference that there is a ten in the hand, and so on in similar problems. We predicted that the incidence of these mistakes in target problems should be far greater than the incidence of invalid inferences in the control problems. 2.1 Method 2.1.1 Participants and Design 241 members of the online Amazon Mechanical Turk worker community (average age 33 years, σ = 10. 2, 94 female, 146 male, 1 other/decline to state) carried out eight reasoning problems, including the same four target and four control problems for each subject. Subjects were randomly assigned to one of two conditions. In condition 1 (126 subjects), subjects saw problems as in (2) and (3) above, in condition 2 (115 subjects), subjects saw the same materials with

304

koralus and mascarenhas

premises in reversed order. In this section we reports the results of condition 1, the canonical order version. Subjects served as their own controls. The order of presentation was randomized for each subject. Both target and control problems each consisted of a disjunctive premise with two disjuncts followed by a premise of either atomic or negated atomic propositional form. In each problem, the premise statements were followed by the question, “what if anything follows?” and a text box to record responses. 2.1.2 Procedure The experiment was carried out over the internet using Qualtrics and participants were anonymously recruited and paid through the Amazon mTurk website. Each participant was rewarded with USD 0.25 for their participation. Participants were invited to engage in a study of reasoning in which they had to say what they can conclude from a set of statements. They were asked not to make notes or use search engines while performing the task. Before the target and control questions were presented, each participant was shown two worked-out sample reasoning problems of an unrelated kind using conditionals. All statements were explained to be about a large hand of cards. Participants typed their responses into text boxes under the premise statements. They were told that the experiment would last approximately 5 minutes but were given as much time as they needed, up to 10 minutes. On average, they took just under five minutes to complete the experiment. 2.2 Results The participants’ written responses were coded as follows into binary categories. For target problems, we coded a response as a “1” in the category of illusory inference (ILL) if and only if at least one illusory inference proposition was a conjunct in the written answer and no other invalid inferences were present. For control problems, we coded a response as invalid (INV) if and only if at least one invalid inference of any sort was present, excepting responses like “nothing follows,” “no,” and the like. In all cases, we made allowance for the fact that some participants may interpret ‘or’ as inclusive and some as exclusive. Two research assistants, unaware of the study’s hypotheses, coded the freeform responses into the above categories, agreeing on 99.2 % of the data points. The few discrepancies were resolved by a third coder. 98 % of participants made one or more illusory inferences, while 22% made one or more invalid inferences in control problems. We rejected the null hypothesis that illusory inferences were as frequent as invalid inferences in control problems (Wilcoxon Matched Pairs Test, V = 83, 400, p < 0. 0001). We summarize the data for the different target and control problems in Table 10.1.

illusory inferences in a question-based theory of reasoning

305

table 10.1 Results of experiment 1, condition 1. “a”, “j”, “k”, “q”, “2”, “8”, and “x” stand for the cards ace, jack, king, queen, two, eight, and ten, respectively.

Type Target Target Target Target Control Control Control Control

Pattern

aq ∨ kx ‖ k j2 ∨ a8 ‖ j q8 ∨ 2a ‖ 2 xk ∨ qa ‖ x ak ∨ qj ‖ ¬a j8 ∨ ax ‖ ¬j 28 ∨ kj ‖ ¬k 8k ∨ qa ‖ ¬q

ILL

INV

116(92%) 115(91%) 116(92%) 114(90%) -

14(11%) 19(15%) 22(17%) 22(17%)

2.3 Discussion Our results show that the pattern of illusory inference is not due to co-reference across premises. The explanation offered by ETR does not depend on co-reference. ETR holds instead that the illusory inference results from treating the first premise of our sample target in (2) as a question, e.g. “am I in an ace and queen situation or in a king and ten situation?,” and treating the second premise as akin to “you’re in a king situation,” interpreted as a maximally strong answer, yielding the fallacious conclusion that there is a king and a ten. 2.4 Order Effect for Illusory Inference from Disjunction One could take the Walsh and Johnson-Laird (2004) notion of “matching” to extend to atomic propositions like “there is an ace,” regardless of co-reference (as opposed to more abstract models). However, it is worth noting that matching does not intrinsically invoke order. Matching curtains to a sofa and matching a sofa to curtains should yield the same outcome if there is only one possible matching sofa-curtain pair (e.g., in the problems we considered, there is only one possible match across the two premises). By contrast, we cannot treat something as an answer without having a question first. Thus, the explanation of the illusory inference offered by ETR, unlike the matching-based account, immediately predicts an order effect. If the premises in the illusory inference problems are reversed, the illusory inference should be mitigated. Condition 2 of experiment 1, where subjects saw the same materials as in condition 1 with the order of the premises reversed, allowed us to test this hypothesis.

306

koralus and mascarenhas

table 10.2 Results of experiment 1, condition 2. “a”, “j”, “k”, “q”, “2”, “8”, and “x” stand for the cards ace, jack, king, queen, two, eight, and ten, respectively.

Type Target Target Target Target

Pattern

k ‖ aq ∨ kx j ‖ j2 ∨ a8 2 ‖ q8 ∨ 2a x ‖ xk ∨ qa

ILL 90(78%) 95(82%) 90(78%) 88(77%)

2.5 Results Participants’ responses were coded into the same categories as condition 1 described above. The two initial coders agreed on 97.7 % of data points, and disagreements were resolved by a third coder. As predicted, fewer illusory inferences were made in the reversed condition compared to the canonicalorder condition in experiment 1. The number of illusory inferences dropped by approximately 10% on average when the premises were reversed. We rejected the null hypothesis that the number of illusory inferences was the same for the two premise orders (Mann-Whitney, W = 130, 474, p < 0. 0001). There was no significant effect of premise order in our controls (disjunctive syllogism, Mann-Whitney, W = 117, 502, p > 0. 5). The results for the target cases are summarized in Table 10.2. 2.6 Discussion A drop in acceptance rates for the reversed illusory inferences in the order of 10% does not entail of course that illusory inferences disappear when premises are reversed. But clearly the reversal has a significant ameliorating effect that was not found in our controls. These results are consistent with the prediction of ETR that reversing the order of premises should mitigate the illusory inference pattern. They cast doubt on explanations of these kinds of inference patterns involving notions like “matching” that are not relevantly asymmetric. Now, mental model theory also provides a more general procedure for conjoining mental models (Johnson-Laird, 2008; Khemlani and Johnson-Laird, 2009). If we apply this to the problems in this experiment, the first premise would generate a single mental model, which would then have to be combined with the set of mental models generated by the second premise according to

illusory inferences in a question-based theory of reasoning

307

the process for “conjoining” mental models (ibid.). However, this process is also designed to make us jump to conclusions in reasoning. As we conjoin the mental model for “there is a king” with the mental models for “there is an ace and a queen, or else a king and a ten,” the procedure, as defined by Johnson-Laird and Khemlani, rules out all of those mental models in which we do not have an ace and a queen. Thus, the mental model conjoin procedure applied to the case at hand would also yield the illusory inference in the reversed case. Moreover, the mental model conjoin procedure would not predict an illusory inference in the canonical order for which the illusory inference is in fact the strongest (see experiment 1). In sum, we do not see a non-ad hoc way to account for the observed order effect with the classical mental model theory.

3

Experiment 2 Illusory Inferences from Two Disjunctions

In experiment 2, we further investigated the nature of illusory inferences from disjunctions. If it is crucial for obtaining an illusory inference that we match a categorical premise to one of the alternative models induced by a disjunctive premise, we would not expect illusory inferences from two disjunctive premises. However, if we take the view that what explains illusory inferences is a process of treating successive premises and questions and maximally strong answers to them, as formally defined by Koralus and Mascarenhas (2013), there is no such restriction. We can clearly treat disjunctions as answers to questions (e.g. Q: Where’s Mary? A: Either at home or at work.). We hypothesized that a significant number of participants would draw an illusory conclusion from premises like the following. (4) Sample Target Problem. There is a queen and an eight, or else a two and an ace. There is a two, or a two and a six. What if anything follows? In the above example, the hypothesized illusory inference would be to say that it follows that there is an ace. The procedures were the same as in experiment 1, using the same control problems. A new set of 124 participants was recruited (average age 36, σ = 10. 8, 46 female, 75 male, 3 other/decline to state).

308

koralus and mascarenhas

table 10.3 Results of experiment 2. “a”, “j”, “k”, “q”, “2”, “8”, and “x” stand for the cards ace, jack, king, queen, two, eight, and ten, respectively.

Type Target Target Target Target

Pattern

aq ∨ kx ‖ k ∨ k5 j2 ∨ a8 ‖ j ∨ j7 q8 ∨ 2a ‖ 2 ∨ 26 xk ∨ qa ‖ x ∨ x9

ILL

INV

62(50%) 62(50%) 63(51%) 61(49%)

-

3.1 Results Free form responses were coded by two blind coders as in the previous experiment, who agreed on 95.4% of the data. Disagreements were resolved by a third coder. 60% of participants made an illusory inference in at least one of the four target problems, while 23% made an invalid inference in at least one control problem. We rejected the null hypothesis that the frequency of illusory inferences on target problems was the same as that of invalid inferences in control problems (Wilcoxon Matched Pairs Test, V = 33, 841. 5, p < 0. 0001). The results for target problems are summarized in Table 10.3 (control problem results omitted since they were nearly identical to those in experiment 1). 3.2 Discussion We found that 60% of participants made at least one illusory inference from two disjunctive premises. This shows that it is not a requirement for illusory inferences from disjunctions that one have a categorical premise to match to an alternative possibility provided by the disjunctive premise. The mental model conjoin procedure described in Khemlani and Johnson-Laird (2009) does not generate this illusory inference either, so it looks like none of the mechanisms in classical mental model theory for generating illusory inferences capture these data. Curiously, the mental model conjoin procedure would generate the illusory inference if the order of the premises was reversed. ETR straightforwardly predicts the observed inference pattern, as can be computed from the formal system in Koralus and Mascarenhas (2013). Intuitively, disjunctions that have categorical entailments can serve as answers. In the case of the example in (4), “there is a two or a two and a six” answers the question of whether there is a two. So far, we have only seen illusory inferences from disjunctions that were driven by hastily eliminating alternatives obtained from a first premise that in

illusory inferences in a question-based theory of reasoning

309

some sense did not overlap with the information in a further premise. Experiment 3 addressed whether this notion needs greater generality.

4

Experiment 3 Illusory Inferences from Triple Disjunctions

In experiment 3, we examined a further novel illusory inference pattern. The question motivating this experiment is whether we make illusory inferences because we detect overlap between an alternative in a first premise with what is established by a second premise, or whether we look for alternatives in a first premise that have the most in common with what is established by further premises. The thought is that if we treat further premises as maximally strong answers, we might expect that only alternatives in the question that have as much in common with the answer survive. This is an appreciably more powerful notion that would correspondingly yield fallacies not predicted by mere “overlap.” Both target and control problems involved three premises, where the first premise consisted of a triple disjunction and the second and third premises consisted of conjunctions or negated atomic statements. The control problems had valid non-disjunctive conclusions that could be reached by two disjunctive-syllogism inferences. The target problems did not have valid nondisjunctive conclusions besides the trivial restatement of the second or third premises. The target and control problems were variants of the following two examples: (5) Sample Target Problem. There is an ace and a jack and a queen, or else there is an eight and a ten and a two, or else there is an ace. There is an ace and a jack, and there is an eight and a ten. There is not a queen. What if anything follows? (6) Sample Control Problem. There is an ace and a king and a queen, or else there is an ace and a jack and a ten, or else there is an eight. There isn’t queen. There isn’t a ten. What if anything follows?

310

koralus and mascarenhas

For the target problem, we predicted that participants would systematically draw the illusory inference that there is a two in the hand. We predicted furthermore that the incidence of these mistakes in target problems should be far greater than the incidence of invalid inferences in the control problems. 4.1 Method The protocol followed was the same as that in experiment 1, with a new set of participants (121, average age 33 years, σ = 10. 5, 70 Female, 51 Male).

4.2 Results Free form responses were coded by two blind coders as in the previous experiment, who agreed on 98.2% of the data. Disagreements were resolved by a third coder. 79% of participants made an illusory inference in at least one of the four target problems, while 25% made an invalid inference in a control problem. We rejected the null hypothesis that illusory inferences were as frequent as invalid inferences in control problems (Wilcoxon Matched Pairs Test, V = 39, 243, p |(⨅ Δ) ⊓ γ|⟩, B, i} What Q-update does is treat Δ as a maximally strong answer to Γ, keeping all of those alternatives in Γ such that no other alternatives in Γ have more in common with all alternatives in Δ. Definition 2 (Update) Let Γ be a question and Δ an answer. The update of Γ with Δ is ⟨Γ, B, i⟩[Δ]Up = ⟨Γ, B, i⟩[Δ]Q [Δ]C . 5 Strictly speaking, the information that there might be a x is also added, yielding {k&j, k&j&x}. The theory’s rule of conjunct simplification applies to every alternative under consideration, so that we are still left with the fallacious conclusion. We omit these details from the main text to simplify exposition, but the procedure is precisely defined in the cited article.

318

koralus and mascarenhas

premise 2. The third alternative of premise 1 shares only one alternative with premise 2, so it is eliminated. Then, premise 3, being contradictory with the first surviving alternative, leaves us with the second alternative, namely {8&x&2}. From here, the observed fallacious conclusion that there is a 2 follows immediately. 5.2 Dynamic Updating of Mental Model Discourses The erotetic theory is dynamic. That is, the operations that update the workspace of reasoning with the mental representations of successive premises are sensitive to the order of those premises. This follows from the natural dynamics of question asking and answering. In particular, the procedure that implements Part I of the erotetic principle will always take a new premise being processed to be an answer to a question that was processed earlier. The procedure itself cannot reverse this order and interpret and old premise as an answer to a new question. This accounts for the order effect in experiment 2. To get a fallacious inference in the reversed case discussed in experiment 2, the reasoner would have to mentally take an extra step to change the order of the premises. It should be noted that besides treating information as questions and answers, our update rule also allows for cases in which we simply accumulate information, as when we are given successive categorical statements. 5.2.1 Simple Deduction Rule Reasoning is not just a matter of update. Once reasoners hear and process each premise, they must then be able to perform simple transformations on the resulting mental model, to check what follows. We assume that there is a rule of disjunct simplification, validating the inference (p ∧ q) ∨ r ⊧ p ∨ r. This rule for disjunct simplification includes conjunction simplification as a special case, as the reader can see and is explained in more detail in Koralus and Mascarenhas (2013). 5.2.2 Default Reasoning Strategy We make a simple postulate describing how reasoning problems are approached by default. Namely: when given a reasoning problem with premises P0 , …, Pn and conclusion C, reasoners update a blank mental model discourse with each premise, in the order the premises were given. They may then apply the simple deductive rule, targeting the conclusion C. If the resulting mental model in discourse is identical to C, then the inference is deemed valid. Otherwise, it is deemed invalid. If no target conclusion is given for evaluation, reasoners simply update their mental model discourse with all the premise

illusory inferences in a question-based theory of reasoning

319

statements and see what holds in the resulting model. The full description of the default reasoning strategy in Koralus and Mascarenhas (2013) includes a model of how background knowledge can influence reasoning performance. 5.2.3 Eliminating Contradictions The theory takes it that reasoners do not immediately see anything wrong with contradictions. However, there must be a process allowing them to look at the representations they are entertaining and check whether they are consistent or not. This comes at a cost and is not part of default reasoning, but it must be a possibility if we want to account for the successes of our reasoning faculty. We therefore define an operation that filters the mental model in discourse, going over each alternative and eliminating all those that are contradictory. This operation also eliminates double negations as described in Koralus and Mascarenhas (2013). 5.2.4 Expanding Possibilities through Inquiry It may seem as though saying that reasoners represent “there is an ace and a queen or there is a jack” as the set of alternative possibilities {a&q, j} means attributing to those reasoners an irrational discarding of information. After all, the disjunctive statement allows for a case in which we have an ace, a jack, but no queen. This case is not one of the alternatives in {a&q, j}, so are we not really attributing an interpretive mistake to reasoners? Not so. In fact, there is no loss of information at the interpretive stage at all, if we take it that the alternatives reasoners represent for premise statements are simply the exact verifiers of those premise statements (Fine 2012). As already noted, the set of exact verifiers of a statement is the set of all and only those things that exactly make the statement true. For example, a&q exactly verifies (a&q or j), as does j. However, while a&j&¬q verifies (a&q or j), it does not verify this disjunction exactly. An exact verifier does not have any superfluous elements that one could take away from it while keeping it a verifier. We can reduce a&j&¬q to j while still having a verifier, so a&j&¬q cannot be an exact verifier. Now, the key to our account of the possibility of correct reasoning is that we can formally recover a full set of “classical” alternatives from a set of exact verifiers. To harness this fact for the erotetic theory of reasoning, we need an operation that expands the mental model under consideration into one that represents every possibility with respect to some propositional atom. We call this operation “inquire.” None of the operations in the erotetic theory of reasoning commits us to fallacious inferences (they are, in a relevant sense, sound in principle), so if we “inquire” enough to represent all alternatives allowed by our premise statements, as we could by inquiring on all propositional atoms that have been mentioned, there

320

koralus and mascarenhas

is no longer any possibility of fallacious inferences. This is the crucial aspect of the theory allowing for classically sound reasoning and it implements Part II of the erotetic principle. In Koralus and Mascarenhas (2013) we prove as a theorem that inquiring on every propositional atom mentioned before updating with new premises guarantees soundness. 5.3 Conclusion In sum, we believe that the erotetic theory of reasoning provides a distinctly favorable combination of advantages. Like Bayesian approaches, the erotetic theory seeks to account for both successes and failures of reasoning as flowing from a core idea about the computational aim of our cognitive capacity for reasoning, though the perspectives offered on that aim by the two theories differ very significantly. According to the erotetic theory, this aim is to answer questions as directly as possible. Like mental model theory, the erotetic theory puts a theory of sparse representations of premises at the center of the explanation of reasoning failures. We hold that, by default, we only represent those alternative possibilities that correspond to exact verifiers of premise statements. Finally, like mental logic, the erotetic theory is formally specified, which makes it possible to calculate predictions and allowed us to prove as a theorem that sufficient inquiry guarantees sound reasoning. Moreover, the erotetic theory shows that technical work in formal semantics can be brought to bear on psychology. Though we do not have the space to present the more general framework here, the erotetic theory is a step toward a general account of reasoning and decision-making. Besides reasoning, the erotetic theory has also been extended to model decision-making (Koralus, 2016; Koralus and Alfano, 2017), and delusional thinking (Parrott and Koralus, 2015).

References Atlas, J. (2005). Logic, Meaning, and Conversation: Semantical Underdeterminacy, Implicature, and Their Interface. Oxford: Oxford University Press. Barrouillet, P., Grosset, N., and Lecas, J. (2000). Conditional reasoning by mental models: chronometric and developmental evidence. Cognition, 75(3):237–266. Braine, M.D.S., Reiser, B.J., and Rumain, B. (1984). Some empirical justification for a theory of natural propositional logic. In Bower, G.H., editor, The Psychology of Learning and Motivation, chapter 18, pages 317–371. New York: Academic Press. Braine, M.D.S. and Rumain, B. (1983). Logical reasoning. In Flavell, J.H. and Markman, E.M., editors, Handbook of Child Psychology: vol. 3, Cognitive Development, pages 263–339. New York: Wiley.

illusory inferences in a question-based theory of reasoning

321

Evans, J.S.B., Newstead, S.E., and Byrne, R.M. (1993). Human reasoning: The psychology of deduction. Psychology Press. Fine, K. (2012). A difficulty for the possible world analysis of counterfactuals. Synthese. Girotto, V., Mazzocco, A., and Tasso, A. (1997). The effect of premise order in conditional reasoning: a test of the mental model theory. Cognition, 63:1–28. Groenendijk, J. (2008). Inquisitive Semantics: Two possibilities for disjunction. ILLC Prepublications PP-2008-26, ILLC. Harman, G. (1986). Change in view: Principles of reasoning. MIT press Cambridge, MA. Heim, I. (1982). The semantics of definite and indefinite noun phrases. PhD thesis, University of Massachusetts Amherst. Johnson-Laird, P., Legrenzi, P., and Girotto, V. (2004). How we detect logical inconsistencies. Current Directions in Psychological Science, 13(2):41–45. Johnson-Laird, P.N. (1983). Mental models: towards a cognitive science of language, inference, and consciousness. Cambridge: Cambridge University Press. Johnson-Laird, P.N. (2008). Mental models and deductive reasoning. In Rips, L. and Adler, J., editors, Reasoning: studies in human inference and its foundations, pages 206–222. Cambridge: Cambridge University Press. Johnson-Laird, P.N. and Savary, F. (1999). Illusory inferences: A novel class of erroneous deductions. Cognition, 71(3):191–229. Khemlani, S. and Johnson-Laird, P.N. (2009). Disjunctive illusory inferences and how to eliminate them. Memory and Cognition, 37(5):615–623. Khemlani, S. and Johnson-Laird, P.N. (2012). Theories of the syllogism: a meta-analysis. Psychological Bulletin, 138:427–457. Khemlani, S., Orenes, I., and Johnson-Laird, P.N. (2012). Negation: A theory of its meaning, representation, and use. Journal of Cognitive Psychology, 24(5):541–559. Koralus, P. (2012). The open instruction theory of attitude reports and the pragmatics of answers. Philosopher’s Imprint, 12(14). Koralus, P. (2016). Decisions, illusory reasons, and erotetic equilibrium. Under review. Koralus, P., and Alfano, M. (2017). Reasons-based moral judgment and the erotetic theory. In: Bonnefon. J.-F. Tremoliere, B. (Eds.). Moral Inferences. Psychology Press Koralus, P. and Mascarenhas, S. (2013). The erotetic theory of reasoning: bridges between formal semantics and the psychology of deductive inference. Philosophical Perspectives, 27:312–365. Kratzer, A. and Shimoyama, J. (2002). Indeterminate pronouns: the view from Japanese. In Third Tokyo Conference on Psycholinguistics. Mascarenhas, S. (2009). Inquisitive semantics and logic. Master’s thesis, ILLC. Mascarenhas, S. (2014). Formal Semantics and the Psychology of Reasoning Building new bridges and investigating interactions. Doctoral Dissertation, New York University. Mascarenhas, S. (2011). Licensing by modification: the case of positive polarity pro-

322

koralus and mascarenhas

nouns. In Guevara, A.A., Chernilovskaya, A., and Nouwen, R., editors, Proceedings of Sinn und Bedeutung 16, pages 417–429. Mascarenhas, S. and Koralus, P. (2015). Illusory inferences: disjunctions, indefinites, and the erotetic theory of reasoning. In: Noelle, D.C., Dale, R., Warlaumont, A.S., Yoshimi, J., Matlock, T., Jennings, C.D., and Maglio, P.P. (Eds.) Proceedings of the 37th Annual Meeting of the Cognitive Science Society. Austin, TX: Cognitive Science Society. Oaksford, M. and Chater, N. (2007). Bayesian rationality: The probabilistic approach to human reasoning. Oxford University Press. Oaksford, M. and Stenning, K. (1992). Reasoning with conditionals containing negated constituents. Journal of experimental psychology. Learning, memory, and cognition, 18(4):835–854. Parrott, M. and Koralus, P. (2015). The erotetic theory of delusional thinking. Cognitive Neuropsychiatry ( forthcoming). Rips, L. (1994). The Psychology of Proof. Cambridge, MA: MIT Press. Sauerland, U. (2004). Scalar implicatures in complex sentences. Linguistics and Philosophy, (27): 367–391. Spector, B. (2007). Aspects of the pragmatics of plural morphology: On higher-order implicatures. eds. Uli Sauerland and Penka Stateva Presuppositions and implicatures in compositional semantics, 243–281. London: Palgrave Macmillan. Walsh, C. and Johnson-Laird, P.N. (2004). Coreference and reasoning. Memory and Cognition, 32: 96–106.

chapter 11

A Commitment-Theoretic Account of Moore’s Paradox* Jack Woods

1

Introduction

Moore’s paradox, the infamous felt bizarreness of sincerely uttering constructions like “I believe grass is green, but it ain’t”, has attracted attention since its original discovery (Moore 1942).1 Moore-paradoxical utterances feel contradictory, even though obviously both conjuncts can be true. It is often taken to be a paradox of belief—in the sense that the locus of the inconsistency is the beliefs of someone who so sincerely utters: If you have an explanation of why a putative content could not be coherently believed, you thereby have an explanation of why it cannot be coherently asserted.2 Shoemaker 1995, 227

* Thanks to Sarah Raskoff for the discussion which prompted this piece, Paul Benacerraf and John Burgess for useful discussions on the topic of Moorean absurdity, members of the New York Philosophy of Language Workshop for early feedback, and Derek Baker, Justin Bledin, Eliot Michaelson, and Ken Turner for very useful comments. 1 Philosophers have often distinguished the omissive form of Moore’s paradox—p, but I don’t believe that p—and the commissive form—p, but I believe that not p. I deal primarily with the omissive form here, but I don’t think there is much philosophical traction to be gained out of the distinction once we reject the idea that we explain Moore’s paradox by means of inconsistent or irrational belief. The commissive form more or less transforms into the omissive by means of the presumption (not always true!) that quite generally people believe not p only when they don’t believe p. 2 Atlas (2007) has argued vigorously (and, I think, successfully) against Shoemaker’s account on grounds that Shoemaker illicitly moves from “manifesting a belief in P” to “manifestly asserting my belief in P”. He also argues against the uniform paradoxicality of constructions of the form “F, but I don’t belief that F” by constructing cases in which they can be sincerely asserted. I agree with the examples, but think that they can be handled by the account I suggest at the close of this paper.

© koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_012

324

woods

This priority thesis, however, is insufficient to give a general explanation of Moore-paradoxical phenomena (it’s also false) as: I promise I’ll marry you, but I won’t has the same paradoxical character as: I believe it’s raining, but it ain’t even though the latter, but not the former, involves expression of the utterer believing thus and so. Moore-paradoxical utterances can occur in non-assertoric constructions and, moreover, can occur in the context of speech acts which do not express mental states at all.3 I will leave aside, for the purpose of this essay, whether we should be giving an independent analysis of Moore’s paradox for thought or, alternatively and as I prefer, whether we should regard Mooreparadoxical thought as derivative from Moore-paradoxical speech.4 I will not address the priority thesis again directly, but it should be clear to the reader that the considerations below tell directly against it.5 The fact that there are Moore-paradoxical constructions which do not involve expressing mental states supports the hypothesis that Moore paradoxicality arises due to speech acts engendering commitments of a particular sort. Some speech acts, like typical assertions, give rise to commitments to believing thus and so.6 Others, such as promises, give rise to commitments to acting thus and so. When we go on to immediately deny that these commitments are met, our purpose in engaging in the speech act becomes mysterious and interpreters quite reasonably become puzzled about what we were attempting to accomplish by means of our utterances. We have seemingly undermined the purpose of making the promise or assertion at all. Basic linguistic competence, presumed of those we typically interpret, requires mastery of the commitments

3 See (Williams 2013) for independent arguments against the priority thesis. Williams focuses on the failure of the priority thesis when we restrict our attention to the case of belief. My interest is in showing that even if the priority thesis holds for belief, it doesn’t suffice to explain other cases of Moore paradoxicality. 4 See (Baker and Woods 2015, § IV.B) for suggestions in this direction. 5 For good examples of linguistically acceptable utterances plausibly manifesting irrationality, like “I shouldn’t do it, but I will anyway”, see the discussion in (Baker 2014, §6). 6 Take ‘typical’ here to be meant seriously; I do not mean to deny that there are contexts where a more relaxed standard for asserting is in force. After all, we’ve all shot the shit a time or two.

a commitment-theoretic account of moore’s paradox

325

incurred by asserting, commanding, and promising, and the like.7 Given this, explicit claimed violation of these commitments gives rise to our inability to successfully interpret the speaker. This is the basic sense of “inconsistency” or “oddness” that we now call Moore paradoxicality.8 Or, anyways, so I will argue. I originally argued for a version of this claim in a paper rejecting expressivism (Woods 2014). It struck me then, as it strikes me now, as the right sort of take on a general phenomenon which had been typically addressed in only its belief-involving manifestations. I was happy to find afterwards that (Atlas 2007) provided further and distinct reasons useful to establishing what I originally took the most natural explanation of Moore paradoxicality. Atlas writes: … it is bizarre to think that the impossibility of the rational, sincere assertion of a Moore sentence could explain the oddity of uttering a Moore sentence assertively (as Moore (1993: 207) would put it), even if the impossibility of a sincere, “rational” assertion explains why either the utterance is insincere, or expresses a “non-rational” belief or a corrigible belief, or is not an assertion—oddities all, but none of insincerity, non-rationality, corrigibility, and an assertoric speech-act “misfire” is the oddity of uttering the Moore sentence assertively. None of those consequences would seem to explain the particular oddity of uttering the Moore sentence assertively. (As Paul Benacerraf once remarked, in reaction to this claim, an adequate account of the oddity must appeal to principles that every speaker/hearer can be expected to “know”, in the sense in which we know our language and its uses. Not every speaker, surely, is natively a Port Royal logician.) Atlas 2007: 142–143

and I completely agree. These sorts of explanation do not get at the root of the problem. Insincere assertions—“I’m pleased to see you, but I ain’t”, when I ain’t—still give rise to Moore paradoxicality, as do irrational and corrigible assertions. The madman who yells that the sky is constantly falling in, but then immediately demurs is not just mad, but linguistically incompetent. And failures to assert at all tend to undermine the phenomena in question since there is a natural and immediate explanation of why the failed assertors have failed 7 This, of course, is a familiar view. For loci classici, see (Searle 1969) and (Brandom 1994). For a more recent discussion treating of related issues that arise with lying, see (Michaelson 2016) and (Stainton 2016). 8 I mean ‘inconsistency’ here in the older and broader sense of “standing in tension with one another”, not the narrower “unable to both be true”.

326

woods

to “contradict” themselves in the relevant sense—they have not actually made an assertion. What we need is a way to chisel down to the class of utterances where Moore paradoxicality arises and, then, to give an actual explanation of why, in many cases, these utterances give rise to such a feeling of oddness. Since insincere promises, commands, and so on can still commit, commitment violation by itself is insufficient to explain Moore paradoxicality. What we need is a general account of why we get Moore paradoxicality in some cases— one which explains why these cases are puzzling for everyone, linguistically savvy or not. We also need to explain why it is so easy to multiply Mooreparadoxical cases for a variety of different speech acts. We can articulate these demands in terms of two desiderata of an explanation of Moore paradoxicality: Uniformity: An explanation of Moore paradoxicality must explain its occurrence in all cases, not merely the somewhat special, if historically important, case of first-personal assertions. Egalitarianism: An explanation of Moore paradoxicality must explain its manifestation to all linguistically competent interpreters of speech (when interpreting speakers themselves presumed linguistically competent.) Since the irrational and the insincere typically count as linguistically competent, it isn’t sufficient explanation of a Moore-Paradoxical utterance that the utterer is seen as irrational or insincere as a result of the seeming inconsistency or irrationality of their utterance. In what follows, I’ll first argue for the breadth of Moore-paradoxical phenomena. I will then argue that this breadth makes implausible Shoemaker’s attempt to derive Moore paradoxicality from the “logic” of belief. Similar views will fall for similar reasons, though I do not argue for this explicitly. I will then turn to developing the commitmenttheoretic explanation of Moore’s paradox I favor. The key to my explanation involves welding a general theory of the commitments arising from speech acts to a constraint on normal interpretation of the utterances of others. The constraint, roughly, is that utterers should generally be taken to reflexively intend to take on the commitments incurred by means of the speech act they are engaged in. In particular, this means that when a speaker says “grass is green, but I don’t believe it is”, they reflexively intend to take on the commitments of asserting that grass is green. Since these commitments involve believing what I assert, when I deny this commitment in the second conjunct, I undercut my presumed intention to take on this commitment

a commitment-theoretic account of moore’s paradox

327

and thereby puzzle the speaker. I seem, in the most literal sense, inconsistent. Likewise, when I say “I promise to give you 5 Turkish lira, but I won’t”, I am presumed to reflexively intend to take on the commitments of promising which include giving you 5 Turkish lira. Only presumed, of course, since I might actually be insincere. So, when I claim that I won’t fulfill this commitment, I give rise in my interpreter the puzzling feeling of directly undermining my earlier promise. This account allows me to meet the constraint that flagrantly insincere assertions, promises, and what have you can be felicitously uttered—as they obviously can be. Irrationality is no excuse for failing to meet a promise, in many cases, and it certainly is no excuse for not believing what you assert. What is needed for cases with the form of Moore’s paradox to be linguistically felicitous is that the default presumption of reflexive intention be waived in that context. I will discuss such cases below (see §2.2, § 5). I conclude that the resulting account satisfies uniformity, egalitarianism, and doesn’t over-predict occurrences of Moore’s paradox. What more could we want from an explanation?

2

Much Moore Paradoxicality

Sincerely uttering any of the following in non-deviant contexts gives rise to the feeling we call Moore paradoxicality: (1a) (1b) (1c) (1d) (1e) (1f) (1g)

It’s raining, but I don’t believe it is. Get me a beer, but I don’t want you to get me a beer.9 Boo Yankees! I have no negative feelings towards the Yankees.10 Would that I could have another drink, but I don’t wish for another. I promise I’ll pick you up at the airport, but I won’t. I swear to you that he’s a spy, but he isn’t. It might be raining, but it’s not.

None of these except the first involve belief in a direct fashion. (1b) involves denying the possession of a desire, (1c) involves denying possessing certain evaluative attitudes, (1d) involves denying that I wish in certain ways, (1e)

9 10

See Condoravdi and Lauer (2012), § 2.2 for defense of the incoherence of this example and discussion of both it and examples like (1d). See Woods (2014) for discussion of this example.

328

woods

involves denying that I’ll act in certain ways, (1f) involves denying that a certain fact holds, and (1g) the same. Note, as evidence of the Moore paradoxicality of these constructions, that switching to the past tense (where it makes sense) results in sensible constructions: (2a) (2d) (2e) (2f) (2g)

It was raining, but I didn’t believe it was. Would that I could have had another drink, but I didn’t wish for another. I promised I’d pick you up at the airport, but I won’t. I swore to you that he’s a spy, but he wasn’t. It might have been raining, but it wasn’t.11

The coherence and lack of paradoxical character of the past-tense versions of present-tense oddness is one of the characteristic marks of Moore paradoxicality. The examples above are also non-paradoxical when we put them into the second or third-person (where such a modification is natural). Are these examples really Moore paradoxical? Yes, it seems so. Searle and Vanderveken (1985) argue that any speech act conjoined with the denial of its sincerity condition will, in typical circumstances, produce the usual reaction we label as Moore paradoxicality. Consider, for example: What time is it? But I don’t want to know what time it is. Williams 2015

Searle and Vanderveken do not discuss cases like (1f), but they feel entirely analogous to me and, I hope, to you. Sincerity conditions are a usual and flagrant case in which we get Moore paradoxical phenomena, but it is entirely unclear why they should be the only such cases.12 The above, especially the promissory cases, seem clearly instances of the phenomenon that Moore described.13

11 12

13

(1b) and (1c) do not have natural past tense constructions. For an entertaining non-standard case, consider Sorensen’s ‘God knows that we are not theists’ (Sorensen 1988: 17). On the presumption that sincere utterance of this commits me to the existence of God, this looks like a similar case. On the other hand, it threatens to collapse a potentially useful distinction between pragmatic paradox and Moore’s. So I shall not further pursue this type of case here. I shall not here argue further that these cases are really Moore paradoxical. My account works even if they’re not—though its motivations become slightly less clear. (Heal 1977) is often cited as claiming that (1b) and like constructions are not Moore paradoxical, but I read her as less concerned with Moore paradoxicality in general and more concerned

a commitment-theoretic account of moore’s paradox

329

If these examples really are Moore paradoxical—and I put it to you that they are—then we need an account of the phenomenon they are all instances of.14 That is, satisfying uniformity demands that our account of Moore paradoxicality encompass assertions, imperatives, ejaculations, optatives, promissory utterances, avowals, and epistemic modals. Moreover, such an account should predict Moore paradoxicality in cases where it occurs, but should not predict Moore paradoxicality in cases where it does not or, importantly, assimilate ordinary cases of having inconsistent beliefs to the same phenomena. extensional adequacy: An explanation of Moore paradoxicality should not overspill to cases of ordinary inconsistent or irrational beliefs. Can we make sense of uniformity if we explain typical cases of Mooreparadoxical utterances on the basis of inconsistent beliefs? In order to do so, we would have to derive inconsistent beliefs from beliefs expressed by the above constructions. Consider, for example, (1f). Why is it that we get Moore paradoxicality from swearing to a fact, then immediately denying that the fact holds. The potential explanation, on the inconsistent belief approach, holds that when we swear that p, we express a belief that p and when we deny p, we express our belief that not p. Presuming that believing that not p rationally entails— requires, on pain of irrationality—not believing p, we get that we do not believe p (at least if we are rational). But believing p and not believing p are inconsistent, so we have displayed that we believe in a way which can’t be satisfied. And this, presumably, accounts for the feeling of contradiction. This explanation also vindicates the first-personal present-tense character of Moore’s paradox. I can coherently say of you that you swore up and down that grass is purple, but it ain’t. I can even coherently say of me that I swore up and down that there was no meat in the dish, but there was. There is, though, a certain certain unloveliness to this type of explanation. It is not at all clear that in swearing to something or promising to do something we express the sort of beliefs necessary to generate Moore paradoxicality. Even

14

with whether wanting is a sincerity condition on commanding. It’s worth noting that my solution below (§ 4) neatly handles her examples as all of them are cases where the presumption of a reflexive intention on the part of the utterer is not in force. I am willing to hedge here and say that for those of us, like myself, Searle and Vanderveken, etc. who find Moore’s paradox in cases outside of first-personal assertions, another account is needed. Of course, I think the data I canvassed above is sufficient to establish that Moore’s paradox is a more general phenomenon, but that is a fight I cannot have here.

330

woods

putting this problem to the side, we also need very contentious principles of rational belief in order to carry out this type of explanation. 2.1 Problem 1: Contentious Principles of Belief Consider, for example, that not only (1f), but also: (1f*) I swear to you that he’s a spy, but I don’t believe it. display Moore-paradoxical character. Our explanation has to somehow handle both. If we explain (1f) by appeal to the fact that swearing that p expresses my belief that p, then presumably we should claim that denying that we believe that p expresses our belief that we do not believe p. Presuming that believing that we do not believe that p rationally entails not believing p, we likewise obtain an inconsistency in our beliefs. We have now invoked a fair number of contentious principles of belief in running our explanation. We need principles of rational belief like: – B(¬p) ⇒r ¬B(p) – B(¬B(p)) ⇒r ¬B(p) – ⇒r ¬(B(p) ∧ ¬B(p))

in order to generate the relevant contradiction. But these are not obvious on their face: suppose, for instance, I believed q and thereby came to be required to believe not p on pain of irrationality. Does it follow, on pain of irrationality, that I need to refrain from believing p? Not obviously, especially if I am unaware of the fact that I’m rationally committed to believing not p. These principles are more plausible when taken to govern obvious rational requirements for explicit belief, but even there they are not obvious. Is it obvious that if I’m required to believe that I believe p that I am thereby obviously required to believe p? No; suppose I’m trying to recover my beliefs after a terrible accident. You, a reliable guide to my previous beliefs, testify that I did not believe p, but you chuckle as you say it. I ask why and you mention that even though I didn’t believe p, I believed that I believed p and that I defended both stances vehemently. Presumably, if I think I had good reason and I trust you, it’s plausible that I am rationally required (given the evidence) to believe that I believe p and, additionally, to not believe p—though here only on the presumed existence of evidence for this stance.15 It would be far better to have 15

This case was suggested to me by thinking over the example of amnesiac Gray Davis in (Atlas 2007: 131).

a commitment-theoretic account of moore’s paradox

331

an account of the problem which does not presume contentious principles of rational belief like the above. 2.2 Problem 2: Irrationality and Moore’s Paradox A more pressing worry, which I articulated in my earlier work, is that irrationality isn’t sufficient to offset Moore-paradoxical reactions (Woods 2014, objection 3). The utterer of a Moore-paradoxical utterance displays linguistic incompetence, not outright irrationality. We should distinguish these reactions accordingly. Compare, for example, the person who asserts one of: – I want a cookie, they’re in the jar, but there’s no reason to go to the jar – István is more likely a libertarian librarian than a librarian – Either Saniye or Sam went to the bank, Sam went if Saniye did, but Sam didn’t go They are explicitly displaying one or another form of irrationality, but it is hard to accuse them of linguistic incompetence. Rather, on the presumption that they mean the same by their words that we would, they fail to be fully rational. Or, alternatively, consider a direct case. Suppose I’m a known lunatic, likely to spurt this or that nonsense at any turn. Does that make my utterance of: I’m made of sugar or my utterance of Fictional characters can’t exist, yet here I am16 incoherent? I submit the answer is no. They are just resoundingly and obviously false. On the other hand, even if I am a known incoherent, my utterance of: I believe that I myself am made of sugar, but I’m not is resoundingly bizarre and feels self-contradictory.17 Why is this? Clearly, it has nothing to do with displaying the fact that I’m incoherent or irrational. Rather,

16 17

This example is plausibly an example of a structural irrationality, unlike the preceding, which is plausibly an example of substantive irrationality. I here adopt Atlas’s trick of putting in an explicit reflexive pronoun so as to force the right reading. See (Atlas 2007: 136–137).

332

woods

it has to do with some feature of these cases over and above the display of irrationality. It should not be surprising that we can find cases of exhibited irrationality that are not Moore paradoxical. Explicit contradictions can be felicitiously uttered, but presuming the connection of belief and assertion, such an utterance will commit the utterer to possessing inconsistent and thereby irrational beliefs.18 Atlas is sensitive to this point: A difficulty with Langford’s explanation is that there are a denumerable number of self-contradictory statements that are perfectly felicitous, so that the logical inconsistency of the total signification of a statement cannot be sufficient for its assertoric oddness. Atlas 2007: 142

and we should be as well. Moore’s paradox is interesting and worth investigating exactly because it is not simply a case of some fancy type of inconsistency or irrationality in our beliefs or the content of what we assert. It is also possible to cook up potential cases in the opposite direction; it is not obviously irrational, though it is obviously Moore paradoxical, to assert that the theory of evolution is correct, but that I don’t believe it (at least in the recherché sense of irrationality involving asserting against the total balance of evidence.) After all, I might have evidence for believing that the theory of evolution is false, but be unwilling to bring this evidence to bear on what I am willing to assert. If these considerations are right, and if irrationality simply doesn’t imply linguistic incompetence, then we need an account of Moore paradoxicality that doesn’t reduce it to exhibition of irrationality. That is, we need an account that respects Egalitarianism. 2.3 The Alternative Even if we could finesse the above problems with rationality and principles of rational belief, we would face another. uniformity already makes plausible the idea that we should look elsewhere than inconsistent beliefs for an explanation of Moore paradoxicality. To see why, consider (1e). Why is it Moore paradoxical to promise while denying that I’ll meet that promise? One explanation claims that when we promise, we express a belief that we will fulfill that promise and when we deny we’ll meet that promise, we express a belief

18

I don’t think all inconsistent beliefs are irrational, but even some manifestly irrational inconsistencies can be felicitiously uttered.

a commitment-theoretic account of moore’s paradox

333

that we won’t. However, even if promising to give you money, then denying I believe that I will gives rise to Moore paradoxicality, the case seems very different than promising to give you money, then explicitly denying that I will. And, since promising to give you money, then denying I intend to fulfill that promise likewise seems Moore paradoxical, we need to invoke the contentious claim that intentions (or lack thereof) require us to believe that we will (or won’t) satisfy them. That is, if we tried to explain the Moore paradoxicality of the last case on the inconsistency of belief model. Similar pressure can be found in reconsidering cases of manifest displays of irrationality which nevertheless do not manifest the sense of linguistic incompetence characteristic of Moore’s paradox. Moran (2001: 84) suggests that we find such cases on the therapist’s couch, where through a combination of external evidence and behavioral observation, a subject realizes that they really do believe such and so, but also affirm that they do not, reporting from “the inside”, so to speak. “My mother didn’t mistreat me, but I believe she did”, in such a context, is a display of a manifest psychological irrationality, but one which does not trigger the sort of feeling of linguistic incompetence we see in the above examples. So there seems to be a mismatch between manifesting psychological irrationality and the jarring phenomenon that Moore described.19 The cases here pile up. We can make some kind of coherent sense of booing a sports team, then later denying that I really believe that they’re terrible, whereas booing a sports team and then denying I dislike or disfavor them seems Moore paradoxical. Consider also that I can explicitly exhibit that I fail to meet a commitment by my actions without any direct commitment to believing that I’m failing to meet it. Suppose, for example, I promise to return your favorite shirt to you while simultaneously intentionally and explicitly setting it on fire. I haven’t asserted that I believed that I won’t satisfy my promise, though I have definitely displayed my lack of intention to do so. Even if we can manage to

19

Thanks to Derek Baker for helpful discussion. Moran nevertheless suggests that there is a problem with the pragmatic picture of Moore’s paradox since “… the prevalent diagnosis of Moore’s Paradox as the paradigm of a “pragmatic paradox” is inadequate at best, for the puzzling quality persists outside the context of utterance and speech act.” (70). However, his examples of this persistence involve internally tokening the Moore-paradoxical thought. This, of course, is an action that may very well bring along commitments all of its own, whether basic or derivative from commitments involved in ordinary utterance or assertion of the corresponding sentence (Baker and Woods, 2015, §4 explores a similar strategy to distinguish Moore-paradoxical incoherence from semantic incoherence). Since I have put aside Moore’s paradox for thought above (and since the pragmatic diagnosis is hardly prevalent), I won’t pursue this point further here.

334

woods

stuff these constructions under the general belief-oriented account, we do so by giving indirect explanations of their Moore-paradoxical character that neglect the character of the particular speech act involved. A better account which respected this difference is preferable. An alternative view is available: we need merely take speech acts, like swearing and promising, to generate commitments in some way and view overt declaration that we will or do violate these commitments as underlying the linguistic oddness of Moore’s paradox. Take our example from above of swearing to the truth of p. This speech act generates a commitment to the truth of p. We could call such a commitment a guarantee. When we deny that we meet it, we cause confusion in the interpreter and thereby the feeling of contradiction. There is no contradiction, of course, just inconsistency between what we have committed to and what we have said is the case. How, that is, could I guarantee the truth of something I am willing to assert is false? This style of explanation promises to vindicate both uniformity and egalitarianism. We might worry, though, that it generates too much Moore paradoxicality and violates extensional adequacy. 2.4 Too Much Moore Paradoxicality? Adopting some version of this account, we will find Moore paradoxicality all over the place.20 Consider, for example, the speech act of promising. A typical promise plausibly commits me to (a) keeping the promise, (b) intending to keep the promise, and (c) believing that I can keep the promise. Correspondingly, we should expect that: (3e) I promise to pick you up, but I won’t. (4e) I promise to pick you up, but I don’t intend to pick you up. (5e) I promise to pick you up, but I don’t think I can. feel Moore paradoxical. Which, I submit, they do. Of course, they all feel slightly different in terms of strength and the conditions under which we can felicitously utter each differ, but this is as we should expect. Commitments come

20

Although, as I argued in (Woods 2014), we do not find it where the expressivist predicts we should. See (Toppinen 2014, 2015) and (Woods 2015) for discussion and (Raskoff ms) for further worries. Baker (2014, 2015) argues that we also do not find it where guise-of-thegood theorists predict we should.

a commitment-theoretic account of moore’s paradox

335

in different strengths. This is, in fact, a virtue of the account; it explains in a natural way why: (1a) It’s raining, but I don’t believe it is. feels more Moore paradoxical than: (6a) It’s raining, but I don’t have reason to believe that. and: (7a) It’s raining, but the evidence (on balance) suggests otherwise. hardly feels Moore paradoxical at all. We are plausibly committed to having some reason for asserting what we believe, we are strongly committed to believing what we assert, but we are not strongly committed to asserting in line with the balance of evidence. Put a slightly different way, sincerely asserting either of (1a) or (6a) is evidence of linguistic incompetence, whereas sincerely asserting (7a) is merely evidence of epistemic incompetence. The explanation for this is that in (7a) we do not overtly deny in the right conjunct commitments incurred by the left conjunct. Whereas, in (1a) and (6a), we do. Of course, there may be more cases of overspill. Consider a case in which I promise to do something that everyone knows that I cannot. Or, alternatively, a case in which I promise to do something, the promising of which entails (in light of other background commitments), that I cannot keep it. Are these cases of Moore’s paradox? Presumably not; a reasonable account of Moore’s paradox should explain this fact. Below, after I discuss in significantly more detail the notion of commitment, I will turn to such cases and show that the addition of a constraint of normal interpretation—that the speaker reflexively intends their utterance to be taken in a certain way—rules out these sorts of cases and allows a more plausible commitment-theoretic account of Moore’s paradox.

3

Commitments

The explanation I want to offer makes crucial use of the notion of commitment.21 What are these commitments though? The rough outline of the story

21

This focus on commitments isn’t entirely novel to me. Commitments have been used

336

woods

I want to tell claims that commitments are the product of a system of formal norms governing our speech behavior. Commitments are not to be reduced to the mental state of feeling committed or some form of intention; rather, they are social facts about what we ought to do given what we have done. I take on a commitment by performing certain actions—those actions the system of norms marks as conditions for being committed. For example, when I promise, I commit myself to satisfying that promise, whether or not I intend to do so or feel like I must. Likewise, when I swear to the truth of something, I am criticizable if it is false, whether or not I believed that it was true when I so swore: I take on a commitment to that thing being true. We may perform these actions knowing full well that in doing so, we take on a commitment. We may perform these actions without knowing any such thing. Sometimes ignorance of the law is no defense, while sometimes ignorance of the law is a full defense; likewise with social facts. Where, though, do these commitments come from? Drawing on my recent work (Woods 2016, forthcoming), I take commitments to be the product of implicit social rules, much like I take legal, moral, and promissory norms to be the product of implicit social rules. These social rules give rise to notions of blame and criticizability which underwrite, in many cases, normative judgments of people who violate them. They also give rise to constitutive rules for analyzing and interpreting the speech of others, analogous to how the social rules governing a game of chess or bar pool give rise to constitutive rules for interpreting the outcome of a game of chess or bar pool. Importantly, in voluntarily taking on a commitment, I bind myself to attempting to satisfy it, at least typically, and can be fairly interpreted as taking that commitment as a constraint on my action. Of course, I can be committed in ways which don’t guide my action and not all commitments are explicitly recognized, but the (interpreted as) voluntary undertaking of a commitment will be important here and below. When I am under a commitment, it will typically be explained by both the system of social rules giving rise to norms and the performance of the action which put me under the commitment.22

22

for analyzing illocutionary acts by Searle (1969), Brandom (1983), Wright (1992), Alston (2000), MacFarlane (2005), Harnish (2005), and others. The focus on using them to explain Moore-paradoxical behavior is relatively novel, being invoked explicitly only by Baldwin (2007) to my knowledge. Baldwin, though, focuses on our undertaking a commitment—note the psychological character here—to the truth of what we judge. My analysis is less mentally oriented and broader, as I countenance commitments to lots of things besides the truth of what we judge or assert. Some commitments, of course, are standing and don’t require any action other than par-

a commitment-theoretic account of moore’s paradox

337

We will start by looking at one of the more common ways of proceeding here, the notion of a sincerity condition, and show that it is inapt as a full analysis of the relevant notion of commitment. I will then turn to a more general picture on which sincerity conditions form a special subclass of the commitments of particular speech acts. It turns out that even commitments are insufficient to generate Moore paradoxicality since we can incur commitments by means of speech acts where we are explicitly incapable of meeting the commitment. I then turn to adapting a proposal of Bach and Harnish (1979) to assuage this problem. 3.1 Sincerity Conditions Consider the paradigm case of a speech-act engendered commitment—the connection of assertion with belief. When we make an assertion, absent special conditions, those who hear us presume us sincere. This means, in particular, that they take us to believe what we assert. If I say “this paper is late”, then unless I’ve given some signal that I’m not to be taken seriously—an upraised eyebrow, a bizarre intonation contour, etc—I am presumed to believe that this paper is late. If I do not believe this, and this comes out, then I am criticizable for not speaking sincerely.23 If I don’t believe this paper is late, and it doesn’t come out that I don’t, I am nevertheless fairly interpreted as believing this paper is late. This connection between belief and assertion is often called a sincerity condition. That is, we take it that what it is to sincerely assert that my paper is late constitutively involves me believing that my paper is late. Sincerity conditions apply to speech acts other than assertions. Plausibly, it is insincere for me to promise that I will give you 5 Turkish lira if I have no intention of doing so and it is likewise insincere for me to request that you get me a beer if I have no desire that you do so.24 Many theorists have taken the expression of a sincerity condition to be fundamental to the explanation of Moore-paradoxical phenomena.

23

24

ticipation in a particular social context in order to be in force. See (Woods 2016, forthcoming) for more detail, especially of how to get reasons to act out of commitments. Strictly speaking, I am criticism liable, not yet criticism worthy. It may be that criticism is misplaced, depending on the conditions under which I failed to believe what I asserted. Nevertheless, someone is licensed to criticize me (even if it would be a moral error for them to do so) merely by the fact that I have violated a norm of correct assertion. Compare the legal notions of being prosecution liable and prosecution worthy. See Condoravdi and Lauer (2012), § 2.2 for defense of this claim and that the resulting request is typically incoherent.

338

woods

… when one performs the speech act one necessarily expresses the sincerity condition and thus to conjoin the performance of the speech act with the denial of the sincerity condition would be to express and deny the presence of one and the same psychological state. Searle and Vanderveken 1985: 19

However, contra these theorists, this does not mean that we fail to commit ourselves in some way when we explicitly insincerely promise or request. According to many promissory conventions, if I promise you to pay 5 Turkish lira, as I pull out my empty pockets, I have still promised. And, likewise, if I request you get me a beer, while telling my friend that I really don’t want you to do it, I have still requested it. If I am in an appropriate position of authority, I can still hold you responsible for failing to heed my request when you don’t, even if I never wanted you to satisfy it in the first place. For some plausible constraints on commanding, requesting, promising, and the like, insincerity is no excuse. The case of assertion and sincerity is not entirely representative of the class of commitments incurred by a particular illocutionary act. This is because we capture nearly all of the commitment of assertion when we talk in terms of sincerity conditions. Since the purpose of assertion is to convey information about what I believe (about the world) and, typically, to get you to believe it as well, when I am insincere, I typically contravene the usual purpose of my asserting in the first place.25 The close connection between sincerity and commitment also holds for some other illocutionary acts, like that of apologizing. An insincere apology is no apology at all since, presumably, the purpose of an apology is to make manifest my feelings of regret. No feelings of regret, no apology. But it is not true for some other common illocutionary acts, like promising, swearing, requesting, demanding and such like. My demand may be insincere, but that doesn’t make it fail as a demand—it just makes me a bit of a jerk. And some illocutionary acts do not have sincerity conditions at all—there is no particular mental state required to back up naming one’s new cottage Dunroamin.26

25

26

Note that this is not to say that we can explain what an assertion is merely in terms of this purpose. That is a further step I am not here endorsing. See (MacFarlane 2011) for discussion. See also (Lackey 2007) for useful examples where we can successfully assert while explicitly violating the sincerity condition. This is relevant to §4 below. Or, anyways, no mental state other than the general knowledge of what I am doing. This, of course, is required for nearly all actions, illocutionary or otherwise, for which we can be held responsible.

a commitment-theoretic account of moore’s paradox

339

So sincerity conditions are inadequate as a full analysis of the relevant notion of commitment. What we need is a more general picture of commitment on which the commitments of illocutionary acts can spill past what is required for an illocution to be sincere. 3.2 Commitments in General Sincerity conditions give rise to a particular class of the commitments. Plausibly, they play the role of constitutive rules specifying what it is to sincerely ϕ, for ϕ a particular speech act we can engage in (Searle 1969). A sincere promise is one we intend to keep. A sincere assertion is one we believe. A sincere declaration of intention is one where we so-intend. And so on. Not all commitments are connected to sincerity, however. An insincere promise is still a promise, as any 12 year old knows. Likewise an insincere assertion counts as an assertion, as any politician skewered on the Daily Show knows all too well. Sincerity is nice, but it isn’t everything. Why do sincerity conditions give rise to commitments at all? Why sincerely promise? Because there are more than constitutive rules in play in a particular conventional enterprise like the practice of promising. There are also regulative rules governing the various practices in play (Searle 1969). Regulative rules, like drive on the right, pay your parking tickets, or don’t wear white shoes before Easter, govern existing behavior, (formally) obliging us to act thus and so, at least to those practices we engage in. The reason that sincerity conditions give rise to commitments is that the practices under consideration are governed by the regulative rule: be sincere!. Since we have a (formal) obligation to be sincere when we assert, we have an obligation to believe what we assert. The reason for the parenthetical ‘(formal)’ in the above is that there is often taken to be a distinction between merely formal obligations, like the obligation to be polite, and obligation tout court, like the obligation to never stab someone just for fun. Merely formal obligations are those which we have to the degree to which we care to engage in a particular conventional activity like that of polite behavior.27 Thankfully, we can bypass this murky dispute in the cases under consideration by noting that the ability to opt out of merely formal obligations is typically muted in cases of illocutionary acts. I can choose not to assert and not to promise, but I cannot simply opt out of the obligations associated with

27

I am speaking here in the voice of the many; I personally believe that etiquette and like “merely formal obligations” are often given undeserved short shrift. See (Woods 2016, forthcoming).

340

woods

asserting and promising by choosing not to care about them. It’s just not that sort of game. This means that the sanctions which are associated with violating my obligations in the cases of asserting and promising cannot be demurred by saying that I don’t care. I’m an involuntary participant in these conventions merely by my participation in the acts of promising and asserting. This is similar to the cases of games in general. I can choose not to play bar pool, but insofar as I play bar pool, I am obliged by the rules tacked on the bar wall. The fact that I don’t care about these rules does not excuse me from whatever sanctions are tied to violation of them (typically loss of the game and general disapprobation of my fellow players. And, at least in Boston, getting my name rubbed off the chalkboard.) 3.3 Commitments and Explaining Moore-Paradoxical Utterances Sincerity conditions are insufficient for explaining Moore paradoxicality since cases like swearing to the truth of something, commissives like promising, declaratives like “you’re fired” and the like can engender Moore paradoxicality without violating sincerity conditions. The conditions required for the successful carrying out of a swearing, a promising, and a declaration have to do with the relevant authority to carry out such an action and not with the possession of a particular mental state. As I’ve said a number of times now, an insincere promise is still a promise and declaring we won’t fulfill it is different from declaring that we don’t intend to fulfill it. Yet both are Moore paradoxical. Put bluntly, the point is that the sincerity condition of a promise is the possession of an intention, not a performance. Yet we cannot adjoin a declaration of non-performance to a promise without Moore paradoxicality, regardless of what mental state we happen to be in. If sincerity conditions won’t suffice, perhaps the general category of commitments will do. Unfortunately, we here face a problem with extensional adequacy. There are cases in which we explicitly and overtly violate our commitments but where no Moore paradoxicality occurs. Consider, for example, a case in which I promise to give you the original deed to my house. I am absentmindedly lighting a cigarette and do so off of a scrap of paper in my pocket I light off of a campfire. You, hearing my promise, watch as I simultaneously light my cigarette off of the original deed to my house. Have I committed myself to giving you the original deed to my house? Yes. Have I displayed, overtly and explicitly, my failure to fulfill this commitment? Yes. Has my assertion given rise to a case of Moore’s paradox? No. Compare a case far closer to a Moore-paradoxical utterance. We’re talking about the deed to my house and it is right there in front of us. I simultaneously set it on fire and

a commitment-theoretic account of moore’s paradox

341

promise to give it to you. In such a case, it seems that I have also explicitly failed to fulfill the commitment I’m making, but in such a way as to actually give rise to the Moore-paradoxical feeling. After all, I’m putting right in your face that I am knowingly destroying my ability to fulfill my commitment. If these behavioral cases don’t resonate, we can get a similar effect by means of case of perjury or insincerity. Suppose I—a prominent mafioso—swear to tell the truth, the whole truth, and nothing but the truth. I then swear that I did not murder Tip on the night of July 17th. Immediately released from my oath, I mutter “of course I did, but what do you expect me to say?” Or, alternatively, suppose in the course of a relationship my partner and I are debating about whether we will always care for each other. I, knowing myself, say that I’m almost certain to have lost interest in 5 or 6 years. My partner, desperate for reassurance, begs me to promise him it won’t happen. I finally cave and say: “Okay, okay. I promise to love you forever … of course, you and I both know that won’t happen.” In such a case, it seems clear to me that I have still committed myself to loving him forever; it’s just that I will not do so.28 Recapping briefly, we have discussed both a general notion of commitment and the special case of commitments which arise from sincerity conditions. We then saw that the commitments arising from sincerity conditions do not suffice as an explanation since (a) some cases do not involve sincerity conditions at all, and (b) we can evince our lack of sincerity without giving rise to Moore paradoxicality. Broadening our commitments resolves (a), but it does little to resolve (b) since we can still find cases of Moore paradoxicality even when we have explicitly violated our commitments. Something else is needed for commitments to explain our target phenomenon.

4

Commitment-Theory Done Right

If commitment and explicit demurral thereof is insufficient to explain Moore’s paradox, then what needs to be added? We find the answer to this in a common thought about illocutionary actions, articulated nicely in principle form by Bach and Harnish: Communicative Presumption The mutual belief in the linguistic community CL to the effect that whenever a member S says something

28

The reader is invited to guess at how this example would actually play out in practice.

342

woods

to another member H, S is doing so with some recognizable illocutionary intent. Bach and Harnish, 1979: 60–61

Bach and Harnish take this presumption, held at least implicitly by all competent speakers and interpreters, to govern successful interpretation of the speech acts of others. Their focus is on communicative acts and, in particular, on the expression of sincerity conditions in carrying out these communicative acts. They take expression of a sincerity condition to be governed by the scheme: For S to express an attitude is for S to R-intend the hearer to take S’s utterance as reason to think S has that attitude. Bach and Harnish, 1979: 15

and to R-intend (reflexively intend) something is to intend that the intention to do it be recognized (typically by the interpreter.) Putting these together, when someone says something with recognizable illocutionary—here expressive— intent, we take them to intend that our recognition of their intention gives us reasons to believe that they believe what they say; we are thereby generally entitled to attribute to them belief in what they have asserted. This picture is geared towards a communicative picture of illocutionary acts. Given our focus on cases where the illocutionary act, such as a promise, gives rise to a commitment to act, not merely to intend to act, we need to look elsewhere than expression to explain how to get Moore’s paradox out of commitments. (See also §3.1 above.) Luckily, we don’t have to look far. Expression is one thing that we can do with illocutionary intent. Another is to take on commitments incurred by participating in an illocutionary practice. We can lay this down by means of a schema, as above: For S to commit themselves by means of an utterance is for S to R-intend the hearer to take S’s utterance as reason to think S is voluntarily engaging in the practice and thereby taking on the commitments engendered thereby. Putting this together with Communicative Presupposition, we obtain that when someone says something with recognizable illocutionary intent, we take them to intend that our recognition of their intention gives us reasons to think S is voluntarily engaging in that practice; we are thereby generally entitled to treat them as recognizing that they are thus and so committed and acting accordingly.

a commitment-theoretic account of moore’s paradox

343

If we tie this into the above discussion of commitments, we can see that treating an utterer as recognizing that they are thus and so committed entitles us to treat them as being bound and acting in accords with their commitments. At least those that are immediately recognizable. We take those who recognize their commitments to also intend to satisfy those commitments any linguistically competent ϕ-er, where ϕ is a speech act like asserting, will immediately recognize.29 When they go on to explicitly deny that they will meet these commitments, by denying they’ll satisfy their commitments, we obtain the sort of self-undermining that characterizes Moore paradoxicality. If they recognize that their voluntary engagement with this practice commits them and requires taking these commitments as guiding their action, why would they deny that they’ll satisfy these commitments? Let’s work though a particular case. Suppose you promise to give me 5 Turkish lira, but then immediately deny that you will do so. By Communicative Presumption, I take your utterance to be knowably intended to be a promise. I also take promising to be a way of committing yourself to keeping the content of the promise. Since you are committing yourself, I take you to intend me to recognize your intention to subject yourself to the commitment to keep your promise. But, given that I recognize this and recognize that you want me to recognize it, I am completely puzzled as to why you would immediately go on to disabuse me of my recognition of your intention to submit yourself to the commitment to keep your promise. Crash. Of course, this does not mean that someone cannot insincerely engage in a practice or, after so engaging, lose their interest or ability to satisfy their intention to keep their commitments. Of course they can. This explains why the third-personal, past-tense, and otherwise non-first-personal cases of Moore’s paradox aren’t paradoxical. There are lots of reasons for taking someone to no longer intend to satisfy their commitments. What’s important here is that the first-personal acts involving committing oneself, like the first-personal acts involving expressing, gives rise to a presumption that we then intend to keep our commitments. Frustration of this presumption causes the paradoxical feeling. When this presumption is not in force, such as when there is a contextual reason to interpret our illocutionary action non-standardly, as in the cases

29

This hedge is necessary to explain why far-out violations of commitments can be coherently tacked onto something like a promise. I might just not recognize that I cannot both have lunch with you and pick up Bob at Atatürk airport within 5 hours. Further discussion of this obviously and easily solvable problem would distract, not illuminate.

344

woods

described in the last section, then we do not have any paradoxical feeling. This is exactly the right result.30 Essentially, this theory ties together a commitment-theoretic picture of (some) illocutionary acts31 with a Bach/Harnishian account of how to interpret speech-acts. This allows us to retain the idea that many illocutionary acts commit us to things other than the possession of mental states, but to also explain why, in typical occurrences of illocutionary acts, we are permitted to attribute to the utterer a particular sort of intention: an intention for us to recognize their intention to subjugate themselves to the norms of the (conventions governing) particular illocutionary acts they have engaged in. This view has the advantages of both the commitment view—it is sufficiently divorced from the theory of verbal communication so as to allow the cases of behavioral Moore’s paradox like the one above involving flagrantly lighting the deed to my house on fire—while retaining the advantageous non-mysteriousness of the Bach/Harnish communicative presupposition account. It allows us to satisfy uniformity as the picture captures all cases of Moore paradoxicality, egalitarianism since the main assumption is that competent speakers and interpreters presume illocutionary intent, and has at least the potential to satisfy extensional adequacy as we can explain why not all commitments incurred by speech acts give rise to Moore’s paradox. 4.1 External Conventions, Attributed Intentions One worry with the above account is that it presumes that speakers commit themselves by actually intending to carry out their illocutionary act. This may strike some—it strikes me as possibly—psychologically unrealistic. It also seems to neglect the fact that we engage in speech acts without thinking through what we are accomplishing by means of them. We all know on reflection that by carrying out various illocutionary acts we commit ourselves thus and so. However, it is nowise clear that this is clear to us, explicitly, at the time that we are actually carrying them out. If this worries you, I will note that I am torn between the above account of commitment and more externalist alternatives on which:

30

31

We can resolve other problems for utterances with Moore-paradoxical form, like those in (Atlas 2007), by noting that context can suggest that we interpret reflexive pronouns, like ‘I’, in non-standard ways in special contexts like those of an amnesiac going over “their” past with a therapist. Compare with the commitment-theoretic picture of assertions discussed in (MacFarlane 2011).

a commitment-theoretic account of moore’s paradox

345

For S to commit themselves by means of an utterance is for S to be fairly attributed the R-intention for the hearer to take S’s utterance as reason to think S is voluntarily engaging in the practice and thereby taking on the commitments engendered thereby. or the combination of the even more externalist schema: For S to commit themselves by means of an utterance is for S’s utterance to subsume them under a norm governing that type of utterance. and the principle: Whenever a member S says U to another member H, absent special conditions, S should be interpreted as R-intentionally committing themselves by means of U. Any of these views would give a recipe for satisfying uniformity, egalitarianism, and potentially extensional adequacy. Which is actually correct is a matter for another time and another paper. Note also that I have spent fairly little time talking about what norms actually govern various illocutionary acts. In the case of assertion, this has been extensively discussed (see (Weiner 2007) for a useful overview) and in the general case of speech acts, a reader can find useful accounts in both (Searle and Vanderveken 1985) and (Bach and Harnish 1979). Again, little turns on this for my point, so long as my above assumptions about particular speech acts and their commitments are close enough. All I need is that some speech acts commit us to things other than possessing various mental states in order to motivate my general picture; which things we are actually committed to is mostly besides the point for this.

5

Conclusion

I have argued that in order to accommodate the wide variety of Moore paradoxical phenomena, we need to abandon the idea that we can explain Moore paradoxicality in speech by means of Moore paradoxicality in belief. I went on to argue that we could not use irrationality in speech or insincerity to run this sort of explanation; rather, we need to move to a commitment-theoretic account of Moore paradoxicality. I then sketched the details of this view, arguing that we needed a presumption about the intentions of speakers (adopted, in modified form, from Bach and Harnish) to separate the commitments incurred

346

woods

by speech acts which gave rise to Moore paradoxical reactions and those which did not. The resulting picture vindicates uniformity, egalitarianism, and potentially vindicates extensional adequacy. (subject to a suitable filling in of the gaps.) I will close on a final advantage of this kind of view. I suggested above— and in my (2014)—that the known irrational can still make promises, declarations, and assertions in ways that give rise to Moore paradoxicality. Suppose we make the presumption, suggested above, that when someone typically says something to someone else, they are doing so with illocutionary intent. If we presume this for ordinary folk, presumably we should also presume it of the irrational as well unless we know that their irrationality is manifesting in lack of ability to successfully communicate. After all, irrationality is not an all or nothing affair and even the raving mad can sometimes successfully request a glass of water. If we presume this sort of interpretive charity in assessing others, then it seems that we should sometimes be startled when even someone known to be irrational deliberately undercuts themselves in requesting, then denying that they want something. We can back-solve to see that they weren’t really voluntarily engaging in an illocutionary act (or, alternatively, deliberately flouting their commitments), of course. But no one, to my knowledge, has ever successfully argued that Moore-paradoxical utterances can’t be later explained. It is their current bizarreness that is so remarkable.

References Alston, W. 2000. Illocutionary Acts and Sentence Meaning. Ithaca and London: Cornell University Press. Atlas, J.D. 2007. What Reflexive Pronouns Tell Us About Belief: A New Moore’s Paradox de Se, Rationality, and Privileged Access. In Moore’s Paradox: New Essays on Belief, Rationality, and the First Person, eds. Mitchell S. Green and John N. Williams, 117–145. New York: Oxford University Press. Bach, K. and R. Harnish. 1979. Linguistic Communication and Speech Acts. Cambridge, Mass. and London: MIT Press. Baker, D. 2014. The Abductive Case for Humeanism Over Quasi-Perceptual Theories of Desire. Journal of Ethics and Social Philosophy 8 (2): 1–29. Baker, D. 2015. Akrasia and the Problem of the Unity of Reason. Ratio 28 (1): 65–80. Baker, D. and J. Woods. 2015. How Expressivists Can and Should Explain Inconsistency. Ethics 125 (2): 391–424. Baldwin, T. 2007. The Normative Character of Belief. In Moore’s Paradox: New Essays on Belief, Rationality, and the First Person, eds. Mitchell S. Green and John N. Williams, 76–89. New York: Oxford University Press.

a commitment-theoretic account of moore’s paradox

347

Brandom, R. 1983. Asserting. Noûs 17 (4): 637–650. Brandom, R. 1994. Making It Explicit: Reasoning, Representing, and Discursive Commitment. Cambridge, Ma: Harvard University Press. Condoravdi, R. and S. Lauer 2012. Imperatives: Meaning and Illocutionary Force. In Empirical Issues in Syntax and Semantics, ed. C. Piñon. 9: 37–58. Harnish, R.M. 2005. Commitments and Speech Acts. Philosophica 75. Heal, J. 1976. Insincerity and Commands. Proceedings of the Aristotelian Society 77: 183– 201. Lackey, J. 2007. Norms of Assertion. Noûs 41 (4): 594–626. MacFarlane, J. 2005. Making Sense of Relative Truth. Proceedings of the Aristotelian Society 105 (3): 321–339. MacFarlane, J. 2011. What Is Assertion? In Assertion, eds. J. Brown and H. Cappelen. OUP Oxford. Michaelson, E. 2016. The Lying Test. Mind and Language 31 (4): 470–499. Moore, G. 2013. G.E. Moore: Selected Writings. London: Routledge. Moore, G.E. 1942. A Reply to My Critics. In The Philosophy of G.E. Moore, ed. P.A. Schilpp. La Salle, IL: Open Court. Moran, R.A. 2001. Authority and Estrangement: An Essay on Self-Knowledge. Princeton: Princeton University Press. Raskoff., S. manuscript. Getting Expressivism out of the Woods. Searle, J. 1969. Speech Acts: An Essay in the Philosophy of Language. Cambridge: Cambridge University Press. Searle, J. and D. Vanderveken. 1985. Foundations of Illocutionary Logic. Cambridge: Cambridge University Press. Shoemaker, S. 1995. Moore’s Paradox and Self-Knowledge. Philosophical Studies 77 (2): 211–228. Sorensen, R.A. 1988. Blindspots. Oxford: Clarendon Press. Stainton, R.J. 2016. Full-On Stating. Mind and Language 31 (4): 395–413. Toppinen, T. 2014. Expressivism and Moore’s Paradox: A Reply to Woods. Journal of Ethics and Social Philosophy: 1–5. Toppinen, T. 2015. Relational Expressivism and Moore’s Paradox. Journal of Ethics and Social Philosophy: 1–7. Weiner, M. 2007. Norms of Assertion. Philosophy Compass 2 (2): 187–195. Williams, J.N. 2012. Moore-Paradoxical Assertion, Fully Conscious Belief and the Transparency of Belief. Acta Analytica 27 (1): 9–12. Williams, J.N. 2013. Moore’s Paradox and the Priority of Belief Thesis. Philosophical Studies 165 (3): 1117–1138. Williams, J.N. 2015. Moore’s Paradox in Speech: A Critical Survey. Philosophy Compass 10 (1): 10–23. Woods, J. 2014. Expressivism and Moore’s Paradox. Philosophers’ Imprint 14 (5): 1–12.

348

woods

Woods, J. 2015. Expressivism Worth the Name: A Reply to Teemu Toppinen. Journal of Ethics and Social Philosophy: 1–6. Woods, J. 2016. The Normative Force of Promising. Oxford Studies in Normative Ethics Vol. 6: 77–101. Woods, J. forthcoming. The Authority of Formality. Oxford Studies in Metaethics Vol. 13. Wright, C. 1992. Truth and Objectivity. Cambridge, Ma: Harvard University Press.

chapter 12

Remarks on Davidson’s Polymorphous Concept of Truth and Its Role in a Theory of Meaning* Ken Turner

I am interested in Davidson exegesis. Atlas, 1988: 164

… … we meet many different conceptions of truth … and we must indicate which conception will be the basis of our discussions. Tarski, 1944: 342

… … the concept of truth must have a connection with how language is used. Davidson, 1993a: 36

… … it seems likely that the content of the word ‘true’ is sui generis and indefinable. Frege, 1984: 353

∵ * Or: ‘The work of a philosopher consists in assembling reminders for a particular purpose’ (Wittgenstein 1958: paragraph 127): Davidson on truth and meaning. I am grateful to Paul Dekker, Christopher Gauker, Larry Horn, Jaroslav Peregrin and Josef Stern for helpful comments on a previous version of this paper; to Ernie Lepore for sending me one of Davidson’s more obscurely published papers; to Michiel Leezenberg for several discussions that helped me enormously and to Jay Atlas who, in a packed lecture theatre over the course of a week in Leuven, during the Second European Summer School in Logic, Language and Information, showed, to me for the first time, how it is possible to reason with perspicuity and orderliness about the semantics/pragmatics interface. (The content of these lectures appears, suitably revised and extended, in Atlas (2005).) © koninklijke brill nv, leiden, 2018 | doi: 10.1163/9789004365445_013

350 1

turner

First remarks

Davidson’s, and more generally, Davidsonian, philosophy of language consists of two, perhaps, three, research programmes that are still occasionally active, in some locations, but not infrequently dormant, in others. The first—the truthmeaning programme (aka the Tarski-style programme; henceforth, Research Programme 1 (RP1))—examines the following hypothesis: ‘truth theories can do duty as theories of meaning’ (Davidson 1984: xiv). The second—the truthinterpretation programme (aka Radical Interpretation; henceforth, Research Programme 2 (RP2))—examines this next hypothesis: ‘the attitude of holding true, directed to sentences under specified circumstances, is the basis for interpretation’ (Davidson 1975: 20). The third—the Primitivist programme (aka, possibly, Relationalism; henceforth, Research Programme 3 (RP3))—attempts to ‘trace the connections between the concept of truth and the human attitudes and acts that give it body’ (Davidson 1996: 276); ‘I believe that what I call the content of the concept of truth can be brought out only by relating it to speech, belief and the evaluative attitudes’ (Davidson 1999i: 18). The precise character of each of these programmes, and the precise nature of the relation, or relations, between them, have always been, and continue to remain, a matter of some discussion, with rather little in the way of exegetical and critical consensus. It would be disappointing if it was still possible to say, at some future point, that, although he is ‘one of the most systematic philosophers in the twentieth century’ (Peréz 2005: 177), ‘Donald Davidson is, notoriously, a philosopher who has appeared as many things to many people’ (Evnine 2007: 121).1 1 These difficulties were noticed early on. Sterelny (1981: 95) remarks that ‘Davidson exegesis is fraught with peril’ and Bennett (1985: 626), in the concluding sentence of his critical notice of Inquiries into Truth and Interpretation, says that ‘[t]his book does not easily yield its content to the reader’. More recently, of an analysis, in Davidson (2005), of the relationship between realism and truth, Burge (2007: 582) says: ‘Davidson somewhat obscures matters in his formulations … Much of the discussion has a coy, nonstraightforward feel to it’. And, of Davidson’s historical account of, and proposed solution to, the problem of predication (aka the unity of the proposition) Burge (2007: 589) adds: ‘Davidson’s discussions … are hampered by sloppy formulations that obscure premises that lead to regresses’. Such are the exegetical and critical challenges that the otherwise magnificent contributions of Lepore and Ludwig (2005, 2007a, 2013) have not, even with a pinch of hindsight, more than modestly improved matters. Cf. Stoutland 2006, 2007; Evnine 2007; Rawling 2007; Amoretti and Vassallo 2008; Grigoriev 2008; Soames 2008; Goldberg 2008, 2009, 2012; Solberg 2009; Smith 2010; Ebbs 2012; Ludwig 2007, 2015; Amoretti and Preyer 2011; Lepore and Ludwig 2007b, 2011; Malpas 2011; Montminy 2011; Glüer 2011; Puolakka 2006, 2011; Bourke 2012/13; Sainsbury 2012; Baghramian 2013; Fallis 2013; Heather and Segal

remarks on davidson’s polymorphous concept of truth

351

1.1 Some Orientation RP1 is introduced and lightly sketched in Davidson (1965, 1967a) and briefly elaborated in Davidson (1968, 1969, 1970, 1973a, 1977). It is revised, in one direction, in Davidson (1986c) and further revised, in another direction, in Davidson (1990a, 1996, cf. 2005). Fragments of RP2 are distributed throughout much of Davidson’s later work, beginning with Davidson (1973b) and continuing with Davidson (1974a, 1975, 1976, 1985a, 1994a, 1994b). There are some later additions and extensions to RP2 that have to do with metaphysical and epistemological consequences of the proposed story about interpretation and not all of these hypotheses are immediately and noticeably coherent. I shall dwell upon some of these in due course. RP3 appears in Davidson (1996) and is assumed in subsequent work. Its slogan (and accompanying rhetorical question) may be taken to be: ‘truth is as clear and basic a concept as we have … Why on earth should we expect to be able to reduce truth to something clearer or more fundamental?’ (Davidson 1990b: 135, 136; cf. Davidson 1990a: 314; 1996: 264–265). The numerical ordering of these research programmes reflects their temporal ordering: RP1 came first; RP2 came second; and RP3 third. Davidson (1976: 33) says, with reference to Davidson (1973b): In a paper first read in Biel, Switzerland, in May 1973, I criticized my own earlier attempts to say exactly what the relation is between a theory of truth and a theory of meaning, and I tried to do better. Emphasis added

And, in reply to Fodor and Lepore (1994a) he adds: I wanted answers to such questions as “what is meaning?”, and became frustrated by the fatuity of the attempts at answers I found in Ogden and Richards, Charles Morris, Skinner and others. So I substituted another question which I thought might be less intractable: What would it suffice an interpreter to know in order to understand the speaker of another language, and how could he come to know it?2 Davidson 1994a: 126. Emphasis added. Cf. Davidson 1999f: 158; Lepore 2004: 258

2013; Hoeltje 2013, 2016; Isaac 2013; Ray 2014; Wheeler 2014, Fennell 2015; Dresner 2006, 2009, 2015; de Rouilhan 2015; Hunter 2015; Armstrong 2016a, b; Bar-On 2016; Begby 2016; Camp 2016; Myers and Verheggen 2016. 2 Davidson is fond of such ‘afterthoughts’ (cf. Davidson 1990b) and there will be more such examples in what follows.

352

turner

Note also the small but perhaps not insignificant matter of the reprint of Davidson (1967a) in Davidson (1984) which refers, on page 27, to ‘radical interpretation’, whilst the original, on page 313, makes reference to the quite distinct notion of radical translation. (See Davidson 1973b: 328fn14; 1974a: 317–318; 1999c: 729–730; 2005: 62–63. Cf. Pearson 2011; Glock 2014.) RP2 was not, initially, a seed in RP1. Davidson concedes as much when he says, with reference to Davidson (1967a): ‘My mistake was to think we could both take a Tarski truth definition as telling us all we need to know about truth and use the definition to describe an actual language. But even in the same essay I (inconsistently) discussed how to tell that such a definition applied to language. I soon recognised the error’ (Davidson 2005: 16fn22). ‘And …’, he might have added, ‘… resorted to radical interpretation as the solution’. To repeat: RP1, RP2 and RP3 are, initially at least, distinct research programmes.

2

Refinements

2.1 RP1 And now for a small refinement. RP1 doesn’t just explore the hypothesis that ‘truth-theories can do duty as theories of meaning’. It examines the more specialist hypothesis that ‘truth-theories, but not correspondence theories, can do duty as theories of meaning’. Call this hypothesis RP1*. The refinement is a result of Davidson’s arguments against ‘facts’ and ‘correspondence’. He says: Correspondence theories rest on what appears to be an ineluctable if simple idea, but they have not done well under examination. The chief difficulty is in finding a notion of fact that explains anything, that does not lapse, when spelled out, into the trivial or the empty. Davidson 1969: 748

I do not think we understand the … claim that there is something in or about the world that makes our thoughts or assertions true when they are true. The fact is that no one has ever been able to say in a nontrivial way what sort of “thing” it is that makes a sentence (or other truth bearer) true. Some sentences or utterances or beliefs are true, and there are a lot of things in the world, but it explains nothing to say that the things “make” the truth vehicles true. Davidson 1997a: 110

remarks on davidson’s polymorphous concept of truth

353

The notion of correspondence would be a help if we were able to say, in an instructive way, which fact or slice of reality it is that makes a particular sentence true. No-one has succeeded in doing this. … there are no interesting and appropriate entities available which, by being somehow related to sentences, can explain why the true ones are true and the others not. Davidson 2000: 66

And he underscores how his rejection of facts ricochets through the rest of his philosophy: … my rejection of facts as entities correspondence to which can explain truth is central to my views of truth and meaning, to my rejections of the scheme-content distinction, representationalism, and much more. Davidson 1999a: 667

Although Davidson repeats similar remarks with some regularity (cf. Davidson 1990a: 303; 1997b: 17; 2001a: 295) and, on occasion—as in Davidson (2005), where he says that the notion of correspondence is ‘unintelligible’ (page 34), ‘without content’ (page 42) and ‘incomprehensible’ (page 47)—appears to let a degree of rhetoric to get the better of him, none of this is particularly persuasive or, indeed, impressive. These remarks are merely assertions and not arguments, and one looks in vain though his other papers for a discussion of what, for example, ‘nontrivial’ and ‘instructive’ accounts of facts would amount to.3 More interesting, perhaps, or certainly more substantiated than the bald denial of the existence of facts, is Davidson’s claim that ‘if we try to provide a serious semantics for reference to facts, we discover that they melt into one; there is no telling them apart’. He goes on: ‘The proof of this claim is given by Alonzo Church, who credits it to Frege. Church thinks this is the reason Frege held that all true sentences name the same thing, which he called The True’ (Davidson 1999b: 106. Cf. Davidson 1996: 266). This Church (and, one might add, Frege, Gödel and Quine) argument has come to be known as … 2.2 The Slingshot It has other names as well. Lycan (1974) calls it the Extensionality Argument; Davies (1981: 210–212), Evnine (1991) and Sainsbury (2001: 291–298) call it Frege’s 3 Blackburn’s (1976: 354) observation, with reference to Davidson (1975), that ‘some of Davidson’s phrases are dangerously balanced between what has been arrived at and what is needed’ would seem to be generalizable to other parts of his work.

354

turner

Argument (or, the Frege Argument); García-Carpintero and Pérez Otero (1998) call it the Frege-Gödel-Church Argument; Burge (1986) calls it the ChurchGödel Argument and Wagner (1986) calls it the GF Argument (for the GödelFrege Argument or, equivalently, the Great Fact Argument). The term ‘the Slingshot Argument’ originated, I believe, with Barwise and Perry (1981a, 1981b) and now has wide circulation. Davidson uses the argument to prove that if sentences refer to facts then all true sentences refer to the same fact (and, as a consequence, presumably, all false sentences refer to the same non-fact, or to the ‘wrong’ fact).4 His initial use of (a version of) the argument is in Davidson (1967a). He uses this version to discredit the referential theory of meaning. He first assumes that (a) ‘logically equivalent singular terms have the same reference’ (Davidson 1967a: 305–306) and that (b) ‘a singular term does not change its reference if a contained singular term is replaced by another with the same reference’ (Davidson 1967a: 306). He continues: But now suppose that ‘R’ and ‘S’ abbreviate any two sentences alike in truth value. Then the following four sentences have the same reference: (1) (2) (3) (4)

R x̂ (x = x.R) = x̂ (x = x) x̂ (x = x.S) = x̂ (x = x) S

For (1) and (2) are logically equivalent, as are (3) and (4), while (3) differs from (2) only in containing the singular term ‘x̂ (x = x.S)’ where (2) contains ‘x̂ (x = x.R)’ and these refer to the same thing if S and R are alike in truth value. Hence any two sentences have the same reference if they have the same truth value. And if the meaning of a sentence is what it refers to, all sentences alike in truth value must be synonymous—an intolerable result. Davidson 1967a: 306

A subsequent use of (a version of) the argument is in Davidson (1969). He uses this version to discredit the correspondence theory of truth. He first introduces 4 Davidson also uses the argument to make a point about the nature of causal connectives in Davidson (1967b) and to make a point against Reichenbach about the logical form, and relevant associated ontology, of action sentences in Davidson (1967c). Neither of these applications is of concern here.

remarks on davidson’s polymorphous concept of truth

355

the principle that ‘if a statement corresponds to the fact described by an expression of the form “the fact that p”, then it corresponds to the fact described by “the fact that q” provided ‘p’ and ‘q’ are logically equivalent, or one differs from the other in that a singular term has been replaced by a coextensive singular term’ (Davidson 1969: 752–753). He continues: The confirming argument is this. Let ‘s’ abbreviate some true sentence. Then surely the statement that s corresponds to the fact that s. But we may substitute for the second ‘s’ the logically equivalent ‘(the x such that x is identical with Diogenes and s) is identical with (the x such that x is identical with Diogenes)’. Applying the principle that we may substitute coextensive singular terms, we can substitute ‘t’ for ‘s’ in the last quoted sentence, provided ‘t’ is true. Finally, reversing the first step we conclude that the statement that s corresponds to the fact that t, where ‘s’ and ‘t’ are any true sentences. Davidson 1969: 753

Davidson returns to the argument(s) in his John Dewey Lectures: … starting from the assumptions that a true sentence cannot be made to correspond to something different by the substitution of co-referring singular terms, or by the substitution of logically equivalent sentences, it is easy to show that, if true sentences correspond to anything, they all correspond to the same thing. But this is to trivialize the concept of correspondence completely; there is no interest in the relation of correspondence if there is only one thing to which to correspond, since, as in any such case, the relation may as well be collapsed into a simple property: thus, “s corresponds to the universe,” like “s corresponds to (or names) the True” or “s corresponds to the facts,” can less misleadingly be read “s is true”. Davidson 2005: 40

He concludes: ‘we may read the result of our argument as showing that there is exactly one fact … The Great Fact’ (Davidson 1969: 753) and therefore ‘the postulation of facts will not explain, define, or even illuminate the concept of truth’ (Davidson 2005: 129). A Davidsonian slingshot can be formulated in a number of ways. Lepore and Ludwig (2005: 50–51), using the notation of set theory rather than Davidson’s class abstracts (this is not an important difference—it is also possible to use iota-operators or definite descriptions), helpfully reconstruct these sketchy remarks into the following template for a clear natural deduction style derivation:

356 (1) Logically equivalent singular terms have the same reference (2) A singular term does not change its reference if a contained singular term is replaced by another with the same reference (3) Sentences are singular terms and refer to their meanings (4) Sentences ‘S’ and ‘R’ are alike in truth value (5) ‘R’ is logically equivalent to ‘{x: x = x & R} = {x: x = x}’ (6) Thus, ‘R’ and ‘{x: x = x & R} = {x: x = x}’ have the same referent (7) ‘{x: x = x & R}’ and ‘{x: x = x & S}’ are co-referring singular terms (8) ‘{x: x = x & R} = {x: x = x}’ and ‘{x: x = x & S} = {x: x = x}’ have the same referent (9) ‘S’ and ‘{x: x = x & S} = {x: x = x}’ are logically equivalent (10) ‘S’ and ‘{x: x = x & S} = {x: x = x}’ have the same referent (11) Thus, ‘R’ and ‘S’ have the same referent (12) Thus, ‘R’ and ‘S’ have the same meaning

turner

(by 1,3,5) (by 4) (by 2,7)

(by 1,3,9) (by 6,8,10) (by 3,11)

And, formal details aside (see Lepore and Ludwig 2005: 51–55), they conclude that the argument equivocates on the interpretation of the expression ‘logically equivalent’ and is open to the charge of being question begging.5 There are those who, like Davidson, (a) attempt to redefine, defend and wield the slingshot,6 and there are others who either (b) find the argument ineffective—essentially invalid or unsound, or deficient in some other way,7 or who (c) believe that there are defences against it—essentially improved theories of facts or more careful definitions of fact identity.8 I do not here seek formally to adjudicate in these matters but just to note, with reference to the correspondence theory of truth, that the correspondence intuition is so evidently strong and that the slingshot’s conclusion is so evidently counterintuitive that I cannot, currently, on the available evidence and argument,

5 Burge (2007: 582), on independent grounds, concurs with this judgment: ‘I find Davidson’s discussion of [correspondence to structured propositions, facts, states of affairs and other relatively fine-grained entities] rather too brisk and ultimately question begging.’ 6 For example, Drai (2002); McGinn (1976): Rodriguez-Pereyra (1998); Shramko and Wansing (2009); Stoutland (2003); Levine 2006; Wallace (2015). 7 For example, Fumerton (2002); Hochberg (1975, 2003); Horgan (1978, 1982); Krüger (1995); Landini (2014); Lee (2002); Mackie (1980); Neale (1994); Widerker (1985); Read (1993); Searle (1996); Taylor (1985); Young (2002); Baumgartner (2010a). 8 For example, Brandl (1991); Englebretsen (2006); Genova (2001); Künne (2003); Manning (2003); Newman (2002); Perry (1996); Rodriguez-Pereyra (1999); Baumgartner (2010b); Rasmussen (2014); Vision (2004).

remarks on davidson’s polymorphous concept of truth

357

surrender my correspondence sympathies. The prudent choice might just be to line up behind Gödel when he says of this argument: ‘I cannot help feeling … that there is something behind it which is not yet completely understood’ (Gödel 1946: 130).9 If, however, with the fullness of time, the arguments endorsed by those in (a), above, are more conclusively shown to be less than adequate and the arguments endorsed by those in (b) or (c) are shown to be more conclusively stronger, then the slingshot will have to undergo a change of name. David’s weapon was successful: Goliath was killed. Davidson’s weapon has not so far been demonstrated to be successful: neither the correspondence theory of truth nor the ontology of facts has been fatally struck. 2.2.1 The Ubiquity of Facts Despite the fact that Davidson has, in fact, attempted to recruit the fact that the Slingshot (allegedly) reduces all facts to one—The Great Fact (aka ‘the world’)—this has not, in fact, prevented him from referring to many very different facts throughout his published record (bold emphasis added in all cases): Corpus A: The ‘Fact’ Facts. The fact that (32) [= (x) (Hx ⊃ Rx) KT] may be correctly interpreted as saying one class is included in another is not due to ‘R’ and ‘H’ coming to refer to classes, but to the fact … that (32) is L-equivalent to (33) [= x̂ (Hx) ⊂ x̂ (Rx). KT]. Davidson 1963: 336

But aside from the fact that Quine declines to draw Strawson’s conclusion, there remains the important difference that Quine at most thinks (1) and (2) are true, while Strawson wants to show they are necessary. Davidson 1965: 385fn1

But of course the occurrence can’t be truth-functional: otherwise, from the fact that I brought about one actual state of affairs it would follow that I brought about every actual state of affairs. Davidson 1967c: 89

9 The go-to place to examine that background and to begin to improve one’s understanding is Neale (1995, 2001). Cf. Preyer and Peter (2006); Oppy (2004); Manning (2004); and Restall (2004).

358

turner

An example will help bring out the fact that the thesis applies not only to translation between speakers of conspicuously different languages, but also to cases nearer home. Davidson 1968: 137

A theory of truth for a natural language must take account of the fact that many sentences vary in truth value depending on the time they are spoken, the speaker, and even, perhaps, the audience. Davidson 1970: 180

Now suppose for a moment that the fact that p doesn’t exist, i.e. that ~ (∃x) (x consists in the fact that p); then the right limb of (11) would be true, and so would the left. But the left side implies that the fact that p does exist. So the fact that p exists, whatever fact it may be. Davidson 1971: 343

Our present trouble springs from the fact that in radical interpretation we cannot assume that a T-sentence satisfies the translation criterion. Davidson 1973b: 326

Well, this earth of ours is part of the solar system, a system partly identified by the fact that it is a gaggle of large, cool, solid bodies circling around a very large, hot star. Davidson 1975: 21

Someone who can interpret English knows, for example, that an utterance of the sentence ‘Snow is white’ is true if and only if snow is white; and he knows in addition that this fact is entailed by a translational theory— that it is not an accidental fact about that English sentence, but a fact that interprets that sentence. Davidson 1976: 37

The fact that satisfaction, which we have been thinking of as recursively characterized, can be given an explicit definition (by the Frege-Dedekind technique) should not lead us into thinking a general concept has been captured. Davidson 1977: 250

remarks on davidson’s polymorphous concept of truth

359

Some stress the special insight metaphor can inspire and make much of the fact that ordinary language, in its usual functioning, yields no such insight. Davidson 1978: 32

These theories draw their strength from the fact that we can, and often do, use indicatives to do the work Dummett believes is conventionally assigned to other moods. … It is a virtue of these theories that they make evident the fact that having a truth value is no obstacle to a sentence’s being used to issue a command or ask a question. Davidson 1979b: 14

If Socrates is right—if such actions are ruled out by the logic of the concepts—then there is nothing puzzling about the facts to be explained. Davidson 1982b: 294

The first two essays on radical interpretation stress the fact that understanding the words of a speaker requires knowing much about what he believes. Davidson 1984: xvii

There is something more about assertion that suggests that convention may be involved, and this is the fact that in making an assertion, the asserter must intend to make as assertion, and he must intend that this intention be recognized by his audience. Davidson 1985c: 14–15

What the judge has done is use the fact that B is willing to pay more as a reason for increasing the value he sets on selling to B; the reference to strength of desire merely acts as a middle term in getting from valued fact to valued action. Davidson 1986b: 198

The fact the beliefs and desires explain actions only when they are described in such a way as to reveal their suitability for causing the action reduces the power of the explanation, and so does the fact that the explanation provides no reason for saying that one suitable belief-desire pair rather than another (which may well also have been present in the agent) did the causing. These two facts are connected, for both are due to the unavailability of accurate laws for reason-explanations. Davidson 1987: 42

360

turner

The ultimate evidence, as opposed to a criterion, for the correctness of a theory of truth we must presume lies in available facts about how speakers use the language. When I say available, I mean publicly available— available not only in principle, but available in fact to anyone who is capable of understanding the speaker or speakers of some languages. Davidson 1988: 152

But as far as I can see, the case can best be made by appealing directly to obvious facts about language learning and to facts about how we interpret words and language with which we are unfamiliar. The relevant facts have already been mentioned. Davidson 1989: 164

But science may without prejudice or circularity note the facts about human nature that reflect interests: the facts about salience, attention and tendencies to generalise in some ways rather than others. Davidson 1990c: 78

In the dark Joyce weaves and reweaves his world; de Todleben is known for the fact that each night at Sevastopol he rebuilt the defences. Davidson 1991a: 7

It follows from the nature of correct interpretation that an interpersonal standard of consistency and correspondence to the facts applies to both the speaker and the speaker’s interpreter, to their utterances and to their beliefs. Davidson 1991b: 158

An utterance of “Ich bin krank” is true in German if it is spoken by a person who is ill at the time of utterance, and this is a correct generalization even if the sentence is in fact never uttered. Davidson 1993c: 21

What a discussion can do, with luck, is dispel some of the fog; it can at least reveal that our aporia is due to the fact, not that we are grappling with a deep metaphysical puzzle, but that we need better or different concepts. Davidson 1994c: 433

remarks on davidson’s polymorphous concept of truth

361

There is no choosing between ways of relating words to things as long as the truth conditions of sentences come out the same; but this an interesting fact because there are empirical constraints on truth conditions. Davidson 1995: 15

We should accept the fact that what makes these concepts so important must also foreclose on the possibility of finding a foundation for them which reaches deeper into bedrock. Davidson 1996: 264

There is also the astonishing fact that children brought up hearing nothing but pidgin, which is a highly simplified invention of adults thrown together and lacking a common language, soon elaborate the pidgin into a Creole as developed and complicated as French or Turkish. Davidson 1997b: 19

The propositional attitudes, such as intentions, desires, beliefs, hopes and fears are every bit as real as atoms and baseball bats, and the facts about them as real as the facts about anything else. Davidson 1998: 98

But this negative point does not entail that there are no facts of the matter; the facts are the empirical relations between a speaker, her sentences, and her environment. Davidson 1998: 100

Neither the fact that conversing speakers may not fully share idioms nor the fact that they often deviate from their own practice has any tendency to show that the semantic properties of their sentences do not depend on the semantic properties of their parts and their structure. Davidson 1999h: 73–74

It is a brute fact—a fact about each brute, including you and me—that each reacts to stimuli as it does; but it is a suggestive fact that many of these brutes react in more or less the same ways. Davidson 2001c: 5

It’s not the facts we disagree about, but about how we ought to describe the facts. Davidson 2003a: 697

362

turner

I depend on two things: the role ostension plays in radical translation, and the emphasis on the fact that all there is to empirical meaning is what is picked up in such situations (aside from what we learn from the interanimation of sentences). Davidson 2003b: 290–291

The fact itself doesn’t, then, consist merely of the two entities, Theaetetus and the property of Sitting. It is a fact because these two entities stand in a certain relationship to each other … Now we see that a fact we can describe in just two words apparently involves three entities. … The fact requires as well that these three entities stand in a certain relation to one another. Davidson 2005: 85

To have this distinction is to have the concept of objectivity, that is, to appreciate the fact that many things are as they are however we think of them. One cannot have beliefs about most aspects of the world without grasping the fact that things may seem, look, or appear to be other than they are. Davidson 2006: 1065

And finally: But in any case, Austin’s discussion of excuses illustrates over and over the fact that our common talk and reasoning about actions is most naturally analysed by supposing that there are such entities. Davidson 1967c: 84

It is difficult to resist, in the context outlined by the above illustrations, that Davidson’s use of the language of facts demonstrates over and over (and, it is tempting to add, over) the case that our common talk and reasoning about facts can be most naturally analysed by supposing that there are such entities.10 The 10

It may be necessary to add, to conform to the requirements of full disclosure, that Davidson also says this: ‘Having banished properties, states, beliefs, and so on, (including facts, presumably, KT) from my explanatory machinery, I will now continue to allow myself to use words that superficially seem to refer to such things, on condition that I will be understood to hold that such talk can be exchanged at boring length for talk that doesn’t use these words as referring to entities … my ontological scruples are due only to my suspicion that these entities do not help explain what I want to explain’ (Davidson 2001a:

remarks on davidson’s polymorphous concept of truth

363

following position is hereby recommended: ‘Davidson cannot just claim there are no necessary truths [or facts, KT]. As long as some sentences using “necessary,” “possible,” and the like [like “facts”, KT] are true, these words must make some contribution to the truth-conditions of sentences’. (Wheeler 2014: 2).11

11

299). And: ‘There should be no harm in talking this way as long as one can, with a bit more stylistic effort, say the same thing without calling on such entities’ (Davidson 1993a: 38). Homework for the interested reader: Do so. It is likely that Davidson is here taking a page out of his mentor’s book: ‘[C]onsider how I have persisted in my vernacular use of “meaning”, “idea”, and the like, long after casting doubt on their supposed objects. True, the use of a term can sometimes be reconciled with rejection of its objects …; but I go on using the terms without even sketching any such reconciliation. What is involved here is simply a grading of austerity. I can object to using a certain dubious term at crucial points in a theory, on the ground that to use it would deprive the theory of its desired explanatory force; but I can still use and condone the term in more casual or heuristic connections, where less profundity of theoretical explanation is professed’ (Quine 1960: 210). Olson (1987) is an instructive discussion of the metaphysics and semantics of facts. Davidson (1999j), in a reply to Neale (1999), says that ‘Neale’s splendid discussion … has encouraged me to stick to the conclusion that facts cannot be incorporated into a satisfactory theory of truth. Neale speaks of Russell’s view of facts as ‘acceptable’, but he does not show that it can be incorporated into a Tarski-style truth theory’ (Davidson 1999j: 89). This is a curious observation to make. Davidson was, by this time, well into RP3 and denying that truth could be defined and therefore be the kind of ‘thing’ for which a theory could be given. The letter, and indeed, spirit, of Tarski disappeared from Davidson’s writings in Davidson (1986c)—‘there is no such thing as a language’. The Russellian view gets some additional support 10 years post-Olson: ‘There are two forms of the slingshot argument. One is due to A. Church and involves abstractions of the form “the class of all x such that x = x & A,” where A is any sentence … The other is due to K. Gödel and involves descriptions of the form “the x such that x = a & Fx” … In both cases, it is essential to the argument that substitutions within certain non-truth-functional contexts involving abstractions and descriptions of these kinds are legitimated by an assumption of “referential transparency” to the effect that singular terms with the same reference can be interchanged salva veritate within these non-truth-functional contexts. Thus it is essential to the argument that it is legitimate to regard these abstractions and descriptions as singular terms; and it is precisely this presumption that is undermined by a Russellian treatment of abstractions and descriptions. In agreeing with Russell on this matter, it is not necessary to accept all the disputable details of his theories of descriptions and abstractions; all that matters is the key insight that these phrases are quantified noun phrases and are no more singular terms than are phrases such as “someone who lives next door”’ (Baldwin 1997: 17–18fn21).

364

turner

2.3 Correspondence Regained? And now for a further small refinement. RP1* doesn’t just explore the hypothesis that ‘truth theories, but not correspondence theories, can do duty as theories of meaning’. It examines the more specialist hypothesis that ‘truth theories, but not fact-correspondence theories, can do duty as theories of meaning’. Call this RP1**. The refinement is the result of Davidson’s arguments for the realignment of correspondence away from facts and towards ‘satisfaction’. Satisfaction is, essentially, a relation between open sentences and an infinite number of arbitrarily constituted infinitely long sequences of objects.12 Although these arguments are backed with the considerable authority of Tarski, Davidson does acknowledge that they are not without cost. He says: The relation, satisfaction, is not … exactly what intuition expected of correspondence; and the … sequences that satisfy may not seem much like facts. Davidson 1969: 758

and Theory of truth based on satisfaction is instructive partly because it is less ambitious about what it packs into the entities to which sentences correspond: in such a theory, these entities are no more than arbitrary pairings of the objects over which the variables of the language range with those variables. Relative simplicity in the objects is offset by the trouble it takes to explain the relation between them and sentences, for every truth-relevant feature of every sentence must be taken into account in describing satisfaction. Davidson 1969: 759–760

This cost is rather considerable. Davidson has previously said that ‘the task of a theory of meaning … is not to change, improve or reform a language, but to describe and understand it’ (Davidson 1967a: 314; cf. Davidson 1979a: 297) and he once teased Quine for his (Quine’s) efforts at regimentation: ‘Like Quine, I am interested in how English and languages like it (i.e., all languages) work,

12

In an effort to bypass a long and technical discussion of this notion I shall here follow Davidson and assert that ‘[t]he reader in search of precision and deeper understanding cannot be too strongly urged to study Tarski’s “The Concept of Truth in Formalized Languages” ’ (Davidson 1969: 757fn11).

remarks on davidson’s polymorphous concept of truth

365

but, unlike Quine, I am not concerned to improve on it or change it. (I am the conservative and he is the Marxist here.)’ (Davidson 1985b: 172). But Davidson is a closet Marxist, and a rather extreme one, at that. Kripke observes: ‘Surely no one really thinks that the ‘true’ structure of quantification over individuals is quantification over sequences’ (Kripke 1976:357). I certainly do not. On the contrary, if we could recover our pre-Tarskian semantic innocence, I think it would seem to us plainly incredible that the truth-theoretic interpretation of any utterance of any sentence bore any relation at all to any objects in an infinite number of infinitely long and arbitrarily constituted sequences. 2.4 An Experiment with Coherence And now for a modest reorientation. RP1 doesn’t just examine the hypothesis that ‘truth-theories can do duty as theories of meaning’, nor just the subsequent more specialist hypotheses RP1* and RP1**. It also includes the slightly tangential hypothesis that ‘coherence truth-theories can do duty as theories of meaning’. Call this hypothesis RP1COH. Davidson says: A coherence theory in its boldest and clearest form declares that beliefs in a consistent set of beliefs are true; coherence is the only possible test for truth, and so coherence must constitute truth. So stated, a coherence theory of truth can be taken as a defence of a philosophical method which claims only to remove inconsistency; for once inconsistencies have been excised, the coherence theory assures us that what remains will be an unadulterated body of truths. Davidson 1993b: 180

Coherence theories have never, though, fared particularly well under examination and Davidson is clear about their limitations: Consistency is not enough, since it leaves no basis on which to choose between various and conflicting consistent theories … it leaves us with no basis for judging truth … [W]e must have … indubitable grounds for choosing some sentences as the true ones rather than others. Davidson 1982a: 478

And: A pure coherence theory of truth would hold, I suppose, that all the sentences in any consistent set of sentences are true. I doubt if anyone ever

366

turner

held such a theory, for it is obviously false.13 … unless something further is added, this view seems as wrong as Schlick held it to be (he called it an ‘outstanding error’).14 Davidson 1988/2001: 185–186

The something further that Davidson adds to the coherence theory is (i) the causal theory of meaning and (ii) the hypotheses of (a) the holism of, and (b) the veridicality of, belief. Both of these additions serve to push any coherence beginnings towards what looks like a correspondence ending. But, at this point, and in this context, to borrow a helpful phrase from Hilary Putnam, ‘I won’t try to explain this further, because this where I get off the boat’ (Putnam 1985: 78), albeit, with a view to re-boarding at some point in the future.15 It is worth mentioning, though, in conclusion that Davidson has signalled that he is deeply unhappy with the claims put forward in Davidson (1986a). His said: ‘There is no paper I have written I would like more to rewrite’ (Davidson 2001b: xv). And he explains: ‘If it had not already been published, I would now change the title of “A Coherence Theory”, and I would not describe the project as showing how “coherence yields correspondence”’ (Davidson 1990b: 134. Cf. Davidson 1999g: 135.)16

13

14

15

16

Davidson was later to become slightly more impatient and to put his point this way: ‘A pure coherence theory of truth would hold, I suppose, that all the sentences in a consistent set of sentences are true. Perhaps no one has ever held such a theory, for it is mad’ (Davidson 1990a: 305). He repeats this version of the judgement in Davidson (2005: 43). ‘The astounding error of the “coherence theory” can be explained only by the fact that its defenders and expositors were thinking only of such statements as actually occur in science, and took them as their only examples. … If one is to take coherence seriously as a general criterion of truth, then one must consider arbitrary fairy stories to be as true as a historical report, or as statements in a textbook of chemistry, provided the story is constructed in such a way that no contradiction ever arises’ (Schlick 1934/1959: 215). ‘… if something like a Davidsonian theory is indeed possible, it is a mighty iceberg about whose overall shape [and especially about whose philosophical consequences, KT] we have very little idea’ (Wright 1993: 218). Davidson’s assessment of his paper is not universally shared: ‘Davidson says this is the paper he would most like to rewrite. He chastises himself for giving the impression that he disallows any role for the senses in “serious commerce between world and mind”. For my part I do not see how any reader of that wonderful paper could miss the fact that Davidson credits the senses with a substantial role in our acquisition of knowledge about the world. His claim is, perfectly clearly, not that the senses play no role in knowledge-acquisition, but that their role is causal and not evidential. I cannot see how rewriting could make that any clearer’ (McDowell 2003: 680).

remarks on davidson’s polymorphous concept of truth

367

3.1 Research Programme 2 RP2 introduces a significant change in orientation. It examines the hypothesis that ‘the attitude of holding true, directed to sentences under specified circumstances, is the basis for interpretation’ (Davidson 1975: 20). The hypothesis is first advanced when, in discussing (a) the interdependence of belief and meaning—Davidson’s Circle—and (b) how it might be possible to break into this Circle, Davidson says: A good place to begin is with the attitude of holding a sentence true, of accepting it as true. … It is an attitude an interpreter may plausibly be taken to be able to identify before he can interpret, since he may know that a person intends to express a truth in uttering a sentence without having any idea what truth.17 Davidson 1973b: 332. Cf. Davidson 1974b: 14, 18

The change is away from an interest in truth theories as a key to the formatting of theories of meaning towards the detection of a propositional attitude to enable—with some other, not uncontroversial, ingredients, such as the principle of charity, triangulation and the assumptions of the veridicality and holism of belief—conversational strategies for the interpretation in context of a speaker. Davidson continues: It seems to me that the basic justification for holding empirical sentences true is given by the conditions under which we observe them to be held true. When we have succeeded in matching up a speaker’s observation sentences with our own, we have discovered what justifies the speaker in holding them true. Of course, such matching is not to be trusted simply on the basis of a few successes; logically connected sentences must come out right, and so must explanations of error. The interpretation of

17

It is possible to argue, although the argument will not be developed here, that this change in perspective was already anticipated in Davidson 1967a: 319. He was later to say: ‘Whether we want to consider theories of this sort [truth theories made sensitive to indexicality and demonstratives, KT] part of semantics is a matter for choice: if theory of truth is semantics, we are still within the domain of semantics; while if any reference to speakers or speech acts is a move to pragmatics, then theory of truth for a natural language belongs to pragmatics’ (Davidson 1970: 180–181fn4). This footnote does not survive in the reprint in Davidson 1984. Instead, there is added the following: ‘Truth conditions are not to be equated with meanings: at best we can say that by giving the truth conditions of a sentence we give its meaning. But this claim too needs clarification’ (Davidson 1984: 56fn3).

368

turner

any sentence depends on placing it in a network of other sentences. But the network must be tied to the world through the kind of triangulation that is fundamental to radical interpretation. Davidson 1999d: 460

The twin assumptions of a speaker holding a sentence true and of an interpreter detecting that a sentence is being held true are among the less plausible assumptions entertained in the philosophy of language. Davidson acknowledges some of the problems: we cannot directly see that someone holds a sentence true: what we can observe are assents and dissents. It is a tricky matter to decide what is theoretical and what observational, what inferred and what is known without inference, especially in this area. Davidson 1999e: 331

But the difficulties go beyond observation. A principal difficulty is the likely impossibility of keeping truth, belief and meaning separate, as the account requires: The claim that we can isolate the attitude of holding a sentence true independently of any knowledge that we have of the meanings of particular sentences deserves pausing over briefly. One might plausibly argue that to say that a sentence is such that it can be held true is already to say at least something about the meaning of sentence. If so, it will follow that Davidson has not made good on his claim to provide non-question begging evidence for his weakened T-sentences. To see that at least some knowledge of meaning is already entailed by our being able to say that on an occasion a speaker holds a sentence true, simply contrast the attitude of holding a sentence to be true with the attitude of holding a sentence to express a command, warning, question, or promise. If we know that a speaker takes a certain sentence to express a truth rather than a command or a question, we would seem to know at least something about the meaning of that sentence. Taylor 1985: 103–104fn4

Ordinarily, there is no difficulty about being able to identify a speaker as holding a sentence true without being able to specify the content of that sentence. That we should be able to do so is an undisputable condition of the possibility of radical interpretation. What is less clear is the extent

remarks on davidson’s polymorphous concept of truth

369

to which this can be done without some presupposition as to the beliefs and desires of the speaker. Interpreting a speaker as holding some unspecified belief true is surely dependent on other assessments of the speaker’s beliefs and desires. Thus, it would seem doubtful to suppose that we can find any place to begin our interpretative project which is unaffected by the holistic nature of interpretation. Insofar as he assumes that we can do just this, Davidson is surely mistaken. Malpas 1992: 44

And the difficulty extends to the identification of intentional behaviour more generally: A theory of radical interpretation should be applicable to giant octopuses or superbeings emerging from the spaceships as well as to newly encountered human beings. But consideration of such non-human cases makes clear that identification of something as a holding true requires ability to distinguish voluntary from involuntary and linguistic from non-linguistic behaviour. There is every reason to suppose that making these distinctions will involve simultaneously making rich hypotheses about the contents of beliefs and purposes. Heal 1997: 184

These assumptions have also been challenged by, among others, Bennett (1976), Peacocke (1979), Holdcroft (1981), Blackburn (1984), Guttenplan (1985), Picardi (1989, 1994), Wiggins (1997), Hacker (1996, 1997) and Zilhão (2003). These challenges have not, though, made too much of an appearance in the exegetical literature. Kemp (2012) doesn’t see a reason to challenge holding true at all, and neither does Weiss (2010: 99). Platts, whilst conceding that there is ‘a major difficulty … of defining and establishing the existence of the appropriate relation, that of holding true’, he goes on, in the next sentence, to say ‘[b]ut we can sidestep discussion of this tortuous issue’ (Platts 1979: 65; 1997: 65). Ramberg (1989: 68–70) acknowledges that the radical interpreter must assume both that s/he is dealing with a speaker who asserts correctly, that is, truthfully, in the relevant circumstances, but he does not deny the interpreter this assumption. Evnine (1991) makes no critical mention of the category of holding true, although he does say, as noted, implausibly, that a radical interpreter can ‘observe’ a speaker holding an utterance true (Evnine 1991: 102). Joseph (2004: 58) also attributes to Davidson the belief that interpreters ‘observe speakers holding sentences true (or false)’, and that they ‘perceive that there are meanings associated with speakers’ utterances, although they do not know what

370

turner

these meanings are’. Glüer (2011) remarks that ‘Davidson never provides much discussion of, or argument for, the assumption that attitudes of holding true are detectable by the radical interpreter’ (Glüer 2011: 72fn51) but she doesn’t dismiss that assumption, on that basis, or attempt to provide an argument to defend it. There is a noticeable smudge in Davidson’s expression when he discusses radical interpretation’s first base. He sometimes speaks à la Quine-ese in that he uses the language of assents and dissents. This way of speaking may have encouraged the mistaken and misleading references to observation above. At other times he is more in character and speaks à la Davidson-ese and he uses the language of holding true. In one of his last papers Davidson uses the language of assents and dissents to speak of radical interpretation’s starting point—the assumption about a speaker holding an utterance true … This is an awesome assumption, since it supposes both that a response of assent can be distinguished from other responses, and that those assents are, often enough, to a sentence taken literally rather than, say, to the informant’s intent to amuse, confuse, metaphorize, or implicate. Awesome or not, I think the assumption is justified. Both adults and threeyear-olds are uncannily skilled at distinguishing the serious from the jocular, the sly from the straight, the honestly instructive from the devious. If they were not, it is hard to see how they would ever uncover meaning at the level at which interpretation must begin. Davidson 2003b: 285

… but it is worth noting that in one of his last papers Quine conceded that ‘[r]adical translation is a near miracle’ (Quine 2008: 345). This is a radical concession. And a parallel concession ought now to be made. It is, I think, time to add that radical interpretation, or at least the initial component of it—holding true—is also, and can only be, a miracle. It just isn’t going to happen. RP2, once a progressive research programme, is now not just a regressive research programme, it is a stalled research programme, one stopped dead by the weight of its own impossibility.18 18

Cf. Fodor and Lepore (1994) and the extraordinary retreat in Davidson (1994b). (For an earlier extraordinary retreat, cf. Davidson (1976): ‘A theory of truth, no matter how well selected, is not a theory of meaning’.) See also Sinclair (2002). Davidson eventually conceded that RP2 was stillborn: ‘One thing that makes it hard to devise a theoretical account of interpretation is the complexity of the interdependencies among the various attitudes, and the extent to which the

remarks on davidson’s polymorphous concept of truth

371

3.2 Truth, via Belief, to Meaning Some of Davidson’s illustrations of the method of interpretation that he is attempting to promote, and how it results in ‘meanings’, provide a good indication of both his focus and his perspective. If you see a ketch sailing by and your companion says, “Look at that handsome yawl,” you may be faced with a problem of interpretation. One natural possibility is that your friend has mistaken a ketch for a yawl, and has formed a false belief. But if his vision is good and his line of sight favorable it is even more plausible that he does not use the word “yawl” quite as you do, and has made no mistake at all about the position of the jigger on the passing yacht. We do this sort of off the cuff interpretation all the time, deciding in favor of reinterpretation of words in order to preserve a reasonable theory of belief. Davidson 1974b: 18

[L]et us suppose we have a speaker and someone else who is trying to understand his words. Each time a mouse appears nearby in good light and with the speaker oriented in the direction of the mouse, etc., the speaker utters what sounds to the interpreter like the same expression: “ratón”. When the lighting is poor or the speaker is inattentive, the response is less firmly correlated with mouse appearances. I think that unless there is a host of evidence against such an interpretation, the competent interpreter will take the speaker to mean by his words, and to believe, that there is a mouse present. What recommends this interpretation is the fact that the presence of a mouse has apparently in each case caused the speaker intentionally to utter the same expression, “ratón”, and to utter it in an affirmative spirit. Of course an interpreter may make a

content of a single thought or expression rests on its place in a network of further thoughts and expressions. These interdependencies entail that the understanding of any particular belief, intention, desire, action, or utterance of an agent is always contingent on knowing or correctly assuming a vast amount about the rest of that agent’s attitudes. If it were possible to discover the contents of thoughts one at a time, it is at least possible to imagine how a general picture of a mind could be built up. But since grasping the content of any one thought or motive or utterance depends on grasping the content of a multitude, there is a problem how an interpreter can get started’ (Davidson 2006: 1057: Emphasis added.). It is perhaps worth noting that when he is in conversation, Davidson retreats to a more intuitive position: ‘Well of course you can’t tell when any sentence is true or false until you know what the words in it mean’ (Davidson in Bergström and Føllesdal 1994: 215).

372

turner

mistake: he may be in error about the intention of the speaker, or about the true regular and exclusive cause of the speaker’s response. The interpreter would be wrong, for example, if a squirrel would provoke the same response in the speaker, for then mice would not be the regular and exclusive causes of utterances of “ratón”. But provided the interpreter makes none of these mistakes … his interpretation is correct. … Suppose the speaker sees a skilfully made mechanical mouse and sounds his “ratón”. The interpreter must decide whether the speaker’s word and concept include mechanical mice, or whether this is a mistake on the speaker’s part. Deciding this is not beyond the interpreter’s power. He may expose the mechanism to the speaker, for example, and note his response. The careful interpreter will have to learn how the speaker responds to many other classes of objects before he arrives at this final interpretive scheme. Davidson 1991c: 194–195

There is an area in the continuum of hues where I say ‘green’ and others say ‘blue’. You may decide either that I am wrong about the color or that I use the words differently: according to how you interpret what I mean you adjust my beliefs to suit. Although the change in your interpretation of my ‘green’ affects an infinity of possible further interpretation, it may be that there is no reason to choose between the interpretations. … there are endless ways in which the predicates and singular terms of a language may be mapped onto the same ontology which yield equivalent truth conditions for all the sentences of that language. … Frege’s dictum that it is only in the context of a sentence that words have a meaning I interpret as saying that semantic theories can postulate what they please about words as long as what they yield for sentences is correct.19 Davidson 2001a: 298

19

Davidson had previously invoked Fregean contextuality in this way: ‘If sentences depend for their meaning on their structure, and we understand the meaning of each item in the structure only as an abstraction from the totality of sentences in which it features, then we can give the meaning of any sentence (or word) only by giving the meaning of every sentence (and word) in the language. Frege said that only in the context of a sentence does a word have meaning; in the same vein he might have added that only in the context of the language does a sentence (and therefore a word) have meaning’ (Davidson 1967a: 308). He would later retreat: ‘Words, Frege emphasized, have a meaning only in the context of a sentence. The basic

remarks on davidson’s polymorphous concept of truth

373

Davidson’s most extreme illustrations of this come from the category of the what-is-mis-said. For example: Corpus B: The What-is-mis-said. In the early 1950s my piano teacher regularly used “eksetera” … Better still, and my favourite, was “sistificates” instead of certificates. I never had the courage to correct her. Letters. The Guardian, 10th October 2013

My elderly neighbour … used to say, “I love Dave Allen. He tells such lovely antedotes.” Letters. The Guardian, 12th October 2013

Appearing to be knocked out by the standard of his completed work in our shower room, our plumber declared it to be “anaesthetically very pleasing” … What more could I say? Letters. The Guardian, 15th October 2013

My mother-in-law always admired the work of the Salivation Army. Letters. The Guardian, 15th October 2013

A friend refers to the legal paperwork we undertake in case we “lose our marbles” as the endearing power of eternity. Letters. The Guardian, 16th October 2013

reason for this is that the work of language, to give information, tell stories, ask questions, issue commands, and so on, is done by sentences; a word that is not being used to convey the content of a sentence cannot do any of these things. It is at the sentential level that language connects with the interests and intentions language serves, and this is also the level at which the evidence for interpretation emerges. But just as words have a meaning only in the context of a sentence, a sentence has a meaning only in a context of use’ (Davidson 1993d: 297–298: emphasis added). Davidson likes to inflect this claim: ‘We have learned from Frege that it is only in the context of a sentence that a singular term comes to life; we should add that it is only in the context of a language that we can tell where the singular terms are’ (Davidson 1971: 347). Frege did not, as far as I have been able to see, refer to the Context Principle after the Grundlagen (Frege 1953). Wittgenstein, however, referred and subscribed to it both Early (Wittgenstein 1922: 50) and Later (Wittgenstein 1958: 24e).

374

turner

I worked with a man who described how he travelled with his family through France by train and slept in a courgette. Letters. The Guardian, 17th October 2013

While others recall nostalgic malapropisms … only days ago BBC 5 Live decried the leak of Hodgson’s half-time team talk as breaking the sanctimony of the dressing room. Letters. The Guardian, 22nd October 2013

No one, however smart, however well-educated, however experienced, is the suppository of all wisdom. Tony Abbott. Australia’s Liberal Party Leader. Quotes of the Week. The Mail on Sunday, 18th August 2013

I am a person who recognises the fallacy of humans. George W. Bush to Oprah Winfrey. The Top Ten Malapropisms. Edited by John Rentoul. The Independent on Sunday, 4th May 2014

In future, all cars will be fitted with ‘Catholic converters’. University of Ulster undergraduate. Sarah Harris, Exam howlers that earn F for failure. Daily Mail, 10th July 2014

Malapropism corner: “She [Kate Winslet] said that at 19 she had initially been intimated by [Alan] Rickman before realising he was ‘a warm and giving man and an utterly phenomenal actor and gifted director’ ” (Truly, madly, deeply: stars mourn a magnificent actor and ‘the kindest, best of men’, 15th January, page 4). Corrections and clarifications. The Guardian, 16th January 2016

If it is a mistake it’s because the rules are not being implicit enough about where that costing should go. Andrew Bridgen, Conservative MP. Newsnight. BBC2, 10.55pm. 20th May 2016

https://www.youtube.com/watch?v=yIHe_HNuw4M Waiter: Me: Waiter: Me: Waiter:

Are you from around here? No, Brighton. And are you happy about Brexit? No. Me neither.

remarks on davidson’s polymorphous concept of truth

Me: Waiter:

375

I can’t believe it. Me neither. It was so unaccepted.

Friday, 24th June 2016, 8.03pm. Sizzling Spice Indian Restaurant, 62 Christchurch Road, Reading RG2 7AZ. www.sizzlingspice.co.uk.

In the declaration from a senior Amazon executive, filed with the Court, he said that Trump’s travel ban immediately and negatively impacted employees, dependents of employees and candidates for employment with Amazon. So it’s very clear that they had a hand in this and wanted this Tramp, this travel ban lifted too. Siobhan Kennedy, Channel 4 News. 4th February 2017

So, Isabel, do you agree that this is basically Spreadshit Phil, Spreadsheet Phil I should say, as he is known in the Treasury, basically holding his nose and saying ‘There’ll be a row but it’ll pass’. Nick Robinson, Today, BBC Radio 4. 8.50am. 9th March 2017

This kind of example illustrates Davidson’s ‘derivations’ very well. The proof would go like this. First, per impossibile, as discussed above, assume the interpreter is able to detect the speaker holding an utterance20 true on an occasion of speaking at a time and in a place; second, assume charity21—that is, some-

20

21

There is room for some further discrimination here. It will be necessary to distinguish, at least, for example, observation, occasion and standing utterances. All of Davidson’s examples are taken from contexts where observable circumstances are assumed to be obvious. There is no discussion, that I have been able to find, of a ‘projection principle’ that would allow correlations beyond the observable circumstances. The absence of such a principle in RP2 restricts its coverage rather severely. There are difficulties associated with the Principle of Charity that are not pursued here (cf. Ramberg 1989: 70: ‘[P]recise articulation of the principle of charity turns out to be an extremely tricky task’). But precise articulation is exactly what is required: ‘Davidson’s Principle of Charity seems to oblige us only to regard most of what the natives say as justified—to regard them as holding mostly beliefs which we regard as true. Should they be unable to offer justification which we regard as reasonable, there will be no coherent set of inferential relationships to be detected between the various string of marks and the noises they produce, and therefore no rationality—no pattern of the sort truth makes’ (Rorty 1995: 286–287). In a sentence added in the reprint of this paper he says, more firmly: ‘Without charity, we cannot detect the pattern truth makes’ (Rorty 1998: 26). For some helpful literature see, amongst many others, Lewis 1974; Davidson 1974c; McGinn 1977; Vermazen 1982; Sundholm 1984–1985; Gauker 1986; Henderson 1987; Malpas 1988; Larson 1990; Warmbröd 1991;

376

turner

thing like, assume the speaker is as rational as the interpreter; thirdly, assume that the speaker cannot, as the interpreter does not, believe that, for example, it is possible to sleep in a courgette; finally, conclude that the interpreter can calculate that the speaker ‘means’, on this occasion of utterance, at this time and in this place, ‘couchette’ by ‘courgette’. QED.22 This kind of proof allows Davidson to say that ‘you can change the meaning provided you believe (and perhaps are justified in believing) that the interpreter has adequate clues for the new interpretation’ (Davidson 1986c: 439), and, more starkly, ‘there is nothing inherently wrong in twisting language any way we want’ (Davidson 1998: 111).23 3.2.1 James Joyce Davidson makes reference to some of the work of James Joyce to illustrate, in an extreme fashion, what he means by ‘twisting language any way we want’. Four examples will suffice: Corpus C: The James Joyce examples. Nobody aviar soar anywing to eagle it. Why do I am alook alike a poss of porter pease? Was he come to hevre with his engiles or gone to hull with the poop. Flunkey Footle furloughed foul, writing off his phoney.

22

23

Sluiter 1998; Wong 1999; Ramberg 1999; McCarthy 2002: 46–51; Jackman 2003; Goldberg 2004; Ludwig 2004; Sorensen 2004; Lee 2006; Saka 2007; Lewiński 2012; Balcerak Jackson 2013; Glüer 2013; Pagin 2013a, b; Båve 2015; Stoutenburg 2016. Note to self: Plan a paper entitled ‘Remarks on Davidson’s polymorphous concept of charity and its role in a theory of meaning’. This derivation schema is more faithful to Davidson’s various remarks than some others currently on the market. Cf. ‘the methodology of radical interpretation is this: First the interpreter applies the principle of charity (i.e., he assumes that sentences held true by the subject are in fact true), and notes correlations between sentences held true by the subject and external circumstances. Then he develops hypotheses about more fine-grained correlations between sentence segments and external objects or events. Finally he tests these hypotheses in his interactions with the subject, and revises them where necessary’ (Talmage 1994: 218). This schema omits the crucial first assumption that, one clearly has to concede, is false. Davidson enters a caveat to these claims: ‘unless a speaker is fairly consistent over time in his use of words, an interpreter has no chance of figuring out what he means’ (Davidson in Pires de Oliveira 1996: 97).

remarks on davidson’s polymorphous concept of truth

377

Of these, and other, examples, Davidson writes: All reading is interpretation, and all interpretation demands some degree of invention. It is Joyce’s extraordinary idea to raise the price of admission to the point where we are inclined to feel that almost as much is demanded of the reader as of the author. Goading his audience into fairly testing creative activity, Joyce both deepens and removes from immediate view his own part in the proceedings. By fragmenting familiar languages and recycling the raw material Joyce provokes the reader into involuntary collaboration, and enlists him as a member of his private linguistic community. Coopted into Joyce’s world of verbal exile, we are forced to share in the annihilation of old meanings and the creation—not really ex nihilo, but on the basis of our stock of common lore—of a new language. All communication involves such joint effort to some degree, but Joyce is unusual in first warning us of this, and then making the effort so extreme.24 Davidson 1991a: 11

3.3 Notes on the RP1/RP2 Interface RP1 is an exploitation and extension of Tarski’s (1944, 1983) work on the definition of truth. But there is a certain number of reconfigurations. Tarski’s focus is on (context-free) formal languages; RP1’s is on (context-sensitive) natural languages: Tarski assumes meaning (synonymy, translation) to define truth; RP1 assumes (some concept or other of) truth to make meaning tractable: Tarski works with sentences; RP1 with utterances: Tarski is essentially doing syntax; RP1 has ambitions to do semantics,25 or, at least, an undisciplined conflation of semantics and pragmatics.26 Tarski’s T-sentences, often presented along the pattern of 24

25

26

It is not difficult to appreciate Davidson’s fascination with James Joyce. A mere flick through something like Finnegans Wake reveals such further examples as ‘pensible’, ‘adulescence’, ‘alcoherence’, ‘sadisfaction’, ‘chaosmos’, ‘earwitness’, ‘intaxication’, ‘intermisunderstanding’, ‘regretitude’, ‘dredgerous’, ‘gossipocracy’, ‘whyacinthinous’ and ‘fleshmonger’. ‘Joyce draws on every resource his readers command (or that he hopes they command, or thinks they should command), every linguistic resource, knowledge of history, geography, past writers and styles. He forces us both to look at and to listen to his words to find the puns and fathom the references’ (Davidson 1991a: 4; emphasis added). ‘Translation is a syntactical process; interpretation (as I understand it) is semantical—it relates language to world, not to another language’ (Davidson in Pires de Oliveira 1996: 101). On the one hand, for example, Davidson says:

378

turner

(a) ‘s is true-in-L if and only if p’ should more accurately be presented as (b) ‘s’ is true in some object language (e.g. English) if and only if p is true in some meta-language (e.g. first-order logic with identity) This formulation is more explicit and more faithful to the spirit of Tarski’s work on truth and it underscores the huge difference between Tarski-truth and more traditional conceptions.27 But (b), though faithful to Tarski, is not of much help to RP1. Suppose s is some sentence in some natural language and p some true proposition in a predicate logic, e.g. ∀x (Man x → Mortal x). Then (b) licences the interpreter to infer that s is true but not to determine what s’s content is. (b) does not distinguish, that is, between ‘rogue’ and ‘interpretive’ T-sentences. It is only the latter that should emerge from the theory. Tarski’s T-sentences only require that each side of the material biconditional be true, not that they ‘mean’ the ‘same thing’. Davidson must impose a means to limit T-sentences to those that are M[eaning]-sentences. RP2 may come to the rescue. RP2, in effect, instructs the interpreter to engage in some fieldwork and to take a look at what real speakers do in real contexts of communication. RP2 converts (b) into (c):

27

‘Thinking of truth, the problem could be put this way: if we were exposed to speakers of a language we did not know, and were given a Tarski-style truth definition, how could we tell whether the definition applied to that language’ (Davidson 1985c: 13; emphasis added). On the other hand, he says: ‘I don’t think of T-sentences as normative in themselves: they don’t, for example, tell us what truth conditions we ought to assign to a sentence, nor do they tell us when we would be “correct” to assert it, unless “correct” here just means “true”. T-sentences are descriptive: I think of them as describing a practice. Using a T-sentence to interpret a speaker whose practice at the moment it correctly describes will yield a correct interpretation, or at least so I have long held, and still do’ (Davidson 1998: 111; emphasis added). ‘Tarski’s name for his theory, “The Semantical Conception of Truth,” is highly misleading! It would have been more revealing to call the theory the “Nonsemantical Conception of Truth.” … If I had to name the theory, I would call it the “Look No Semantics! Conception of Truth”; then no one could miss the startling claim Tarski made … the concern of philosophy is precisely to discover what the intuitive notion of truth is. As a philosophical account of truth, Tarski’s theory fails as badly as it is possible for an account to fail … It just isn’t truth at all’ (Putnam 1985: 63, 64). McGee (1993) also, more diplomatically, and via an interesting explication and defense of what he calls ‘idiolectic disquotationalism’, comes to pretty much the same conclusion.

remarks on davidson’s polymorphous concept of truth

379

(c) S[peaker] holds: u is true in some object language, relative to a context and a time if and only if p is true in some relevant metalanguage.28 But the-little-Fregean-that-ought-to-be-in-everybody alerts us to the fact that (b) and (c) are not equivalent (or even related):29 being true is different from being taken to be true, whether by one or many or everybody, and in no case is to be reduced to it. There is no contradiction in something’s being true which everybody takes to be false … If it is true that I am writing this in my chamber on the 13th July 1893, while the wind howls out-of-doors, then it remains true even if all men should subsequently take it to be false. Frege 1964: 13

Somewhere along the way (mid 1980s), Davidson dispenses with the concept of a language: ‘I conclude that there is no such thing as a language, not if a lan28

29

The move from (b) to (c) also changes the object of inquiry from sentences to utterances. Davidson is sometimes relaxed about this: ‘Verbal felicity apart, there is no reason not to call the utterance of a sentence under conditions that make the sentence true, a true utterance’ (Davidson 1990a: 310). On other occasions he is more tense: ‘[A]n interpreter (correctly) interprets an utterance of a speaker only if he knows that the speaker intends the interpreter to assign certain truth conditions to his (the speaker’s) utterance. A full account of this thesis would require an explanation of the idea of “assigning truth conditions” to an utterance, and this idea is no doubt as difficult to understand in relevant respects as the concept of meaning itself’ (Davidson 1992: 258). In a footnote, he adds: ‘I do not argue here for the very large assumption that knowledge of truth conditions is adequate for interpretation’ (Davidson 1992: 265fn5). On the more general problem, Rojszczak 2005 is an enlightening historical account. The-little-Fregean-in-all-of-us will find much helpful instruction in Künne 2008. This paper has done much to convince me that we should all look after our own little Fregeans as well as we can. And in the absence of a little Fregean, a little Peircean will do: ‘You certainly opine that there is such a thing as Truth. … What do you mean by there being such a thing as Truth? You mean that something is SO—is correct, or just—whether you, or I, or anybody thinks it is so or not. … The essence of the opinion is that there is something that is SO, no matter if there be an overwhelming vote against it’ (Peirce 1960a: volume 2: paragraph 135); ‘Every man is fully satisfied that there is such a thing as truth, or he would not ask any question. That truth consists in a conformity to something independent of his thinking it to be so, or of any man’s opinion on that subject’ (Peirce 1960b: volume 5: paragraph 211).

380

turner

guage is anything like what many philosophers and linguists have supposed’ (Davidson 1986c: 446).30 This conclusion is based on the examples of the what is mis-said but it can also have a phenomenological basis: We all talk so freely about language or languages that we tend to forget that there are no such things in the world; there are only speakers and their various written and acoustical products. This point, obvious in itself, is nevertheless easy to forget, and it has consequences that are far from universally recognized … Our practical, as opposed to our purely theoretical, interest in linguistic phenomena is this: we want to understand the actual utterances of others, and we want our utterances to be understood. What has language to do with this interest? Davidson 1992: 256

Also: We forget there is no such thing as a language apart from the sounds and marks people make, and the habits and expectations that go with them. ‘Sharing a language’ with someone else consists in understanding what they say, and talking pretty much the way they do. There is no additional entity we possess in common any more than there is an ear we share when I lend you an ear. Davidson 1997b: 18; cf. Davidson 1985c: 13; 1994b: 14; 1994c: 435

30

This remark has provoked almost no end of commentary. Cf. Hacking 1986; Dummett 1986, 1994, 2007; George 1990; Bar-On and Risjord 1992; Pietroski 1994; Gustafsson 1998; Pereda 1998; Green 2001; Cash 2004; Ludlow 2006; Lepore and Ludwig 2007c: Hornsby 2008; Knowles 2015; Bar-On 2016; Stainton 2011, 2016. The remark, though, should not occasion such surprise (nor, in some cases, such outrage). Recall Quine: ‘What I find … disturbing … is the need of dragging in the language concept at all. What is a language? What degree of fixity is supposed? When do we have one language and not two?’ (Quine 1956: 187). And Bloomfield is something of a proto-Davidsonian: ‘If we observed closely enough, we should find that no two persons—or rather, perhaps, no one person at different times—spoke exactly alike. … the linguist is forced to consider [these differences] very carefully, even though in some of his work he is forced provisionally to ignore them. When he does this, he is merely employing the method of abstraction, a method essential to scientific investigation’ (Bloomfield 1933: 45). Both McGuire (2003: 360) and Haack (2008: 23fn13) think that this remark signals the end of Davidson’s Tarski interest and ushers in the Later Davidson. I think the change comes much earlier in Davidson 1973b. This is why Davidson’s 1984 terminological revision of ‘radical translation’ to ‘radical interpretation’ (Section 1.1 above) is a nonsense. ‘True-in-L’ is incompatible with ‘holding true’.

remarks on davidson’s polymorphous concept of truth

381

These considerations require the conversion of (c) into (d): (d) S holds: u is true on some occasion of utterance, relative to a context and a time if and only if p is true in some relevant meta-language.31 (d) is a long way from (a)–(b). Tarskian considerations would appear to have dropped out of the picture altogether.32 RP3, as we are about to see, increases the distance between Tarski- and Davidson-truth.

4

Research Programme 3

The principal claim of the third research programme, and the final for the purposes here, can be stated briefly, but firmly: ‘truth is as clear and basic a concept as we have’ (Davidson 1990b: 135. Cf. ‘Truth is beautifully transparent … and I take it as primitive’ (Davidson 1986a: 308)). Davidson continues: Any … attempt to explain, define, analyse or explicate the concept will be empty or wrong: correspondence theories, coherence theories, pragmatist theories, theories that identify truth with warranted assertability (perhaps under ‘ideal’ or ‘optimum’ conditions), theories that ask truth to explain the success of science or serve as the ultimate outcome of science or the conversations of some elite, all such theories either add nothing to our understanding of truth or have obvious counter-examples. Why on earth should we expect to be able to reduce truth to something clearer or more fundamental? Davidson 1990b: 135–136. Italics added to represent what I imagine to be Davidson’s emphasis to be if he ever gave this paper as a talk.

Elsewhere he says, a little more briefly:

31 32

There is further discussion to be had here, which is reserved for another paper: ‘Remarks on Davidson’s polymorphous concept of triangulation and its role in a theory of meaning’. It is perhaps, with the benefit of hindsight, not entirely clear why they were ever in the picture in the first place. One firm intuition is that truths are cross-linguistic and that one expressed in English can equally well be expressed in any other language. ‘Truein-L’ is a very strange predicate and, like ‘satisfaction’, must be placed in a category of those Tarskianisms that have little or no demonstrable use in natural language semantics (and/or pragmatics).

382

turner

It is a mistake to look for a behavioristic definition, or indeed any other sort of explicit definition or outright reduction of the concept of truth. Truth is one of the clearest and most basic concepts we have, so it is fruitless to dream of eliminating it in favour of something simpler or more fundamental. Davidson 1990a: 314. Italics added; ditto

And again: the concepts philosophers single out for attention, like truth, knowledge, belief, action, cause, the good and the right, are the most elementary concepts we have, concepts without which … we would have no concepts at all. Why then should we expect to be able to reduce these concepts definitionally to other concepts that are simpler, clearer, and more basic? We should accept the fact that what makes these concepts so important must also foreclose on the possibility of finding a foundation for them which reaches deeper into bedrock. … Truth is … an indefinable concept. This does not mean we can say nothing revealing about it: we can, by relating it to other concepts like belief, desire, cause, and action. Davidson 1996: 264, 265

4.1

Some Replies

(a) Claim: ‘truth is as clear and basic a concept as we have’. Reply: This is very unlikely. In fact, I am going to say that this claim is false. The motivations of this reply I take as obvious. (b) Rhetorical question: ‘Why on earth should we expect to be able to reduce truth to something clearer or more fundamental?’ Answer: Because that is what philosophers do. Because that is what philosophy is. (c) Claim: ‘Truth is one of the clearest and most basic concepts we have, so it is fruitless to dream of eliminating it in favour of something simpler or more fundamental’. Reply: Analysis does not try to eliminate the concept, but to understand it. (d) Claim: ‘We should accept the fact that what makes these concepts so important must also foreclose on the possibility of finding a foundation for them which reaches deeper into bedrock.’ Reply: The words ‘foundation’ and ‘bedrock’ are part of an elaborate metaphor that often enters in abstract discussion. They are more suited to an architectural or geological discourse that to the philosophical. (e) Claim: ‘Truth is … an indefinable concept’. Reply: Davidson equivocates.

remarks on davidson’s polymorphous concept of truth

(f)

383

In (b) he takes about reduction, in (c) about elimination, here about definition. But of the many different kinds of definition, he argues only against two. Cf. Gupta (2015). He has not yet established that it is a folly to attempt to define truth. Claim: ‘[W]e can … relat[e] [truth] to other concepts like belief, desire, cause, and action’. Reply: Not without discussing the character of truth first (and the character of belief, desire and the rest).

In the light of (a)–(f) we must conclude that the case for RP3 has not been convincingly established.

5

Final Remarks

My remarks serve as elucidations in the following way: anyone who understands them eventually recognizes them as preliminary, when he or she has used them—as steps—to climb up beyond them. (He, or she, must, so to speak, throw away the ladder after he or she has climbed up on it.)

References Amoretti, Maria Cristina and Gerhard Preyer, eds. 2011. Triangulation: From an Epistemological Point of View. Paris/Frankfurt/Lancaster/New Brunswick: Ontos Verlag. Amoretti, Maria Cristina and Nicla Vassallo, eds. 2008. Knowledge, Language, and Interpretation: On the Philosophy of Donald Davidson. Paris/Frankfurt/Lancaster/New Brunswick: Ontos Verlag. Armstrong, Josh, 2016a. Coordination, triangulation, and language use. Inquiry: An Interdisciplinary Journal of Philosophy 59: 80–112. Armstrong, Josh, 2016b. The problem of lexical innovation. Linguistics and Philosophy 39: 87–118. Atlas, Jay D., 1988. Review of Lycan 1984. Language 64: 158–167. Atlas, Jay D., 2005. Logic, Meaning, and Conversation: Semantical Underdetminacy, Implicature, and Their Interface. Oxford: Oxford University Press. Baghramian, Maria, ed. 2013. Donald Davidson: Life and Words. London/New York: Routledge. Balcerak Jackson, Brendan, 2013. Metaphysics, verbal disputes and the limits of charity. Philosophy and Phenomenological Research 86: 412–434. Baldwin, Thomas, 1997. Frege, Moore, Davidson: The indefinability of truth. Philosophical Topics 25: 1–18.

384

turner

Bar-On, Dorit, 2016. Sociality, expression, and this thing called language. Inquiry: An Interdisciplinary Journal of Philosophy 59: 56–79. Bar-On, Dorit and Mark Risjord, 1992. Is there such a thing as a language? Canadian Journal of Philosophy 22: 163–190. Barwise, Jon and John Perry, 1981a. Semantic innocence and uncompromising situations. In Peter A. French, Theodore E. Uehling, Jr., and Howard K. Wettstein, eds. Midwest Studies in Philosophy Volume VI: The Foundations of Analytic Philosophy. 387–403. Minneapolis: University of Minnesota Press. Barwise, Jon and John Perry, 1981b. Situations and attitudes. The Journal of Philosophy 78: 668–691. Baumgartner, Michael, 2010a. Causal slingshots. Erkenntnis 72: 111–133. Baumgartner, Michael, 2010b. Shallow analysis and the slingshot argument. Journal of Philosophical Logic 39: 531–556. Båve, Arvid, 2015. Charity and error-theoretic nominalism. Ratio 28: 256–270. Begby, Endre, 2016. Deranging the mental lexicon. Inquiry: An Interdisciplinary Journal of Philosophy 59: 33–55. Bennett, Jonathan, 1976. Linguistic Behaviour. Cambridge: Cambridge University Press. Bennett, Jonathan, 1985. Critical Notice: Inquiries into Truth and Interpretation. By Donald Davidson. Mind 94: 601–626. Bergström, Lars and Dagfinn Føllesdal, 1994. Interview with Donald Davidson in November 1993. Theoria 60: 207–225. Blackburn, Simon, 1976. Critical study: Mind and Language. The Philosophical Quarterly 26: 354–362. Blackburn, Simon, 1984. Spreading the Word: Groundings in the Philosophy of Language. Oxford: Clarendon Press. Bloomfield, Leonard, 1933. Language. London: George Allen & Unwin Ltd. Bourke, Michael, 2012/13. Davidson’s account of truth and fictional meaning. Praxis 3: 1–27. Brandl, Johannes, 1991. Some remarks on the “slingshot” argument. In Gerhard Schurz and Georg J.W. Dorn, eds. Advances in Scientific Philosophy: Essays in Honour of Paul Weingartner on the Occasion of the 60th Anniversary of his Birthday. 421–437. Amsterdam/Atlanta, GA: Rodopi. Burge, Tyler, 1986. Frege on truth. In Leila Haaparanta and Jaakko Hintikka, eds. Frege Synthesized: Essays on the Philosophical and Foundational Work of Gottlob Frege. 97– 154. Dordrecht/Boston/Lancaster/Tokyo: D. Reidel Publishing Company. Burge, Tyler, 2007. Review essays: Predication and truth. The Journal of Philosophy 104: 580–608. Camp, Elizabeth, 2016. Conventions’ revenge: Davidson, derangement and dormativity. Inquiry: An Interdisciplinary Journal of Philosophy 59: 113–138. Cash, Mason, 2004. Unconventional utterances? Davidson’s rejection of conventions in

remarks on davidson’s polymorphous concept of truth

385

language use. ProtoSociology: An International Journal of Interdsciplinary Research 20: 285–319. Cook, John, 2009. Is Davidson a Gricean? Dialogue 48: 557–575. Davidson, Donald, 1963. The method of extension and intension. In P.A. Schilpp, ed. The Philosophy of Rudolf Carnap. 311–349. La Salle, Illinois: Open Court. Davidson, Donald, 1965. Theories of meaning and learnable languages. In Yehoshua Bar-Hillel, ed. Logic, Methodology and Philosophy of Science: Proceedings of the 1964 International Congress. 383–394. Amsterdam: North-Holland Publishing Company. Davidson, Donald, 1967a. Truth and meaning. Synthese 17: 304–323. Davidson, Donald, 1967b. Causal relations. The Journal of Philosophy 64: 691–703. Davidson, Donald, 1967c. The logical form of action sentences. In Nicholas Rescher, ed. The Logic of Decision and Action. 81–95. Pittsburgh: University of Pittsburgh Press. Davidson, Donald, 1968. On saying that. Synthese 19: 130–146. Davidson, Donald, 1969. True to the facts. The Journal of Philosophy 63: 748–764. Davidson, Donald, 1970. Semantics for natural languages. In (No Editor Cited). Linguaggi nella Società e nella Tecnica. 77–88. Milan: Edizioni di Comunità. Davidson, Donald, 1971. Eternal vs. ephemeral events. Nous 5: 335–349. Davidson, Donald, 1973a. In defense of Convention T. In Hughes Leblanc, ed. Truth, Syntax and Modality: Proceedings of the Temple University Conference on Alternative Semantics. 76–86. Amsterdam/London: North-Holland Publishing Company. Davidson, Donald, 1973b. Radical interpretation. Dialectica 27: 313–328. Davidson, Donald, 1974a. Belief and the basis of meaning. Synthese 27: 309–323. Davidson, Donald, 1974b. On the very idea of a conceptual scheme. Proceedings and Addresses of the American Philosophical Association 47: 5–20. Davidson, Donald, 1974c. Replies to David Lewis and W.V. Quine. Synthese 27: 345–349. Davidson, Donald, 1975. Thought and talk. In Samuel Guttenplan, ed. Mind and Language: Wolfson College Lectures 1975. 7–23. Oxford: Clarendon Press. Davidson, Donald. 1976. Reply to Foster. In Gareth Evans and John McDowell, eds. Truth and Meaning: Essays in Semantics. 33–41. Oxford: Clarendon Press. Davidson, Donald, 1977. Reality without reference. Dialectica 31: 247–258. Davidson, Donald, 1978. What metaphors mean. Critical Inquiry 5: 31–47. Davidson, Donald, 1979a. The method of truth in metaphysics. In Peter French, Theodore Uehling and Howard Wettstein, eds. Contemporary Perspectives in the Philosophy of Language. 294–304. Minneapolis: University of Minnesota Press. Davidson, Donald, 1979b. Moods and performances. Avishai Margalit, ed. Meaning and Use: Papers Presented at the Second Jerusalem Philosophical Encounter April 1976. 9– 20. Dordrecht: D. Reidel Publishing Company. Davidson, Donald, 1982a. Empirical content. In Rudolf Haller, ed. Schlick und Neurath— Ein Symposion. 471–489. Amsterdam: Rodopi. Davidson, Donald, 1982b. Paradoxes of irrationality. In Richard Wollheim and James

386

turner

Hopkins, eds. Philosophical Essays on Freud. 289–305. Cambridge: Cambridge University Press. Davidson, Donald, 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press. Davidson, Donald, 1985a. A new basis for decision theory. Theory and Decision 18: 87– 98. Davidson, Donald, 1985b. Reply to Quine on events. In Ernest Lepore and Brian McLaughlin, eds. Actions and Events: Perspectives on the Philosophy of Donald Davidson. 172–176. Oxford: Basil Blackwell Ltd. Davidson, Donald, 1985c. Communication and convention. In Marcelo Dascal, ed. Dialogue: An Interdisciplinary Approach. 11–25. Amsterdam/Philadelphia: John Benjamins Publishing Company. Davidson, Donald, 1986a. A coherence theory of truth and knowledge. In Ernest Lepore, ed. Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson. 307–319. Oxford: Basil Blackwell. Davidson, Donald, 1986b. Judging interpersonal interests. In Jon Elster and Aanund Hylland, eds. Foundations of Social Choice Theory. 195–211. Cambridge: Cambridge University Press. Davidson, Donald, 1986c. A nice derangement of epitaphs. In Ernest Lepore, ed. Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson. 433–446. Oxford: Basil Blackwell. Davidson, Donald, 1987. Problems in the explanation of action. In Philip Pettit, Richard Sylvan and Jean Norman, eds. Metaphysics and Morality: Essays in Honour of J.J.C. Smart. 35–49. Oxford: Basil Blackwell. Davidson, Donald, 1988. Epistemology and truth. Congreso Internacional Extraordinario de Filosofía del 20 al 26 de Setiembre de 1987. 147–161. Universidad Nacional de Cordoba, República Argentina. Davidson, Donald, 1989. The myth of the subjective. In Michael Krausz, ed. Relativism: Interpretation and Confrontation. 159–172. Notre Dame, Indiana: University of Notre Dame Press. Davidson, Donald, 1990a. The structure and content of truth. The Journal of Philosophy 87: 279–328. Davidson, Donald, 1990b. Afterthoughts, 1987. In Alan Malachowski, ed. Reading Rorty: Critical Responses to Philosophy and the Mirror of Nature (and Beyond). 134–138. Oxford: Basil Blackwell. Davidson, Donald, 1990c. Meaning, truth and evidence. In Robert Barrett and Roger Gibson, eds. Perspectives on Quine. 68–79. Oxford: Blackwell. Davidson, Donald, 1991a. James Joyce and Humpty Dumpty. In Peter A. French, Theodore E. Uehling and Howard K. Wettstein, eds. Philosophy and the Arts. 1–12. Notre Dame, Indiana: Notre Dame University Press.

remarks on davidson’s polymorphous concept of truth

387

Davidson, Donald, 1991b. Three varieties of knowledge. In A. Phillips Griffiths, ed. A.J. Ayer Memorial Essays. 153–166. Cambridge: Cambridge University Press. Davidson, Donald, 1991c. Epistemology externalized. Dialectica 45: 191–202. Davidson, Donald, 1992. The second person. In Peter A. French, Theodore E. Uehling, Jr. and Howard K. Wettstein, eds. Midwest Studies in Philosophy Volume XVII: The Wittgenstein Legacy. 255–267. Notre Dame, Indiana: University of Notre Dame Press. Davidson, Donald, 1993a. Reply to Richard Schantz. In Ralf Stoecker, ed. Reflecting Davidson: Donald Davidson Responding to an International Forum of Philosophers. 36–39. Berlin/New York: Walter de Gruyter. Davidson, Donald, 1993b. Plato’s philosopher. Apeiron: A Journal for Ancient Philosophy and Science 26: 179–194. Davidson, Donald, 1993c. Reply to Wolfgang Künne. In Ralf Stoecker, ed. Reflecting Davidson: Donald Davidson Responding to an International Forum of Philosophers. 21–23. Berlin/New York: Walter de Gruyter. Davidson, Donald, 1993d. Locating literary language. In Reed Way Dasenbrock, ed. Literary Theory after Davidson. 295–308. University Park, Pennsylvania: The Pennsylvania State University Press. Davidson, Donald, 1994a. Radical interpretation interpreted. Philosophical Perspectives 8: 121–128. Davidson, Donald, 1994b. The social aspect of language. In Brian McGuiness and Gianluigi Oliveri, eds. The Philosophy of Michael Dummett. 1–16. Dordrecht/ Boston/ London: Kluwer Academic Publishers. Davidson, Donald, 1994c. Dialectic and dialogue. In Gerhard Preyer, Frank Siebelt and Alexander Ulfig, eds. Language, Mind and Epistemology: On Donald Davidson’s Philosophy. 429–437. Dordrecht/Boston/London: Kluwer Academic Publishers. Davidson, Donald, 1995. Pursuit of the concept of truth. In Paolo Leonardi and Marco Santambrogio, eds. On Quine: New Essays. 7–21. Cambridge: Cambridge University Press. Davidson, Donald, 1996. The folly of trying to define truth. The Journal of Philosophy 93: 263–278. Davidson, Donald, 1997a. Indeterminism and antirealism. In Christopher B. Kulp, ed. Realism/Antirealism and Epistemology. 109–122. New York/Oxford: Rowman & Littlefield Publishers, Inc. Davidson, Donald, 1997b. Seeing through language. In John Preston, ed. Thought and Language. 15–27. Cambridge: Cambridge University Press. Davidson, Donald, 1998. Replies. Crítica: Revista Hispanoamericana de Filosfia 30: 97– 112. Davidson, Donald, 1999a. Reply to Stephen Neale. In Lewis Edwin Hahn, ed. The Philosophy of Donald Davidson. 667–669. Chicago and La Salle, Illinois: Open Court. Davidson, Donald, 1999b. The centrality of truth. In Jaroslav Peregrin, ed. Truth and

388

turner

its Nature (If Any). 105–115. Dordrecht/Boston/London: Kluwer Academic Publishers. Davidson, Donald, 1999c. Reply to Dagfinn Føllesdal. In Lewis Edwin Hahn, ed. The Philosophy of Donald Davidson. 729–732. Chicago and La Salle, Illinois: Open Court. Davidson, Donald, 1999d. Reply to Pascal Engel. In Lewis Edwin Hahn, ed. The Philosophy of Donald Davidson. 460–461. Chicago and La Salle, Illinois: Open Court. Davidson, Donald, 1999e. Reply to Deborah Hansen Soles. In Lewis Edwin Hahn, ed. The Philosophy of Donald Davidson. 330–332. Chicago and La Salle, Illinois: Open Court. Davidson, Donald, 1999f. General comments. In Urszula M. Ẓegleń, ed. Donald Davidson: Truth, Meaning and Knowledge. 157–160. London and New York: Routledge. Davidson, Donald, 1999g. Reply to Roger F. Gibson. In Urszula M. Ẓegleń, ed. Donald Davidson: Truth, Meaning and Knowledge. 134–135. London and New York: Routledge. Davidson, Donald, 1999h. Reply to Peter Pagin. In Urszula M. Ẓegleń, ed. Donald Davidson: Truth, Meaning and Knowledge. 72–74. London and New York: Routledge. Davidson, Donald, 1999i. Is truth the goal of inquiry? Discussion with Rorty. In Urszula M. Ẓegleń, ed. Donald Davidson: Truth, Meaning and Knowledge. 17–19. London and New York: Routledge. Davidson, Donald, 1999j. Reply to Stephen Neale. In Urszula M. Ẓegleń, ed. Donald Davidson: Truth, Meaning and Knowledge. 87–89. London and New York: Routledge Davidson, Donald, 2000. Truth rehabilitated. In Robert B. Brandom, ed. Rorty and His Critics. 65–74. Oxford: Blackwell Publishing. Davidson, Donald, 2001a. Comments on the Karlovy Vary papers. In Petr Kotatko, Peter Pagin and Gabriel Segal, eds. Interpreting Davidson. 285–307. Stanford: CSLI Publications. Davidson, Donald, 2001b. Introduction. In his Subjective, Intersubjective, Objective. xiii– xviii. Oxford: Clarendon Press. Davidson, Donald, 2001c. Externalisms. In Petr Kotatko, Peter Pagin and Gabriel Segal, eds. Interpreting Davidson. 1–16. Stanford, California: CSLI Publications. Davidson, Donald, 1988/2001. Epistemology and truth. Reprinted in Donald Davidson, 2001. Subjective, Intersubjective, Objective. 177–191. Oxford: Clarendon Press. Davidson, Donald, 2003a. Responses to Barry Stroud, John McDowell, and Tyler Burge. Philosophy and Phenomenological Research 67: 691–699. Davidson, Donald, 2003b. Quine’s externalisms. In Hans-Johann Glock, Kathrin Glüer and Geert Keil, eds. Fifty Years of Quine’s “Two Dogmas”. 281–297. Amsterdam/New York: Rodopi. Davidson, Donald, 2005. Truth and Predication. Cambridge, MA/London: The Belknap Press of Harvard University Press. Davidson, Donald, 2006. The perils and pleasure of interpretation. In Ernest Lepore and Barry C. Smith, eds. The Oxford Handbook of Philosophy of Language. 1056–1068. Oxford: Clarendon Press.

remarks on davidson’s polymorphous concept of truth

389

Davies, Martin, 1981. Meaning, Quantification, Necessity: Themes in Philosophical Logic. London: Routledge and Kegan Paul. de Rouilhan, Philippe, 2015. Putting Davidson’s semantics to work to solve Frege’s paradox on concept and object. In Theodora Achourioti, Henri Galinon, José Martínez Fernández and Kentaro Fujimoto, eds. Unifying the Philosophy of Truth. 119–142. Dordrecht: Springer. Drai, Dalia, 2002. The slingshot argument: An improved version. Ratio 15: 194–204. Dresner, Eli, 2006. Davidson’s philosophy of communication. Communication Theory 16: 155–172. Dresner, Eli, 2009. Radical interpretation, the primacy of communication, and the bounds of language. Empedocles: European Journal for the Philosophy of Communication 1: 123–134. Dresner, Eli, 2015. Davidson’s interpretations: The step not taken. Nous 49: 698–712. Dummett, Michael, 1986. A nice derangement of epitaphs: Some comments on Davidson and Hacking. In Ernest Lepore, ed. Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson. 459–476. Oxford: Basil Blackwell. Dummett, Michael, 1994. Reply to Davidson. In Brian McGuiness and Gianluigi Oliveri, eds. The Philosophy of Michael Dummett. 257–267. Dordrecht: Kluwer Academic Publishers. Dummett, Michael, 2007. Reply to Ernie Lepore and Kirk Ludwig. In Randall E. Auxier and Lewis Edwin Hahn, eds. The Philosophy of Michael Dummett. 215–227. Chicago and La Salle, Illinois: Open Court. Ebbs, Gary, 2012. Davidson’s explication of meaning. In Gerhard Preyer, ed. Donald Davidson on Truth, Meaning, and the Mental. 76–104. Oxford: Oxford University Press. Englebretsen, George, 2006. Bare Facts and Naked Truths: A New Correspondence Theory of Truth. Aldershot: Ashgate. Evnine, Simon, 1991. Donald Davidson. Cambridge: Polity Press. Evnine, Simon, 2007. Comments on Donald Davidson: Meaning, Truth, Language, and Reality. Florida Philosophical Review 7: 121–127. Fallis, Don, 2013. Davidson was almost right about lying. Australasian Journal of Philosophy 91: 337–353. Fennell, John, 2015. Davidson: Normativist or anti-normativist? Acta Analytica 30: 67– 86. Fodor, Jerry and Ernie Lepore, 1994. Is radical interpretation possible? Philosophical Perspectives 8: 101–119. Frege, Gottlob, 1953. The Foundations of Arithmetic: A Logico-mathematical Enquiry into the Concept of Number. 2nd Edition. (English translation by J.L. Austin.) Oxford: Basil Blackwell. Frege, Gottlob, 1964. The Basic Laws of Arithmetic: Exposition of the System. (Translated

390

turner

and Edited, with an Introduction by Montgomery Furth.) Berkley/Los Angeles/London: University of California Press. Frege, Gottlob, 1984. Collected Papers on Mathematics, Logic, and Philosophy. (Edited by Brian McGuiness.) Oxford: Basil Blackwell. Fumerton, Richard, 2002. Realism and the Correspondence Theory of Truth. New York/ Oxford: Rowman & Littlefield Publishers, Inc. García-Carpintero, Manuel and Manuel Pérez Otero, 1998. Davidson, correspondence truth and the Frege-Gödel-Church Argument. History and Philosophy of Logic 19: 63– 81. Gauker, Christopher, 1986. The principle of charity. Synthese 69: 1–25. Genova, A.C., 2001. How Wittgenstein escapes the slingshot. Journal of Philosophical Research 26: 1–22. George, Alexander, 1990. Whose language is it anyway? Some notes on ideolects. Philosophical Quarterly 40: 275–298. Glock, Hans-Johann, 2014. The relation between Quine and Davidson. In Gilbert Harman and Ernie Lepore, eds. A Companion to W.V.O. Quine. 526–551. Oxford: John Wiley and Sons, Inc. Glüer, Kathrin, 2011. Donald Davidson: A Short Introduction. Oxford: Oxford University Press. Glüer, Kathrin, 2013. The status of charity I: Conceptual truth or a posteriori necessity? In Maria Baghramian, ed. Donald Davidson: Life and Words. 144–166. London: Routledge. Goldberg, Nathaniel, 2004. The principle of charity. Dialogue 43: 671–683. Goldberg, Nathaniel, 2008. Tension within triangulation. The Southern Journal of Philosophy 46: 363–383. Goldberg, Nathaniel, 2009. Triangulation, untranslatability, and reconciliation. Philosophia 37: 261–280. Goldberg, Nathaniel, 2012. Davidson, dualism, and truth. Journal for the History of Analytical Philosophy 1: 1–19. Gödel, Kurt, 1946. Russell’s mathematical logic. In Paul Arthur Schilpp, ed. The Philosophy of Bertrand Russell. 125–153. Evanston, Illinois: George Banta Publishing Company. Green, Karen, 2001. Davidson’s derangement: On the conceptual priority of language. Dialectica 55: 239–258. Grigoriev, Serge, 2008. Beyond radical interpretation: Individuality as the basis of historical understanding. European Journal of Philosophy 17: 489–503. Gupta, Anil, 2015. Definition. The Stanford Encyclopedia of Philosophy (Summer Edition.) Edward N. Zalta (ed.), URL = http://plato.stanford.edu/archives/sum2015/ entries/definitions/. Gustafsson, Martin, 1998. Systematic meaning and linguistic diversity: The place of meaning-theories in Davidson’s later philosophy. Inquiry 41: 435–453.

remarks on davidson’s polymorphous concept of truth

391

Guttenplan, S.D., 1985. Review of Davidson 1984. Philosophy 60: 408–411. Haack, Susan, 2008. The whole truth and nothing but the truth. In Peter A. French and Howard K. Wettstein, eds. Truth and its Deformities. 20–35. Oxford: Blackwell Publishing. Hacker, P.M.S., 1996. On Davidson’s idea of a conceptual scheme. The Philosophical Quarterly 46: 289–307. Hacker, P.M.S., 1997. Davidson on first-person authority. The Philosophical Quarterly 47: 285–304. Hacking, Ian, 1986. The parody of conversation. In Ernest Lepore, ed. Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson. 447–458. Oxford: Basil Blackwell. Heal, Jane, 1997. Radical interpretation. In Bob Hale and Crispin Wright, eds. A Companion to the Philosophy of Language. 175–196. Oxford: Blackwell Publishers. Heather, Nick and Gabriel Segal, 2013. Understanding addiction: Donald Davidson and the problem of akrasia. Addiction Research and Theory 21: 445–452. Henderson, David, 1987. The principle of charity and the problem of irrationality (Translation and the problem of irrationality). Synthese 73: 225–252. Hochberg, Herbert, 1975. Explaining facts. Metaphilosophy 6: 277–302. Hochberg, Herbert, 2003. Introducing Analytic Philosophy: Its Sense and Nonsense 1879– 2002. Frankfurt/London: Ontos Verlag. Hoeltje, Miguel, 2013. Lepore and Ludwig on ‘explicit meaning theories’. Philosophical Studies 165: 831–839. Hoeltje, Miguel, 2016. ‘Meaning and truth’ and ‘Truth and meaning’. Dialectica 70: 201– 215. Holdcroft, David, 1981. Principles of conversation, speech acts, and radical interpretation. In Herman Parret and Jacques Bouveresse, eds. Meaning and Understanding. 184–203. Berlin and New York: Walter de Gruyter. Horgan, Terrence, 1978. The case against events. Philosophical Review 87: 28–47. Horgan, Terrence, 1982. Substitutivity and the causal connective. Philosophical Studies 42: 47–52. Hornsby, Jennifer, 2008. Davidson and Dummett on the social character of language. In Maria Cristina Amoretti and Nicla Vassallo, eds. Knowledge, Language, and Interpretation: On the Philosophy of Donald Davidson. 107–122. Frankfurt etc: Ontos Verlag. Hunter, David, 2015. Davidson on practical knowledge. Journal for the History of Analytical Philosophy 3: 1–19. Isaac, Joel, 2013. Donald Davidson and the analytic revolution in American philosophy, 1940–1970. The Historical Journal 56: 757–779. Jackman, Henry, 2003. Charity, self interpretation, and belief. Journal of Philosophical Research 28: 145–170. Joseph, Marc, 2004. Donald Davidson. Chesham: Acumen.

392

turner

Kemp, Gary, 2012. Quine versus Davidson: Truth, Reference, and Meaning. Oxford: Oxford University Press. Knowles, Jonathan, 2015. Davidson versus Chomsky: The case of shared languages. In Ulf Zackariasson, ed. Action, Belief and Inquiry: Pragmatist Perspectives on Science, Society and Religion. 300–319. Helsinki: Nordic Pragmatism Network. Kripke, Saul, 1976. Is there a problem about substitutional quantification? In Gareth Evans and John McDowell, eds. Truth and Meaning: Essays in Semantics. 325–419. Oxford: Clarendon Press. Krüger, Lorenz, 1995. Has the correspondence theory of truth been refuted? From Gottlob Frege to Donald Davidson. European Journal of Philosophy 3: 157–172. Künne, Wolfgang, 2003. Conceptions of Truth. Oxford: Clarendon Press. Künne, Wolfgang, 2008. Frege on truths, truth and the True. Studia Philosophica Estonica 1: 5–42. Landini, Gregory, 2014. Russellian facts against the slingshot. Axiomathes 24: 533–547. Larson, David, 1990. The implications of error for Davidsonian charity. Philosophia 20: 311–320. Lee, Byeong, D., 2002. Davidson’s slingshot argument revisited. Dialogue 41: 541–550. Lee, Carole, J., 2006. Gricean charity: The Gricean turn in psychology. Philosophy of the Social Sciences 36: 193–218. Lepore, Ernie, 2004. An interview with Donald Davidson. In Donald Davidson, Problems of Rationality. 231–265. Oxford: Clarendon Press. Lepore, Ernie and Kirk Ludwig, 2005. Donald Davidson: Meaning, Truth, Language, and Reality. Oxford: Clarendon Press. Lepore, Ernie and Kirk Ludwig, 2006. Ontology in the theory of meaning. International Journal of Philosophical Studies 14: 325–335. Lepore, Ernie and Kirk Ludwig, 2007a. Donald Davidson’s Truth-Theoretic Semantics. Oxford: Clarendon Press. Lepore, Ernie and Kirk Ludwig, 2007b. Radical misinterpretation: A reply to Stoutland. International Journal of Philosophical Studies 15: 557–585. Lepore, Ernie and Kirk Ludwig, 2007c. The reality of language: On the Davidson/Dummett exchange. In Randall E. Auxier and Lewis Edwin Hahn, eds. The Philosophy of Michael Dummett. 185–214. Chicago and La Salle, Illinois: Open Court. Lepore, Ernie and Kirk Ludwig, 2011. Truth and meaning redux. Philosophical Studies 154: 251–277. Lepore, Ernie and Kirk Ludwig, eds. 2013. A Companion to Donald Davidson. Oxford: Wiley Blackwell. Levine, James, 2006. Analysis, abstraction principles, and slingshot arguments. Ratio 19: 43–63. Lewiński, Marcin, 2012. The paradox of charity. Informal Logic 32: 403–439. Lewis, David, 1974. Radical interpretation. Synthese 27: 331–344.

remarks on davidson’s polymorphous concept of truth

393

Ludlow, Peter, 2006. The myth of human language. Croatian Journal of Philosophy 6: 385–400. Ludwig, Kirk, 2004. Rationality, language, and the principle of charity. In Alfred R. Mele and Piers Rawling, eds. The Oxford Handbook of Rationality. 343–362. Oxford: Oxford University Press. Ludwig, Kirk, 2007. The problem of radical interpretation: Reply to Simon Evnine and Piers Rawling. Florida Philosophical Review 7: 138–146. Ludwig, Kirk, 2015. Was Davidson’s project a Carnapian explication of meaning? Journal for the History of Analytical Philosophy 3: 1–55. Lycan, William G., 1974. The extensionality of cause, space and time. Mind 83: 498– 511. Lycan, William G., 1984. Logical Form in Natural Language. Cambridge, MA: MIT Press. Mackie, J.L. 1980. The Cement of the Universe: A Study of Causation. Oxford: Clarendon Press. Malpas, J.E. 1988, The nature of interpretative charity. Dialectica 42: 17–36. Malpas, J.E. 1992. Donald Davidson and the Mirror of Meaning: Holism, Truth, Interpretation. Cambridge: Cambridge University Press. Malpas, Jeff, ed. 2011. Dialogues with Davidson: Acting, Interpreting, Understanding. Cambridge, MA./London: The MIT Press. Manning, Richard N., 2003. All facts great and small. In Gerhard Preyer, Georg Peter and Maria Ulkan, eds. Concepts of Meaning: Framing an Integrated Theory of Linguistic Behavior. 199–221. Dordrecht/Boston/London: Kluwer Academic Publishers. Manning, Richard N., 2004. Facing Facts with Davidsonian semantics. Philosophical Books 45: 111–127. McCarthy, Timothy, 2002. Radical Interpretation and Indeterminacy. Oxford: Oxford University Press. McDowell, John, 2003. Subjective, intersubjective, objective. Philosophy and Phenomenological Research 67: 675–681. McGee, Vann, 1993. A semantic conception of truth? Philosophical Topics 21: 83–111. McGinn, Colin. 1976. A note on the Frege Argument. Mind 85: 422–423. McGinn, Colin, 1977. Charity, interpretation, and belief. Journal of Philosophy 74: 521– 535. McGuire, John Michael, 2003. Davidson on metaphorical meaning: A reply to Stainton. Dialogue: Canadian Philosophical Review 42: 355–361. Montminy, Martin, 2011. Indeterminacy, incompleteness, indecision, and other semantic phenomena. Canadian Journal of Philosophy 41: 73–98. Myers, Robert H. and Claudine Verheggen, 2016. Donald Davidson’s Triangulation Argument: A Philosophical Inquiry. London: Routledge. Neale, Stephen, 1994. The place of language. Proceedings of the Aristotelian Society XCIV: 215–227.

394

turner

Neale, Stephen, 1995. The philosophical significance of Gödel’s slingshot. Mind 104: 761– 825. Neale, Stephen, 1999. From semantics to ontology, via truth, reference and quantification. In Urszula M. Ẓegleń, ed. Donald Davidson: Truth, Meaning and Knowledge. 77–87. London and New York: Routledge Neale, Stephen, 2001. Facing Facts. Oxford: Clarendon Press. Newman, Andrew, 2002. The Correspondence Theory of Truth: An Essay on the Metaphysics of Predication. Cambridge: Cambridge University Press. Olson, Kenneth Russell, 1987. An Essay on Facts. Stanford: Centre for the Study of Language and Information. Oppy, Graham, 2004. Facing facts? Australasian Journal of Philosophy 82: 621–643. Pagin, Peter, 2013a. The status of charity II: Charity, probability, and simplicity. In Maria Baghramian, ed. Donald Davidson: Life and Words. 167–189. London: Routledge. Pagin, Peter, 2013b. Radical interpretation and the principle of charity. In Ernie Lepore and Kirk Ludwig, eds. A Companion to Donald Davidson. 225–246. Oxford: Wiley Blackwell. Peacocke, Christopher, 1979. Holistic Explanation: Action, Space, Interpretation. Oxford: Clarendon Press. Pearson, James, 2011. Distinguishing W.V. Quine and Donald Davidson. Journal for the History of Analytical Philosophy 1: 1–22. Peirce, Charles Sanders, 1960a. Why study logic? In his Collected Papers of Charles Sanders Peirce; Volume I, Principles of Philosophy and Volume II, Elements of Logic. (Edited by Charles Hartshorne and Paul Weiss.) Vol. II: 67–125. Cambridge, Massachusetts: The Belknap Press of Harvard University Press. Peirce, Charles Sanders, 1960b. Lectures on pragmatism. In his Collected Papers of Charles Sanders Peirce; Volume V, Pragmatism and Pragmaticism and Volume VI, Scientific Metaphysics. (Edited by Charles Hartshorne and Paul Weiss.) Vol. V: 11–222. Cambridge, Masschusetts: The Belknap Press of Harvard University Press. Pereda, Carlos, 1998. Is there such a thing as a language? Crítica: Revista Hispanoamericana de Filosofía 30: 73–91. Perez, Diana J., 2005. Is thought without language possible? Principia: An International Journal of Epistemology 9: 177–191. Perry, John, 1996. Evading the slingshot. In Andy Clark, Jesús Ezquerro and Jesús M. Larrazabal, eds. Philosophy and Cognitive Science: Categories, Consciousness, and Reasoning. 95–114. Dordrecht/Boston/London: Kluwer Academic Publishers. Picardi, Eva, 1989. Davidson on assertion, convention and belief. In Johannes Brandl and Wolfgang L. Gombocz, eds. The Mind of Donald Davidson. 97–107. Amsterdam: Rodopi. Picardi, Eva, 1996. Convention and assertion. In B. McGuinness and G. Oliveri, eds. The Philosophy of Michael Dummett. 59–77. Dordrecht: Kluwer Academic Publishers.

remarks on davidson’s polymorphous concept of truth

395

Pietroski, Paul M., 1994. A defense of derangement. Canadian Journal of Philosophy 24: 95–117. Pires de Oliveira, Roberta, 1996. Metaphor, logical form, and event: A linguist talking to the philosopher Donald Davidson. Cadenos de Estudos Lingüísticos 31: 91–107. Platts, Mark de Bretton, 1979. Ways of Meaning: An Introduction to a Philosophy of Language. London: Routledge and Kegan Paul. Platts, Mark de Bretton, 1997. Ways of Meaning: An Introduction to a Philosophy of Language. London: The MIT Press. Preyer, Gerhard and George Peter, eds. 2006. Facts, slingshots and anti- representationalism: On Stephen Neale’s Facing Facts. Protosociology: An International Journal of Interdisciplinary Research 23: 5–171. Puolakka, Kalle, 2006. There’s a nice knockdown argument for you: Donald Davidson and modest intentionalism. Postgraduate Journal of Aesthetics 3: 15–24. Puolakka, Kalle, 2011. Relativism and Intentionalism in Interpretation: Davidson, Hermeneutics, and Pragmatism. Lanham: Lexington Books. Putnam, Hilary, 1985. A comparison of something with something else. New Literary History 17: 61–79. Quine, Willard van Orman, 1956. Quantifiers and propositional attitudes. Journal of Philosophy 53: 177–187. Quine, Willard van Orman, 1960. Word and Object. Cambridge, Massachusetts: MIT Press. Quine, Willard van Orman, 2008. Confessions of a Confirmed Extensionalist and Other Essays. (Edited by Dagfinn Follesdal and Douglas B. Quine.) London: Harvard University Press. Ramberg, Bjørn T., 1989. Donald Davidson’s Philosophy of Language: An Introduction. Oxford: Basil Blackwell. Ramberg, Bjørn T., 1999. The significance of charity. In Lewis Edwin Hahn, ed. The Philosophy of Donald Davidson. 601–618. Illinois: Open Court. Rasmussen, Joshua, 2014. Defending the Correspondence Theory of Truth. Cambridge: Cambridge University Press. Rawling, Piers. 2007. Comments on Donald Davidson: Meaning, Truth, Language, and Reality. Florida Philosophical Review 7: 128–137. Ray, Greg, 2014. Meaning and truth. Mind 123: 79–100. Read, Stephen, 1993. The slingshot argument. Logique & Analyse 143–144: 195–218. Restall, Greg, 2004. One way to face facts. The Philosophical Quarterly 54: 420–426. Rodriguez-Pereyra, Gonzalo, 1998. Searle’s correspondence theory of truth and slingshot. The Philosophical Quarterly 48: 513–522. Rodriguez-Pereyra, Gonzalo, 1999. Truthmaking and the slingshot. Beiträge des 22 Internationalen Wittgenstein Symposiums, 15–21 August. 177–184. Kirchberg am Wechsel.

396

turner

Rojszczak, Artur, 2005. From the Act of Judging to the Sentence: The Problem of Truth Bearers from Bolzano to Tarski. Edited by Jan Woleński. Dordrecht: Springer. Rorty, Richard, 1995. Is truth a goal of enquiry? Davidson vs. Wright. The Philosophical Quarterly 45: 281–300. Rorty, Richard, 1998. Is truth a goal of inquiry? Donald Davidson versus Crispin Wright. In his Truth and Progress: Philosophical Papers, Volume 3. 19–42. Cambridge: Cambridge University Press. Sainsbury, Mark, 2001. Logical Forms: An Introduction to Philosophical Logic. 2nd Edition. Oxford: Blackwell Publishers. Sainsbury, Mark, 2012. A very large fly in the ointment: Davidson truth theory contextualized. In Richard Schantz, ed. Prospects for Meaning. 223–257. Berlin/Boston: Walter de Gruyter. Saka, Paul, 2007. Spurning charity. Axiomathes 17: 197–208. Schlick, Moritz, 1934/1959. The foundation of knowledge. (Translated by David Rynin.) Reprinted in Alfred J. Ayer, ed. Logical Positivism. 209–227. Westport, Connecticut: Greenwood Press, Publishers. Searle, John, 1996. The Construction of Social Reality. London: Penguin Books. Shramko, Yaroslav and Heinrich Wansing, 2009. The slingshot argument and sentential identity. Studia Logica 91: 429–455. Sinclair, Robert, 2002. What is radical interpretation? Davidson, Fodor, and the naturalization of philosophy. Inquiry 45: 161–184. Sluiter, Ineke, 1998. Metatexts and the principle of charity. In Peter Schimmer and Marijke van der Wal, eds. Metahistoriography: Theoretical and Methodological Aspects of the Historiography of Linguistics. 11–27. Munster: Nodus Publikationen. Smith, Imogen, 2010. Interpreting Mrs Malaprop: Davidson and Communication without Conventions. MA Thesis. Department of Philosophy. University of Birmingham. Soames, Scott, 2008. Truth and meaning: In perspective. In Peter A. French and Howard K. Wettstein, eds. Truth and its Deformities. (Midwest Studies in Philosophy 32.) 1–19. Oxford and Boston: Blackwell Publishing. Solberg, Mariann, 2009. Belief, Truth and the Enigma of Error: On the Epistemology of Donald Davidson. Saarbrücken: VDM Verlag Dr. Müller. Sorensen, Roy, 2004. Charity implies meta-charity. Philosophy and Phenomenological Research 68: 290–315. Stainton, Robert J., 2011. In defense of public languages. Linguistics and Philosophy 34: 479–488. Stainton, Robert J., 2016. A deranged argument against public languages. Inquiry: An Interdisciplinary Journal of Philosophy 59: 6–32. Sterelny, Kim, 1981. Davidson on truth and reference. The Southern Journal of Philosophy 19: 95–116. Stoutenburg, Gregory, 2016. Principles of interpretive charity and the semantics of knowledge attributions. Acta Analytica 31: 153–168.

remarks on davidson’s polymorphous concept of truth

397

Stoutland, Frederick, 2003. What philosophers should know about truth and the slingshot. In Matti Sintonen, Petri Ylikoski and Kaarlo Miller, eds. Realism in Action: Essays in the Philosophy of the Social Sciences. 3–32. Dordrecht/Boston/London: Kluwer Academic Publishers. Stoutland, Frederick, 2006. Critical Notice: A mistaken view of Davidson’s legacy. International Journal of Philosophical Studies 14: 579–596. Stoutland, Frederick, 2007. Radical misinterpretation indeed: Response to Lepore and Ludwig. International Journal of Philosophical Studies 15: 587–597. Sundholm, Göran, 1984–1985. Brouwer’s anticipation of the principle of charity. Proceedings of the Aristotelian Society 85: 263–276. Talmage, C.J.L., 1994. Literal meaning, conventional meaning and first meaning. Erkenntnis 40: 213–225. Tarski, Alfred, 1944. The semantic conception of truth and the foundations of semantics. Philosophy and Phenomenological Research 4: 341–376. Tarski, Alfred, 1983. Logic, Semantics, Metamathematics: Papers from 1923 to 1938. (Translated by J.H. Woodger. Second edition edited and introduced by John Corcoran.) Indianapolis, Indiana: Hackett Publishing Company, Inc. Taylor, Barry, 1985. Modes of Occurrence: Verbs, Adverbs and Events. Oxford: Basil Blackwell. Taylor, Kenneth, 1985. Davidson’s theory of meaning: Some questions. Philosophical Studies 48: 91–105. Vermazen, Bruce, 1982, General beliefs and the principle of charity. Philosophical Studies 42: 111–118. Vision, Gerald, 2004. Veritas: The Correspondence Theory and its Critics. London: MIT Press. Wagner, Steven J., 1986. California semantics meets the Great Fact. Notre Dame Journal of Formal Logic 27: 430–455. Wallace, Meg, 2015. Rearming the slingshot. Acta Analytica 30: 283–292. Warmbröd, Ken, 1991. The need for charity in semantics. Philosophical Review 100: 431– 458. Weiss, Bernard, 2010. How to Understand Language: A Philosophical Inquiry. Durham: Acumen. Wheeler III, Samuel C., 2014. Neo-Davidsonian Metaphysics: From the True to the Good. New York/London: Routledge Taylor and Francis Group. Widerker, David, 1985. Davidson on singular causal sentences. Erkenntnis 23: 223–242. Wiggins, David, 1997. Meaning and truth conditions: From Frege’s grand design to Davidson’s. In Bob Hale and Crispin Wright, eds. A Companion to the Philosophy of Language. 3–28. Oxford: Blackwell Publishers. Wittgenstein, Ludwig, 1922. Tractatus Logico-Philosophicus. (With an Introduction by Bertrand Russell, F.R.S.) London: Routledge and Kegan Paul Ltd.

398

turner

Wittgenstein, Ludwig, 1958. Philosophical Investigations. (Translated by G.E.M. Anscombe.) 2nd Edition. Oxford: Basil Blackwell. Wong, Wai-Hung, 1999. Interpretive charity, massive disagreement, and imagination. Canadian Journal of Philosophy 29: 49–74. Wright, Crispin, 1993. Theories of meaning and speakers’ knowledge. In his Realism, Meaning and Truth. 2nd Edition. 204–238. Oxford: Blackwell Publishers. Young, James, 2002. The slingshot argument and the correspondence theory of truth. Acta Analytica 17: 121–132. Zilhão, António, 2003. From radical translation to radical interpretation and back. Principia 7: 229–249.

Index of Names Abelard 7 Aristotle x, xiii, 86 Atlas, Charles 4 (see also Siciliano, Angelo) Atlas, Jay David xiv–xv, 83–84, 87–89, 90–91, 92–93, 94–95, 96–98, 195–196, 100–103, 149–150, 151–152, 153–155, 158, 159–160, 165–166, 170–174, 210, 212, 215, 261, 301, 323, 325, 332, 349 Austin, John L. 233 Bach, Kent xi, 341–342 Bloomfield, Leonard 197–198, 209, 380 Burley, Walter 7, 10 Carston, Robyn 195, 204 Clark, Andy 285–286, 287 Davidson, Donald 170, 174–176, 178–180, 181– 183, 186–187, 191, 350–383 (passim) Evans, Gareth xiii Frege, Gottlob ix, x–xi, xvii, 83, 99–100, 353, 372, 379 Grice, Paul 184, 187, 188, 203 Harnish, Robert 341–342 Haspelmath, Martin 112–113, 127 Hegel, Georg Wilhelm Friedrich 19 Horn, Laurence 55–56, 60–62, 69–70, 72–73, 78, 80, 92, 93, 200, 201, 202, 205 Huge, Harold (see also Snowden, Bruce) 205–206

Jespersen, Otto 42, 108, 115, 282 Joyce, James 376–377 Kaplan, David 156, 200, 214–215, 217–218, 200, 234, 235–237, 254, 255 Moore, G.E. 323–346 (passim) Morgan, Jerry 95, 200, 201 Peano, Giuseppe 198, 209 Peirce, Charles Sanders 379 Perry, John 214–215 Peter of Spain 7 Quine, Willard van Orman 370, 380 Recanati, François 150–151, 184–185, 203 Rorty, Richard 174, 175–178, 179, 191 Russell, Bertrand xiii–xiv, 82–83 Shoemaker, Sidney 323, 326 Siciliano, Angelo (See also Atlas, Charles) 4 Snowdon, Bruce (see also Huge, Harold) 205–206 Stern, Josef 170 Strawson, Peter 83, 93 Tarski, Alfred 352, 363, 364, 377–378, 380– 381 William of Sherwood 7, 17

Index of Subjects Acquisition, first language 291–292, 293–294 Adaptive, language as 289–290 Ambiguity xiv–xv, xvii, 84–88, 91–94, 155, 158–159, 162–165 (See also Disambiguation; Generality; Occam’s Razor (Modified)) Anaphora 33–35 Assertoric inertia 12, 17–18, 27, 31, 41 Atlas-Kempson Thesis 18 At least, at most 17, 23 Belief 330–331, 371–372 Brain As predictive engine 285–286 Shared 286 Social 294–295 Cancellation 7–8, 19–22, 24, 27–31, 36–38, 41, 43 Cardinality 198–199, 204 Clefts 24–31, 36, 40–41 Coherence 158, 160–161, 162–164 (See also Rhetorical relations) Commitments (see Speech acts) Compositionality 21, 99, 180, 201, 223, 268, 273, 290, 294 Compset vs. refset 34–35 Conjunction Reduction Identity Test (CRIT) xv–xvi, 93–94 Content Evolving 266–270 Propositional 150, 173, 175, 182, 191, 268 Context 157 Dependence x, xv, 16–17, 19, 28, 38, 155 Dynamic 262–264 Evolving 157–158 Counting 199, 209 (See also Number words) Denoting, Theory of xiii Definite descriptions 82–83, 87, 89–91, 92– 93, 100, 103, 355 Theory of xiii–xiv Demonstrative 218–219, 223–224 Disambiguation 155, 159–160, 162–165 (See also Ambiguity) Discourse 153, 157–159, 160–166

Disjunction x, xiii, 301, 309–313 Dynamic update 318–320 Entailment 6–8, 14, 22–24, 27, 36, 38, 40–41 Downward 8, 11–12 Erotetic Theory 300, 301–302, 308, 313–320 Existence xiv, 83–84, 91, 93–94, 96–97, 99– 101 Nonexistence 99–103 Fallacies 301, 310 Few 31–41 Generality Semantic xiv–xvi, xvii, 83–84, 88–89, 170, 171, 190–191, 202–203 (See also Ambiguity) Goldilocks Problem xi Horn clauses 55–56, 57–60, 62–63, 67, 70, 71–72, 76–78 Implicature Conventional xi, 8, 14, 22, 24, 29–30 Conversational xi, 6–8, 14, 19, 22, 25, 27, 29–30, 36–38, 40–41, 96–98, 173, 184–185 Scalar 22, 36–38, 41 Indefinites 107, 109–111, 116, 118, 121, 125–130, 133, 135–137, 139, 303 Indexical x–xii, 152–153, 155–157, 159, 169, 181, 183–184, 212–215, 218–219, 235–237, 248, 254, 278, 303 Inference Abductive 153, 162–163, 173, 292 Higher order 270, 278, 284, 288, 289–290 Intellectual acupuncture 45 Intention 151, 156–158, 160, 162–163, 165, 175– 176, 179, 183–184, 188, 217, 218–222, 225, 232, 240–242, 269–270, 278, 294, 327, 333, 336–340, 342–345 Intentionalism 152–153, 181–182, 187, 216 Interaction Conversational 288 Grammar as a tool for 267, 270 Interpretation, Radical 350, 352, 367–370 Islands 56–57, 71–72, 77

401

index of subjects Know Non-factive

55–56, 60–62, 69–70

Logic Bivalent xiii Gapped bivalent xiii, 3 Extended Term 42–43 Fregean ix, xi, xvii Post-Fregean xiv, xvi, 42 Trivalent xiii Logical form xiv–xvii, 82, 87–89, 96, 150–151, 153, 159, 162, 183, 354 Meaning Linguistic 149–152, 156, 166, 213–215, 219, 223–231, 234, 239–254, 255 Mein Herr Problem xii Metaphor 169–192 (passim), 201 Minimalism Semantic 151–152, 153, 154, 170, 181–183, 190–191 Models Mental models 300, 301–302, 308 Moore’s Paradox 323–346 (passim) Negation Classical NEG Raising 54–80 (passim) External (wide-scope) xvi, 43 Internal (narrow scope) xvi Klima diagnostics 31 Logic of 42–44 Metalinguistic 3, 28, 41–42 NEG deletion 66–69 Negative concord 43, 107–109, 111, 113, 115 Negative First 115 Negative indefinites 107–140 (passim) Negative inversion 7, 12, 57–59, 63 Negative polarity items (NPIs) 4–5, 7–14, 18, 25, 27, 35–36 Negative quantification 107–140 (passim) Number words 195–196, 197–199, 204–206, 209–210 (See also Counting) Occam’s Razor (Modified) 3, 155, 160, 162 (See also Ambiguity) Only 4–24, 26–36, 38–41

Parentheticals 32, 72–73, 75–76 Position Postverbal 120 Pragmatics ix, x, xvii, 3, 17, 18, 97, 103, 149– 151, 154–155, 180, 191–192, 198, 200, 203, 212, 255, 268, 289, 311, 377, 381–383 Radical xvii, 195, 197 Prediction Error 285, 288 Of content 271, 285–289 Presupposition xiii, xvii, 3, 7–8, 10–12, 14, 16, 18, 29, 36–37, 43, 88, 94, 98–100, 185, 189, 342, 344 Procedural Knowledge 261 Proper names 87, 99–100, 103, 216, 303 Questions 313, 316 Rationality 161, 184, 191 Irrationality 324, 326–327, 329, 330, 331– 333, 345–346 Reasoning x, xiii, 150–152, 156, 159–160, 162– 163, 165, 187, 203, 270, 300–320 (passim) Rhetorical relations 157–159 (See also Coherence) Scale reversal 202, 204–209 Scope 12, 14, 16, 25, 42, 54, 59, 66, 84–89, 95– 96, 98–99, 154, 269 Semantics ix, xii, xvii, 3, 5, 12, 17, 40, 149–150, 164–165, 169, 180–183, 191–192, 195, 200, 212–219, 221–229, 231–235, 237–241, 245– 247, 253–255, 268, 313, 315, 320, 363, 377, 381 (See also Minimalism, Semantic) Alternative 313, 315 Inquisitive 313, 315 Tarski-style 350 Slingshot Argument 353–357 Speech acts 201–202, 233–234, 269–270, 324–325, 334 Assertion 12, 14, 16–18, 38, 170, 325–326 Demanding 338 Promising 332–333, 334–335, 336, 343 Swearing 329–330, 334, 338, 340 Performatives 12, 201 Felicity conditions 337–339 Commitments 324–325, 334, 335–337, 339–344

402

index of subjects

Insincerity 338, 341, 345 Strawson-entailment 11–12, 14 Synonymy 213, 221, 247, 250–251, 253, 255

Tarski-style definition of Value gaps xiii, 219 Typology 107, 120, 121, 129

377–381

Topic Sentence 90–91, 100–102 Truth Coherence theory of 365–366 Correspondence theory of 352–353, 364– 365

Underspecification, (Underspecificity) 150– 151, 153–155, 166, 271–272, 273–274 Structural 273 Utterances (and their semantics) 212–255 (passim)