Philosophical and Formal Approaches to Linguistic Analysis 9783110320244, 9783110320015

Citation preview

Piotr Stalmaszczyk (Ed.) Philosophical and Formal Approaches to Linguistic Analysis

Piotr Stalmaszczyk (Ed.)

Philosophical and Formal Approaches to Linguistic Analysis

Bibliographic information published by Deutsche Nationalbibliothek The Deutsche Nastionalbibliothek lists this publication in the Deutsche Nationalbibliographie; detailed bibliographic data is available in the Internet at http://dnb.ddb.de

North and South America by Transaction Books Rutgers University Piscataway, NJ 08854-8042 [email protected] United Kingdom, Ire, Iceland, Turkey, Malta, Portugal by Gazelle Books Services Limited White Cross Mills Hightown LANCASTER, LA1 4XS [email protected]

Livraison pour la France et la Belgique: Librairie Philosophique J.Vrin 6, place de la Sorbonne ; F-75005 PARIS Tel. +33 (0)1 43 54 03 47 ; Fax +33 (0)1 43 54 48 18 www.vrin.fr

2012 ontos verlag P.O. Box 15 41, D-63133 Heusenstamm www.ontosverlag.com ISBN 978-3-86838-142-9 2012 No part of this book may be reproduced, stored in retrieval systems or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use of the purchaser of the work Printed on acid-free paper ISO-Norm 970-6 FSC-certified (Forest Stewardship Council) This hardcover binding meets the International Library standard Printed in Germany by CPI buch bücher.de

Contents Philosophical and Formal Approaches to Linguistic Analysis. Preface Piotr Stalmaszczyk ............................................................................... 1 Boghossian’s Implicit Definition Template Ben Baker .......................................................................................... 15 The Two One Fallacy Theory Theory David Botting..................................................................................... 37 Modal Games in Natural Language Elżbieta Chrzanowska-Kluczewska ................................................... 57 A Problem with Structured Propositions Tadeusz Ciecierski ............................................................................. 81 Does Intensional Semantics Account for ‘Travis Cases’? Alex S. Davies .................................................................................... 87 Chomsky’s Methodological Naturalism and the Mereological Fallacy Florian Demont ............................................................................... 113 Kripke and Mill on Natural Kind Terms Luis Fernández Moreno ................................................................... 127 In Defence of Axiomatic Semantics Chris Fox, Raymond Turner ............................................................ 145 Identity of Linguistic Expressions and Lexical Synonymy in the Fields of Logical Semantics, Linguistic Semantics, and ‘Pragmatic Semantics’ Barbora Geistová Čakovská ............................................................ 161 Structured Propositions and Shared Content Thomas Hodgson ............................................................................. 177 In Defence of Singular Propositions Filip Kawczyński.............................................................................. 197

ii Context and Compositionality: the Challenge of Conversational Dialogue Ruth Kempson, Eleni Gregoromichelaki, Ronnie Cann ................... 215 Deferred Reference and Descriptive Indexicals. Mixed Cases Katarzyna Kijania-Placek ................................................................ 241 Proper Names: Denying Semantic Uniformity David Kirkby .................................................................................... 263 Taking (It) Seriously: Normativity of Meaning Joanna Klimczyk .............................................................................. 273 Accommodating Abstracta in Naturalist Accounts of Meaning Katarzyna Kobos.............................................................................. 295 Ambiguous Conditionals Karolina Krzyżanowska ................................................................... 315 Defining Linear Dynamic Syntax Agnieszka Kułacka ........................................................................... 333 Towards a Non-referential Semantics Janusz Maciaszek ............................................................................. 357 Are Propositions Facts? Andrei Mărăşoiu .............................................................................. 385 Judgement-Dependence, Tacit Knowledge and Linguistic Understanding Alexander Miller .............................................................................. 405 Making Events Redundant: Adnominal Modification and Phases Ulrich Reichard ............................................................................... 429 A Philosophical Reflection on Language: Is Ontology Needed in Semantics? Mª Uxía Rivas Monroy ..................................................................... 457 In Defense of Implicit Knowledge in a Full-Blooded Theory of Meaning Ali Saboohi....................................................................................... 477

iii On Pragmatic Regularities Arthur Sullivan ................................................................................ 491 A New Perspective for the Theory of Intentionality Mieszko Tałasiewicz ........................................................................ 513 A Conceptual Role Semantics for Attitude Reports Tomoo Ueda .................................................................................... 527 Index of Names .................................................................................. 551 Index of Subjects ............................................................................... 553

Piotr Stalmaszczyk University of Łódź [email protected]

Philosophical and Formal Approaches to Linguistic Analysis. Preface 1. Introduction The Chair of English and General Linguistics (University of Łódź, Poland) organized in May 2009 the first International Conference on Philosophy of Language and Linguistics, PhiLang2009. The principal aim of that conference was to bring together linguists, philosophers and logicians with interest in analysis of natural language. Papers presented during that conference investigated different turns in contemporary philosophy of language: ‘formal’, ‘philosophical’, and ‘cognitive’.1 The name most often referred to during the conference was undoubtedly that of the German logician, Gottlob Frege. His legacy has proved to be of unquestionable importance for contemporary philosophers of language and linguists alike. Also the continuing influence of Bertrand Russell, Ludwig Wittgenstein (early and late), Kazimierz Ajdukiewicz, Alfred Tarski, William Orman Van Quine, Donald Davidson, Noam Chomsky and Saul Kripke was clearly visible in numerous presentations and discussions. The second edition of the conference, PhiLang2011, was organized in May 2011, with contributions centering once again around different philosophical and formal approaches to language and linguistic analysis. Articles gathered in the present volume tackle several issues highly relevant for traditional and contemporary debates within the philosophy of language, and the interfaces between linguistics, philosophy and logic. 1

See the contributions in Stalmaszczyk, ed. (2010a, 2010b, 2011).

2 The topics include analytic accounts of the a priori and implicit definitions (Baker), medieval and contemporary theories of fallacy (Botting), game-theoretical semantics, modal games in natural language and literary semantics (Chrzanowska-Kluczewska), possible-world theories and paradoxes involving structured propositions (Ciecierski), identity sentences and intensional semantics (Davies), Chomsky’s methodological naturalism (Demont), John Stuart Mill and Saul Kripke on general and natural kind terms (Fernández Moreno), axiomatic semantics (Fox and Turner), logical and linguistic semantics (Geistová Čakovská), structured propositions and shared content (Hodgson), singular propositions (Kawczyński), Dynamic Syntax (Kempson, Gregoromichelaki and Cann), uses of indexicals (Kijania-Placek), semantics of proper names (Kirkby), normativity of meaning (Joanna Klimczyk), ambiguous conditionals (Krzyżanowska), Linear Dynamic Syntax (Kułacka), non-referential semantics (Maciaszek), unity of propositions (Mărăşoiu), judgement-dependence, tacit knowledge and linguistic understanding (Miller), event semantics and phasal architecture of grammar (Reichard), ontology in semantics (Rivas Monroy), implicit knowledge and theory of meaning (Saboohi), pragmatic regularities (Sullivan), theory of intentionality (Tałasiewicz), conceptual role semantics for attitude reports (Ueda).2 As this multitude of topics shows, the convergence of linguistic, philosophical, formal and cognitive approaches opens exciting new research perspectives within contemporary philosophy of language. 2. Detailed contents of the volume Ben Baker discusses Paul Boghossian’s implicit definition template. In Boghossian’s account, a priori knowledge of basic logical principles is available by inference from knowledge of their role in determining the meaning of the logical constants by implicit definition together with knowledge of the meanings so-determined that we possess through our 2

Additionally, selected papers from the conference appear in Lodz Papers in Pragmatics 7.2 (Special issue on context and contextualism), and in Kwartalnik Neofilologiczny LVIII/3.

3 privileged access to meaning (cf. Boghossian 1997). As noted by Baker, some commentators have objected that if the thesis of implicit definition on which Boghossian relies were true, knowledge of the meaning of the constants would presuppose knowledge of the very logical principles knowledge of which the account purports to explain. A consequence would seem to be that implicit definition is incompatible with privileged access. Baker argues that whilst it is possible for Boghossian to defend against these objections, the form of argument he proposes does exhibit a subtle form of question begging such that it exhibits a transmission of warrant-failure. David Botting describes two theories of fallacy. One is the medieval theory of Jean Buridan, the other the “one fallacy theory” set forth by Lawrence Powers (1995), with the basic claim that all fallacies are instances of the fallacy of equivocation. Analysis of the two theories leads Botting to his main claim that fallacy theory has from the very beginning treated two questions as if they were one: “what is a fallacy?” and “why do fallacies occur?” The Aristotelian division of fallacies into those depending on language and those not depending on language is reinterpreted as being addressed to these different questions. Powers’ theory that all fallacies are equivocations is correct when reinterpreted as an answer to the second question. The first question is also to be answered in terms of one fallacy, namely ignorantia elenchi, following Aristotle’s own procedure in the Sophistical Refutations. According to Botting, there is a one fallacy theory for each question, hence his contribution is titled “The Two One Fallacy Theory Theory”. Elżbieta Chrzanowska-Kluczewska discusses Modal Games in natural language. She observes that Ludwig Wittgenstein never approached the issue of language-games played with modal expressions in much detail. The problem was taken up in the late 1960’s by Jaakko Hintikka and his collaborators within Game-Theoretical Semantics (GTS), which created appropriate logical environment for the discussion of modality. However, even within the framework of GTS, modality has never received a holistic treatment. The focus of GTS has been predominantly on epistemic modal games with a propositional scope. Discourse games, in the shape proposed by Lauri Carlson (1983), have not centred on modal issues. The article tries to complement this

4 theoretical gap by focusing on the concept of narrative modalities as initiated by Lubomir Doležel (1976), and practised ever since within the framework of literary semantics and text-world theories. Tadeusz Ciecierski’s paper is concerned with structured propositions and some possible paradoxes. He shows that the paradox of the totality of propositions rests on assumptions characteristic of some theories of structured contents. The threat of possible paradoxes, contrary to the widespread opinion, is not solely connected with possible-world theories of content. Moreover, possible-world theories – due to the rejection of structural constraints put on contents – can avoid the paradox in quite an elementary way. Alex Davies concentrates on a certain type of identity statements, known as the ‘Travis Cases’ (cf. Travis 2008), in which a change in truth-value occurs. Several accounts have suggested that the appropriate response to this kind of linguistic phenomenon is to produce a formal semantic analysis that eliminates the phenomenon. However, the empirical debate has stymied because analyses are inevitably incomplete: they are not specific enough for it to be checked whether Travis cases can be constructed for them. So it cannot be checked whether Travis cases have really been eliminated. Davies adopts a different approach. He sketches a genealogy of Travis cases and assumes that if the genealogy is correct then the elimination of Travis cases is not possible nor is it a means to a better understanding of the phenomenon. As a test case, he discusses an account of Travis cases that invokes features of intensional semantics. Florian Demont provides a critical analysis of Noam Chomsky’s methodological naturalism. He observes that ever since the publication of Cartesian Linguistics (1966), Chomsky has been eager to show that his vision of psychology in general and theoretical linguistics in particular is conceptually sound. The claim has been opposed from various sides, including more philosophically-oriented objections by Baker and Hacker (1984) and Wright (1989). Demont’s paper considers one particular question in relation to these efforts: Is Chomsky’s methodology in psychology and theoretical linguistics conceptually sound? The answer he arrives at suggests that whereas Chomsky’s

5 methodological naturalism appears to be sound, his particular brand of internalism is shown to be pernicious. Luis Fernández Moreno compares in his contribution Saul Kripke’s and John S. Mill’s views on natural kind terms. In a famous passage of the third lecture of Naming and Necessity (1980) Kripke summarizes the core of his criticism to the description theory of natural kind terms, taking into account the theory of general terms proposed by Mill’s System of Logic, insofar as it is applied to natural kind terms, as a paradigmatic version of that sort of theory. The aim of Fernández Moreno’s paper is to argue that Mill’s generic theory of general terms does not coincide with his theory concerning the type of general terms that natural kind terms are and that the main thesis of the latter is not subject to Kripke’s objections put forward in the aforementioned passage. Chris Fox and Raymond Turner observe that many semantic theories implicitly attribute a foundational status to set theory, and settheoretic characterisations of possible worlds in particular. The goal of a semantic theory is then to find a translation of the phenomena of interest into a set-theoretic model. The authors contrast this kind of theory with an axiomatic approach in which “new” primitives and ontological categories can be formulated, and logical rules and axioms that capture the appropriate inferential behaviour can devised. However, they also observe that this alternative approach might be criticised as being mere “descriptivism”, lacking predictive or explanatory power. In their paper, they defend the axiomatic approach. Any formal account must assume some intended interpretation, but there is a sense in which such theories can provide a more honest characterisation. In contrast, the set-theoretic approach tends to conflate distinct ontological notions. Mapping a pattern of semantic behaviour into some pre-existing set-theoretic behaviour may lead to certain aspects of meaning being overlooked, or ignored. According to Fox and Turner the insights, and conceptual categories used in developing a set-theoretic interpretation, should themselves all be considered part of the semantic analysis. If these behavioural and ontological insights were to be internalised into an adequate formal analysis, then it would be more appropriate to adopt an axiomatic approach.

6 Barbora Geistová Čakovská focuses on two basic semantic notions: linguistic expression (word) and relation of synonymy (lexical synonymy). She analyses these notions with respect to three semantic fields: logic, linguistics, and “pragmatic semantics” (as advocated, for instance, by Recanati). She further shows that the pragmatic approach fails to satisfactorily explain both the notion of linguistic expression and the notion of synonymy – two very basic semantic concepts – whereas satisfactory explanations can be derived from the approaches of logical and linguistic semantics. Thomas Hodgson devotes his contribution to structured propositions and shared content. A theory recently advanced by Jeffrey C. King (2007) identifies the structure of a structured proposition with the logical form of a sentence expressing it. Hodgson argues that a theory such as King’s cannot work. He demonstrates that any theory entailing that no two sentences that have different logical forms can express the same proposition is vulnerable both to an objection based on entailment relations between attitude-reports, and to a more general objection based on the notion of shared content. Filip Kawczyński discusses singular propositions and compares two approaches, those offered in Kaplan (1977/1989) and Stanley (2002). Stanley argues that propositions in general are not the bearers of modal properties, and thus he refutes one of the major arguments in favour of singular propositions (offered by Kaplan). Kawczyński’s aim is to preserve singular propositions from the attack carried out by Stanley; in doing so he demonstrates that Stanley’s reasoning is fallacious, since the Expression-Communication Principle which is the basis for his argument suffers from being circular. Ruth Kempson, Eleni Gregoromichelaki and Ronnie Cann present an innovative linguistic framework, Dynamic Syntax, in which natural-language syntax is defined as procedures for context-dependent interpretation. The framework provides a formalism where both representations of content and context are defined dynamically and structurally, with time-linear monotonic growth across sequences of partial trees as the core structure-inducing notion. Application of this framework to the resolution of elliptical phenomena, both inter- and intra-sentential, provides an integrated account of ellipsis construal.

7 In addition, this intrinsically dynamic perspective extends naturally to the modelling of dialogue exchanges with free shifting of role between speaking and hearing (split-utterances). The authors argue that the success in capturing dialogue patterns of ellipsis within an overall account of ellipsis leads to a novel composite concept of compositionality and to reconsideration of the relation between the language capacity and more general cognitive capacities. Katarzyna Kijania-Placek is concerned in her contribution with descriptive and deferred uses of indexicals. She argues that, contrary to prevalent opinion, the former do not form a class of the latter. She proposes an analysis of descriptive uses by a postulated mechanism of descriptive anaphora. Having separated these two phenomena, she further focuses on cases that seem to require an analysis combining both deferred reference and descriptive anaphora in the interpretation of one use of a pronoun. David Kirkby concentrates on the semantics of proper names. However, rather than arguing in favour of a particular position regarding their semantics, he points out that the whole debate rests upon two relatively unrecognised presuppositions. One of these, implicit in Matushansky (2008), explicit in Fara (2010), is that proper names are semantically uniform, and that they all make the same kind of propositional contribution. Kirkby argues that this presupposition is false. Joanna Klimczyk argues for rejecting the narrow understanding of normativity adopted by semantic antinormativists. She suggests an alternative wide reading of normativity that much more fits our commonsensical views on the nature of prescriptivity. The main idea lying behind the author’s understanding of normativity is that this notion does not primarily concern what ought to be done but rather what is implied by having categorical reason for some behaviour. Under the proposed view, the putative normativity is primarily a phenomenon of thought, not of action. On this account, the thesis that meaning is normative commits one to nothing over and above taking meaning seriously, where this expression is to be read as a special and mostly tacit attitude towards the social fact that words or expressions have fixed correctness conditions. Further on, Klimczyk shows that the core arguments put forward by semantic antinormativists are based on a tacit

8 acceptance of several assumptions that are either mistaken or in principle inaccessible to antinormativists. Katarzyna Kobos analyses explanations of abstractness from two ends of the spectrum of naturalist views, and points to the difficulties involved. An insufficient account of abstracta is the main charge against extreme-end empiricists who endorse a view of perception-based concepts. Their views are the main focus of the paper as what they have to say on abstractness is deflationary but not much of an eye-opener as it follows logically from the initial assumptions. After considering two eliminativist takes on abstractness, Kobos contributes to the debate by providing the rationale for a novel naturalist approach to the notion. Karolina Krzyżanowska focuses on ambiguous conditionals. According to the principle of Conditional Non-Contradiction (CNC), unless p is impossible, conditionals “If p, then q” and “If p, then not q” are jointly inconsistent. Although intuitively appealing, CNC gives rise to serious problems that semantic theories of conditionals validating it have to face. Most notably, an example of violation of CNC, as presented by Allan Gibbard (1981), may lead to the conclusion that conditionals do not express propositions at all. Krzyżanowska proposes a new analysis of Gibbard’s argument, showing that the violation of CNC is only apparent. Subsequently, she suggests a new way of defining truth conditions for conditional sentences. On her proposal, conditionals, although context-dependent, are not more subjective or obscure than sentences with scope-ambiguous quantifiers or indexicals. Agnieszka Kułacka presents a new model of interpreting natural language phrases based on Labelled Deductive Systems (as in Gabbay 1996) and a Montague grammar, as described in Kułacka (2011). She calls her model Linear Dynamic Syntax, and presents it as a logic which models natural language processing. In the paper Linear Dynamic Syntax rules are exemplified by both grammatical and ungrammatical sequences. Janusz Maciaszek argues in favour of non-referential semantics of natural language. He indicates some ontological difficulties that face referential theories and argues that a satisfactory theory of meaning of natural language should not be referential because it does not explain the meaningfulness of natural language expressions. As an illustration of a

9 possible way of constructing semantics without reference, the author considers Russell’s theory of descriptions, which is referential but can be considered as the first step towards non-referential semantics, and three classical non-referential solutions: Ajdukiewicz’s directival theory of meaning, Quine’s theory of meaning in terms of a manual of translation, and Davidson’s theory of interpretation based on Tarski’s theory of truth. Andrei Mărăşoiu explores in his paper Jeffrey King’s theory of propositions (King 2007, 2009). He first presents the problem of unity of the proposition, next King’s doubtful identification of propositions with certain existentially generalized facts and the objections to that identification read as a substantive metaphysical claim. Finally, he argues in favour of a deflationist approach: we can better understand propositions by attempting to ramsify them – to provide substitutes that do all the explanatory work propositions do, but are not metaphysically dubious. Mărăşoiu argues that King’s claim to identify propositions with facts is better interpreted as an example of such a ramsification project, and not in a metaphysically substantive way. Alexander Miller devotes his contribution to issues involved in judgement-dependence, tacit knowledge and linguistic understanding. He notes that Crispin Wright (1989) argued that a judgement-dependent conception of meaning may be in some tension with the project of giving a cognitive-psychological explanation of speakers’ abilities to understand novel utterances. Miller broadens Wright’s bipartite distinction between judgement-independent and judgement-dependent properties into a tripartite distinction between judgement-independent, strongly judgement-dependent and weakly judgement-dependent properties. He further explores the consequences for this idea of the so-called “Doctrine of Essential Linguistic Structure”. Ulrich Reichard discusses adnominal modification and phases in the context of Davidson’s (1967) event semantics. He observes that in the last two decades, Davidson’s event-argument hypothesis has become very popular in natural language semantics. His article questions that event-based analyses actually add something to our understanding of the respective phenomena: Reichard argues that they already find their explanation in independently motivated grammatical assumptions and

10 principles which apply to all kinds of modification. Apart from a short discussion of Davidson’s original arguments in favour of his hypothesis, Reichard also addresses Larson’s event-based account of the distinctions between stage-level vs. individual-level modification and adverbial vs. adjectival modification in the nominal domain. He argues that his analysis of the former reduces straightforwardly to the grammatical structure of the nominal phase. Reichard provides reasons which motivate a re-description of the phenomenon in terms of sensitivity to descriptive content rather than events. So described the phenomenon can be explained in terms of interface conditions, given a phasal architecture of grammar. María Uxía Rivas Monroy provides a philosophical reflection on language and investigates the need for ontology in semantics. She observes that Frege was concerned about the ontological status of senses, which he located in the third realm, opening thus the way to reifying notions such as meaning. Late Wittgenstein, Quine and Davidson did not accept such reification, and they referred to pragmatic features in order to explain meaning. Rivas Monroy focuses on the justification of the ontological status of senses provided by Frege. Her point of view is that pragmatic accounts of meaning are possible just because semantic proposals, such as Frege’s, could explain the cognitive content of language. Ali Saboohi defends the role of implicit knowledge in a full-blooded theory of meaning. Dummett’s “full-blooded” theory of meaning (e.g. Dummett 1993) is grounded in Frege’s work in the philosophy of language, particularly in Frege’s criticism of psychologism in logic and language. Saboohi defends this theory against McDowell’s (1998) objections. In particular, he argues that McDowell’s claims against Dummett’s invocation of the notion of implicit knowledge – the central notion in a full-blooded theory – cannot be sustained. Arthur Sullivan discusses pragmatic regularities, since, as he claims, pragmatic regularities may constitute an interesting and useful pragmatic natural kind which merits further investigation. The aim of his contribution is to help illuminate a relatively neglected aspect of a Gricean picture of the semantics-pragmatics interface: namely, the notion of a pragmatic regularity, or, to use Bach’s (2004) terminology,

11 “standardized non-literality”. Sullivan focuses on the debate between Bach (2004, 2007) and Devitt (2004, 2007) over whether the phenomenon of referential uses of descriptions is amenable to a pragmatic explanation. The guiding idea of the paper is that better charting the notion of a pragmatic regularity will have significant effects on some ongoing debates. Mieszko Tałasiewicz offers a new perspective for the theory of intentionality. The aim of his paper is to initiate a naturalistic approach to the concept. According to the author, intentionality in the primary sense is a property of some physical, perceptible objects such as arrows and sticks, but also of gestures or glances (gaze), relative to subjects that perceive these objects – much as colors are properties of some physical objects relative to perceptual abilities of subjects. Human beings are naturally equipped with ability to perceive some objects as directed somewhere, or intentional. This ability is prior to and required by first language acquisition. Tałasiewicz argues that intentionality of linguistic expressions is derived from this natural primitive intentionality, or more exactly: human ability to perceive some material entities – collections of ink molecules or sounds – as meaningful expressions is derived from ability to perceive glances and gestures as directed. Tomoo Ueda presents a conceptual role semantics for attitude reports. Conceptual role semantics (CRS) is a semantic and pragmatic theory determining truth-evaluable contents of each utterance. Reports of propositional attitudes have two components. To use Sternberg’s terminology (1982), one is a frame and the other is an inset. The paper focuses on the relation between them and aims at individuating the content expressed by insets. Ueda examines examples from English and Japanese where the indirectness of speech reports is characterised in terms of deixis. Next, he argues that there are two sorts of opacity. One is based on Frege’s discussion of modes of presentation, the other arises from evaluating attitude reports in the discursive context. Finally, both sorts of opacity are explained in terms of one-factor CRS. Ueda’s version of CRS postulates the inferential and functional nexus in the discursive context.

12 Acknowledgments Papers gathered in this volume were submitted and, in most cases, presented at the second International Conference on Philosophy of Language and Linguistics, PhiLang2011. The conference was held in Łódź in May 2011, and organized by the Chair of English and General Linguistics at the University of Łódź. I am grateful to all invited participants for stimulating presentations and discussions. I am especially grateful to Mr Ryszard Rasiński for comprehensive editorial assistance, and to Dr Piotr Duchnowicz who prepared the final manuscript. Finally, I wish to thank Dr Rafael Hüntelmann for advice and accepting the volume for publication with Ontos Verlag. References Bach, Kent 2004. Descriptions: Points of Reference. In: Anne Bezuidenhuit and Marga Reimer (eds.), Descriptions and Beyond. Oxford: Oxford University Press, 189-229. Bach, Kent 2007. Referentially Used Descriptions. European Journal of Analytic Philosophy 3, 33-48. Baker, Gordon and Hacker, Peter 1984. Language, Sense and Nonsense. Oxford: Blackwell. Boghossian, Paul 1997. Analyticity. In: Bob Hale and Crispin Wright (eds.), A Companion to the Philosophy of Language. Oxford: Basil Blackwell, 331368. Carlson, Lauri. 1983. Dialogue Games. An Approach to Discourse Analysis. Dordrecht: Reidel. Chomsky, Noam 1966. Cartesian Linguistics. New York and London: Harper and Row. Davidson, Donald 1967. The Logical Form of Action Sentences. In: Essays on Actions and Events. Oxford: Oxford University Press (2001), 105-122. Devitt, Michael 2004. The Case for Referential Descriptions. In: Anne Bezuidenhuit and Marga Reimer (eds.), Descriptions and Beyond. Oxford: Oxford University Press, 280-305. Devitt, Michael 2007. Referential Descriptions and Conversational Implicatures. European Journal of Analytic Philosophy 3, 7-32.

13 Doležel, Lubomir 1976. Narrative Modalities. Journal of Literary Semantics V/1, 514. Dummett, Michael 1993. The Seas of Language. Oxford: Oxford University Press. Fara, Delia Graff 2010. Names as Predicates (unpublished draft). Gabbay, Dov M. 1996. Labelled Deductive Systems. Oxford: Oxford University Press. Gibbard, Allan 1981. Two Recent Theories of Conditionals. In: William L. Harper, Robert C. Stalnaker and Glenn Pearce (eds.), Ifs. Dordrecht: Reidel, 211-247. Kaplan, David 1977/1989. Demonstratives. In: Joseph Almog, John Perry, Howard Wettstein (eds.), Themes From Kaplan. Oxford: Oxford University Press, 481563. King, Jeffrey 2007. The Nature and Structure of Content. New York & Oxford: Oxford University Press. King, Jeffrey 2009. Questions of Unity. Proceedings of the Aristotelian Society 109, 257-277. Kripke, Saul 1980. Naming and Necessity (revised and enlarged edition). Oxford: Blackwell. Kułacka, Agnieszka 2011. Syntax of a Montague Grammar. LingVaria 1(11), 9-23. Matushansky, Ora 2008. On the Linguistic Complexity of Proper Names. Linguistics and Philosophy 21, 573-627. McDowell, John 1998. Meaning, Knowledge and Reality. Cambridge, MA: Harvard University Press. Powers, Lawrence 1995. The one fallacy theory. Informal Logic 17(2), 303-314. Stalmaszczyk, Piotr (ed.) 2010a. Philosophy of Language and Linguistics. Volume 1: The Formal Turn. Frankfurt am Main: Ontos Verlag. Stalmaszczyk, Piotr (ed.) 2010b. Philosophy of Language and Linguistics. Volume 2: The Philosophical Turn. Frankfurt am Main: Ontos Verlag. Stalmaszczyk, Piotr (ed.) 2011. Turning Points in the Philosophy of Language and Linguistics. Frankfurt am Main: Peter Lang. Stanley, Jason 2002. Modality and what is said. Philosophical Perspectives 16, 321344. Sternberg, Meir 1982. Proteus in Quotation-land: Mimesis and the Forms of Reported Discourse. Poetics Today 3(2), 107-156. Travis, Charles 2008. Pragmatics. In: Charles Travis, Occasion-sensitivity: Selected Essays, Oxford: Oxford University Press, 109-129. Wright, Crispin 1989. Wittgenstein’s Rule-Following Considerations and the Central Project of Theoretical Linguistics. In: Alexander George (ed.), Reflections on Chomsky. Oxford: Blackwell, 233-264.

Ben Baker Watford [email protected]

Boghossian’s Implicit Definition Template Abstract: In Boghossian’s 1997 paper, “Analyticity” he presented an account of a priori knowledge of basic logical principles as available by inference from knowledge of their role in determining the meaning of the logical constants by implicit definition together with knowledge of the meanings so-determined that we possess through our privileged access to meaning. Some commentators (e.g. BonJour 1998, Glüer 2003, Jenkins 2008) have objected that if the thesis of implicit definition on which he relies were true, knowledge of the meaning of the constants would presuppose knowledge of the very logical principles knowledge of which the account purports to explain. A consequence would seem to be that implicit definition is incompatible with privileged access. I argue that whilst it is possible for Boghossian to defend against these objections the form of argument he proposes does exhibit a subtle form of question begging such that it exhibits a transmission of warrant-failure. Key words: knowledge of logic, analyticity, implicit definition, privileged access, warrant transmission

1. Boghossian on implicit definition and a priori justification In his paper “Analyticity” (1997a) Paul Boghossian attempted to rescue analytic accounts of the a priori from the disrepute into which they had fallen due to objections pressed by Quine and others.1 Central to such accounts is the claim that there is a route from implicit definition to a priori knowledge and in particular that such a route is available for the central case of basic logical knowledge. Accounts of how implicit definition could explain direct a priori knowledge seem to require that it 1

A shorter version had appeared as Boghossian (1996) together with a response by Gilbert Harman.

16 involve the explicit stipulation of sentences that express a priori truths.2 However, it is one lesson of Quine’s “Truth By Convention” (1936) that we cannot take the meaning of our logical constants to be fixed by the explicit stipulation of basic logical laws, at least if this is to be a fundamental account of their meaning, for any explicit stipulation would have to presuppose logic. Boghossian’s response is that one may save the idea that the meaning of the logical constants is fixed by implicit definition by saying that it is by our acceptance, in some suitably basic way, of certain inference rules, and that this may be seen as amounting to a tacit stipulation of the validity of those rules. His proposal regarding how such a semantic thesis can help in explaining basic logical knowledge is that knowledge of the meaning of the logical constants together with knowledge of how that meaning is fixed by implicit definition puts us in a position to know premises from which we may infer that the inference rules that serve to determine that meaning are valid. In more detail the account proceeds along the following lines. Boghossian characterises the thesis of implicit definition, as applied to the case of logic, as follows: Implicit Definition3: It is by arbitrarily stipulating that certain sentences of logic are to be true, or that certain inferences are to be valid, that we attach a meaning to the logical constants. More specifically, a particular logical constant means that logical object, if any, which makes valid a specified set of sentences and/or inferences involving it. (Boghossian 1997: 348)

Now, call one of the specified sentences or, as I will henceforth assume, inferences for a logical constant, C, an implicit definer of C. Now suppose that we know that the above thesis is true and that M(C) is one of the implicit definers of C, as it might be one of its introduction or elimination rules. Then we know that C means what it does only if M(C) is valid. For according to Implicit Definition it is a condition of C having the meaning that it does that there is some (unique) logical object which 2 3

Harman (1996). I give the thesis as Boghossian expresses it initial capitals here and in what follows so as to be clear it is this specific version of the thesis that is in play.

17 makes its implicit definers, including M(C), valid. Thus we are in a position to knowledgeably state as a first premise: (1) If C means what it does then Μ(C) is valid. But we are generally supposed to have privileged access to the meaning of terms in our language, so given that C is a meaningful term and that one understands it, one is in a position to knowledgeably state its meaning; that is to state that (2) C means what it does. But now it seems that we are in a position to infer from (1) and (2) by modus ponens that (3) Μ(C) is valid. Boghossian argues that an argument of this form can justify a thinker in believing that M(C) is valid, this justification being sufficient for knowledge. Since knowledge of these facts about the meaning of C is sufficient for knowledge of the premises of such an argument, Boghossian says that the conclusion is epistemically analytic. He argues that this can be the case without any implication that it is metaphysically analytic, true in virtue of meaning alone. Boghossian’s focus in this paper is the case of logic, but a similar argument is available whatever kind of term Implicit Definition applies to. When a sentence, S(f), is an implicit definer for its ingredient term f we can run through a similar argument template: (1*) If f means what it does then S(f) is true (2*) f means what it does Therefore (3*) S(f) is true

18 This argument however does not give us the conclusion we need. For (3) is a metalinguistic claim about the sentence S(f) whereas what needs to be explained is knowledge of the truth that sentence expresses. Accordingly when Boghossian discusses this case in his (2003b) the argument he presents is extended as follows:4 (1#) S(f) means that p (2#) If S(f) means that p then S(f) is true (3#) S(f) is true (4#) If S(f) means that p then S(f) is true if and only if p. (5#) S(f) is true if and only if p (6#) p As can be seen the premises also differ. Boghossian claims this form is a better representation of the relevant template but that “it is not materially different” (2003b: 21, fn.3). Similarly, since C is a linguistic expression, the conclusion of the original template, that M(C) is valid, is also metalinguistic, so in this case too we need a further step, now to reach a conclusion stating the validity of a pattern of inference at the level of propositions.5 It is an interesting question just what the required conclusion is and what might be required to reach it from Boghossian’s (3) but I will not pursue it here. Following Boghossian we can call the general form of argument the Implicit Definition Template. In this paper I consider whether Boghossian is right to think that this form of argument can be a route to a priori knowledge of its conclusion. In particular I will consider the worry that since, if the thesis of Implicit Definition is correct, that C means what it does, indeed that it means anything at all, depends on there being a meaning that makes M(C), being justified in believing the minor premise of the original template – that is premise (2) – requires being justified in believing the conclusion. That this is the case is argued by BonJour (1998) and Glüer (2003). If it were then not only would the template be viciously epistemically circular but it would also seem to 4 5

I have added the suffix to Boghossian’s numbering. Ebert (2005).

19 entail that implicit definition as Boghossian characterises it is incompatible with privileged access to meaning insofar as this is thought to be groundless. Boghossian replies to these objections in his (2003b). Whilst I don’t find Boghossian’s response altogether satisfactory I will argue for the compatibility of implicit definition with privileged access. I will argue, though, that the template exhibits a more subtle form of question begging, but one which nonetheless entails that it fails to transmit warrant from its premises to its conclusion. 2. Rule-circularity Boghossian says that A is epistemically analytic for T if “T’s knowledge of the meaning of A suffices for T’s justification for A” (1997: 356). Given that the template is supposed to explain how the proposition that M(C) is valid is epistemically analytic two points should be noted. First, as Boghossian acknowledges, knowledge of the premises of the template involves knowledge of meaning in a fairly broad sense (1997: 357) in which it includes knowledge of how the meaning of C is fixed. That is not knowledge that speakers who understand C typically possess, certainly not if its meaning is fixed by tacit stipulation. In this case it then would not be true to say that understanding the proposition that M(C) is valid is sufficient to be justified in believing it. But presumably speakers who do not know how the meaning of C is fixed have been warranted in believing that the rule is valid. To accommodate this Boghossian draws on Tyler Burge’s distinction between justification in a narrow sense, in which being justified requires having reasons to which one has access, and entitlement, which is a form of warrant which does not.6 With this distinction in place he can allow that speakers who understand it can be entitled to believe that M(C) is valid. In (1997a) the suggestion is that the facts about how meaning is determined suffice for this; in (2003b) it is the availability of the relevant instance of the template.

6

Burge (1993).

20 Secondly, the justification (in the narrow sense) the template provides is inferential.7 Thus knowledge of the meaning of the conclusion will be sufficient for justification to believe it only in the sense that it makes available such justification; to be justified in believing it one must actually draw the inference.8 Further one must meet whatever conditions are required to acquire justification by doing so. But consider the case in which C is ‘if’ and M(C) is modus ponens. As noted, modus ponens is the rule of inference appealed to in the step from the premises, (1) and (2), to the conclusion, (3). Thus in the case in question the relevant instance of the schema will use the very rule which the conclusion states to be valid; it will be rule-circular as Boghossian puts it. Clearly if in order to be justified in inferences drawn by modus ponens one needed to be justified in believing that rule to be valid it would be impossible to acquire such justification by a rule-circular argument. Boghossian, however, argues that in certain cases we can be entitled to infer according to a rule, and thereby in a position to extend our knowledge when we do so, without needing antecedently to be warranted in believing it to be valid. Roughly his suggestion is that we can be entitled to infer in accordance with rules which are meaningconstituting for concepts whose constitutive rules do not build in unavoidable commitments.9 Such concepts will include our basic logical constants. On this account the facts about how meaning is fixed explain our entitlement to infer according to a basic rule.10 So Boghossian’s account takes rules of inference to be basic in determining how the meaning of the logical constants is determined and appeals to this fact in explaining how an inferential account of basic logical knowledge can avoid vicious epistemic circularity. Boghossian’s 7

8

9

10

Or so I will assume. It seems to be Boghossian’s intention though it is perhaps not entirely clear in (1997a). This seems to point the way to a response to Ebert’s (2005) objection to the template. See Jenkins (2008). The account is developed in Boghossian (2000, 2001, 2003a, 2003b). The last two papers present a defence of the account in the form described in the text. Entitlement to infer according to a rule should be distinguished from entitlement to believe it valid.

21 meaning-based account of entitlement to rely on a rule is arguably open to a number of objections but alternative accounts of such entitlement have been proposed.11 Here I will take it that some such account is available and that rule-circularity is compatible with the acquisition of warrant by inference.12 3. The justification of the major premise As we have noted Boghossian’s case for our knowledge of premise (1), the major premise of the modus ponens inference, is based on his thesis of Implicit Definition. This thesis combines two claims, both of which are required for the explanation of how we can know the first premise independent of knowing the conclusion to go through. The first part, which we will call Stipulation, states: It is by arbitrarily stipulating that certain sentences of logic are to be true, or that certain inferences are to be valid, that we attach a meaning to the logical constants.

As we have also noted, in the case of our basic logical constants we must understand stipulation in a somewhat broad sense so as to allow that stipulation can be tacit. The important property of stipulation for the purposes of the template is that what makes an inference rule an implicit definer of a logical constant is that it is accepted in some suitably basic way, or as we may say has a certain acceptance property. That being so we can recognize which rules are implicit definers by the fact that they have that acceptance property, hence independently of knowing that those rules are valid. If implicit definition were so understood that only 11

12

See Williamson (2003), Schechter and Enoch (2006) for criticism. The latter paper proposes an alternative account, as does Wright (2004). Given that rule-circular justification is taken to be possible then it may seem to be preferable to appeal to other forms of inferential justification whose premises do not rest on controversial semantic theses (or which rest on no undischarged premises at all as in Wright 2004). Even if alternative forms of argument are available however, the question remains whether one could acquire justification through the template.

22 rules which determine a meaning such that they are valid could be implicit definers then knowledge of the first premise would not be independent of knowledge of the conclusion. Quine and others following him have argued that it is not possible to draw a principled division between those sentences and/or rules that a defender of the analytic theory of the a priori would claim to be implicit definers for a term and other obvious principles in which that term features, so that there is no principled basis on which to say that a sentence or rule is an implicit definer. Boghossian spends a large part of (1997a) arguing that the determinacy of meaning entails that there must be such a division. I will not assess this part of his case here. Rather I will simply assume for the sake of the discussion that there is not good reason to believe that such a division cannot be drawn and the relevant acceptance property identified.13 Even granting the possibility of some such division, there may be some question whether our knowledge of what the implicit definers of a term are would be a priori. In cases in which there is (and perhaps also if it is as if there is) explicit stipulation this might be a case of knowledge of our intentions. But what of cases in which stipulation could only be tacit? It will rather depend on what the relevant acceptance property is but perhaps we can identify the rules by some combination of introspection (what we would say about cases) and theory. Again I propose to grant that we can know what the implicit definers are a priori. The second claim is implicit in the second part of the characterisation of Implicit Definition: “a particular logical constant means that logical object, if any, which makes valid a specified set of sentences and/or inferences involving it”. What is implied is that if there is no such logical object then the term will lack a meaning. Let us call this claim Determination. It is this that secures the link between C meaning what it does and the validity of its implicit definers, our knowledge of which, together with our knowledge of what those implicit definers are, puts us in a position to know the first premise. It should be noted that neither Stipulation nor Determination are inevitable features of anything that might be called implicit definition. 13

There are also issues concerning just what acceptance of the rule involves.

23 To take the Stipulation first, this posits a certain primacy of use in determining what the implicit definers of the logical constants are. But one might think of a specification of its introduction and elimination rules as a definition of a logical constant that we use, given that they do determine a meaning, but view those rules as norms of correct use, to which actual use must approximate. As regards Determination, this is a thesis Boghossian himself had come to reject in cases of tacit stipulation, and so for the basic logical constants, by the time of (2003b). He had come to be convinced by examples such as the pejorative ‘Boche’ as described by Dummett (1981) that the meaning of a term can be fixed by inference rules without those rules determining a semantic value. So he could no longer appeal to the template to account for our justification for believing basic inference rules such as modus ponens valid. His implicit definitionbased account of entitlement to rely on the rules however does not depend on Determination, and other inferential routes to such justification may be available given that we have such entitlement. However, in the rest of this paper I will assume for the sake of the discussion that the thesis of Implicit Definition understood as Boghossian stated it in (1997a), entailing both Stipulation and Determination, holds in the cases of interest, and in particular in the case of logic, together with whatever else is required for the sort of justification for the major premise of the Template we have considered to be available. The question I will consider is whether, supposing the meaning of C to be fixed by Implicit Definition, we may assume that we do have privileged access to the meaning of C, and whether if we do the relevant instance of the template can be a route to justification of its conclusion in the way Boghossian supposes. 4. The justification of the minor premise Premise (2) of the template, the minor premise of the modus ponens step, is (2) C means what it does.

24 I will assume that we are to read this as saying that C means what it in fact does, so that when C is ‘if’ we can read it as something like ‘if’ means if, where if is the concept or meaning ‘if’ expresses. This is something to which we would ordinarily be thought to have privileged access. But can we assume that we have such privileged access when the meaning of C is fixed by Implicit Definition? That there might be a problem emerges when we consider that there is not in general any guarantee that stipulating that certain rules are to be valid will determine a semantic value which makes those rules valid; there may be no semantic value that will do. A set of rules which we might follow can exhibit various defects that prevent their determining a meaning such that they are valid. At the extreme a pair of introduction and elimination rules may jointly engender inconsistency. If the inconsistency is not obvious it is possible that this could go unnoticed by those using the term in accordance with its meaning-constituting rules, for whom it could appear to be meaningful, to be used to make determinate claims, and so on.14 But Determination says that C only has a meaning if the stipulated inference rules do determine a meaning. That C is an apparently meaningful term in one’s language then does not guarantee that it has a meaning. It depends on certain non-trivial constraints being met.15 This may raise a concern that knowledge of premise (2) is quite substantial and so is not something to which we can simply assume that we have privileged first-person access. But further, since according to Implicit Definition it is a condition on C meaning what it does that there is a meaning, C, which makes its implicit definers valid, that M(C) is valid when C means C is a condition of C meaning what it does given that it is defined in the way it is. But that M(C) is valid when C means C is in effect the required conclusion. This may raise the worry that being justified in believing (2) requires being justified in believing that 14

15

Consider an operator introduced as defined by rules corresponding to the left and right directions of Frege’s Axiom V. Prior to the discovery that it leads to contradiction it seems that it would have been possible to use this operator at least apparently meaningfully. For present purposes we need not consider how those constraints should be characterised.

25 condition met so that the template is viciously epistemically circular. We turn now to objections according to which this is the case. 5. BonJour on implicit definition I start by considering an objection due to Laurence BonJour against any attempt to derive a priori knowledge from implicit definition. The idea is that even if one could fix the meaning of an expression by implicit definition, knowledge that the meaning is fixed in this way would not be enough by itself to be in a position to know what that expression meant; one would have to know that the implicit definers are true in order to know that it meant. Taking his model of implicit definition from Butchvarov (1970), he claims that: […] offering a form of words as an “implicit definition” amounts to a stipulation that any previously unknown terms it contains are to be interpreted in such a way as to make the proposition expressed under that interpretation come out true (or perhaps, necessarily true). […?] Thus, for example, one might stipulate that the sentence ‘40@8=5’ is to count as a (partial) implicit definition of the symbol ‘@’. This, along with other stipulations of the same kind, might prove a useful way of conveying that ‘@’ is to stand for the operation of long division (assuming that the other symbols in the sentence are already understood). But if this is the right account of implicit definition, then the justification of the proposition that 40 divided by 8 is equal to 5 (as opposed to that of the linguistic formula ‘40@8=5’ is not a result of the implicit definition, but is rather presupposed by it: if I were not justified in advance, presumably a priori, in believing that forty divided by eight is equal to five, I would have no reason for interpreting ‘@’ in the indicated way. (BonJour 1998: 50-51)

This line of objection is applied to Boghossian’s template by Carrie Jenkins (2008). In this case, the claim would be that justification of the proposition that C means what it does presupposes justification of the required conclusion so that the argument schema cannot explain that justification. Note that if this objection is correct then it would seem to entail that the meaning of C being fixed by implicit definition is incompatible with privileged access to the meaning of C. Whilst one could still be justified

26 a priori in believing that C means what it does, assuming one could be justified a priori in believing that the logical object C means has the necessary properties to make its implicit definers valid, according to this objection this justified belief would be required as a ground for one’s belief that C means that logical object, whereas groundlessness is generally held to be a mark of privileged access. At least privileged access would hold only in a qualified form. The objection as stated does not directly apply to those cases in which stipulation is tacit, such as that of the basic logical constants. In such cases there does not seem to be any question of our needing to interpret the constant. However it might perhaps be argued that something similar can be said in this case. The thought would be mere competence in using C in accordance with its meaning constituting rules is not by itself enough to be in a position to know which logical object C means. To be able to identify a logical object is the one C means I would have to know that if C is interpreted in that way its implicit definers would be valid. If that is the objection then it seems to rest on a conception of implicit definition according to which it amounts to a form of referencefixing. On that conception even if an implicit definition does fix a meaning it does not tell us what that meaning is; one needs some independent grasp of that meaning and its properties to know that it is what C means. Though Boghossian’s characterisation may invite that interpretation it is not obviously compulsory. A defender of implicit definition may say that in at least some cases an implicit definition can fully convey the meaning of the defined term – in the case of a logical constant this might be its inferential role – and that one who understands the term in accordance with its implicit definition can thereby grasp that meaning.16 Understanding an implicitly defined term, at least sufficiently to be able knowledgeably to state its meaning, involves more than merely being disposed to use it in accordance with the rules; it requires some appreciation of how it contributes to the content of sentences in which it occurs. But that need not require that one be able to identify an

16

Hale and Wright (2000).

27 independently specifiable logical object as the meaning as this objection assumes. As far as I can see the characterisation of implicit definition on which the objection rests has not been shown to be compulsory; nor has the position of the defender of Implicit Definition been shown to be untenable. More may need to be said to give an account of how knowledge of meaning is possible given such a position but BonJour has not given us reason to think that no such account could be given. I will shortly try to sketch how such an account might go. 6. Glüer’s objection Kathrin Glüer (2003) presents an objection that does not seem to rely on any tendentious conception of Implicit Definition, but simply on the dependence of meaning on the existence of a suitable semantic value that the thesis entails. Referring to the template for sentences she argues: That f means what it does depends on there in fact being something that makes S(f) true, according to premise [(2#)]. Only if there is such a fact does f have any meaning. Being justified in believing premise [(1#)] therefore, requires being justified in believing that there is something that makes S(f) true. Moreover, it requires being justified in believing that S(f). (Glüer 2003: 57)

What this objection does seem to assume is that since the validity of M(C) is a precondition of C meaning what it does then possessing justification for the former presupposes the possession of justification for the latter. Since the distinction between justification and entitlement is in play in Boghossian’s original paper I will take it for now that justification in this passage should be understood in the narrow sense in which it requires reason for belief to which the subject has access.17 If so then 17

I will argue in effect that the objection can be sustained if it is interpreted as claiming that being justified (in the narrow sense) in believing premise (2) requires possessing justification (in the wide sense) to believe the conclusion, for I will argue that it requires being entitled to believe the conclusion.

28 this objection too would seem to entail a denial of the compatibility of Implicit Definition with privileged access.18 At any rate Boghossian interprets it as having that entailment. For in reply he says: […] most philosophers simply assume that meaning facts are first-person accessible in some privileged way, regardless of what the supervenience base for meaning facts is taken to be. For example, many philosophers believe that even if facts about meaning and concept possession were to supervene on facts that are external to the mind that that would have no tendency to undermine our privileged access to first-person facts about meaning. (Boghossian 2003b: 22-23)

He goes on to complain that Glüer has given no special reason to believe that this assumption should be suspended when the meaning of a term is fixed by implicit definition. To be relevant to Glüer’s objection the relevant externalist theses will have to include those according to which terms which fall under them may lack a meaning; such would be certain views concerning natural kind terms or an Evans-style view of Russellian proper names. Many philosophers do think that such views are perfectly compatible with privileged access. So if, as seems to be the case, Glüer is appealing a principle incompatible with that position the onus may be thought to be on her to argue for it rather than assuming it. However, given that Boghossian is invoking the template to explain a priori knowledge this response does not seem altogether satisfactory. For the fact that the validity of M(C) is a precondition of C meaning what it does gives us some reason to wonder whether an account of privileged access is available which will make justification for (2) epistemically independent of the conclusion, as is required if the template is to explain how we can be justified in believing it. For one thing it is hard to see how anything to which the implicit definition theorist can legitimately appeal could explain how a speaker who understands an implicitly defined term could thereby be in a position to directly apprehend that C 18

Perhaps the minimal claim would be that one could not be justified in believing (2) unless one were also justified in believing that S(f). That is not obviously in conflict with privileged access.

29 has that meaning. To do that it would seem to be necessary to have some sort of direct access to the meaning. To assume that one had such access would be illegitimate in the present context, for the template is supposed to explain a priori knowledge without invoking an unexplained capacity for insight into necessary features of reality. But the sort of capacity for privileged access with which this assumption would credit speakers would seem to be no less mysterious, if indeed it is distinct. On the other hand the fact that C is not guaranteed to have a meaning would seem to rule out the sort of simple disquotational account of our knowledge of its meaning that might be appealed to in order explain privileged access to meaning.19 For instances of the meaning disquotation schema ‘t means t’ – in which t is first mentioned and then used – are not guaranteed to express truths even when t is restricted to terms which are apparently meaningfully used. Our knowledge of the minor premise thus seems to be more substantial than such an account would make it. This is so far just to say that two extreme positions do not seem to be available; but if the availability of the template is to explain how we can acquire a priori justification in the relevant cases then it seems we need some assurance that a satisfactory account can be given. If it is thought that we must suppose that we have privileged access to the meaning of our terms then it would seem that there is some onus on a theorist either to explain how a semantic thesis that is prima facie in tension can be compatible with it or to show that the semantic theory is the only tenable one. But it does not seem that Boghossian is in a position to claim the latter. Accordingly at least a sketch of an account of privileged access seems to be called for.20 7. The compatibility of Implicit Definition with privileged access For the reasons given it seems that if we can have privileged access to the meaning of an implicitly defined term, then this must be possible 19 20

Boghossian (1997: 359-360). Boghossian offers an explanation for the case of logic in an Appendix to his (1997a). We will consider this explanation in section 8.

30 even though our grounds for supposing that the term does have a meaning leave open the possibility that it is meaningless. How might that be? It would seem to require that there is, at least in favourable cases, a default presumption in favour of taking the appearance of the meaningfulness of a term at face value. Crispin Wright (2000) suggests such a position concerning the compatibility of privileged access with externalist theses that raise the possibility that apparently meaningful terms of certain kinds do not in fact have a meaning. On such views it is an open possibility, for all that a speaker has special first-person access to, that the relevant external conditions are not met. Nevertheless, Wright’s thought is, the ordinary a priori presumption that seemingly meaningful terms in a speaker’s language are meaningful could still apply. The mere possibility of perceptual illusion is not generally thought to undermine our entitlement to take perceptual appearances at face value; it is only if there is some special reason to believe it actual that the usual presumption in favour of the appearances being veridical is suspended and one might be required to justify the belief that the possibility is not actual. The suggestion in the case of meaning is that neither here is the mere possibility of what Wright calls an illusion of content sufficient to defeat the a priori presumption that an apparently meaningful term is meaningful. If there is special reason to believe the possibility actual, or perhaps reason to think that our practice is defective in some way, then the default presumption would be defeated, but otherwise it is in force and we can be justified in asserting the relevant instances of the meaning disquotation schema. Supposing that is plausible in the externalist cases, can we say something similar in cases of terms whose meaning is fixed by implicit definition? It might perhaps be suggested that we are not. In the externalist cases the possibilities in which the terms are meaningful are scenarios such as Boghossian’s Dry Earth (1997b) which are similar to sceptical scenarios in perceptual cases and it may be this sort of feature that lends plausibility to Wright’s view. However in a case in which we fix the meaning of a term by implicit definition we in effect presuppose that certain non-trivial constraints are satisfied and in doing so run the risk that they are not. It might seem that it is less plausible that we are entitled to a default presumption of meaningfulness in this case.

31 Nevertheless, I think that it is possible to argue that we are entitled to such a presumption in at least some cases. Again I take a suggestion from Wright, though it seems to be in the spirit of Boghossian’s later views on the relation of implicit definition to entitlement.21 The general idea is that we need terms governed by certain kinds of rules to engage in rational enquiry. This involves a certain unavoidable risk, but suppose that the use of a given term is necessary to our cognitive projects, that our stipulation satisfies certain general constraints intended to ensure that we do not run unnecessary risks in framing implicit definitions,22 and that the term is apparently meaningful on the basis of that stipulation; then I think it is plausible that we could be entitled to the presumption that the term is meaningful, absent special reason to believe the definition defective. The above sort of account may draw some plausibility from the fact that ordinary speakers will not typically know what the conditions are on which the meaningfulness of the terms in question depend if the relevant semantic theses are true, and they will typically blameless in this. If that is so then in the absence of reason to doubt the term meaningful it would seem inappropriate to demand that speakers be able to produce grounds for their attributions of meaning. But what if not only is the relevant semantic thesis true but also a speaker does know that it obtains so knows that the precondition must hold if the term is to have a meaning, as would seem to be called for if the meaning is fixed by an explicit stipulation, and is certainly the case if the speaker is justified in believing (1)? Would that affect what would be required to be justified in the belief that the term was meaningful? James Pryor (2007) argues that it would in the case of the externalist theses. In his view given that one was justified in believing such a thesis one would need to produce some reason to believe that the external conditions were met if one were at the same time to be justified in believing that the term was

21 22

See Boghossian (2003a, 2003b). See Boghossian (2003a and 2003b) and Hale and Wright (2000) for attempts to formulate such constraints.

32 meaningful.23 That might be easily done given that one did stand in the necessary relations, but it would involve a qualification of privileged access. Against this it might be observed that the possibility of perceptual illusion is generally known, but that it remains the case that unless there is special reason to believe one might be suffering an illusion one need not produce reason to believe one is not in order to be justified in the perceptual beliefs one forms.24 I am not altogether sure that this is effective in the externalist cases for in those cases speakers may be expected to have the necessary evidence and the challenge to produce it does not seem to lead to a regress. But it does seem to be relevant in the cases of implicit definition given that the above account is available. One of the motivations for an implicit definition based account is after all the thought that no relevant “evidence” is antecedently available to us.25 If such a position is tenable then the thesis that the meaning of C is fixed by implicit definition is compatible with our having privileged access to the meaning of C; one would need no antecedent justification to believe that M(C) is valid in order to be justified in believing that C means what it does. It seems to me then that at least the sketch of an account of how implicit definition can be compatible with privileged access is available. 8. Transmission of warrant failure I have argued that assuming that the thesis of Implicit Definition is true of C we can be justified in believing both premises of the Implicit Definition Template together, and that in neither case does our justification depend on our being antecedently justified in believing the 23

24 25

He actually considers the case of thought contents but similar considerations would seem to apply to the meaning case. Wright (2000). One might try to justify the claim that the implicit definers of a basic logical constant are valid by some argument. But since the argument will be rule-circular one must pragmatically presuppose that the constant is meaningful. It is thus doubtful that the argument can amount to a justification of the belief that it is.

33 conclusion, at least if justification is understood as having reasons for belief. Does this lay to rest the worry that the argument begs the question in some way incompatible with its being a route to justified belief in its conclusion? No, because there can be more subtle ways in which an argument can beg the question in such a way that a kind of warrant for its premises fails to transmit to its conclusion. I think there is good reason to think that this is the case with the template. To take apparent meaningfulness of C at face value is in effect to presuppose that the conditions on C having a meaning are satisfied, since apparent meaningfulness does not intuitively suffice to establish that it has a meaning. As Glüer points out, according to Implicit Definition those conditions require that there is a meaning, C, which makes C’s implicit definers, which include M(C), valid. Since C is the meaning C is determined to have they require that the “disquoted” rule corresponding to M(C) is valid. But that is just the required conclusion. So in judging that premise (2) is true we in effect presuppose that conclusion. Thus there is an epistemic dependence of that premise on the conclusion – we must be epistemically entitled to presuppose it if we are to be warranted in believing (2) – that seems to be incompatible with our warrant for the premises of the template being transmitted to the required conclusion. It seems then that a version of Glüer’s claim may be sustained, if we read her as claiming that being justified in believing premise (2) requires being entitled to believe that there is something that makes M(C) valid.26 It is also worth stressing that the objection here is not merely that our warrant for the second premise depends on our being entitled to the conclusion. It is not merely that doubt concerning the conclusion would be incompatible with being justified in believing (2) so that to be justified in believing (2) we must be entitled not to doubt the conclusion absent special reason to do so.27 Rather it is that our first-person reasons for accepting (2) in themselves fall short of justifying it; they only do so in a context in which we are entitled to take it on trust that the necessary 26

27

This is essentially the objection Wright (2000) makes to McKinsey-style reasoning. In the terminology of Davies (2004) this would be a negative entitlement.

34 conditions are met, and thus that the required conclusion is true. We can thus say that premise (2) rests on the conclusion and so cannot provide support for it, at least not in the way the account requires. Against this it might perhaps be objected that it is sufficient to be entitled to discount the general possibility of meaninglessness, not anything more specific. And in an Appendix to his (1997a) Boghossian presented an account of how we can be justified in believing (2) for the special case in which C is a basic logical constant. The suggestion was that in such a case we can know premise (2) because doubt concerning the meaningfulness of such a term is self-defeating. (Some logical vocabulary is required to state the denial, so the denial states of itself that it is meaningless. This is not a contradiction but it would be pragmatically self-defeating to assert it.) He claimed that this is of more than merely pragmatic relevance. The suggestion would seem to be that one is in effect entitled to the presupposition that the preconditions of C being meaningful are satisfied and so to accept the relevant instance of the meaning disquotation schema in this case. If we only need to cite this fact in explaining why we don’t need to eliminate the possibility that C lacks a meaning, then perhaps it is unnecessary also to require an entitlement to believe M(C) valid if we are to be justified in believing (2). I think we can say a number of things in response. First, so far as Boghossian’s argument goes the most that seems to follow is that as it happens we lack the expressive resources to entertain a doubt, which seems too parochial a fact about our concepts to explain our justification for (2).28 Secondly, even if something stronger may be claimed,29 it may be doubted whether the fact that doubt regarding a claim would be selfdefeating is sufficient for us to be entitled to believe or trust that it is true. Perhaps if it could be claimed that it could not be doubted that our logical constants are meaningful then this could explain why we need not take the possibility of defeat seriously. 28 29

See Jenkins (2008) for discussion. Hale (2002) argues that for some basic logical rules we cannot doubt that they are valid. Insofar as we can view those rules as the meaning-constituting for the relevant logical constants one might extend that argument to claim that it cannot be doubted that those constants are meaningful.

35 Then given a default presumption in favour of meaningfulness we might have an explanation of our justification to instances of the disquotation schema, but the default presumption seems to depend on the implicit definition meeting certain conditions, which are precisely those needed to explain our entitlement to believe it successful. Finally, the fact remains that in judging that (2) is true one must presuppose that those conditions are met insofar as one’s basis for that judgement does not eliminate the possibility that they are not. I would suggest that this is likely to remain true whatever particular account is given of privileged access compatible with Implicit Definition given that a defender of the template is not in a position to assume a notion of privileged access which gives us conclusive justification believing things like (2). If so then it seems that he will be forced to something like the view I have suggested or one similar enough to be open to the same sort of objection. If he thinks that an alternative account is possible he will need to provide it if he is to be in a position to claim to have explained any case of a priori knowledge. References Boghossian, Paul 1996. Analyticity Reconsidered. Noûs 30, 360-391. Boghossian, Paul 1997a. Analyticity. In: Bob Hale and Crispin Wright (eds.), A Companion to the Philosophy of Language. Oxford: Basil Blackwell, 331-368. Boghossian, Paul 1997b. What the Externalist Can Know “a Priori”. Proceedings of the Aristotelian Society 97, 161-175. Boghossian, Paul 2000. Knowledge of Logic. In: Paul Boghossian and Christopher Peacocke (eds.), 229-254. Boghossian, Paul 2001. How are objective epistemic reasons possible? Philosophical Studies 106, 1-40. Boghossian, Paul 2003a. Blind Reasoning. Proceedings of the Aristotelian Society Supplementary Volume 77, 225-248. Boghossian, Paul 2003b. Epistemic Analyticity: A Defence. Grazer Philosophische Studien 66, 15-35. Boghossian, Paul and Peacocke, Christopher (eds.) 2000. New Essays on the A Priori. New York: Oxford University Press. BonJour, Laurence 1998. In Defense of Pure Reason. Cambridge: Cambridge University Press. Burge, Tyler 1993. Content Preservation. The Philosophical Review 102, 457-488.

36 Butchvarov, Panayot 1970. The Concept of Knowledge. Evanston: Northwestern University Press. Davies, Martin 2004. Epistemic entitlement, warrant transmission and easy knowledge. Aristotelian Society Supplementary Volume 78, 213-245. Dummett, Michael 1981. Frege: Philosophy of Language. London: Duckworth. Ebert, Philip 2005. Transmission of Warrant-Failure and the Notion of Epistemic Analyticity. Australasian Journal of Philosophy 83, 505-521. Glüer, Katrin 2003. Analyticity and Implicit Definition. Grazer Philosophische Studien 66, 37-60. Hale, Bob 2002. Basic Logical Knowledge. In: Anthony O’Hear (ed.), Logic, Thought and Language. Cambridge: Cambridge University Press. Hale, Bob and Crispin Wright 2000. Implicit Definition and the A Priori. In: Paul Boghossian and Christopher Peacocke (eds.), 286-319. Harman, Gilbert 1996. Analyticity Regained? Noûs 30, 392-400. Jenkins, Carrie 2008. Boghossian and Epistemic Analyticity. The Croatian Journal of Philosophy 8, 113-27. Pryor, James 2007. What’s Wrong With McKinsey-style Reasoning? In: Sanford Goldberg (ed.), Internalism and Externalism in Semantics and Epistemology. Oxford: Oxford University Press. Quine, Willard Van Orman 1936. Truth by Convention. In: O. H. Lee, (ed.), Philosophical Essays for A. N. Whitehead. New York: Longman’s, 90-124. Schechter, Joshua and Enoch, David 2006. Meaning and Justification: The Case of Modus Ponens. Noûs 40, 687-715. Williamson, Timothy 2003. Understanding and Inference. Proceedings of the Aristotelian Society Supplementary Volume LXXVII, 249-293. Wright, Crispin 2000. Cogency and Question-Begging: Some Reflections on McKinsey’s Paradox and Putnam’s Proof. Philosophical Issues 10, 140-163. Wright, Crispin 2004. Intuition, Entitlement and the Epistemology of Logical Laws. Dialectica 58, 155-175.

David Botting Universidade Nova de Lisboa [email protected]

The Two One Fallacy Theory Theory∗ Abstract: In this paper I will describe in outline two theories of fallacy. One is the medieval theory of Jean Buridan, the other the “one fallacy theory” set forth by Lawrence Powers. Analysis of these will lead to my main claim that fallacy theory has from the very beginning treated two questions as if they were one: “what is a fallacy?” and “why do fallacies occur?” The Aristotelian division of fallacies into those depending on language and those not depending on language are reinterpreted as being addressed to these different questions. Powers’ theory that all fallacies are equivocations is correct when reinterpreted as an answer to the second question. The first question is also to be answered in terms of one fallacy, namely ignorantia elenchi, following Aristotle’s own procedure in the Sophistical Refutations. There is a one fallacy theory for each question, hence the title of this paper. Key words: fallacy theory, equivocation, the One Fallacy Theory, Powers, Buridan, Aristotle

0. Buridan on the causes of illusion A fallacy is not simply a bad argument, but a bad argument that appears to be good by which men “of a certain stamp” (those not well-enough trained in logic, Aristotle 1965: §9) are deceived. This is an important qualification, because it points to an objective resemblance of the bad argument to the good, and not just a subjective relation. The purpose of such treatises as Aristotle’s Sophistical Refutations and Buridan’s On Fallacies1 is to help the arguer see through this appearance, and not (because they are infinite) arm the arguer against every possible way of making a mistake, appearances being always in the eye of the beholder.

∗ 1

Project “Argumentation, Communication and Context” PTDC/FIL-FIL/110117/2009 This is the seventh treatise in Buridan’s (2001) Summulae de dialectica.

38 The fallacies listed by Aristotle and commented on by Buridan describe certain characteristic ways that a bad argument may appear good; these ways are called by Buridan the ‘causes of illusion’. The ‘badness’ of the argument he calls the ‘cause of defectiveness’ (Buridan 2001: 504). Buridan goes through each of Aristotle’s fallacies and describes the cause of defectiveness and the cause of illusion, and I will briefly do the same: Fallacy of Equivocation (Buridan 2001: 516) Cause of Illusion: An incomplex utterance (word) is identical in matter and form with another token of the same utterance.2 Cause of Defectiveness: In their use, the tokens of this utterance have different significations. Fallacy of Amphiboly (Buridan 2001: 516-17) Cause of Illusion: A complex utterance (spoken expression) is identical in matter and form with another token of the same utterance. Cause of Defectiveness: In their use, the tokens of this utterance have different meanings. Fallacies of Composition and Division (Buridan 2001: 524-25) Cause of Illusion: Two complex utterances (spoken expressions) are identical in matter. Cause of Defectiveness: In their use, these utterances have different meanings. Fallacy of Accent (Buridan 2001: 532-34) Cause of Illusion: An incomplex utterance (word) is identical in matter with another token of the same utterance.3 2

3

I do not mean to suggest that there actually exist two tokens of an utterance that are being conflated, but rather that the one token is being taken as if it were the other. Similarly, when I talk about two utterances, one of those should be taken as a fiction. In the Greek there is no difference in spelling between accented and unaccented words, although there is a difference in pronunciation. This is why Aristotle (1965: §4) says that an example of the fallacy of accent is difficult to construct in

39 Cause of Defectiveness: In their use, the tokens of these utterances have different significations. Fallacy of Figure of Words (Buridan 2001: 534-35) Cause of Illusion: Two utterances (whether words or spoken expressions) are similar in matter. Cause of Defectiveness: In their use, these utterances have different significations. • Commentary on the fallacies in dictione o What Buridan calls an utterance is what is uttered rather than its uttering. It is a spoken sentence. It is not the propositional content or meaning of the sentence, nor is it individuated by its content or meaning – the meaning is a feature of the intention with which it was spoken, which is to say, it individuates the acts of uttering. Utterances have different meanings in so far as they are uttered with different intentions. This is indicated in each case by my phrase “in their use.” Each uttering of the sentence produces another token of that utterance that is numerically, but not necessarily materially or formally, distinct, and each numerically distinct utterance may possibly have a distinct signification and meaning in an ‘analogical’ sense.4 An utterance is individuated not by its meaning or signification but by its matter and form.

4

speech. In many languages, e.g., French, the accent is indicated, for instance with an acute, grave, or circumflex. A word written with an acute accent over one letter is not materially identical to one that does not; on the definition given, then, fallacies of accent would not occur in French. However, broadly speaking we may still say that a fallacy of accent occurs whenever, because of their material similarity, an accented word is confused with an unaccented word. In other words, sounds derive whatever intentionality they have from acts of consciousness, and it is only in this derivative sense that they can be said to be meaningful; the uttering imposes meaning on the terms uttered ad placitum or ‘at will’. This is not to deny that words are associated with what they are taken to mean by conventions, but a word that does not have a double meaning by convention may still be used by a speaker equivocally.

40 o With regards to spoken language, material identity is homonymy. To see what formal identity amounts to, it is better to consider a couple of examples. “Speaking of the silent is possible” is amphibolous because it might mean that it is possible for the silent to speak or that it is possible to speak about the silent. Today we would call this syntactic ambiguity and perhaps represent its two readings as “Speaking-of-the-silent is possible” and “Speaking ofthe-silent-is-possible”; “the silent” is a patient (what is spoken about) in the first and an agent (what is doing the speaking) in the second. Compare this with a fallacy of division: “A man can walk while sitting.” This also seems to be syntactically ambiguous, having “A man can walk-while-sitting” and “A man can walk while-sitting” (equivalent to “While sitting, a man can walk”) as its alternative readings, “while sitting” modifying the predicate in the first (impossible) reading and the subject in the second (possible) reading (Aristotle 1965: §4). However, according to Buridan the two readings in the division case amount to utterances that are materially but not formally identical, and in a similar vein Aristotle (1965: §20) denies that there is ambiguity in the latter, his reason being that “the spoken words are not the same” although the written ones may be. o This suggests that formal identity is connected to features of speech such as aspiration and inflexion. This means that complex utterances that are not formally identical sound different, i.e., are not homonymous. This raises the question of how material identity and formal identity differ. Material identity, I think, amounts to homonymy on the level of words and syllables, formal identity on the level of sentences. What makes formal identity ‘formal’ is that it concerns the manner with which something is uttered but is nevertheless a holistic feature of what is uttered, unlike the intention with which the sentence is uttered which is a feature of the act of uttering.5 5

It is not altogether clear from this what the formal identity of words amounts to. A word that that is a sequence of syllables may be spoken in different ways, of course, in much the same way as a sequence of words, but if the word is

41 o The fallacies of equivocation and amphiboly concern utterances with double meanings that are numerically distinct but are materially and formally identical. This is ambiguity properly speaking.6 o The fallacies of composition, division, and accent concern utterances that are numerically and formally distinct but are materially identical. The fallacy of figure of words concern utterances that are distinct in all three ways. o Because of their identity or similarity, different (tokens of) utterances are taken to have been uttered with identical intentions, or we would say in modern terms that they are taken to have the same sense. This belief is the cause of their illusion. That they do not have the same sense is the cause of their defectiveness. Fallacy of Accident (Buridan 2001: 548-550) Cause of Illusion: The argument resembles one that has a valid logical form. Cause of Defectiveness: The argument does not have a valid logical form because its middle term does not render the inference necessary (this being, according to Buridan, all that is meant in this context by the term ‘accident’). Fallacy of Secundum Quid et Simpliciter (Buridan 2001: 554) Cause of Illusion: A complex predicate is partially materially identical with the incomplex predicates of which it is composed. Cause of Defectiveness: The complex and incomplex predicates are distinct, and the necessary inference is not preserved when one is substituted for the other in a valid argument.

6

monosyllabic then I think that different occurrences of it must be formally identical by default, as it were. However, Buridan (2001: 532) also calls the fallacy of accent an ambiguity despite the fact that the utterances in question are not said to be formally identical. If it is an ambiguity, then it seems to be merely a special case of the fallacy of equivocation. Buridan’s account seems inconsistent here.

42 Fallacy of Ignorantia Elenchi (Buridan 2001: 563) Cause of Illusion: A false contradiction satisfies some but not all of the conditions necessary for a true contradiction. Cause of Defectiveness: Some but not all of the conditions necessary for a true contradiction are satisfied. Fallacy of Petitio Principii (Buridan 2001: 570-571) Cause of Illusion: There is a connection7 between the thesis and a premise introduced by the respondent such that in conceding the premise and conclusion the proponent’s concessions are logically inconsistent. Not noticing the connection, the premise is conceded. Cause of Defectiveness: The connection between the premise and the negation of the thesis. Fallacy of the Consequent (Buridan 2001: 573-574) Cause of Illusion: A valid conversion is partially identical with an invalid conversion. Cause of Defectiveness: These conversions are distinct, and the necessary inference is not preserved after an invalid conversion as it would be after a valid conversion. Fallacy of non-cause as cause (Buridan 2001: 579-580) Cause of Illusion: A premise that is logically compatible with the other premises is taken to be a cause of the conclusion but is not, i.e., it is redundant. Cause of Defectiveness: The falsity of the conclusion implies the falsity of one or more of its causes but implies nothing at all about a non-cause. Fallacy of many questions as one (Buridan 2001: 584-585) Cause of Illusion: A complex question is partially identical with a single question. Cause of Defectiveness: A complex question is distinct from, and must be answered differently from, a single question. 7

This unspecified connection seems to be contrariety – the premise and conclusion cannot both be true but may both be false.

43 • Commentary of the fallacies extra dictionem o With the exception of the fallacy of petitio principii, the causes of illusion all come down to partial material identity with a valid refutation. In this sense, they are just like fallacies in dictione. In the case of the fallacy of accident and the fallacy of non-cause as cause, the partial identity in question is between arguments, but otherwise the partial identity is between words and expressions as before. o The arguer’s mistake, and what makes the fallacy one occurring “apart from words”, seems to be not in taking these utterances to have been uttered with identical intentions, but in taking utterances that have been uttered with disparate intentions as functioning in the same way in the refutation. Thus, in the fallacy of accident the material difference between a valid and an invalid syllogism is a different middle term; the error is in supposing that all middle terms (or at least the middle terms in question) can equally render the inference necessary. In the fallacy of the consequent, it is not that the arguer believes that an invalid conversion is valid so much as he fails to realise that whereas a sentence and validly converted counterpart can be substituted for each other without altering the necessity of the inference, this is not the case for a sentence and an invalidly converted counterpart. o The difference between the fallacies extra dictionem and the fallacies in dictione is that the former occurs because of a mistaken belief about things and the latter because of a mistaken belief about how words were used. o In the fallacy of ignorantia elenchi the conclusion of the refutation is not a ‘true contradiction’, which is to say, not the contradictory of the thesis. This is typically because of equivocation in the conclusion; for instance, a proof that it is possible for a man to speak about the silent will not have refuted the arguer who has claimed that it is not possible for the silent to speak, since obviously these are not contradictories. This being so, it appears that all fallacies in dictione are also instances of ignorantia

44 elenchi, and indeed Aristotle (1965: §6) reduces the former to the latter in just this way. o Aristotle (1965: §6) also reduces the other fallacies extra dictionem to ignorantia elenchi, but in these cases a broader concept of refutation is being used. A refutation is a proof of a true contradiction, so something can fail to be a refutation either because what it proves fails to be a true contradiction (as illustrated above) or because it fails to be a proof. It is by failing to be proofs that the fallacies extra dictionem are instances of ignorantia elenchi.8 1. Lawrence Powers and the One Fallacy Theory In a series of papers Lawrence Powers has argued for the One Fallacy Theory: all fallacies are instances of the fallacy of equivocation.9 This involves, as he frankly admits, a modification of what is to count as a fallacy. A fallacy has to have both a virtue and a vice; in the absence of one of these a mistake does not count as a fallacy.

8

9

These two types of failure are called respectively deductive errors and errors of contradiction in Hansen (1996: 321). It would seem that fallacies extra dictionem are thus reduced to and identified with deductive errors and fallacies in dictione with errors in contradiction. Admittedly, it is sometimes only by appeal to some technical stipulations of Aristotle’s about proofs that the fallacies of petitio principii and many questions fail to be proofs and are classed as deductive errors. Hansen (1996: 321-324) seems to embrace the former identification, but for reasons that are not entirely clear he rejects the identification of fallacies in dictione with errors in contradiction, as does Buridan (2001: 566) on the grounds that errors in contradiction do not always depend on language and may arise “apart from words” (regrettably, he does not give an example). Perhaps they would say that all instances of in dictione fallacies are also instances of errors in contradiction (ignorantia elenchi), but not all instances of ignorantia elenchi are also instances of in dictione fallacies. In principle, Buridan goes on, Aristotle could have gone on to divide ignorantia elenchi into language-dependent and language-independent divisions, but did not do so for the sake of brevity. My presentation will follow that given in “The one fallacy theory” (Powers 1995).

45 The vice of a fallacy is that it can lead from truth to falsehood – or in other words, it instantiates a logical form that is invalid – and its virtue is that it appears to be valid.10 It might seem strange to consider this a virtue, but Powers explains this as that the appearance of being valid is in itself a genuine reason to believe the conclusion. This seems the same as what is often called in the literature doxastic or personal justification. Imagine what would be the case if this were lacking: you would entertain a valid argument but would not believe in its validity, or would not be justified in believing in its validity because, for instance, your belief is based on unreliable testimony. In such a case it would surely be wrong to say that you, from your own point of view, are justified in believing the conclusion – you have got lucky. The kind of justification just advertised comes about in this case from the fact that each inferential step appears – when attention is directed to it – self-evident. The inferential step that goes under the name of the fallacy of affirming the consequent is not self-evident and, because it is “nakedly invalid” and lacking any virtue, not a fallacy (Powers 1995: 309). The only way something that is not self-evident can appear selfevident is because of equivocation: “every invalid step is covered by an appropriate ambiguity” (Powers 1995: 305). Resolving the ambiguity amounts to revealing the true logical form; one cannot put the argument into a logical form first and then ask oneself whether such a form is invalid. One of the most interesting things in Powers’ account (and on my interpretation echoes Aristotle’s account but becomes diluted in Buridan’s, where it is the mistaken belief that is the psychological cause and whether it is about words or things seems to be the basis for distinguishing what fallacy is committed) is that “appears to be” does not denote a subjective relation; indeed, one of Powers’ motivations is to reject a psychological criterion of fallacy. There is any number of reasons why an argument may appear to a particular person to be good, 10

It would be more precise to say something like “It does not instantiate any logical form that is valid” since every argument whatever is an instantiation of the invalid logical form p├ q.

46 and likewise there is any number of reasons why an argument may appear to a particular person to be fallacious, most common of which being that the conclusion is quite obviously absurd, as in the argument Rivers run Whatever runs have feet Therefore, rivers have feet However, saying that the premises are true and the conclusion is false does not explain (because it is a virtus dormitiva saying in other words simply what it means to be invalid) why it appears to instantiate a valid argument form. The answer to that question is the answer to the question what the virtue is in this argument (Powers 1995: 308-310), namely equivocation on “runs”. • Commentary on the one fallacy theory o There is clearly a strong parallel between Powers virtue/vice distinction and Buridan’s illusion/defectiveness distinction. With one exception, the causes of illusion are identity or partial identity with an utterance that would have made the argument valid. In the case above, this is an utterance of “Whatever runs have feet” where “runs” has the same sense as in “Rivers run.” This utterance is obviously false, but I reiterate that it is not its falsity as such that is problematic but rather its lack of doxastic justification. o The exception seems to be petitio principii where the reasoner did not notice a connection between the premise and the conclusion. But even in this case, it is plausible that the reason that this connection is overlooked is not because an utterance has two meanings as that two utterances have the same meaning or are logically equivalent, but this is not noticed because of a difference in the form of words. For instance, the virtus dormitiva I described above that could be written as “An argument is invalid because its premises are true and its conclusion false” is a petitio principii in this sense, amounting to “An argument is invalid because it is invalid.” Although the latter is not “nakedly invalid” (indeed, it is

47 nakedly valid) it is nakedly circular and non-probative, and for that reason would probably not be counted as a fallacy by Powers; in contrast, it is quite easy to imagine someone being taken in by the former. As Powers says, it is a question of revealing the true logical form, so I do not think that it is too much of a stretch to consider this also as a fallacy of equivocation. o On the other hand, I find the accounts of defectiveness and vice unsatisfactory. In Buridan’s description of the fallacies in dictione the cause of defectiveness does not seem to expand at all on the cause of illusion, saying circularly that the illusion is illusory. This also seems to be case with Powers’ vice. It seems to me that the question to which Buridan’s causes of illusion and Powers’ virtue is addressed is not “What is a fallacy?” but “Why do fallacies occur?” Provisionally we might grant (although I do not think this is quite right, as I will show) that fallacies always arise because (and this is a causal ‘because’) of equivocation, but I think we should not grant that fallacies are thereby always fallacies of equivocation. In fact, I think that what fallacies are relates to their true logical form, and this does not contain ambiguity, as Powers admits. o The true logical form is given by what is asserted in the argumentation. It has been entered as a criticism of logocentric approaches to argumentation by Gilbert (1997) and others that it assumes that language is clearer than gestures and other nonlinguistic but apparently meaningful acts, while in fact both language and non-linguistic acts equally require contextual information for their disambiguation and interpretation. This much I think we can grant, but it does not lead to the conclusion Gilbert wants, which is that these different acts comprise different types of argumentation and are governed by different norms, norms other than deductive validity. Linguistic and non-linguistic acts alike can be interpreted as assertives and these are not ambiguous. It is the propositional content of the assertives that comprise the argument, and the same propositional letter can obviously not be used for different propositional contents, and the same propositional content should not have different propositional letters used for it

48 (as in the petitio principii). Although the ‘raw data’ contains ambiguity, the argument itself does not. Thus we only need one norm: validity.11 2. The Two One Fallacy Theory Theory Powers argues that there is only one fallacy: equivocation. I claim that this is not quite right: “equivocation” is always the answer to the question “Why do fallacies occur?” but is never the answer to the question “What is a fallacy?” A fallacy is an invalid argument, and formal arguments are never ambiguous. This seems consistent with Buridan’s exposition. Although Buridan insists that fallacies extra dictionem arise out of a confusion between things rather than words, it still seems the case that arguments fallacious in these ways have the appearance of being good – in Powers’ sense of ‘appears’ – because of the identity or partial identity of two (tokens of) utterances with different meanings or significations. Nevertheless, I wish to retain the notion that there is only one fallacy and one answer to the question “What is a fallacy?” and this answer is that it is an ignorantia elenchi. Putting to one side the problematic fallacies of petitio principii and many questions, all fallacies are invalid arguments or become invalid arguments when the thesis the argument is intended to refute is included as a premise. For instance, an argument that rivers do not run because they do not have feet would be perfectly valid but would not, as a point of logic, conclude the contradictory of “Rivers run” in the meaning that we would normally take that sentence to have. Including this as a premise the argument would then take the form of a reductio arguing from the self-contradictory conclusion “Rivers run and rivers do not run” to the falsity of “Rivers run.” But this 11

I do not commit myself here to considering only formal validity, but am ready to include material and inductive validity in this category. Working out what is asserted is the same thing as working out the true logical form, or as Powers (1995: 312) puts it, “logical form is linguistic form – minus fallacy” and “[t]he determination of fallacy and of logical form are two sides of the same process.”

49 conclusion only appears to be a self-contradiction because of the ambiguity of “runs” so the reductio is invalid. Something failing to be a genuine contradiction is ignorance of refutation or ignorantia elenchi. All instances of fallacies in dictione seem to be instances of ignorantia elenchi for just this reason.12 This also supports the view that the fallacies in dictione are actually answering a different question from that answered by the fallacies extra dictionem. The confusion of these two questions has infected fallacy theory from Aristotle onwards. What about the other fallacies extra dictionem: accident, secundum quid, consequent, and non-cause as cause? All of these are fallacies. I disagree with Powers’ rejection of the fallacy of consequent as a fallacy; I do not think that the fact that this is obviously invalid is to the point. These can all, as Aristotle argues in the Sophistical Refutations (1965: §6), be reduced to ignorantia elenchi. There are two parts of a refutation – proof and contradiction – and a narrower and a broader sense of ignorance of refutation. In its narrow 12

See especially Aristotle (1965: §19) where, after describing that one should correct an equivocating argument by saying of the equivocation “in one sense it is so, but in another not so” he nonetheless concludes by saying “even though he draws his conclusion in a quite unambiguous manner, one should contend that what he has negated is not the fact which one has asserted but only its name; and that therefore there is no refutation”, which looks very much like ignorance of refutation. Note what is not being claimed here. Unless the arguer is being deliberately deceptive or has the attention span of a goldfish, it is not that the arguer uses the ambiguous term in one sense in one premise and in another sense in the other premise, thus creating a fallacy of four terms that does not even appear to be a good argument; it is that although the arguer argues validly and without ambiguity he uses a premise (or in a reductio, the apparent contradiction in the conclusion) that he is not entitled to because it is not doxastically justified for him in the sense that he wants to use it. But one can imagine cases where a proposition is doxastically justified for the arguer and it is in fact true, yet the evidence that justifies the proposition doxastically for the arguer does not justify it propositionally. Despite the argument being valid and having true premises, we are nevertheless inclined to accuse the arguer of equivocating. In calling such an event an equivocation and a mistake we ipso facto make use of our own knowledge in determining the logical form.

50 sense it means that the refutation fails to be a contradiction; this is the sense given it in the previous paragraph and the sense first given it by Aristotle (1965: §5). But the fallacies of accident, secundum quid, consequent, and non-cause as cause are instances of ignorantia elenchi because they fail to be proofs – they include an invalid inference step that does not guarantee the transmission of truth, which is not the same, of course, as saying that what is inferred must be false, but only that it is not true by logical necessity. 3. Objections It will be objected that there are a number of issues that I have simply put aside. I have put aside the fallacies of petitio principii and many questions. I have discussed these elsewhere and given them nonAristotelian treatments such that they do not constitute counter-examples to the thesis, common to my position and the Standard Treatment, that fallacies are invalid arguments (and not dialectical mistakes, as the modern literature often takes them to be). I leave them aside here. The other objection I have put aside is the consideration of counterexamples to Powers’ thesis that a fallacy is always due to an equivocation. Zuckero (2003) provides three such examples: (1) Dividing by zero i) Let a = b ii) a2 = ab iii) a2 – b2 = ab – b2 iv) (a+b)(a-b) = b(a-b) v) (a+b) = b vi) 2b = b vii) 2 = 1 Cancelling at (v) involves division by zero so this proof is invalid. (2) The Monty Hall paradox

51 (3) The Wason selection test.13 Commenting on Powers paper “Dividing by zero and other mathematical fallacies” Zuckero says that Powers agrees that dividing by zero is a fallacy and explains this fallacy as equivocating between two senses of ‘number’ – the argument form illustrated by this fallacy is valid when applied to the positive rationals but not when applied to all rationals. Zuckero (2003: 288) replies that the arguer does not equivocate but simply does not recognize zero in its disguise as (a-b); if he had he certainly would not have divided by it. This is a much more plausible explanation of why this fallacy occurred than Powers appeal to equivocation. I think that Zuckero has been misled by Powers’ talk of ‘appearance’ – it is symptomatic of this misunderstanding that he refers to Powers ‘phenomenological test’ of a fallacy (Zuckero 2003: 287). Zuckero has taken Powers to have made the causal claim that fallacies are always caused by equivocation, and has correctly pointed out that as a point of psychological fact, Zuckero’s explanation of why the reasoner made this mistake is much more plausible. I think that this much can be conceded, but it is not this that Powers is trying to explain (as witnessed by his rejection of a psychological approach to fallacy) but rather why the argument has the objective appearance of a good argument. Does Zuckero’s alternative explanation explain this? Now the reasoner could argue inductively that most arguments with this appearance are good arguments and so he is quite entitled to believe that this argument is good also, but he does not explain thereby why there are some exceptions to this generalization or how they differ from those instances that are good arguments – he does not explain the objective resemblance 13

Monty Hall paradox is well-known, but the Wason selection test maybe needs some exposition, especially since Zuckero’s use of it is original. A subject is asked to turn over cards in order to verify the conditional statement “If a card has a vowel on one side, then it must have an even number on the other side.” There are four cards whose sides facing the subject have written on them “A”, “F”, “3”, “4”. The correct solution is to turn over the cards with “A” and “3” since the conditional is false if the antecedent is true and the consequent is false. However, the vast majority of subjects turn only the card with “A”.

52 of this argument to one that is good. I don’t think that Zuckero’s alternative does explain this, and his objection argues beside the point. In the Monty Hall paradox, there is an argument that one is not justified in changing one’s mind without learning anything new, and intuitively Monty’s opening a door and showing that there is nothing behind does not tell you anything that you didn’t already know since you knew already that one of the doors Monty has to choose from has nothing behind it, and the only thing you have found it is which door that is. That one does not change one’s mind is due to the reasoner failing to realize that this is relevant evidence and not due to any kind of equivocation (Zuckero 2003: 289-290). It seems to me that when it is presented as it is above there is quite clearly an equivocation, namely between (what is prior) “I know that there exists a door such that nothing is behind it” to (what is posterior) “There exists a door such that I know that nothing is behind it.”14 Since the former is entailed by the latter, but not the converse, it is obvious that the latter contains information that the former does not. True, it is not at all obvious what the relevance of this additional information is, and Zuckero may be right in attributing to the failure to note its relevance the reasoner’s mistake. But this, once more, is not the point. The Wason selection test is harder to judge because it seems that there are several candidates for the mistake. Is it simply that you cognitively lock on to whatever comes first in the sentence? One can imagine reading “If a card has a vowel on one side…”, noting which cards in front of you have a vowel, namely the card with “A”, next reading “…then it must have an even number on the other side”, and then turning over the card with “A”. This amounts to treating the hypothetical as two propositions instead of one, of thinking that because “A card has a vowel on one side” is a proposition then it is also a proposition when embedded in a conditional. This would be, I think, a fallacy of form of expression, and hence under Powers’ concept of

14

These could be symbolized as Ka(Ex)(the door with nothing behind it = x) and (Ex)Ka(the door with nothing behind it = x). See Hintikka (2005: 103-105).

53 equivocation.15 It could be that the reasoner simply did not think about the contrapositive, and it is worth remembering that contrapositives are only logically equivalent and are not necessarily equivalent in meaning – this is demonstrated by the fact that they have different contraries. This also seems interpretable as an equivocation. Or it could be that the reasoner does not make either of these mistakes, but simply fails to realize that turning the card marked “3” verifies this contrapositive. But in principle this could happen equally with “A” so I am not convinced that this would amount to a fallacy. 4. Conclusion A fallacy is something that appears to be a valid argument but is not. This has led to a concept of fallacy with two aspects: its appearance and the falsity of that appearance. This is called the cause of illusion and the cause of defectiveness by Buridan. I believe that the concept of fallacy does not have two aspects but only one, viz., that its true logical form is invalid. An invalid logical form does not have even the appearance of a valid logical form, so for instance if an argument equivocates on a middle term then the argument commits the fallacy of four terms that is self-evidently invalid. The concept of fallacy is to be identified with its vice or cause of defectiveness, where this is, in the case of the fallacies 15

There is a syntactical similarity (the same form of expression) between “If a card has a vowel on one side, then it must have an even number on the other side” and “A card has a vowel on one side and an even number on the other side” because of which someone may take the embedded sentences as being propositions in both cases, or as we would probably say today, taking them as having been uttered with the same illocutionary force. Although we should probably not read a force/content distinction into Aristotle and Buridan, the fact that they see conjunctions – e.g. those involved in the fallacy of many questions – not as propositions but as a plurality of propositions amounts to much the same thing. It is worth noting that not all of the fallacies in dictione are due to double meaning. The fallacy of division and the fallacy of form of expression do not involve ambiguous terms, but do involve linguistic confusion of some kind (Aristotle 1965: §20 and §22). They are nonetheless equivocations in Powers’ liberal usage of this term.

54 in dictione, a failure in contradiction, an ignorantia elenchi. This provides a positive characterization of the cause of defectiveness in these cases. The question “what is a fallacy?” is always to be answered “ignorance of refutation, in the broad sense”. This can be divided again into those that fail because of a failure in contradiction and those that fail because of an error in proof. These are both logical mistakes. Fallacies are fallacies extra dictionem; there are no fallacies in dictione strictly speaking because arguments themselves are not ambiguous. However, as Powers argues, calling an argument fallacious would be vacuous unless it is explained why the argument goes wrong, and I think there is good reason to think this amounts to pointing out an ‘equivocation’ liberally construed. If so, then the one fallacy theory can be reinterpreted as a theory about how to answer the question “why do fallacies occur?” meaning by this not a psychological/etiological question but rather “why do fallacies appear objectively to be good arguments?” This corresponds to the virtue or cause of illusion. This is not another aspect of the same concept. “What is a fallacy?” and “why do fallacies occur?” are separate questions that fallacy theory has, from its inception and to its disadvantage, stuck together. There is a one fallacy theory for each of these questions. Hence I call this the two one fallacy theory theory. References Aristotle 1984. Sophistical Refutations (translated by W. A. Pickard-Cambridge). In: Jonathan Barnes (ed.) The Complete Works of Aristotle. Princeton: Princeton University Press. Buridan, Jean 2001. Summulae de dialectica (translated by Gyula Klima). New Haven and London: Yale University Press. Gilbert, Michael 1997. Coalescent Argumentation. New York: Routledge. Hansen, Hans V. 1996. Aristotle, Whateley, and the taxonomy of fallacies. Practical Reasoning 1085, 318-330. Hintikka, Jaako 2005. Knowledge and Belief. Reprint edition. London: King’s College London Publications. Powers, Lawrence 1995. The one fallacy theory. Informal Logic 17 (2), 303-314.

55 Zuckero, Matthew 2003. Three potential problems for Powers’ one-fallacy theory. Informal Logic 23 (2), 285-292.

Elżbieta Chrzanowska-Kluczewska Jagiellonian University [email protected]

Modal Games in Natural Language Abstract: Ludwig Wittgenstein never approached the issue of language-games played with modal expressions in much detail. The problem was taken up in the late 1960s by Jaakko Hintikka and his collaborators within Game-Theoretical Semantics (GTS), which – as a possible-worlds semantics – creates an obvious logical environment for the discussion of modality. However, even within the framework of GTS modality has never received a holistic treatment. The focus of GTS has been predominantly on epistemic modal games with a propositional scope. Discourse games, in the shape proposed by Lauri Carlson (1983), have not centred on modal issues. Hence, the article tries to complement this theoretical gap by bringing in the discussion of the concept of narrative modalities as initiated by Lubomir Doležel (1976) and practised ever since within the framework of literary semantics and text-world theories. Key words: modal games, Game-Theoretical Semantics (GTS), possible-worlds semantics, propositional/sentential games, discourse games, narrative modalities

0. Introduction The aim of this article is to analyse in some detail the concept of Wittgensteinian language-game as formalized in Game-Theoretical Semantics (GTS), propagated uninterruptedly since 1968 by Jaakko Hintikka, his collaborators and followers. GTS is a combined system which brings together elements of possible-worlds semantics, the mathematical theory of games and Wittgenstein’s pragmatic conception of language-game as a link between language and reality. The fact that GTS assumes the existence of possible worlds in its model axiomatically makes it appropriate for the analysis of modality. Thus, I intend to focus on modal expressions as operative in natural language and consider whether modal games first suggested by Hintikka in 1969 (discussed also in Hintikka and Kulas 1983) make sense for the

58 description of various modalities in English. According to Hintikka and Kulas, modal operators can be treated as quantifiers ranging over possible worlds, so each move in a game indicates a journey between two worlds. The modal branch of GTS still lies open to further research, despite a number of studies devoted to or only touching on this subject. In what follows, I intend to compare two types of modal games: 1) propositional games (with clausal and sentential scope) and 2) discourse games (first proposed by Carlson 1983, discussed also in Chrzanowska-Kluczewska 2004). A standard logical subdivision of modalities (and, consequently, of possible worlds) into alethic, epistemic, deontic, boulomaic and axiological is applicable to both categories. Discourse games, of special interest to me as a literary semanticist, are related not only to dialogue-games of interest to Discourse Analysis but also to narrative modalities of literary semantics (Doležel 1976, Chrzanowska 1990b, Ryan 1991, Simpson 1993) in the sense of global story-forming (macrostructural, text-organizing) constraints. The juxtaposition of two different theories, one coming from the tradition of formal semantics, the other from text and discourse studies, is a consequence of the logicians’ preoccupation with propositional rather than textual phenomena and the resulting lack of anything like a developed apparatus to describe the latter. 1. Main tenets of Game-Theoretical Semantics (GTS) Game semantics in general can be traced back to dialogical logic initiated by Paul Lorenzen early in the 1960s (Dialogspiele, 1960), probably best known from the subsequent study written with Kuno Lorenz (1978). Game-Theoretical Semantics (henceforth GTS), in turn, has been a system propagated by Hintikka since the publication of his seminal Language, Language-Games and Information (1968/1973),1 first in application to formal languages and then to fragments of natural language (English). Hintikka has always been very decided in claiming 1

In 1982, in one of his essays Risto Hilpinen suggested that Charles S. Peirce could, actually, be treated as a precursor of GTS (cf. Hintikka and Kulas 1983: 255).

59 that GTS is an authentic extension of Wittgenstein’s semantical ideas, that is of his concept of language-games. Hintikka, together with Erik Stenius (1967), have initiated a long and fruitful line of research on the formalization of the informal Wittgensteinian notion of game played with language, which has included – chronologically – such logicians and philosophers as Esa Saarinen, Risto Hilpinen, Lauri Carlson, Jack Kulas, Gabriel Sandu, Johan van Benthem, and many others. Throughout that time their ideas have been applied in various logical systems and programming languages (computer science, AI studies). 1.1. GTS as a model-theoretic and truth-functional semantics GTS is based on the model encompassing not only the real world (actual world, aw) but also possible worlds (pw’s), according to the relation of accessibility awRpwn, as formulated by Saul Kripke (1963/1971), and developed by Nicholas Rescher (1975). Since possible worlds are accepted axiomatically in GTS, the system can be classified as an instance of possible-worlds semantics. Accessibility among possible (alternative) worlds should be understood as imaginability/conceivability from the point of view of actuality, with worlds being either proximate or remote to our experience. Unsurprisingly, GTS is also truth-functional (partly in the spirit of Alfred Tarski and Donald Davidson but with several departures from their ideas). The utmost goal is to assign to propositions, whenever possible, a truth-value in the model. All the games are, ultimately, subordinate to the supergame of evaluation, that is they are used to verify or falsify propositions (Hintikka calls it a “dynamic” semantics). In this respect, GTS comes close to modal logics in which truthvaluation is made possible by the verification/falsification in an appropriate pw (or all pw’s in the case of necessity): Thus the suggestion yielded by the historical background of possible-worlds semantics is that it should join forces with game-theoretical semantics. Both are aspects of the same general approach, and hence can be expected to be well suited not only to complement each other but to join forces. Roughly speaking, the idea is this: PWS is the static theory of possible worlds,

60 whereas GTS is the abstract theory of possible-world exploration. For the activities included in the moves in the semantical games of GTS are intuitively speaking [...] of the nature of moves in a “game of exploring the world”. (Hintikka 1989: 56)

1.2. GTS and the mathematical theory of games In its main outline GTS accepts the main tenets of the mathematical theory of games as formulated by John von Neumann and Oskar Morgenstern (1947), John Nash (1950) as well as R. Duncan Luce and Howard Raiffa (1958). It is basically concerned with propositional games as played on natural language clauses/sentences. These are, typically, 2-player games in a dialogue schema of question-answer. Hintikka has given various names to his players, for instance I/Myself as playing against Nature. This personification stands for a non-cooperative player, omniscient but malicious, and is reminiscent of the Gadamerian activity of “posing questions to Mother Nature” (Gadamer 1960), seen also as Luck or Fortune. Other names given to the players at various stages of the development of GTS are: Proponent/Defender vs. Opponent, Verifier vs. Falsifier, Eloise vs. Abelard, Builder vs. Critic (van Benthem 2006) or System vs. Environment (this de-personified schema has been applied in theoretical computer science, cf. Abramsky 2006). The duality shown by the players’ behaviour can also be interpreted as instantiating provocation vs. reaction (cf. Caillois 1958/1967, or attack vs. defence schema, cf. Rahman and Dégremont 2006), thus highlighting the agonal (competitive) dimension of human games. It bears also a certain resemblance to Manfred Eigen and Ruth Winkler’s (1975) conception of biochemical processes seen as regulated by two counter-players: Life and Death. On a less grand scale, van Benthem (2006) compares these games of evaluation to a case in court. Later in his theory, Hintikka uses sometimes a 3-player arrangement, where the protagonists are now: Eloise, Abelard and a mysterious Oracle, substituting Nature, who tries to hide information from Eloise and Abelard. The idealization applied in the mathematical theory of games is that what is being played are full information games, in which strategies are

61 well known to both players, who also boast an excellent memory of all the previous moves, both on their own and their opponent’s side. These games should also be 0-sum, where the gain of player1 is equal to the loss of player2. Ideally, these games are also closed, that is the ultimate verification/falsification should proceed in a finite number of moves. The play of each game can be visualized in the game-matrix (a tableau or a game-tree). What is of utmost importance for each game is its strategy, viz. a set of optimal moves, called often a winning strategy by Hintikka. In this sense, a strategy is a superstructure of the game, understood by Hintikka as a quantification over rules (Hintikka 1996: 294) or a function of functions (viz. a functional, calling for a higherorder logic as its representation, cf. Hintikka and Carlson 1979: 183, Hintikka and Kulas 1983: 47). In brief: A sentence will be true in the model if the Proponent’s strategy is a winning strategy [...] a sentence will be false in the model if the Opponent’s strategy is a winning strategy. (Rahman and Dégremont 2006: 112)

In other words: … someone must show that the formula is true, and someone else tries to show that it is false. (van Benthem 2006)

A prototypical sub-game in GTS is the game of seeking and finding, according to Hintikka a natural activity for the evaluation of, e.g. quantificational expressions in natural language. Since this game has become a classic in the literature, we will quote it here as an example of a semantical game (propositional game). Let us begin with an English sentence which contains an existential quantifier in its semantic representation, realised as a 0-article at the grammatical level: (1a) There are black swans in this country. (1b) ∃x (Swan (x) and Black (x)) (1c) There exists at least one x such that x is a swan and x is black. At this point Hintikka proposes introducing the game of looking for and, hopefully, finding at least one black swan to validate this proposition:

62 (1d) One can find a black swan, if one searches/seeks/looks for it. The game of seeking and finding is a new way of expressing the fact that quantificational expressions depend on valuation, that is a substitution of an entity for the variable(s) bound by the quantifier(s). Proponent has a winning strategy if s/he founds an appropriate swan as a designation, a proper “member of our universe of discourse” (Hintikka 1996: 278). Similarly, if an English sentence happens to contain a universal quantifier (like in 2a) below, where it is realized grammatically as the definite article the in the generic sense): (2a) The lion is a friendly beast. (2b) (x) ((Lion (x) → ((Beast (x) a Friendly (x))) (2c) For all x’s, if x is a lion, then it is a friendly beast (‘beast’, the hyperonym of ‘lion’, is taken here as its synonym), in Hintikka’s parlance, the gamesome character of the evaluation of this proposition will be made manifest in the following formulation: (2d) One cannot find a lion such that it would not be a friendly beast. Proponent (Verifier) has to demonstrate that all substitutions of particular lions as referents, found in the game of seeking and finding, work for this sentence (Chrzanowska-Kluczewska 2004: 40-41, after Hintikka 1968/1973). 1.3. GTS and the Wittgensteinian conception of language-games As has been mentioned above, Hintikka considers his theory to be a direct formalization of Wittgenstein’s crucial conception of languagegames as genuine links between human language and reality. These connections are typical semantic relations and, according to Hintikka (2000: 40-41), Wittgenstein’s ideas are in fact the “thesis of the universal role of language games as mediators of meaning”. Wittgenstein sketched the initial concept of his games in The Blue and

63 Brown Books (which originated in the years 1933-1935), but expounded it, famously, in The Philosophical Investigations (henceforth PI, 1953: §7): “I shall also call the whole, consisting of language and the action into which it is woven, the ‘language-game’”. Rephrased by Hintikka (1996: 307): “A language-game is a rulegoverned [linguistic] interaction with our non-linguistic environment”, in which vertical links relate words to objects and sentences to facts. Hintikka (1996: 285, 304) also refers to games so conceived as tools in our dealings with the world, thus instruments for carrying out nonlinguistic activities; at yet another place he refers to them metaphorically as a bridge between language and the outside world (Hintikka 1996: 284). He also dubs them outdoor activities, pointing to the fact that language-games in GTS are extra-, rather than intralinguistic, phenomena, embedded in the practical dealing with reality. In this sense, GTS, as an inheritor of the late Wittgensteinian approach, forms a system open towards pragmatics and has always invoked the need of context-sensitivity in interpreting sentences of natural language. Hence follows the subdivision of games into I-games, “governing such more internal processes as deduction or checking of coherence in discourse”, and E-games, “involving interaction between a cognitive agent and some independent outside environment” (Hintikka cited in van Benthem 1990: 241). Yet, there exists one major difference between Wittgenstein’s idea of language-games and its formal development in GTS, connected with the issue of rule-following. For Wittgenstein games were always primary with respect to rules (he describes it as a conceptual priority of languagegames over rules, cf. PI I, §138-242, also Hintikka 1996: 273, 297) and, consequently, rules could only be deduced from the practical exercise of games. This idea of the relative order of rules and practice is called the holism of language-games. This is not the case in GTS, where strict rules are specified in advance and in accordance with ordering relations, of which the most important ones are the Progression Principle (O.LR) and the Command Principle (O.Comm). According to the former, games are played from left to right, whereas according to the latter top-down within a sentence, which can be seen as a gradual unpacking or decomposition of complex sentences into atomic clauses. This

64 assumption raises the question about the (non)compositionality of GTS, an issue to which we return in 2.2. 1.4. Discourse games (dialogue/dialogic games) of GTS Already in his enumeration of language-games in PI (§ 23) Wittgenstein listed the games that definitely went beyond the limits of a phrase or sentence, such as reporting an event, making conjectures, making jokes, guessing riddles, telling a story, etc. They all can be connected with a production of a text (which requires minimally a sequence of two coherent utterances). A minimal text is, obviously, an exchange of a question-answer between two interlocutors. This observation prompted Carlson (1983) to take up the problem of a GTS representation of largescope games, which he called discourse games (cf. also Hintikka 1984). From the perspective of our approaching discussion of narrative modalities, it has to be emphatically stated that this particular development of GTS has been, undoubtedly, a correct theoretical decision. Human communication does not rely on single, isolated sentences but is carried out through text/discourse formation, the fact that has oftentimes been forgotten or ignored not only by logicians and formal semanticists but also by practising linguists. Carlson’s concern was mainly with dialogues seen as n-player games, typically everyday conversations (which are nowadays the object of investigation of Discourse Analysis). Each sentence (S) is a move in a particular game (G), thus we can distinguish between an initial move, continuation move(s) (by the same player) and counter-move(s) (response/reaction by the other player(s)). In turn, every text is perceived by Carlson as a hidden dialogue in the underlying structure, with a gappy surface realization that requires textual interpolations (cf. Carlson 1983 and a more extensive discussion in ChrzanowskaKluczewska 2004). What is important, this kind of textual incompleteness should not be confused with the interpretative gappinesss of a text, which undergoes the completion by the reader in the process of interpretation, in the sense of Roman Ingarden’s phenomenological notion of concretization (cf. Ingarden 1931/1973).

65 It is a purely technical description of a text within the GTS format, which must proceed according to the rigid question-answer pattern, in analogy to sentential games of truth-valuation played between Myself and Nature, for instance. Since every text is construed according to two major text-forming principles (called sometimes the criteria of textuality): 1) cohesion (overt, grammatically signalled connectedness of sentences) and 2) coherence (covert, logical sequence of ideas that build the text-world), GTS compares these two phenomena to the game-rules and to the game-strategy, respectively. In this way coherence is raised to a higher status as a holistic organization of a text. In my monograph on games (Chrzanowska-Kluczewska 2004: 50) I provide a critique of the Carlsonian idea that every text (including narration or description) must be expanded into a dialogue. The Carlsontype of the filling of gaps with apparently missing portions of the text seems to be not only inelegant but also artificial. The attempt at imposing a unified dialogic structure onto all kinds of discourse genres and sub-genres, though theory-induced, seems not only unfeasible but also missing the descriptive richness of text-creating means in natural language. On the other hand, in my opinion one of the greatest theoretical assets of introducing the notion of dialogue or discourse games into GTS was the fact that evaluation (in the sense of verification/falsification) has been suspended for such games, in contradistinction to propositional/ sentential games, in which the assignment of a truth-value to a given sentence is an ultimate goal. This agrees with a fundamental observation about the functioning of human language, in which the concern for truthfulness or falsity of utterances is very often immaterial. Actually, only a limited area of any natural language, comprising unmodalized declarative sentences (and modalized propositions only on condition a possible-worlds semantics will support their evaluation), can be subjected to the verification against the factual data from the world. As the Speech Act Theory has convincingly demonstrated, apart from assertives, all other speech act types do not rely on the notion of truth/falsity but rather on the idea of felicity/infelicity of execution. Text studies and literary semantics will also point to the fact that the question of truth is irrelevant or does not arise for fictional discourse (where

66 we can talk only about ‘truth-in-fiction’ or ‘truth-by-convention’ as a secondary phenomenon). The most important aspect of dialogue games, underscored by Carlson, is the cooperation between the participants in this kind of gamesome activity, contrary to sentential games played between Myself and Nature, which are basically competitive if not straightforwardly conflictive (agonal). I agree with Carlson on this point, seeing cooperation as a prevalent feature of linguistic communication. The conclusion which follows from the above considerations is that a gametheory that aims at describing human linguistic behaviour must take both its cooperative and agonal character in consideration. The same applies to human gamesome transactions that are only partly linguistic or nonlinguistic in character. Not without reason do Avinash K. Dixit and Barry J. Nalebuff claim that: We have since come to the full realization of the important part that cooperation plays in strategic situations, and how good strategy must appropriately mix competition and cooperation. (Dixit and Nalebuff 2010: ix-x)

Thus co-opetition is a smart blend that catches the essence of the ambiguity of human gamesome behaviour. 2. Modalities and modal games in natural language Although it can be claimed that the Wittgensteinian idea of logischer Raum (understood as an infinite container for all modal notions) contributed considerably to the later development of possible-worlds semantics, we owe a full-blown version of this theory to analytical philosophers and modal logicians, with Hintikka being one of its founding fathers: In its different forms, possible-worlds semantics has been by and large an eminently successful logico-semantical theory. It began its life as a theory of logical (conceptual) necessity and possibility, inspired by the old idea of necessity as truth in all possible worlds. Soon it was broadened into a framework for a general theory of meaning, based on the idea of meanings

67 as functions from possible worlds to extensions. Possible-worlds semantics can also, and in my judgement much more naturally, be used to explicate the semantics of such propositional attitudes as knowledge, belief, perception, etc. For what could be a more natural explication of what someone knows than to spell it out in terms of the possible scenarios (“possible worlds”) which the information he or she possesses enables him or her to disregard? (Hintikka 1989: 52)

2.1. Taxonomy of modalities and their respective possible worlds Roughly speaking, the idea behind possible-worlds semantics is that modalized proposition, which in actual world escape truth-valuation, can be assigned truth-values in alternative (imaginable) scenarios, states of affairs, or worlds. While necessity requires travels to all possible worlds, other modalities will be satisfied with a discovery of at least one possible world in which the propositions on which they function as modal operators will come true. In accordance with the taxonomy of logical modalities mentioned in Introduction, we can draw a parallel subdivision of modal worlds: (a) Alethic worlds correspond to alethic modalities of classical logic running along a scale, from objective necessity, via natural possibility (this term comes from Hintikka 1979: 337) measured by various degrees of probability, to impossibility. The dictum (3b) of the modalized proposition (3a) that expresses impossibility: (3a) He can’t be my father (It is impossible that he is my father) (3b) He is my father will come false in all possible worlds (impossibility is thus the opposite of necessity). (b) Epistemic worlds will correspond to a cline of such notions as knowledge, different degrees of belief, different degrees of doubt and ignorance. Belief is sometimes classified as a separate doxastic modality.

68 (c) Deontic worlds will be such worlds in which duties, obligations, permissions and prohibitions come true. (d) Boulomaic worlds reflect the most subjective group of volitional and emotive modalities such as wants, wishes, desires, hopes and fears. (e) Axiological worlds come in two versions: as supporting either aesthetical modalities (focusing on such key concepts as beautiful vs. ugly, preferable) or else ethical modalities (good vs. bad, proper vs. improper).2 2.2. Modal games in GTS Modal games, albeit not yet under this official name, were already alluded to by Wittgenstein (1956/1994), who had an intuition that sentences that contain modal expressions have a peculiar use and distinct application in language and that they deserve a separate treatment. Like many of his ideas, this one had to wait for further elaboration.3 Hintikka (2000: 44-47) discusses an important difference between two types of games distinguished by Wittgenstein. He calls them primary and secondary (Wittgenstein, apparently, never applied such terms). The former catch basic human transactions with the environment mediated through sense-experiences, hence basically “spontaneous reactions” to the surrounding world connected with e.g. bodily and emotional 2

3

This taxonomy of modalities is not followed unanimously by all the researchers. Georg von Wright (1951), for instance, mentions four basic modal categories: alethic, epistemic, deontic and existential (the last one, closer to quantification, covers universality, existence and emptiness, cf. also van Benthem 2006: 130, who refers to universal and existential modality). Hintikka favours epistemic and doxastic modalities, Rescher was particularly interested in the Value Theory, hence axiological modalities. Some logicians and philosophers have also singled temporality as a candidate for a modal category. The five-fold classification I have accepted for the purposes of this article follows Doležel’s taxonomy of 1976 (cf. also Chrzanowska 1990a, 1990b). In Blue and Brown Books Wittgenstein refers cursorily to a family resemblance between such modalities as possibility and capability/ability.

69 reactions (feelings of pain, cold, pleasure, anger, etc.). The latter, which Hintikka (2000: 45) calls, somewhat pejoratively, “parasitic” on the primary games, deal with, for instance, epistemic and alethic notions such as doubt, certainty, justification, knowledge, belief, disbelief, etc. (cf. Hintikka 2000: 46).4 Consequently, we have to extrapolate this qualification on the remaining types of modalities. The marriage of possible-world semantics with game-theoretical principles started to be executed by Hintikka already in 1969. Modal operators (in English realized as modal auxiliaries, modal adverbs, modal adjectives, modal nouns, modalized main verbs of propositional attitudes or a grammatical category of mood) are treated in GTS as quantifiers ranging over pw’s. Necessity is a quantification over all pw’s, other modalities resemble an existential quantification over at least one pw. These modal games, particular moves of which can be seen as travels across the Wittgensteinian logischer Raum, help in the verification of modalized propositions. The competitive aspect of such games, played by Myself against Nature, is revealed in the successful selection of an appropriate world that would support a given modality: At each stage of the extended games, the two players are considering a sentence S’ and a world ω’, beginning with S and ω. The game rules for epistemic and other modal notions (in the wide sense of the term) can then be formulated as rules for stepping from one world to another. (Hintikka 1979a: 41-42).

The broad sense in which modality is understood in GTS refers to the fact that conditionals and all intensional, opaque contexts might be subsumed under this category, in brief all those sentences which escape a regular truth-value assignment in a simple single-world model, calling for a more complex way of evaluation. Hintikka (1979b: 68, 78 fn. 37) signals in this connection a very important theoretical issue, namely that the treatment of modality as a quantification over pw’s implies the need for second-order logic. In this way, GTS – similarly to Montague-type grammars which rely on higher-order functions (cf. Chrzanowska 4

Interestingly, lying belongs with secondary games (cf. Hintikka after Wittgenstein 2000: 44-45).

70 1990a) – goes beyond the description of natural language executed with the apparatus of first-order (FO) logic, as practised in Generative Semantics or generative grammar in general (through the construct called Logical Form). Hintikka (1990: 319) states bluntly that “FO languages are not the right representation of natural-language semantics”.5 For the sake of brief illustration, let me adduce an example of a topdown (hence non-compositional) analysis of a mini-text, consisting of two complex sentences, that pivots on the doxastic verb believe. Truth, a crucial concept in this link of language with reality, equals a possession of a winning strategy by Myself (Verifier, Proponent): (4a) Joan believes that someone stole her car. (4b) She believes he tried to break into her house. This in Hintikka’s (1962, 1989) notation translates as: (4c) BJoanp1. BJoanp2 (B stands for the doxastic Belief operator). In (4a) the supergame G consists of two subgames G(S1), G(S2). G(S1) starts from the matrix clause: (G. believes that) is a modal game triggered by a doxastic modalized main verb. Nature (Falsifier, Opponent) selects a pwr; Myself (Verifier) wins if this world is the one in which Joan’s belief comes true and moves to the subordinate clause, to start G(S2). Now (G. pro) applies twice: the anaphoric her has to be related to its antecedent in order to specify its reference and Myself selects an individual, say Jack, from the domain D valid for pwr, to valuate the quantificational expression someone. If the atomic proposition Jack stole 5

Two additional issues are related to this important theoretical problem. The first is an implication of the definition of the game-strategy as a superstructure of the game proper (cf. 1.2.). This means that we need a third-order logic to treat game strategies adequately. The second is the point raised by Hintikka in several of his writings: natural language quantification far exceeds the classical quantification of first-order calculus and requires a higher-order description. The recourse to a higher-order logical description for natural language is thus motivated by at least two different linguistic phenomena.

71 Joan’s car comes out as true in pwr, Myself wins and proceeds to play another supergame on sentence (4b). Likewise, if Myself has a winning strategy in G (S1), the game continues for G(S2). The modal game is repeated and in both clauses of (4b) (G. pro) has to be applied three times as the subgame of finding referents for all the pronominal expressions (after Hintikka and Kulas 1983). Hintikka and Kulas assume the following ordering relation of the moves by Myself and Nature: the stronger modal operators, such as necessity, knowledge, belief, obligation, etc. mark Nature’s moves, the weaker ones, such as natural and epistemic possibility, permission, etc. mark Myself’s moves. A move, as mentioned before, consists of a selection of one of pw’s (what van Benthem 2006: 130 calls the “opening of a new world”). An interesting issue was raised by Hintikka and Kulas (1983: 123 -124) in connection with the following example: (5) Joan believes that a burglar might have broken into her apartment yesterday, and she believes that he might have stolen her pearls (italics mine). Similarly to 4a) and 4b), this is a sequence of two complex sentences, joined by the conjunction and to form a coordinate construction (compound sentence). We have to assume, then, that the supergame consists this time of two sentential games G(S I) and G(S II). G(S I), in turn, will proceed top-down as a sequence of G(S1) and G(Ss) played on the matrix and subordinate clause, respectively. G(S II) will then be analysed as a hierarchy of G(S3) and G(S4), on analogy to the first sentence of the compound. The question arises whether we need to repeat the modal game on the doxastic main verbs believe in both S1 and S3 and on the modal constructions might have broken (modal auxiliary + perfect infinitive) in S2 and S4, which can be treated as exponents of either alethic (objective, natural) or epistemic possibility. At this point Hintikka and Kulas (1983: 124) suggest implementing the economy principle which says: “Don’t multiply the entities you need in a semantical game beyond necessity!” Since, in fact, S I and S II refer in their predicates to one and the same belief and possibility, respectively, there is no need to play the game of picking appropriate pw’s every time

72 we come across a modal predicate. Once the appropriate pw has been found for S1, there is no need to repeat the game again. This commonsensical decision, however, is related to a fundamental theoretical issue about the compositionality or lack thereof in GTS. Hintikka and his school have discussed reasons for the acceptance of non-compositionality on several occasions, claiming that we need a semantics that works outside-in, contrary to the inside-out compositional systems along the lines of Gottlob Frege, Alfred Tarski or Richard Montague. Moreover, GTS attempts at accounting for the links and relationships holding between elements distant in the S-space (like our modal expressions in (5) above). It aspires to be a system capable of looking backward and forward (accounting for long-distance syntactic and semantic phenomena), as well as outside, to the context. Saarinen (1979: 252) aptly summarises this problem of ‘crossing scopes’ with the recourse to backwards-looking operators which “enable reference back to a possible world [like in our example (5)] (or, more generally, to a point of reference or a feature of a point of reference) that has been considered earlier”. Thus, modal games do not need to rely on one-way journeys between aw and various pw’s but allow for returns to the previously visited worlds. By no means do modal games operate in isolation; to the contrary, together with other subgames that take care of quantification proper, temporal and anaphoric relationships, sentential focus, etc., they serve in the evaluation of modalized propositions within logical space. Additionally, Hintikka has introduced into game-theoretical description the so-called partial-information games (based on branching quantifiers) that reflect information-independence of human linguistic behaviour. In this respect GTS departs from the strict formalism of the mathematical theory of games, which assumes full-information games as primary, and comes closer to variants of game-theory elaborated for other types of human activities, such as economic behaviour (cf. Dixit and Nalebuff 2010).6 6

The interesting and important issue whether GTS is really non-compositional, or rather partly compositional, goes far beyond the scope of this paper and bears on fundamental questions of the construction of semantic representation for natural

73 2.3. Discourse modalities (beyond GTS to Text-World Theories) Contemporarily with the development of GTS in the 1970s, Lubomir Doležel, while working on text theory and narratology, observed that: Modalities seem to display several properties which make them suited for the role of global story-forming constraints. (Doležel 1976: 6)

He suggested that we should pay attention not only to what he called “particular manifestations of modalities” (propositional modalities in our terminology) but also to “general logical modalities” (viz. discourse modalities). The latter “can transcend the scope of individual actions and assume a role of global (macrostructural) constraints dominating a whole sequence of actions” within a narrative. He drew his ideas both from the structural semantics of Algirdas Greimas and from Georg von Wright’s (1951) considerations on the global role of deontic modalities. A similar idea, actually, is presaged by Hintikka, who states that: The scope of a quantifier word can apparently extend arbitrarily far in a sentence and even in a text or discourse. (Hintikka 1979b: 81-82)

Remembering that modal operators are treated as quantifiers ranging over possible worlds, this extension does not come as a theoretical surprise. In a similar vein, Hintikka and Carlson (1979: 213) talk about generalizing certain regularities from sentence semantics to text semantics, which finds its ultimate development in the Carlsonian schema of discourse games of 1983. The propositional modal games can sometimes merge smoothly into discourse games. Already our examples (4) and (5) above, taken from colloquial, everyday language, illustrated minimal two-sentence epistemically-oriented texts. Let us consider for a change the following short passage taken from a literary source, Bhagavad-Gita. The Song of God, one of the greatest religious documents and expositions of the

language and of its “psychosemantic realism” (for more details cf. Chapter 10 of Hintikka and Kulas 1983).

74 Vedanta philosophy. This epic poem, originally in Sanskrit, has been rendered in English as a prosimetrum (prose with poetical inserts): (6a) A man should not hate any living creature. Let him be friendly and compassionate to all. He must free himself from the delusion of ‘I’ and ‘mine’. He must accept pleasure and pain with equal tranquillity. He must be forgiving, ever-contented, self-controlled, united constantly with me in his meditation. His resolve must be unshakable. He must be dedicated to me in intellect and in mind. Such a devotee is dear to me. (Krishna instructing Arjuna, Bhagavad-Gita 1944/1972: 99, italics mine)

(6b) Oxp1. Oxp2. Oxp3. Oxp4. Oxp5. Oxp6. Oxp7 (where O is a generalized Obligation operator, cf. Chrzanowska 1990a after von Wright 1951) The above-quoted passage, closing with a refrain repeated in some variation in the following paragraphs of the text, consists of seven sentences that consequently realize the deontic modality. According to the principle of economy postulated by Hintikka, instead of playing repeatedly seven consecutive games of locating a pw in which the situation prescribed by Krishna does occur, it is enough to invoke one common game of searching for an appropriate alternative reality. Yet, the deontic pw in which the truth-valuation of the particular propositions forming this passage can occur, in fact encompasses the entire narration of Bhagavad-Gita, which is predominantly obligation-based, an instance of religious instruction as a textual genre. Over the years the idea of narrative modalities, supported by appropriate possible worlds (fictional worlds, text worlds, discourse worlds) has been uninterruptedly developed by researchers in the field of literary semantics, viz. Doležel himself (1989), Nils Erik Enkvist (1989), Marie-Laure Ryan (1991), Paul Simpson (1993). I took up this subject twice (Chrzanowska 1990b, Chrzanowska-Kluczewska 2004), first describing some discourse modalities after the fivefold logical taxonomy accepted by Doležel and next, trying to extend the Carlsonian gametheoretical paradigm to the analysis of written non-literary and literary texts.

75 As a brief exemplification, the fabulae describing the possible or rather impossible worlds of madness or dream will be classified as alethic stories (cf. Alice’s Adventures in Wonderland by Lewis Carroll, Finnegan’s Wake by James Joyce, or The Unconsoled by Kazuo Ishiguro). All mystery and detective stories can be classified as epistemic, since they show a gradual transformation of ignorance or false belief into knowledge (Hamlet or Ursula Le Guin’s cycle The Earthsea Trilogy and Tehanu). A classical scheme of long tradition in literature that shows a violated prohibition that cannot escape punishment as well as stories of penance and expiation will be characterized as predominantly deontic (Macbeth, Tess of the d’Urbervilles by Thomas Hardy, Lord Jim by Joseph Conrad, The Lord of the Rings by John R. R. Tolkien or The Name of the Rose by Umberto Eco). Boulomaic fabulae are those connected with strong emotions and passions, like love, greed or lust for power, as well as hope, with romance stories being their most typical representatives (Romeo and Juliet is an exemplar of this type, but also Miguel Cervantes’s Don Quixote). Finally, axiological stories are for instance all moralistic productions dealing with a lack of value and its acquisition. In reality story structures are seldom governed by a single modal system. Typically, a narrative is a modally molecular (heterogeneous) rather than atomic (homogeneous) structure. Thus The Bible is an intricate combination of a deontic and axiological discourse, the Conradian Heart of Darkness is basically epistemic and deontic in its modal make-up, Ishiguro’s The Unconsoled represents all five basic modalities, etc. Within the cognitive paradigm, Paul Werth (1999) started the discussion about possible worlds and modalities as conceptualizations reflected in discourse and his theory, known today as Text World Theory, has been developing as an important part of cognitive poetics and stylistics. However, the typology of modalities accepted as a descriptive tool by Peter Stockwell (2002) or Joanna Gavins (2005) does not square with the fivefold distinction accepted in this article. Stockwell relies basically on the typology of text worlds propounded by Ryan (1991: 19-20, 114-123), who distinguishes the following categories: K-worlds (epistemic), O-worlds (obligation-worlds, deontic), W-worlds

76 (wish-worlds, boulomaic), I-worlds (intention-worlds, not clearly fitting our typology) and F-universes (worlds of dreams, fantasies, hallucinations, fictions, thus alethic). In turn, Gavins bases her description on three modalities only (alethic, epistemic and deontic), following Paul Simpson’s (1993) proposal of modal grammar for the narrative. The modal structuring of texts has thus been widely approved as one of their important characteristics. It is also unquestionable that producing and interpreting a text is a gamesome activity. GTS, as a highly formalized system, has definitely proved to be descriptively more adequate to coping with propositional rather than discourse modalities. The latter are still awaiting a more detailed game-theoretical elaboration. At the present stage of development, the application of the concept of games to texts/discourses follows rather the informal patterns propounded by literary semanticists, narratologists and theorists of literature (cf. Ryan 1991, Chrzanowska-Kluczewska 2004). 3. Suggested directions of the future development of game-theory An analysis of young children’s games (Schwartz 1991) proves that the dyadic pretend play so characteristic of a certain stage in the psychological development of a child is ultimately focused on the construction of a plot (narrative). This amounts to saying that children fashion their communication on the basis of coherent discourses and their play, and later rule-governed games, are narration-advancing. These activities are also highly mimetic and help children to practice their pragmatic (contextual) and social abilities. Since games and narratives are developmentally related activities, the game-theoretic approach to natural language should focus more on the text- and discourse-forming strategies, with modality as one of its important aspects. In what concerns the applications of game-theory, a growing attention has been turned nowadays to multi-player games (n-agent games, non-0-sum games called coalitions), especially in economics and business, but also in politics, sports and social interactions (cf. Dixit and Nalebuff 2010). This signifies a departure in emphasis from the theory

77 of classical 2-player 0-sum games that was the foundation of GTS. A future direction of development of game-semantics for natural language might then require an adaptation to the above-mentioned framework, along the lines that will show a convergence of several linguistic, social and economic transactions. Finally, a growing field of logics of communication and interaction (cf. Abramsky 2006) proves that contemporary logical systems have to reach beyond classical formalizations in order to capture the nature of natural language communication in its rich semantico-pragmatic aspect. Human strategies and variants of gamesome behaviour that can be classified as unfair, deceptive or even pathological (ChrzanowskaKluczewska 2004, Dixit and Nalebuff 2010) far exceed the idealization of the mathematical game-theory. And although the computer Deep Blue managed to win a chess match with the world champion Gary Kasparov in 1997, the unpredictability of human reactions, tactics and strategies remains a continuing challenge to game and AI theorists.7 References Abramsky, Samson 2006. Socially Responsive, Environmentally Friendly Logic. In: Tuomo Aho and Ahti-Veikko Pietarinen (eds.), 17-45. Aho, Tuomo and Ahti-Veikko Pietarinen (eds.) 2006. Truth and Games. Essays in Honour of Gabriel Sandu (Acta Philosophica Fennica, Vol. 78). Helsinki: University of Helsinki. Allén, Sture (ed.) 1989. Possible Worlds in Humanities, Arts and Sciences. Berlin and New York: de Gruyter. van Benthem, Johan 1990. Computation versus Play as a Paradigm for Cognition. In: Leila Haaparanta et al. (eds.), 236-251. van Benthem, Johan 2006. Logical Construction Games. In: Tuomo Aho and AhtiVeikko Pietarinen (eds.), 123-137. Bhagavad-Gita. The Song of God 1944/1972. Trans. by Swami Prabhavananda and Christopher Isherwood. Introduction by Aldous Huxley. New York and Scarborough, Ontario: New American Library, A Mentor Book. Caillois, Robert 1958/1967. Les jeux et les homes. Paris: Éditions Gallimard.

7

Hintikka (1990: 325) claims that a computer can win a chess party due to its vast memory and computing speed, without reproducing human gamesome reasoning.

78 Carlson, Lauri. 1983. Dialogue Games. An Approach to Discourse Analysis. Dordrecht: Reidel. Chrzanowska, Elżbieta 1990a. Modality as Represented in the Logical Form of Language. Unpublished PhD dissertation. Jagiellonian University of Krakow. Chrzanowska, Elżbieta 1990b. Semantics of Discourse: Narrative Modalities. Zeszyty Naukowe UJ, Prace Historycznoliterackie 73, 153-160. Chrzanowska-Kluczewska, Elżbieta 2004. Language-Games: Pro and Against. Kraków: Universitas. Dixit, Avinash K. and Barry J. Nalebuff 2010. The Art of Strategy. A Game Theorist’s Guide to Success in Business and Life. New York & London: W. W. Norton & Co. Doležel, Lubomir 1976. Narrative Modalities. Journal of Literary Semantics V/1, 514. Doležel, Lubomir 1989. Possible Worlds and Literary Fictions. In: Sture Allén (ed.), 221-242. Eigen, Manfred and Ruth Winkler 1975. Das Spiel. München: R. Piper Co. Verlag. Enkvist, Nils Erik 1989. Connexity, Interpretability, Universes of Discourse, and Text Worlds. In: Sture Allén (ed.), 162-186. Gadamer, Hans-Georg 1960. Truth and Method. New York: Crossroad. Gavins, Joanna 2005. (Re)thinking Modality: A Text-World Perspective. Journal of Literary Semantics 34(2), 79-93. Haaparanta, Leila, Kusch, Martin and Ilkka Niiniluoto (eds.) 1990. Language, Knowledge and Intentionality. Perspectives on the Philosophy of Jaakko Hintikka (Acta Philosophiva Fennica Vol. 49). Helsinki: University of Helsinki. Hintikka, Jaakko 1962. Knowledge and Belief. Ithaca: Cornell University Press. Hintikka, Jaakko 1968/1973. Logic, Language-Games and Information. Kantian Themes in the Philosophy of Logic. Oxford: Clarendon Press. Hintikka, Jaakko 1969. Models for Modalities. Selected Essays. Dordrecht: Reidel. Hintikka, Jaakko 1979a. Quantifiers in Logic and Natural Language. In: Esa Saarinen (ed.), 27-47. Hintikka, Jaakko 1979b. Quantifiers vs. Quantification Theory. In: Esa Saarinen (ed.), 49-79. Hintikka, Jaakko 1979c. Quantifiers in Natural Languages: Some Logical Problems. In: Esa Saarinen (ed.), 81-117. Hintikka, Jaakko 1984. Rules, Utilities and Strategies in Dialogical Games. In: Jaakko Hintikka and L. Vaina (eds.), Cognitive Constraints on Communication. Dordrecht: Reidel, 277-294. Hintikka, Jaakko 1989. Exploring Possible Worlds. In: Sture Allén (ed.), 52-73. Hintikka, Jaakko 1990. The Languages of HI and AI. In: Leila Haaparanta et al. (eds.), 307-330.

79 Hintikka, Jaakko. 1996. Ludwig Wittgenstein: Half-Truths and One-and-a-HalfTruths (Selected Papers of J. Hintikka 1). Dordrecht, Boston & London: Kluwer. Hintikka, Jaakko 2000. On Wittgenstein. Belmont, CA: Wadsworth/Thomson Learning. Hintikka, Jaakko and Lauri Carlson 1979. Conditionals, Generic Quantifiers and Other Applications of Subgames. In: Esa Saarinen (ed.), 179-214. Hintikka, Jaakko with Jack Kulas 1983. The Game of Language. Studies in GameTheoretical Semantics and Its Applications. Dordrecht: Reidel. Ingarden, Roman 1931/1973. The Literary Work of Art. Evanston: Northwestern University Press. Kripke, Saul 1963/1971. Semantical Considerations on Modal Logic. In: Leonard Linsky (ed.), Reference and Modality. Oxford: Oxford University Press, 63-72. Lorenzen, Paul and Kuno Lorenz 1978. Dialogische Logik. Darmstadt: Wissenschaftliche Buchgesellschaft. Luce, R. Duncan and Howard Raiffa 1958. Games and Decisions. Introduction and Critical Survey. New York: John Wiley and Sons. Nash, John 1950. The Bargaining Problem. Econometrica 18, 155-162. von Neumann, John and Oscar Morgenstern 1947. Theory of Games and Economic Behavior. Princeton: Yale University Press. Rahman, Shahid and Cédric Dégremont 2006. The Beetle in the Box: Exploring IF Dialogues. In: Tuomo Aho and Ahti-Veikko Pietarinen (eds.), 91-122. Rescher, Nicholas 1975. A Theory of Possibility. Pittsburgh: University of Pittsburgh Press. Ryan, Marie-Laure 1991. Possible Worlds, Artificial Intelligence and Narrative Theory. Bloomington and Indianapolis: Indiana University Press. Saarinen, Esa (ed.) 1979. Game-Theoretical Semantics. Dordrecht, Boston and London: D. Reidel. Saarinen, Esa 1979. Intentional Identity Interpreted. In: Esa Saarinen (ed.), 245-327. Schwartz, Ursula V. 1991. Young Children’s Dyadic Pretend Play. A Communication Analysis of Plot Structure and Plot Generative Strategies. Amsterdam and Philadelphia: John Benjamins. Simpson, Paul 1993. Language, Ideology and Point of View. London: Routledge. Stenius, Erik 1967. Mood and Language-Game. Synthese 17(3), 254-274. Stockwell, Peter 2002. Cognitive Poetics. An Introduction. London and New York: Routledge. Werth, Paul 1999. Text Worlds: Representing Conceptual Space in Discourse. London: Longman. Wittgenstein, Ludwig 1953. Philosophical Investigations. Trans. by G. E. M. Anscombe. Oxford: Basil Blackwell.

80 Wittgenstein, Ludwig 1956/1994. Remarks on the Foundations of Mathematics. Ed. by G. H. von Wright, R. Rhees, G. E. M. Anscombe. London: Basil Blackwell. Wittgenstein, Ludwig 1958/1969. The Blue and Brown Books. Preliminary Studies for the “Philosophical Investigations”. Oxford: Basil Blackwell. von Wright, Georg H. 1951. An Essay in Modal Logic. Amsterdam: North-Holland Publishing Company.

Tadeusz Ciecierski University of Warsaw [email protected]

A Problem with Structured Propositions* Abstract: The paper shows that the paradox of the totality of propositions rests on assumptions characteristic of some theories of structured contents. Thus it is shown that the threat of possible paradoxes, contrary to the widespread opinion, is not solely connected with possible-world theories of content. Moreover, possible-world theories – due to the rejection of structural constraints put on contents – can avoid the paradox in quite an elementary way. Key words: structured propositions, Russell-Kaplan paradox

It has been claimed by many authors1 that, among various accounts of propositions, these putting no constraints on internal structure of content (e.g. conceiving contents as sets of possible worlds) are much more problematic than those treating contents as structured entities. One of several problems2 mentioned in the philosophical literature (e.g. Lewis 1986: 104-108, Kaplan 1995) is the threat of inconsistency that results from the possibility of constructing, within unstructured approaches, counterparts of set-theoretical paradoxes. The minimal moral to draw from this eventuality, it has been claimed, is that philosophers sympathetic towards them ought to rethink the foundations of their theory – maybe even explicitly admit that it is committed to a particular variant of set theory or intensional logic (cf. Oksanen 1999, Cocchiarella 2000). *

1 2

I would like to extend my thanks to Witold Hensel, Tomasz Puczyłowski, Mieszko Tałasiewicz and Bartosz Wcisło for valuable comments on an earlier version of this paper. Robert Stalnaker being a prominent exception. Of which the most infamous is still that of identifying propositions expressed by logically equivalent sentences.

82 Nobody, to my knowledge, pointed out, however, that a similar problem may arise for theories of propositions that individuate contents in a way that relies too closely on the structure of sentences. Below I argue that this is a serious mistake. The argument that shows this to be the case was formulated in 1984 by Patrick Grim (Grim 1984). Although Grim’s reasoning was intended primarily to address the issue of the existence of the set of all truths (he also mentions its possible application to ersatz theories of possible worlds), it relies on some assumptions concerning the structured nature of propositions, i.e. theories engendering the paradox are precisely those committed to those assumptions. Slightly modified,3 Grim’s argument goes as follows. Let P be the set of all propositions: P = {P1, P2, P3, ….} and 2P be its power set: 2P = {Ø, {P1}, {P2}, {P3}, ….} We may assign a true sentence to each element of 2P as follows: Ø → “P1 ∉ Ø” {P1} → “P1 ∈{P1}” {P2} → “P1 ∉{P2}” {P3} → “P1 ∉{P3}” . . {P1, P2} → “P1 ∈{P1, P2}” {P2, P3} → “P1 ∉{P2, P3}” . . 3

Grim starts with the set of all truths while we are starting with the set of all sentences/propositions – not excluding the possibility that it contains also false ones. We are also formulating the argument in terms of sentences and (possibly) their contents to make its structural assumptions explicit.

83 i.e. for each element x of 2P we assign a sentence of the form “P1 ∈ x” if P1 ∈ x or of the form “P1 ∉ x” if P1 ∉ x (the choice of this particular element of P is irrelevant). Other lists, similar to this one, are also possible. For example, one may use the identity sign instead of setmembership and its negation – the result will be a list of sentences like “P1 = Ø”, “P1 = {P1}”, “P1 = {P1, P2}” etc. (false, if the axiom of foundation is assumed). Now, if each sentence on our list expresses a different proposition, we will be able to show that there are at least as many propositions as subsets of the power set of the class of all propositions. This, of course, contradicts the well known theorem of Cantor. Note that the assumption to the effect that each sentence on our list expresses a distinct proposition (call it (A)) is essential to our argument – if someone, for example, takes propositions to be sets of possible worlds, then her move is to deny precisely this claim by noting that all sentences on our list really express a single necessary proposition (at least, if you believe that all sentences like “x ∈ {x}” etc. are necessary). In recent years Jeffrey King (King 1995, 1996, 2007) has defended a structured theory of propositions (he calls it “the new account of structured propositions”) which entails that all sentences on our list differ in contents. One of the claims of his theory is: “Sentences with different syntactic structures at LF ipso facto express different propositions” (King 2007: 95). (LF is the level of syntax that is the output to semantics). Moreover, the relation that connects elements of the structured proposition is the result of composing sentential relations abstracted from LF of some particular sentence with semantic relations holding between terminal elements of this sentence and their semantic values. This results, among other things, in consequences like: (...) sentences “Laura is happy and Scott is sad” and “Scott is sad and Laura is happy” express different propositions; similarly for “1 = 2” and “2 = 1”. In both cases, the propositions expressed by the sentence pairs will have different constituents occurring at corresponding positions (terminal nodes) in the propositions. (King 2007: 95)

To see that King’s view entails (A) note that all sentences on our list have one of two general forms (a) “P1 ∈ x” or (b) “P1 ∉ x” and that our

84 list could be divided into two sub-lists – each containing either only (a)sentences or (b)-sentences.4 Each sentence on an appropriate sub-list will differ only with respect to a particular instantiation of the variable x. Now, x takes as particular instantiations various set-denoting expressions (e.g. “{P1, P2}”) and there are generally three approaches that one may adopt towards them: (1) One may claim – though this is not very likely – that they are like proper names, i.e. singular terms without relevant internal structure. In such a case, since they denote distinct objects, ({P1, P2} is certainly distinct from {P2, P3} etc.), they have different semantic values, which means that all propositions on our list would have “different constituents occurring at corresponding positions (terminal nodes) in the propositions”. (2) One may – this is more likely – take them to be abbreviations for complex abstract terms like “{x: x ≠ x}” (which we abbreviate as “Ø”) or “{x: x = P2} ∪ {x: x = P3}” (which we abbreviate as “{P2, P3}”). That would, of course, not matter (when compared with (1)) insofar as we are assuming that a difference in reference implies a difference in semantic value. But, even if we reject this assumption (I do not think that King is committed to that), we will arrive at the very same conclusion due to the fact that the syntactic dissimilarities between our propositions would be quite noticeable (sometimes as huge as syntactic distinctions between “{x: x ≠ x}”, “{x: x = P2} ∪ {x: x = P3}” and “{x: x = P1} ∪ {x: x = P2} ∪ {x: x = P3} ∪ {x: x = P4} ∪ {x: x = P5} ∪ {x: x = P6}”, for example). (3) One may claim that they are incomplete symbols that can be eliminated, when placed in the sentential context, by means of contextual definitions; for example, “P1 ∈ {P1}” could be replaced by “Ex (P1 ∈ x) a Az (z ∈ x d z = P1)”. Again, we may repeat here the line of reasoning from previous points: event if one rejected the view 4

Note that our reasoning may be simplified, if one used identity sign instead of set-membership. In such a case we would have only one list with different arguments on the right side of identity statement.

85 that a difference in reference implies a difference in semantic value, the syntactic differences between propositions cannot be ignored; e.g., “(Ex (P1 ∈ x) a Az (z ∈ x d z = P1)” and “Ex (P2 ∈ x) a Az (z ∈ x d z = P2)” would differ exactly due to the presence of distinct singular terms on corresponding syntactical places. Furthermore, no sentence from the (a)-list would express the same proposition as the sentence from (b)-list (and vice versa): (b)-sentences are negations while (a)-sentences are not. Thus, “the new account of structured propositions” cannot consistently speak about classes of all propositions, all false propositions falsehoods, all true propositions etc. The same can be said of some of its predecessors (cf. Ajdukiewicz 1967a, 1967b). I must admit that while I see the reason why one may want to ignore the problem in the first case (the class of all propositions) I agree with Grim’s opinion that the reasoning derives its philosophical significance from other cases (especially the case of the set of all true propositions). On the other hand, the problem is avoided by theories that are more liberal in the way they individuate contents. References Ajdukiewicz, Kazimierz 1967a. Proposition as the connotation of sentence. Studia Logica 20, 87-98. Ajdukiewicz, Kazimierz 1967b. Intensional expressions. Studia Logica 20, 63-86. Cocchiarella, Nino 2000. Russell’s paradox of the totality of propositions. Nordic Journal of Philosophical Logic 5, 25-37. Grim, Patrick 1984. There is no set of all truths. Analysis 44, 206-208. Lewis, David 1986. On the Plurality of Worlds. Oxford: Blackwell Publishers. Kaplan, David 1995. A problem in possible world semantics. In: Walter SinnottArmstrong, Diana Raman and Nicholas Asher (eds.), Modality, Morality and Belief: Essays in Honor of Ruth Barcan Marcus. Cambridge: Cambridge University Press, 41-52. King, Jeffrey 1995. Structured propositions and complex predicates. Nous 29, 516535. King, Jeffrey 1996. Structured propositions and sentence structure. Journal of Philosophical Logic 25, 495-521.

86 King, Jeffrey 2007. The Nature and Structure of Content, Oxford: Oxford University Press. Oksanen, Mika 1999. The Russell-Kaplan paradox and other modal paradoxes: a new solution. Nordic Journal of Philosophical Logic 4, 73-93.

Alex S Davies King’s College London [email protected]

Does Intensional Semantics Account for ‘Travis Cases’? Abstract: Several have supposed that the appropriate response to a linguistic phenomenon known as a ‘Travis case’ is to produce a formal semantic analysis that eliminates the phenomenon. However the empirical debate has stymied because analyses are inevitably incomplete: they are not specific enough for it to be checked whether Travis cases can be constructed for them. So it cannot be checked whether Travis cases have really been eliminated. I adopt a different approach. I sketch a genealogy of Travis cases. If the genealogy is correct then the elimination of Travis cases is not possible nor is it a means to a better understanding of the phenomenon. As a test case, I discuss an account of Travis cases that invokes features of intensional semantics. Key words: the ballistic requirement, calibration, intensional semantics, formal semantics, logico-syntactic truth-conditions, Predelli, substantial truth-conditions, Travis

0. Introduction Travis cases have the following form. A question is raised as to whether it is true to say of an object O, “O is F”. Despite no relevant ambiguity, ellipsis, or indexicality in the sentence, it is possible for the sentence to be true and it is possible for the sentence to be false of the same unchanged object. For instance, Pia’s maple has russet leaves but she has just painted them green.1 Pia says of her plant, “There. Now the leaves are green”. What she says is true. But suppose further that Pia’s friend asks after some green leaves for a chemistry experiment. Pia says of the same plant, “The leaves (on my plant) are green. You can have those”. 1

The example is taken from Travis (2008: 111).

88 But this time what she says is false. Call the exhibition of such a change in truth-value a Travis case. Some suppose that it is the responsibility of formal semantics to “account for” Travis cases.2 To account for Travis cases is to provide a formal semantic analysis of the sentences afflicted such that, once all the elements of that analysis have been held constant, different instances of the same sentence cannot differ in truth-value. Travis cases have attracted attention because on “standard” formal semantic analyses Travis cases remain unaccounted for. In this paper I present a reason why we should not expect any account of Travis cases to be correct. As a test case I apply this to an intensional account.3 I will proceed in the following stages. First, I describe a toy formal semantics. Second, I raise the question “what does a formal semantics describe?” I then introduce two distinctions which allow us to identify different answers to this question. Third, I explain how an attempt to “account for” Travis cases requires adoption of one of these answers: what I will call the strong answer. Fourth, I introduce five requirements on the use of linguistic expressions for reasoning validly. Fifth, I introduce a further requirement (the ballistic requirement) that must be met to reason validly with linguistic expressions if the strong answer is correct. Sixth, I give some indication of why (being without space to argue that) this requirement is unlikely to be met. Seventh, I describe an approach to reasoning with linguistic expressions which forfeits commitment to the strong answer. The approach can enable valid reasoning merely by meeting the five requirements without meeting the unmanageable ballistic requirement. Eighth, I explain why, if speakers did pursue this alternative in their language use then Travis cases would arise as a straightforward epiphenomenon. All this constitutes a candidate genealogy of Travis cases. If the genealogy is correct then one cannot (in our technical sense) account for Travis cases nor should one want to: to try to do so betrays a far from innocuous misunderstanding of 2

3

The view is pervasive, but for examples see Kennedy and McNally (2010), Pietroski (2003), Predelli (2005a, 2005b), Stanley (2007), Szabó (2001, 2010). For the intensional account see Predelli (2005a, 2005b). See also MacFarlane (2009).

89 the relation of the structures described by a formal semantics to ordinary speech behaviour. Ninth, I turn to the attempt to account for Travis cases with an intensional semantics and I explain why this is no exception to the rule. I close with a discussion of some objections to the line developed in the previous sections. 1. Formal semantics A formal semantics can be divided into two parts: the syntactic part and the semantic part. The syntactic part describes what expressions are in question. The semantic part is defined for given expressions relative to their syntactic categories. It describes what the (semantic) properties of expressions can be, given the (semantic) properties of other expressions. Both syntactic and semantic parts have a compositional component. These take the form of rules such as those listed in the following two subsections. 1.1. Compositional syntactic rules a. S → NP Pred b. S → S conj S c. S → neg S d. S → If S then S e. Pred → INFL VP f. INFL → (neg)

PAST PRES

g. [s NP INFL X] → [s INFL [s NP X]] h. conj → and, or i. neg → it is not the case that

90 1.2. Compositional semantic rules a. If β is a trace then I(β)M,w,i,c,g = g(β). Otherwise I(β)M,w,i,c,g = F(β)(w,i) unless β is an indexical or demonstrative in which case, I(β)M,w,i,c,g = F(β)(c). b. If A and B are any categories, I([A B])M,w,i,c,g = I(B)M,w,i,c,g c. I([s NP Pred])M,w,i,c,g = 1 iff I(NP)M,w,i,c,g ∈ I(Pred)M,w,i,c,g and 0 otherwise. d. I([s S1 conj S2])M,w,i,c,g = I(conj)M,w,i,c,g ()4 e. I([s If S1 then S2])M,w,i,c,g = I(If...then...)M,w,i,c,g () g. I([s neg S])M,w,i,c,g = I(neg)M,w,i,c,g (I(S)M,w,i,c,g) h. I([[INFL PRES] S])M,w,i,c,g = I(S)M,w,i,c,g i. I([[INFL PAST] S])M,w,i,c,g = 1 iff where i' ∈ I and there is some i'1, is called a label. The first element, which I will call a feature, l1 is a type. Following Richard Montague, I defined a type in Kułacka (2011a) in the following way. Let B be a set of base types. S is the set of simple types, which elements are constructed inductively: (i) they are elements of B, (ii) if a and b are types, so is (iii) nothing else is a type. The type is related to the category of a word and determines all the other features in a label. For instance, following Ajdukiewicz’s approach, as discussed in Kułacka (2011c), a noun is of the category n, and the category of a sentence is s. The related types are e and t (Kułacka 2012), and they constitute the set of base types. The type e determines all the other grammatical information that is represented by other features such as number, person, gender, case. As one can see all these features are also determined by the natural language we intend to describe, in this case Polish. Thus the word ‘kawę’ (coffee) has the following label . We may also include other grammatical information if needed, e.g. I omitted the gender in this example. The abbreviations used in the label stand for standard grammatical notation. Labels may be even more complex as for verbs; e.g. on the condition that it is not followed by the negative particle, the word ‘piję’ (I-drink) has the following label: , where in curly brackets I included information about the arguments that the verb takes. Since the subject has already been included in the word itself, the first curly bracket contains a labelled variable formula for ‘I’, which will be

337 discussed in full later in this section. The second curly bracket contains the grammatical requirements for the direct object of the verb, which in this case is a labelled variable formula. As previously, we may also include aspect or other grammatical information in this label. It is worth noticing that labels containing labelled formulae are an innovation that I have introduced to the model. How to generate these labels? The first feature, as mentioned above, is the type, then we have the list of grammatical information (a list of other features) that are included followed by a sequence of labelled formulae that represent information of the arguments of the word. The number of required arguments (labelled variable formulae) is reflected in the complexity of the type. The number of labelled constant formulae is the difference between the number of arguments determined by the language and the number of labelled variable formulae. The next element in the algorithmic system S├ is L, a logical language. I will simplify the definition that I provided in Kułacka (2011a). The notation vτ means a variable vτ is of type τ, cτ means a constant cτ is of type τ and MEτ is a set of meaningful expressions of type τ. The type e is the base type that represents an individual and the type t is the base type that represents a sentence. The language L is defined recursively. Individual variables or constants vτ and cτ belong to MEτ. If α ¤ MEa and x ¤ MEb and x is a variable, then λx.α ¤ ME. If α ¤ ME and β ¤ MEa, then αβ ¤ MEb. If α, β ∈ MEτ, then the following also belong to MEτ: Nα, Kαβ, Aαβ, where the logical constants N, K, A stand for negation, conjunction, disjunction, respectively. I used Polish notation for these logical functors. (5) If α, β ∈ MEt, and x is a variable, then the following also belong to MEt: Nα, Kαβ, Aαβ, Cαβ, Eαβ, Lα, Πx. α, Σx.α, where the logical constants N, K, A, C, E, L stand for negation, conjunction, disjunction, implication, bi-conditional, necessity, respectively, and Π and Σ are universal and existential quantifiers. The sentence Lα should be read ‘it is necessary that α’. As above, I used Polish notation for these logical functors. (1) (2) (3) (4)

338 (6) If α, β ¤ MEτ, then (α=β) ¤ MEt. (7) Nothing else is in the language L. The logical language L will be used to represent the semantic content of the word. For the studied examples above, we can have ce,1 representing the semantic content of ‘kawę’ (coffee), and λve,1.c,1ve,1ve,2 to represent the meaning of ‘piję’ (I-drink), where ve,2 stands for the variable representing the pronoun ‘I’, c,1 is the semantic content of ‘drink’, and the variable ve,1 is the place holder for a semantic content of the direct object. The numbers written in the subscript next to the type of a variable of a constant are used to distinguish between different semantic contents of the same type. One may find more on lambda expressions in Kułacka (2011a). To complete defining the algorithmic system S├, we need to describe the discipline ℝ that connects labels with meaningful expressions of the logical language. This will be done by connecting the labels with the first feature, the type τ and a meaningful expression of the same type. For instance: the word ‘kawę’ (coffee) will be represented by the labelled formula (a declarative unit) :ce,1. The verb ‘piję’ will be represented by a more complex declarative unit: (1) : λve,1.c,1ve,1ve,2 if it follows the negative particle, or (2) : λve,1.c,1ve,1ve,2, otherwise. In Table 1, I provide more examples of declarative units.

339 Word kawa (coffee) kawy (coffee) kawie (coffee) kawę (coffee) kawą (coffee) piłam (drink) piłem (drink) piło (drink) negative particle ‘nie’

its declarative unit :ce,1 :ce,1 :ce,1 :ce,1 :ce,1 :ce,1 : λve,1.c,1ve,1ve,2 if it follows the negative particle. : λve,1.c,1ve,1ve,2 if it follows the negative particle. : λve,1.c,1ve,1ve,2 if it follows the negative particle. : λvτ,1.Nvτ,1, where τ is a type (with some restrictions on prepositions, other particles). Table 1. Examples of declarative units

It is worth noticing that some words have two declarative units, the choice of one or the other is context sensitive. What I have not discussed here is the semantic restrictions, which are also included as features of the n-tuple label. Also, in positional languages, among the features in the labels, one needs to include information about the position that the word holds in a phrase. Polish is statistically an SVO language that also has some tendency to free order; in other words the majority of the sentences are SVO, but occasionally under certain conditions the order in the sentence can be different. To be able to refine our model to include information about the order we may include other features in the label:

340 where i means ‘immediately to the left’, p1 – the probability of i, and r11,…,r1k1, (k1≥0) represent other restrictions, (2) , where r means ‘immediately to the right’, p2 – the probability of r, and r21,…,r2k2, (k2≥0) represent other restrictions, (3) , where i+ means ‘to the left’, p3 – the probability of i+, and r31,…,r3k3, (k3≥0) represent other restrictions, (4) , where r+ means ‘to the right’, p4 – the probability of r+, and r41,…,r4k4, (k4≥0) represent other restrictions. (1)

,

For instance, the declarative unit that represent the negative particle ‘nie’ : λvτ,1.Nvτ,1, where τ is a type, should actually be : λvτ,1.Nvτ,1, as the negated word or a phrase always immediately follows the particle. The probability may be found on the system for an abstract language speaker (see more on assumptions in Kułacka 2011b). The relative frequencies found on corpora may vary from one corpus to another. 3. Extended Lambda Calculus Using the algorithmic system as defined in the previous section, we are able to construct a declarative unit. The purpose of this section will be to provide tools for combining two declarative units into one if one is the argument of the other given that the features in their labels match; for example ‘nie’ and ‘kawa’ represented as : λvτ,1.Nvτ,1, where τ is a type, and :ce,1, can be combined if the second word follows the first as indicated by the feature . The declarative units: :ce,1 and :ce,1 cannot be combined as neither of them takes an argument. The declarative units: :ce,1 and: λve,1.c,1ve,1ve,2 cannot be combined as the case feature in the first word does not match the case features in the labels of

341 any of the arguments of the second word. On the other hand, the second declarative unit can be combined with :ce,1 as both the type and the case feature in the label of the argument match the relevant features in the declarative unit : λve,1.c,1ve,1ve,2. 3.1 Substitution Before defining the substitution that is needed for reductions on lambda terms, which are the formulae of the logical language L, I will provide two other definitions of identity between two logical expressions (formulae) of language ℒ and of the set of free variables. Two expressions of L, P and Q, are identical, written P≡Q, if and only if Q is an exact copy (symbol by symbol) of P, correct to conventionally disregarded parenthesis. For instance, two formulae λvτ,1.Nvτ,1 and λvτ,1.N(vτ,1) are identical, but two formulae (λvτ,1.c,1)vτ,1 and λvτ,1.c,1vτ,1 are not identical as the application binds more tightly than the abstraction. The set of the free variables of a logical expression P, denoted as FV(P), is defined recursively as follows: (1) (2) (3) (4) (5) (6) (7)

FV(c)= ∅ if c is a constant, FV(x)={x} if x is a variable, FV(QM)=FV(Q)∪FV(M), FV(λx.Q)=FV(Πx.Q)=FV(Σx.Q)=FV(Q)-{x}, FV(Kαβ)=FV(Aαβ)= FV(Cαβ)=FV(Eαβ)=FV(αβ), FV(Nα)=FV(Lα)=FV(α), FV(α=β)=FV(αβ).

In these notations Q and M are arbitrary terms, α and β are terms of relevant types as defined for the language. The preceding letters in the logical expressions are either logical constants or quantifiers. Let us consider the following example:

342 FV((λvτ,1.c,1)vτ,1)=FV(λvτ,1.c,1)∪FV(vτ,1)=FV(c,1){vτ,1}∪{vτ,1} ={vτ,1} as c,1 is a constant, and FV(λve,1.c,1ve,1ce,2)=FV(c,1ve,1ce,2)-{ve,1}=FV(c,1)∪ FV(ve,1)∪FV(ce,2)-{ve,1}=∅ as c,1 and ce,2 are constants. Another example is FV(ce,1)=∅ as ce,1 is a constant. Let us define the substitution for the logical expressions. For any M, P, Q and x, the notation [P/x]M means that P is substituted for x in the logical expression M. It is important to notice that P and x must have all their overt features in the label equal, in the case below we see that they are of the same type. The substitution is defined as follows: (1) [P/x]x≡P, e.g. [ce,1/ve,1]ve,1≡ce,1; (2) [P/x]a≡a for all constants and variables aÔx, e.g. [ce,1/ve,1]ve,2≡ve,2; (3) [P/x](MQ)≡ ([P/x]M)([P/x]Q), e.g. [ce,1/ve,1](c,1ve,1)≡ ([ce,1/ve,1]c,1)([ce,1/ve,1]ve,1)≡c,1ce,1; (4) [P/x](Kαβ)≡K([P/x]α)([P/x]β), [P/x](Aαβ)≡A([P/x]α)([P/x]β), [P/x](Cαβ)≡C([P/x]α)([P/x]β), [P/x](Eαβ)≡E([P/x]α)([P/x]β), [P/x](Nα)≡N([P/x]α), [P/x](Lα)≡L([P/x]α), where α and β are of relevant types, e.g. [ce,1/ve,1](Nve,1)≡N([ce,1/ve,1]ve,1)≡Nce,1; (5) [P/x](λx.M)≡λx.M, [P/x](Πx.M)≡Πx.M, [P/x](Σx.M)≡Σx.M, e.g. [ce,1/ve,1](λve,1.ct,1)≡λve,1.ct,1; (6) If yÔx, and y ∉ FV(P) or x ∉ FV(M), then [P/x](λy.M)≡λy.[P/x]M, [P/x](Πy.M)≡Πy.[P/x]M, [P/x](Σy.M)≡Σy.[P/x]M, e.g. [ct,1/vt,1](λve,2.ct,1)≡λve,2.[ct,1/vt,1]ct,1≡λve,2.ct,1; (7) If yÔx, and y ∈ FV(P) and x ∈ FV(M), then we need to introduce a new variable of the same type as x and y: z FV(PM) and [P/x](λy.M)≡λz.[P/x][z/y]M, [P/x](Πy.M)≡Πz.[P/x][z/y]M, [P/x](Σy.M)≡Σz.[P/x][z/y]M, e.g. [vt,2/vt,1](λvt,2.Nvt,1)≡λvt,3.[vt,2/vt,1][vt,3/vt,2](Nvt,1)≡λvt,3.[vt,2/vt,1](N vt,1) ≡ λvt,3.Nvt,2; (8) [P/x](α=β)≡(([P/x]α)=([P/x]β)), e.g. [vt,2/vt,1](vt,1=ct,1)≡(([vt,2/vt,1]vt,1)=([vt,2/vt,1]ct,1))≡(vt,2=ct,1).

343 In the following section I will provide the definitions for some of the reductions used in the lambda calculus. Since I extended lambda terms to cover also quantified expressions and expressions with logical constants, there is a need to extend the reductions to be able to apply them to all relevant terms. 3.2 Reductions Let us start with the α-reduction, which is another name for a change of bound variable. As the description suggests, it can only be performed on abstractions and quantified expressions. For any x, M and y∉FV(M) we can replace λx.M, Πx.M, Σx.M with λy.[y/x]M, Πy.[y/x]M, Σy.[y/x]M, respectively, and this replacement is denoted: λx.Miαλy.[y/x]M, Πx.MiαΠy.[y/x]M, Σx.MiαΣy.[y/x]M. It is worth noticing that both x and y are of the same type. For instance: λve,2.ct,1iαλve,1.[ve,1/ve,2]ct,1≡ λve,1.ct,1. Two terms of the logical language L, M and P, are α-congruent, written as M≅P if one of the following holds: (1) M≡P, or (2) MiαP, (3) P is obtained from M by replacing a subterm S of P by a subterm T of M such that SiαT, (4) there is some term of the logical language L, R, such that M≅R and R≅P. The terms λve,1.c,1ve,1ce,2 and λve,1.(c,1ve,1)ce,2 are α-congruent by (1) and the terms λve,1.c,1ve,1ce,2 and λve,2.c,1ve,2ce,2 are α-congruent by (2). I will show now that the terms λve,1.(c,1Πve,2.ct,1)ve,1 and λve,2.(c,1Πve,3.ct,1)ve,2 are α-congruent. A subterm Πve,2.ct,1 of the first term can be α-reduced to Πve,3.ct,1, a subterm of the second term, therefore by (3) the two terms λve,1.(c,1Πve,2.ct,1)ve,1 and λve,1.(c,1Πve,3.ct,1)ve,1 are α-congruent. Since the latter two terms are α-congruent, and we can α-reduce λve,1.(c,1Πve,3.ct,1)ve,1 to λve,2.(c,1Πve,3.ct,1)ve,2, is α-congruent to therefore λve,1.(c,1Πve,2.ct,1)ve,1 λve,2.(c,1Πve,3.ct,1)ve,2 by (4). I will now turn to defining β-reduction. The terms of the logical language L that can undergo this simplification are of the form (λx.M)P, called β-redex, and represent an operator λx.M applied to its operand P.

344 As for α-reduction, we only can β-reduce the application (λx.M)P if both x and P are of the same type. I will make sure that this is the case in the definition of the consequence relation. For any x, M, P and Q, a β-redex (λx.M)P, being a subterm of Q, can be β-reduced to [P/x]M, which is called a contractum. The notation associated with this reduction is (λx.M)Piβ[P/x]M. For instance, (λve,1.c,1ve,1ce,2)ce,1iβ[ce,1/ve,1](c,1ve,1ce,2)≡ c,1ce,1ce,2. The application (λve,1.c,1ve,1ce,2)ct,1 is not a β-redex as the terms ve,1 and ct,1 are not of the same type, therefore we cannot apply β-reduction to it. The last reduction that I will apply to the terms of the logical language L is called leftmost β-reduction. The leftmost β-reduction of M to P, written as MhP, is defined recursively as follows: (1) MhP if M≅P, (2) MhP if MiβP, (3) If MhP, for some M and P, then for any term of the logical language L of the relevant type, Q, we have both QMhQP and MQhPQ, (4) If MhP, for some M and P, then for any variable x, λx.Mhλx.P, Πx.MhΠx.P and Σx.MhΣx.P, (5) If MhQ and QhP, then MhP, (6) MhP only in the cases (1)-(5), as defined above. For instance, (λve,1.c,1ve,1)ce,2ce,1hc,1ce,2ce,1 by applying (3) to the subterm of the application. Also, we can obtain λve,2.(λve,1.c,1ve,1ve,2)ce,1h λve,2.c,1ce,1ve,2 by (4), or similarly applying the same rule to the term Πve,2.(λve,1.c,1ve,1ve,2)ce,1, we may leftmost β-reduce it to Πve,2.c,1ce,1ve,2. The example of application rule (5) is as follows: Πve,2.(λve,1.c,1ve,1ve,2)ce,1hΠve,2.c,1ce,1ve,2hΠve,1.c, 1ce,1ve,1. The second reduction in the sequence are justified by (1), as it is α-reduction.

345 4. Consequence relation and the language In the approach discussed in this paper, I will only provide the method of recognising a set of sentences as opposed to a coherent discourse. However, the model may be easily expanded to cover also a set of sentences that are joined logically. The declarative unit is given by :αf11,1, where f11,…,f1n are features of the label, n≥1, f11 is a type and αf11,1 is a formula of the logical language L. The consequence relation is defined between a set of declarative units or a declarative unit and a declarative unit in the following way (on the left-hand side of the turnstile there is a database or a declarative unit from which a new declarative unit can be derived): (1)

:αf11,1├:αf11,1 we

add a new declarative unit

to the database. (2) If the features in the labels of one of declarative units of the database and the new declarative unit match in the sense that the label of one declarative unit matches the label of the declarative unit that is a feature of the other declarative unit, then {…,:αf11,1,…}, :αf21,2├ :αf31,1, or {…,:αf21,2,…}, :αf11,1├ :αf31,1, where f11≡ and αf31,1 is obtained to applying the leftmost β-reduction to αf11,1αf21,1. It is worth noticing that the left-hand side of the turnstile is the total database, but for the sake of clarity, I separate the old database from a new declarative unit. The component declarative units that participated in the reduction are deleted from the database with leaving a trace. They can be retrieved if the hearer needs to perform a reanalysis of the reduced declarative units. (3) In case the new declarative unit matches the argument of one of the features of a declarative unit in the database, then the

346 operation is performed within the label of the declarative unit in question in the following way: {…,:αf11,1,…}, :αf21,2├ :αf31,1 if αf21,1 is a variable. The term αf31,1≡[αf21,2/αf21,1]αf11,1. Let us notice that (2) and (3) differ in the type of the derived declarative unit. This may be the case of overwriting a pronoun with a noun. (4) If as a result we obtain :αf31,1 and f31≡f21, then : αf31,1├:αf31,1. (5) In other cases the new declarative unit becomes an element of the database without making any alternations. In real-life situations, the hearer will try to approximately match the new declarative unit with any declarative units in their temporary database. Let us consider an example analysis based on the lexicon given in Table 1. We want to derive ‘Nie piłem kawy’ (I didn’t drink coffee). (1) : λvτ,1.Nvτ,1, ├ : λvτ,1.Nvτ,1,; the first word is included into the database of an abstract hearer. (2) : λve,1.c,1ve,1ve,2├ : λve,1.c,1ve,1ve,2; the second word is included into the database of an abstract hearer; and also

347 (3) {: λvτ,1.Nvτ,1}, : λve,1.c,1ve,1ve,2├ :Nλve,1.c,1ve,1ve,2 as (λvτ,1.Nvτ,1)(λve,1.c,1ve,1ve,2)hNλve,1.c,1ve,1ve,2; the first word constitutes a coherent unit with the second word. (4) : Nλve,1.c,1ve,1ve,2├ : Nλve,1.c,1ve,1ve,2. This is an application of the part (4) in the definition of consequence. (5) :ce,1├:ce,1; the third word is included into the database of an abstract hearer; and also (6) {: Nλve,1.c,1ve,1ve,2},:ce,1├ : Nc,1ce,1ve,2. The latter result is obtained, because (λve,1.c,1ve,1ve,2)ce,1h c,1ce,1ve,2 by (2) in the definition of leftmost β-reduction and (Nλve,1.c,1ve,1ve,2)ce,1h Nc,1ce,1ve,2 by (4) in the same definition. In the above databases I only included the necessary elements that are needed to derive a new declarative unit. This is done for simplicity of the notation and will be continued in the same manner in the rest of the paper. This consequence relation mimics a natural language understanding or rather the process of building the basis for the interpretation of the sequence of words as they are provided to the abstract hearer word by

348 word. The language recognised is a set of declarative units. Subsets of the language constitute a database for a given utterance. We have to bear in mind that the speaker may switch to the role of only a hearer, and enrich their database by new information in terms of declarative units provided by another speaker. 5. A Fragment of Polish Word kawa (coffee) kawy (coffee) kawie (coffee) kawę (coffee) kawą (coffee) ja (I) dziecko (child) dziecka (child) dziecku (child) dzieckiem (child) piłam (drink)

its declarative unit :ce,1 :ce,1 :ce,1 :ce,1 :ce,1 :ce,1 :ve,2 :ce,3 :ce,3 :ce,3 :ce,3 :ce,3 :ce,3

: λve,1.c,1ve,1ve,2 if it follows the negative particle; : λve,1.c,1ve,1ve,2; otherwise. piła (drink) : λve,1.c,1ve,1ve,4 if it follows the negative particle; : λve,1.c,1ve,1ve,4; otherwise.

349 piło (drink) : λve,1.c,1ve,1ve,5 if it follows the negative particle; : λve,1.c,1ve,1ve,5; otherwise. negative : λvτ,1.Nvτ,1,where τ is a type. particle ‘nie’ Table 2.The lexicon of a fragment of Polish

In this section I introduce the lexicon of a fragment of Polish and I show the analysis of some sequences of words. In Table 2, I gathered the lexicon. One needs to remember that in this first version of Linear Dynamic Syntax, I have not included any information about the semantic restrictions in the labels of the declarative units, such as the direct object of the verb drink is a drinkable thing. In the following sections, I analyse the following sequences: ‘Kawa piła dziecko’ (the coffee drank a child), ‘Nie kawa piła dziecko’ (not coffee drank a child), ‘Dziecko piła nie kawa’ (not coffee drank a child), ‘Piłam nie dziecko’ (I drank not a child), ‘Nie piłam dziecka’ (I didn’t drink a child), ‘Ja piłam kawę’ (I-emphasised drank coffee), ‘piłam kawie’ (ungrammatical), ‘piła nie’ (ungrammatical). 5.1 ‘Kawa piła dziecko’ (the coffee drank a child) (1) :ce,1├:ce,1. (2) : λve,1.c,1ve,1ce,4├ : λve,1.c,1ve,1ce,4. (3) {:ce,1}, : λve,1.c,1ve,1ve,4├ : λve,1.c,1ve,1ce,1. (4) :ce,3├:ce,3.

350 (5) {: λve,1.c,1ve,1ce,1},:ce,3├ : c,1ce,3ce,1. 5.2 ‘Nie kawa piła dziecko’ (not coffee drank a child) (1) : λvτ,1.Nvτ,1 ├ : λvτ,1.Nvτ,1. (2) :ce,1├:ce,1. (3) {: λvτ,1.Nvτ,1}, :ce,1├ :Nce,1. (4) :Nce,1├ :Nce,1. (5) : λve,1.c,1ve,1ve,4├ : λve,1.c,1ve,1ve,4. (6) {:Nce,1}, : λve,1.c,1ve,1ve,4├ : λve,1.c,1ve,1Nce,1. (7) :ce,3├:ce,3. (8) {: λve,1.c,1ve,1Nce,1}, :ce,3├ : c,1ce,3Nce,1. 5.3 ‘Dziecko piła nie kawa’ (not coffee drank a child) (1) :ce,3├:ce,3. (2) :

351

(3)

(4)

(5) (6) (7)

(8) (9)

λve,1.c,1ve,1ve,4├ : λve,1.c,1ve,1ve,4. The system cannot join the declarative units in the database, it may try the reanalysis: :ce,3├:ce,3. {: λve,1.c,1ve,1ve,4}, :ce,3├ : c,1ce,3ve,4. : λvτ,1.Nvτ,1 ├ : λvτ,1.Nvτ,1. :ce,1├:ce,1. {: λvτ,1.Nvτ,1}, :ce,1├ :Nce,1. :Nce,1├ :Nce,1. {: c,1ce,3ve,4}, :Nce,1├ : c,1ce,3Nce,1.

The number of steps clearly follows the time necessary to analyse a less frequent order of words. However, the result is the same as in Section 5.2. 5.4 ‘Piłam nie dziecko’ (I drank not a child) (1) : λve,1.c,1ve,1ve,2├ : λve,1.c,1ve,1ve,2.

352 (2) : λvτ,1.Nvτ,1 ├ : λvτ,1.Nvτ,1. (3) :ce,3├:ce,3. (4) {: λvτ,1.Nvτ,1}, :ce,3├ :Nce,3. (5) :Nce,3├ :Nce,3. (6) {: λve,1.c,1ve,1ve,2}, :Nce,3├ : c,1Nce,3ve,2. 5.5 ‘Nie piłam dziecka’ (I didn’t drink a child) (1) : λvτ,1.Nvτ,1 ├ : λvτ,1.Nvτ,1. (2) : λve,1.c,1ve,1ve,2├ : λve,1.c,1ve,1ve,2. (3) {: λvτ,1.Nvτ,1}, : λve,1.c,1ve,1ve,2├ : Nλve,1.c,1ve,1ve,2. (4) : Nλve,1.c,1ve,1ve,2├ : Nλve,1.c,1ve,1ve,2. (5) :ce,3├:ce,3. (6) {: Nλve,1.c,1ve,1ve,2}, :ce,3├

353 : Nc,1ce,3ve,2. 5.6 ‘Ja piłam kawę’ (I-emphasised drank coffee) (1) :ve,2├:ve,2. (2) : λve,1.c,1ve,1ve,2├ : λve,1.c,1ve,1ve,2. (3) {:ve,2}, : λve,1.c,1ve,1ve,2├ : λve,1.c,1ve,1ve,2. (4) :ce,1├:ce,1. (5) {: λve,1.c,1ve,1ve,2}, :ce,1├ : c,1ce,1ve,2. 5.7 ‘piłam kawie’ (ungrammatical) and ‘piła nie’ (ungrammatical) (1) : λve,1.c,1ve,1ve,2├ : λve,1.c,1ve,1ve,2. (2) :ce,1├:ce,1. The system cannot join these two declarative units into a coherent declarative unit. The system will behave similarly with the second pair. (1) : λve,1.c,1ve,1ve,4├

354 : λve,1.c,1ve,1ve,4. (2) : λvτ,1.Nvτ,1 ├ : λvτ,1.Nvτ,1. Both sequences are not a part of a language, though every element of each of them is an element of the language. 6. Conclusion and further research I have based my model on the fundamentals of Dynamic Syntax, but due to consistency with Labelled Deductive System, I defined a new model, Linear Dynamic Syntax, and showed that it is a kind of logic. I also defined some properties of the consequence relation, but more properties may be defined or the ones given modified to model well the process of building the basis for interpretations of linguistic expressions from the point of view of the hearer. The declarative units are not interpretations yet, they constitute the basis for these interpretations. The mechanism of arriving at interpretations of the expressions of a language of logic was extensively defined in Kułacka (2011a). One needs to bear in mind that unlike a Montague Grammar, Linear Dynamic Syntax does not have translation rules defined yet. At this stage of research I translated the lexicon of a fragment of Polish into expressions of a logic word by word, and used consequence relation to derive more complex expressions of this logic from less complex expressions. In the forthcoming paper, “Mechanism of a Montague Grammar”, I will explain how a Montague grammar generates interpretations from the expressions of a fragment of English and this will influence the framework of Linear Dynamic Syntax. Also, it worth noticing that the lexicon of the fragment of Polish discussed in this paper may be expanded to include more lexemes of different types, and this will entail expanding the properties of the consequence relation.

355 References Gabbay, Dov M. 1996. Labelled Deductive Systems. Oxford: Oxford University Press. Kempson, Ruth M., Wilfried Meyer-Viol and Dov M. Gabbay. 2001. Dynamic Syntax: The Flow of Language Understanding. Oxford: Blackwell Publishers. Kułacka, Agnieszka 2011a. Intensional Logic for a Montague Grammar. LingVaria 2(12), in print. Kułacka, Agnieszka 2011b. Metodologiczne założenia semantyki komputerowej. In: Piotr Stalmaszczyk (ed.), Metodologie Jezykoznawstwa. Łódź: Wydawnictwo Uniwersytetu Łódzkiego, 111-122. Kułacka, Agnieszka 2011c. Syntax of a Montague Grammar. LingVaria 1(11), 923. Kułacka, Agnieszka 2012. Translation Rules in a Montague Grammar. LingVaria 2(14), forthcoming. Montague, Richard 1974. Formal Philosophy. Selected Papers of Richard Montague. Edited by Richmond H. Thomason. New Haven: Yale University Press.

Janusz Maciaszek University of Łódź [email protected]

Towards a Non-referential Semantics Abstract: The aim of the paper is to argue in favor of non-referential semantics of natural language. In the first section I indicate some ontological difficulties that face referential theories and argue that a satisfactory theory of meaning of natural language should not be referential because it does not explain how it is possible that the expressions of natural language are meaningful. As an illustration of a possible way of constructing semantics without reference I consider Russell’s theory of descriptions which is referential but can be considered as the first step towards nonreferential semantics, and three classical non-referential solutions: Ajdukiewicz’s directival theory of meaning, Quine’s theory of meaning in terms of a manual of translation, and Davidson’s theory of interpretation based on Tarski’s theory of truth. Key words: sense, meaning, reference, belief, theory of truth, holism, semantics, translation, indeterminacy of translation, inscrutability of reference

0. Introduction The word “semantics” has at least two meanings. In the analytic tradition semantics is just the part of semiotics that deals with the relation between language and extra-linguistic reality. This kind of semantics is by definition referential. The modern referential semantics was initiated by Mill, elaborated by Frege, and developed by Carnap, Montague and Kripke among others. The very idea of referential semantics can be found in early Wittgenstein. The referential semantics makes one main presupposition: a particular sound is meaningful because it is anchored

358 in reality. This presupposition is petrified in form of the relation of denotation or reference.1 In the traditional sense semantics deals with meaning of linguistic expressions. Though these two types of semantics seem to have nothing in common, after Frege’s seminal article On Sense and Reference (1892) meanings of linguistic expressions have been often considered as their special referents, and in consequence any theory of reified meaning can be conceived as part of a theory of reference. 1. Referential semantics and its drawbacks The prehistory of referential semantics could be traced to Aristotle who introduced the notion of truth as correspondence between thoughts, language and reality. Aristotle was a metaphysician and his theory of language can be only reconstructed (see Modrak 2006). In spite of this, Aristotelian notion of form (idea) can be considered as a prototype of the notion of meaning. Aristotle and his followers started with metaphysics, and semantics can be considered as its by-product. The analytic tradition after the linguistic turn has inverted this direction – metaphysics or ontology has been considered as a by-product of referential semantics. But even this approach does not explain how natural language can be conceived as a system of meaningful expressions, and how it hooks onto the world. The main philosophical problem of any referential semantics of natural language deals with the very nature of the relation of reference. According to Aristotle, a word refers to its referent because the referent has a particular form, and the form constitutes the reference relation. But the form is connected only with the category of substance, and it is not clear how to constitute the reference relation between objects of other categories and the corresponding words. One can consider this difficulty as a result of the drawbacks of Aristotelian ontology. These drawbacks were criticised by Ingarden (1947, 1948) who formulated very elaborate 1

The reference relation and the relation of denotation are sometimes distinguished, like in significant titles of Russell’s On denoting (1905) and Strawson’s On referring (1950).

359 ontology and even proposed his own theory of meaning. But even Ingarden’s ontology cannot serve as a basis for adequate semantics of natural language, because the semantics proposed by Ingarden is not compositional. The compositional referential semantics was proposed in various works of Frege (mainly in 1892) who introduced the principle of compositionality of reference and the distinction between function and its argument.2 But when the principle in Frege’s formulation is accepted, the sentences must also have their referents. Contrary to ontologically oriented Husserl and Ingarden who treated referents of sentences as states of affairs, Frege considered truth-values as abstract referents of sentences. The problem of referents of sentences arises in case of fiction. According to Frege, the sentences about fiction do not have truth values, because some parts of it do not have referents. Nonetheless, the sentences about fiction are meaningful, because they have senses.3 The theory of sense was developed by Frege in the analysis of nonextensional contexts (e.g. belief-contexts). In these contexts, linguistic expressions do not have normal referents, but some special ones – their normal senses. This move enabled Frege to save the compositionality of reference in intensional contexts. He was not only determined to introduce an infinite hierarchy of senses but also to face problems of ontological status of senses.4 The very idea of every referential semantics is that linguistic expressions are meaningful because they refer to something. This assumption, together with the principle of compositionality expressed in function – argument form, generates an ontology that faces some serious problems.5 Let us mention at least some of them:

2 3 4 5

For Frege compositionality and the functor-argument distinction are inseparable. I use in the whole article Frege’s term sense for reified meaning. In the standard interpretation the senses are ideal objects. One can object this contention stating that in case of model theory of formal languages the ontology based on set theory is not controversial. But I maintain that a formal language can meaningful not because it has a model-theoretic semantics but because it can be interpreted in natural language.

360 1.

2.

3.

4.

The ontology of references and senses for various kinds of linguistic expressions is rather unrealistic. It has to comprise elaborated and complicated functions, fictional objects, abstract, and intentional ones. The challenge for a referential semantics are references of sentences in general, sentences about fiction, and nondeclarative sentences.6 It faces serious problems in some contexts, e.g. belief-contexts, because the rule of substitution of co-referring terms is the consequence of the rule of compositionality. The problem with the classical concept of truth as correspondence arises. If a sentence is true because it corresponds to the reality, one has to explain why many sentences that correspond to nothing are considered as true.

Referential semantics assumes some kind of idealization. It can be accepted in case of idealized formal languages of mathematics and logic but it does not work in case of natural language because this idealization looses some aspects of natural language that are crucial to the fact of understanding. In particular it cannot explain the phenomenon of our mutual communication which is usually successful though we frequently refer our words to something different, i.e. it is difficult to conceive the referential semantics of natural language as the source of meaningfulness. One of the first philosophers who was aware of this fact was Bertrand Russell who tried to solve some of the problems mentioned above. 2. Russell’s theory of descriptions The first step towards non-referential semantics was made by Bertrand Russell in some papers dealing with definite descriptions which are counterparts of English phrases containing definite article the. In his seminal text On denoting (1905) Russell made the first attempt to avoid 6

According to Frege and his followers non-declarative sentences have no logical values, i.e. are neither true nor false, and have no referents at all.

361 certain ontological problems that appear in referential semantics of natural language. After rejecting idealism7 and the theory of denoting concepts,8 he adopted direct realism, i.e. the view that one can be acquainted with existing objects without any mediation of internal structures (e.g. Fregean senses).9 From this point of view Russell’s solution is in fact referential, but it is an important step towards nonreferential semantics because it solves one of its main problems, namely, how we understand sentences containing terms that do not refer to anything. In particular, Russell tried to solve the problem of how we can sensibly talk using meaningful definite descriptions that do not refer to anything (e.g. the present king of France).10 Russell’s solution consists in treating individual terms as apparent expressions that appear in surface forms of sentences only, and the sentences containing them as elliptical. In consequence, Russell proposed that the logical structure of naturallanguage sentences with individual terms can be expressed in a rather complicated way in the language of predicate calculus. This proposal enabled Russell to avoid some problems of referential semantics (see Russell 1905; Lycan 2000): 1. 2.

7

8

9

10

The apparent reference to non-existent objects in sentences like in the sentence The present king of France is bald. The negative existential sentences, e.g. The present king of France does not exist, which are paradoxical because though they are obviously true, they ascribe the property of non-

The name “idealism” is highly ambiguous. For our purpose it suffices to mention that during his idealistic period Russell admitted the existence (or better subsistence) of non-existing objects. From this point of view sentences about non-existing objects are meaningful because they are anchored in an “ideal” reality. Denoting concepts can be considered as counterparts of Fregean senses. In consequence sentences about non-existing objects are meaningful because empty terms express denoting concepts though they do not refer to anything. The evolution of Russell’s philosophical views is briefly described in Hylton (2003). Moreover we can make mistakes and talk about the objects that do not exist though we believe in their existence.

362

3.

4.

existence to something that does not exist. And if something does not exist you cannot ascribe any property to it. In consequence the sentences seem false. Frege’s puzzle (formulated in Frege 1892) of cognitive status of identity sentences of the form a = a and a = b, where a and b refer to the same object. The problem of co-referential substitution in believe contexts.

Russell’s proposal has been strongly criticized, mainly by Strawson (1950). The main drawback of the Russell’s theory of descriptions is that contrary to our intuitions the sentences about non-existing objects as well as their apparent negations must be false, e.g.: (1) The present king of France is bald (1’) The present king of France is not bald are false, but the latter is not the negation of the former. According to Russell the sentences are equivalent to the conjunction of the following three sentences: (a) There is at least one present king of France or (∃x) present-king-of-France(x) (b) There is at most one king of France or (∀x)(present-king-of-France(x) ⇒ (∀y)(present-king-of-France(y)⇒ (y = x)) * (c ) Whatever is the present king of France it is bald or (∀x)(present-king-of-France(x) ⇒ (bald(x)) ** (c ) Whatever is the present king of France it is not bald or **(∀x)(present-king-of-France(x) ⇒ ∼bald(x)) which are equivalent to: (1*) (∃x)(present-king-of-France(x) ∧ (∀y)(present-king-ofFrance(y) ⇒ (x = y)) ∧ bald(x))

363 (1**) (∃x)(present-king-of-France(x) ∧ (∀y)(present-king-ofFrance(y) ⇒ (x = y)) ∧ ∼bald(x)) Russell’s analysis enables one to discover hidden ambiguities in surface forms of the sentences. Let us consider the following belief sentences: (2) Columbus believed that Castro’s island was not far from India (2’) Columbus believed that Cuba was not far from India11 In the intuitive ascription of truth values (2) seems rather false, and (2’) rather true, because everybody knows that Columbus did not use the term Castro’s island, and it is not clear if he used the name Cuba either. The reason for the mentioned discrepancy of truth value ascription lies not in Columbus’ beliefs about the island he discovered but rather in what we know or suppose to know about Columbus’ beliefs and the words he could have used to express them. In Russell’s approach both sentences have two readings: (2a) Columbus believed that (∃x)(Castro’s island(x) ∧ (∀y)(Castro’s island(y) ⇒ (x = y)) ∧ was-not-far-from-India(x)) (2b) (∃x)(Castro’s island(x) ∧ (∀y)(Castro’s island(y) ⇒ (x = y)) ∧ Columbus believed that was-not-far-from-India (x)) where (2a) is obviously false, and (2b) is true, and: (2’a) Columbus believed that (∃x)(an-island-called-Cuba(x) ∧ (∀y)(an-island-called-Cuba (y) ⇒ (x = y)) ∧ it was-not-farfrom-India(x)) (2’b) (∃x)(is-an-island-called-Cuba(x) ∧ (∀y)(an-island-calledCuba(y) ⇒ (x = y)) ∧ Columbus believed that it was-not-farfrom-India(x))

11

This example is adopted from Lycan (2000: 57).

364 where (2’a) is probably false thought many people would think that it is true, and (2’b) is obviously true.12 Russell’s theory of definite descriptions has one obvious merit. If we express natural language sentence in predicate calculus without individual constants we make only one rather reasonable ontological commitment that there is at least one object. It explains why we understand sentences with definite descriptions whether they have referents or not. But when we treat proper names as individual constants of the language of predicate calculus we are obliged to make a rather dubious supposition that every constant has its referent (like objects called numbers for numerals in platonic philosophy of mathematics). To avoid this problem Russell intended to reduce proper names to definite descriptions.13 In spite of the obvious merits Russell’s theory provokes some problems. The most serious one concerns meaningfulness. Russell did not use the notion of sense or meaning conceived as an object of a certain kind that can be grasped (like in Frege’s terminology). The surface forms sentence with individual terms are meaningful because they are equivalent to three existential and general sentences that are meaningful independently of the existence of the referents of the (apparent) individual terms. But it is not clear how existential and general sentences can be meaningful. One possible solution is that these sentences are meaningful because they are true or false in various circumstances which are represented by different models. But the sentences about fictional objects with empty names are regarded as false in all models, i.e. permanently false, what does not fit our linguistic intuitions. Everybody who read the book Harry Potter considers the sentence Harry Potter was a wizard as true though the theory of descriptions would treat it as permanently false. This provokes another question. How is it possible to understand the sentences The present king of France is bald and Harry Potter was a wizard in a different way 12

13

Here I have implemented in an oversimplified way Russell’s suggestion that proper names are in fact definite descriptions. In his descriptivist approach to proper names Russell opposed John S. Mill and initiated still unresolved controversy between Millianists and descriptivists.

365 though they are false in all circumstances, i.e. in all models? The same problem appears in case of analytic sentences that are true in all models. Despite of the mentioned drawback of Russell’s theory, it explains the fact that the meaningfulness of a sentence containing an individual term does not depend on the existence of its referent. Another merit of the theory demands a slight reinterpretation. Let us substitute the notion of equivalence of a surface form of a natural language sentence to three sentences of the language of the predicate calculus with the notion of translation from one language to another. The source language is meaningful if it can be translated into the target language which is already meaningful. This idea was developed later by Quine and Davidson. 3. Some classical arguments against referential semantics Russell intended to overcome some ontological problems of the referential semantics of language on referential ground. In this section, I give three classical arguments against any referential semantics for natural language conceived as a source of its meaningfulness.14 The first one called the Argument from the learnability of language was presented by Davidson (1966). The argument starts with a reasonable assumption that every natural language can be learned in a finite time. This assumption was obvious to Frege who introduced the principle of compositionality in terms of function and argument. The referent or sense of a compound expression (e.g. a sentence) is a function of the referents or senses of its parts. Given the compositionality principle and knowing references or senses of finite number of simple expressions we can “calculate” references or senses of 14

I deliberately omit here the famous but very controversial Slingshot argument. The aim of the argument is to show that the sentences can only have two semantic properties: truth and falsehood, and cannot have any referents (e.g. states of affairs) or senses conceived as special referents. The typical but informal presentation can be found in some articles of Davidson, e.g. in (1967). The formal presentation of the argument can be found in Shramko and Wansing (2009).

366 infinitely many compound expressions. Let us consider a sentence John smokes and let us assume that the reference of John is an individual j and the sentence is true, i.e. its reference is just truth. If the compositionality principle holds the reference of smokes can be any function from individuals to truth values such that it ascribes truth to j.15 In belief-contexts like Mary believes that John smokes the expressions John smokes, John and smokes do not have their normal references but special ones which are their senses. If we iterate the contexts, e.g. John suspects that Mary believes that John smokes, etc., we have to introduce infinite hierarchy of references:16 normal senses (senses of the first level), special senses (senses of the second level), senses of the third level, etc. Davidson’s Argument from the learnability of language goes as follows. If the Frege-style referential semantics were correct, i.e. if it were adequate semantics of natural language, natural language would not be learnable. The problem is that there is no rule of constructing the higher-level sense of a word from the lowerlevel one. To learn a language we should have grasped infinite senses for every word.17 As the infinite hierarchy of senses is a serious problem in itself, it was amended by Rudolf Carnap, who in Meaning and necessity (1947) introduced two types of referents only: extension and intension. But Davidson argues, giving various examples, that if we take sufficiently rich fragment of natural language, any referential semantics violates the demand of learnability. The argument against Frege’s form of the compositionality principle is sometimes called the Third man argument.18 Let us consider the previous example John smokes. The problem is how the reference or sense of the sentence is constituted by the references or senses of both 15

16 17

18

But if the reference of smoke were the source of its meaningfulness, we should know this reference. This is impossible as we simply do not know all smokers in the world. I will come back to this difficulty little later. References can be considered as senses of level “zero”. We can say that Fregean semantics is “horizontally” compositional on every level of sense but is not “vertically” or “cross-level” compositional. The argument resembles original Plato’s Third man argument against his own theory of ideas pronounced by Parmenides. The argument was presented in Davidson (1967/1984: 19-21), see also Lepore and Ludwig (2005: 44-48).

367 words. On the syntactic level we apply the operation of concatenation. But concatenation is a syntactic notion, and if we want to understand the sentence the concatenation must have some semantic counterpart (its referent), i.e. we need a semantic device that permits us to apply a function (the reference or sense of smokes) to its argument (the reference or sense of John). Let us imagine that we construct a new function of two arguments which is the reference or sense of the word concatenation in the sentence Concatenation (John smokes). It is easy to see that the concatenation cannot be eliminated because now the sentence contains another concatenation represented by parenthesis.19 On the semantic level we need another referent for the new occurrence of concatenation, and this procedure can be extended ad infinitum. The conclusion is that the referential semantics cannot be de adequate for natural language sentences because the sentences contain meaningful devices that do not have neither referents nor senses. One can object to this argument and say that the problem was solved by Frege who introduced saturated expressions (possible arguments) and unsaturated expressions (possible functions).20 The sense of unsaturated expression smokes is the function from senses of individual constants to the senses of sentences. But here the problem of learnability comes back. If the referential semantics of this sort were right we would never be able to learn our first language because in order to understand a syntactically primitive expression like a verb smoke we should have grasped a function whose arguments were senses or references of possibly infinitely many other expressions. The last argument against referential semantics can be found in Quine’s Word and Object (1960). In his popular thesis of inscrutability of reference, Quine argues that the referent, i.e. whatever we understand as it, is always an arbitrary construct. The speakers use the notion of object as the result of language understanding, and not as a condition of language understanding. In consequence, the words are not meaningful 19

20

The concatenation can also be represented on syntactic level by rather complicated and artificial expression “belongs to” like in Smoking belongs to John. Davidson says that by introducing this distinction Frege only masked the problem (Davidson 1967/1984: 20).

368 because they refer to objects, but they refer to objects because they are already meaningful. Moreover different speakers of the same language can refer in the same expressions to different referents without a possibility of grasping any differences. In consequence, referential semantics of natural language has no explanatory force.21 The thesis of inscrutability of reference is connected with Quine’s theory of meaning in terms of translation manual and with the thesis of indeterminacy of translation. I will come back to this theory later. 4. Ajdukiewicz’s semantics without reference In the early thirties, Kazimierz Ajdukiewicz introduced the directival theory of meaning, which can be conceived as an early version of conceptual role semantics.22 The theory was based on the notion of belief understood as a disposition to approve sentences in certain circumstances. The inspirations of Ajdukiewicz were threefold: 1.

2.

3.

21

22

Frege maintained that sense was a vehicle of knowledge. Ajdukiewicz supported the contention of the close relation between meanings and knowledge but inverted the relation. In his approach the knowledge is the vehicle of linguistic meanings. Ajdukiewicz like Husserl considered the mental act of understanding a sentence as an act of thinking what the sentence says, and the act of perceiving its physical shape. Ajdukiewicz supported French conventionalism. The conventionalists argued that the scientific theory does not describe the

This should not be treated as presentation of Quine’s theses from Word and Object but rather a remark about it. The original presentation of the theses is very controversial and demands an elaborate explication. The informal presentation with references to natural language can be found in Ajdukiewicz (1931), the formal one in Ajdukiewicz (1934a), and the philosophical background was presented in Ajdukiewicz (1934b).

369 empirical data in an unique way but depends on conventions.23 In consequence there are various independent but coherent descriptions of the same reality. Ajdukiewicz maintained that for a person to speak a language is just to actualize his dispositions to the language use, and he understood any act of language use as an act of accepting sentences in relevant circumstances.24 These dispositions, i.e. the beliefs, are holistic. It is not possible to have one belief (belief holism), and in consequence to understand one sentence (meaning holism). Some sentences of a language express beliefs of the language user because he is motivated to accept them by three kinds of meaning directives: axiomatic, inferential and empirical. The axiomatic directives motivate the user to accept sentences in all circumstances; the inferential directives motivate the user to accept sentences if other sentences were previously accepted, and the empirical directives motivate the user to accept certain sentences if empirical data of the relevant sort are accessible.25 The meaning directives make every sentence that expresses a belief meaningful. In order to define meaning, Ajdukiewicz imposed very strict rules on translation between languages. The rules cannot change the structure of beliefs expressed in the language. Then he introduced the notion of a matrix representing the structure of beliefs. Every matrix corresponds to a class of mutually translatable languages. Translatable languages are just the languages that 23

24

25

French conventionalists were mainly philosophers of science. Ajdukiewicz tried to apply their ideas in the semantics of natural language. Ajdukiewicz anticipated some ideas from Wittgenstein’s Philosophical Investigations (1953), especially the idea of meaning as use. Contrary to Wittgenstein, he managed to precisify the term “use”. As Ajdukiewicz’s theory of meaning is based on use it is sometimes referred to as pragmatic. In my opinion, it is the main drawback of Ajdukiewicz’s theory. First of all Ajdukiewicz was obliged to name empirical data introducing unwillingly another language. Another problem is that the empirical beliefs actualized by occasional sentences do not in fact constitute meanings. This feature of Ajdukiewicz’s theory can be considered as the remainder of the verificationist approach typical for analytic philosophy of the early thirties.

370 differ in physical shape of linguistic expressions and do not differ in the role of their expressions in meaning directives. Every matrix consists of three parts corresponding to three types of meaning directives. Every part consists of a sentence that expresses a belief of a relevant sort, and its parts from compound to simple ones enumerated in a predefined order. The meaning of an expression (simple or compound) in a language is (or is represented by) its position in the matrix of this language.26 Though the general idea of Ajdukiewicz’s theory lets one avoid the main problems of referential semantics it has some obvious and mutually connected drawbacks: 1. 2. 3.

4.

26

27

The sentences of a language that do not express speaker’s beliefs have no meaning.27 Ajdukiewicz’s theory is not compositional. Ajdukiewicz treated natural language as a class of mutually nontranslatable languages conceived as incommensurable theories. The fact of communication between two people who speak two different natural languages can be explained in the following way: natural languages consist of mutually translatable languages, i.e. languages that have the same matrices but differ in the physical shape of sentences that actualize beliefs. This demand is highly unrealistic. The change of one belief results in changing of the meanings of all expressions of the language. The consequence of this

The position is defined in a rather complicated way as a family of sequences of natural numbers. I will omit the details here. One can maintain that every sentence s of a language must have its position in the relevant matrix if we accept a tautology with it, e.g. s ∨ ~s. But s occurs here not as a meaningful sentence but a sentential variable. In consequence, it does not have a position in the matrix, though the logical constants ∨ and ~ do have their positions (see Maciaszek 2008: 307-330 for the details). Ajdukiewicz developed his theory for the family of closed languages of scientific theories indicating that it can be applied to natural languages. In the languages of scientific theories it suffices to say that the theorems are meaningful.

371 contention is highly improbable: the condition of mutual understanding or translation should be the identity of all beliefs. Ajdukiewicz was not interested in intensional contexts, speech acts, propositional attitudes, and performative sentences. In case of intensional contexts Ajdukiewicz’s theory of meaning would be trivial and not satisfactory. But there is one point worth noticing. The notion of coextensional substitution makes no sense in Ajdukiewicz’s approach, for instead of it we have another kind of substitution – the substitution of expressions that do not change the matrix of a language. Let us assume that Peter believes in some sentences where the names Tully and Cicero appear. Each of the names can be substituted by another in all his beliefs, and in consequence they are synonyms. Let us add to his beliefs two additional ones: John believes that Tully was Roman orator and John does not believe that Cicero was Roman orator. In the extended language of Peter Cicero and Tully are no more synonyms. This disadvantage of the theory can be overcome if we separate the beliefs ascription and meaning ascription. This was done by Donald Davidson. 5. Tarski’s theory of truth At the same time when Ajdukiewicz published his papers on the directival theory of meaning, Alfred Tarski published his seminal paper “The Concept of Truth in Formalized Languages” (1933). Logicians and mathematicians usually interpret this article in a referential mode as an introduction to the model theory. But there are other interpretations of Tarski’s paper. In this section I will focus on translational interpretation of Tarski’s theory of truth.28 In the formulation of his theory Tarski uses typical notions of any referential semantics like object or infinite sequence of objects. For a given language (object language) one can formulate many different theories of truth. Every theory of truth is formulated in a meta-language and sometimes the meta-language can comprise the object language. 28

This interpretation is close to, but not identical with, Davidson’s interpretation.

372 The technicalities are roughly as follows: for a given theory of truth a sequence ascribes objects to individual constants and variables. In the first case, every sequence ascribes the same object to a constant and it does not matter if the object belongs to the sequence. Another theory of truth can ascribe another object to the same constant.29 In case of a variable, e.g. xi, every sequence ascribes its i-th object to it. Another important notion is the notion of satisfaction of a formula by a sequence, and probably this notion is the source of different interpretations of the Tarski’s theory. In the original formulation a formula A(a) where a is a variable or individual constant and A is a predicate is satisfied by a sequence s if the object ascribed to a by s is A, where A is the metalanguage translation of the predicate A.30 In a referential interpretation of this rule we can substitute the phrase “the object ascribed to a by s is A” by “the object ascribed to a by s belongs to the set A”.31 A formula is true if it is satisfied by all sequences. The most popular notion of Tarski’s theory is the notion of T-sentences, e.g. z is true if and only if tr(p) where z is a “structural description” of the sentence p of an object language and tr(p) is its translation to the meta-language. Let us imagine that we build a theory of truth in Polish as a meta-language for a fragment of English. The structural description of the sentence There are moments of great luxury in the life of a secret agent can be “the first sentence of Ian Fleming’s Live and let die”, and its translation into Polish: W życiu tajnego agenta zdarzają się momenty wielkiego luksusu. But in case of the meta-language comprising the object language z is “p” and tr(p) is p. In consequence, the T-sentence would be:

29

30

31

In a referential interpretation different theories of truth represent different models. In a translational interpretation they are just different translations. The original Tarski’s formulation of the theory uses the notion of translation. Nevertheless the translational interpretation differs from the original formulation. I will explain the difference later. I omit here other rules of satisfaction, e.g. for quantifiers.

373 “p” is true if and only if p32 The T-sentences are logical consequences of the theory. Tarski treated the translation from object language to meta-language as given. In consequence he maintained that he defined the very concept of truth. In the translational interpretation the notion of truth is assumed, and the theory of truth is just a mean of translation. In consequence in order to translate one language into another one has to construct a whole theory of truth. It means that neither a single rule of the theory nor a single T-sentence is not a translation of an appropriate expression. If we state that the we have constructed a theory of truth for a fragment of Polish in English as a meta-language with a rule stating that every sequence of objects ascribes Plato to Polish Platon, it just means that the whole theory, and not the mentioned rule on its own, translates Platon into Plato. The reason is very simple: the meanings of Platon and Plato depend on the meanings of other expressions of relevant languages. One can object to this interpretation stating that the theory of truth is referential because it uses objects, i.e. references of individual names. I think that this is a misunderstanding. The theory is constructed in a meta-language and it comprises meta-linguistic names of objects and not objects themselves. From this point of view the notion of the sequence of objects is just a device used in the formulation of the theory. As I have mentioned earlier there are many possible theories of truth or translations from an object language into a meta-language. The example given above shows that T-sentences play a very important role in our choice of the theory. We would probably reject the theory whose consequence is the following T-sentence: “Platon był filozofem” is true if and only if Aristotle was a painter The rejected theory would comprise the rule which ascribes Aristotle to Platon by every sequence, and the rule satisfaction: “a był filozofem” is

32

Some philosophers maintain that Tarski’s theory leads to the “disquotational” concept of truth. I do not share this opinion.

374 satisfied by a sequence s if and only if an object ascribed to a by s was a painter. The translation interpretation of Tarski’s theory presupposes the notion of truth. But the notion of truth comprised in a classical Aristotelian way presupposes the existence of a direct relation between language and reality or even isomorphism of the structures of language and reality. As we have no direct, i.e. extralinguistic, access to this relation, the classical notion of truth is not very useful. Nonetheless, the concept of truth we use reflects the idea that there is a connection between the language we understand, and reality. The nature of this connection will be discussed in the last section. The theory of truth was formulated for formal languages and Tarski expressed skepticism about the possibility of its application to natural languages. If we treat theory of truth as a compound definition of the semantic notion of truth given the notion of translation as granted, the prospects of its application to natural languages are not promising as we do not have the strict notion of translation for natural languages at hand. In the translational interpretation we can be more optimistic but we have to reconstruct the syntactic structure of natural language. This structure should be extensional as Tarski’s theory is extensional. 6. Quine’s theory of meaning in terms of the translation manual Before presenting the semantics of Donald Davidson it is worth mentioning the non-referential semantics proposed by Quine in Word and Object (1960). This semantics takes the form of a translational manual which gives the translation of all sentences of one language to the sentences of another language we already understand. In his exposition of the theory Quine refers to the “language of the jungle” and the translation is called “radical”, as the language is totally unknown. Nevertheless the results are valid even for two idiolects of the same natural language. The only basis of radical translation are stimuli that are accessible to both the speaker and the author of the translation manual. In consequence, an English speaking linguist recognizes step by step the

375 system of beliefs of the speaker expressed in his own language and translates them into English. The central role in this procedure have occasional sentences accepted by the speaker in the presence of certain stimuli. These sentences can be translated immediately into English (famous Gavagai translated as “Lo, a rabbit”). Then there are analytic (or stimulus-analytic) sentences which are accepted in the presence of all stimuli, and pairs of synonymous sentences accepted and rejected in the presence of the same stimuli. These sentences can be recognized be the linguist but cannot be immediately translated into English. Additionally, the linguist can translate logical sentential connectives. The rest of the translation depends o the analytic hypotheses that are tested in the presence of empirical data (stimuli). As Quine states, the translation manual, and in consequence the meanings are not determined by all possible stimuli. The thesis of indeterminacy is connected with the Dihem-Quine thesis which states that the scientific theory is not determined by all empirical data,33 and the system of beliefs expressed in natural language can be treated like a primitive scientific theory. In consequence, the same reality can be described in different ways. Even if the physical shape of the words of the “theories” is identical they can differ in meaning. The main problem that was not solved in a satisfactory way either by Ajdukiewicz or by Quine was the problem of the separation of meanings from beliefs. Another problem that Ajdukiewicz’s and Quine’a theories faced was the problem of compositionality. Neither the directival theory of meaning nor the theory of meaning in terms of the translation manual is compositional. Both problems were solved by Donald Davidson in his theory of interpretation where he achieved compositionality by applying Tarski’s theory of truth.

33

Pierre Duhem was one of the most important French conventionalists. His famous thesis was slightly reformulated by Quine in Two Dogmas of Empiricism (1961). The thesis was accepted by Ajdukiewicz who drew rather controversial conclusions for the theory o meaning of natural language.

376 7. Donald Davidson’s theory of interpretation Donald Davidson proposed the holistic theory of interpretation that consisted of two parts: the beliefs ascription and the meaning ascription. The meaning ascription consists in translating the words of a speaker in the context of the whole language (or idiolect) into the interpreters language (or idiolect). This translation is based on Tarski’s theory of truth conceived as an extensional and recursive means of translating the sentences of an object language (the speaker’s idiolect) into metalanguage (the interpreter’s idiolect). In consequence the meaning ascription is compositional. The procedure of translation can be verified empirically by the totality of T-sentences of the form: (The speakers sentence) “s” is true if and only if p (where “p” is an interpreters sentence) In case of two idiolects of the same language s and p usually have the same physical shape. The process of empirical verification depends on the ascription of beliefs to the speaker and confronting them with the beliefs of the interpreter. In consequence the interpreter can reject many possible theories of truth (meaning ascriptions). It does not mean that we can formulate the only “right” theory of truth. In fact, various combinations of belief ascription and meaning ascription remain possible but usually one theory is accepted for pragmatic reasons.34 As stated earlier, the formulation of Tarski’s theory of truth uses the notion of reference but the theory is not based on the relation of reference which requires existence of a referent for every expression, e.g. in Tarski’s semantics we can use the notion of an object when we apply the following rule: The sequence s ascribes (object) Harry Potter to the name “Harry Potter”

34

This indeterminacy of interpretation is in fact a counterpart Quine’s theses of the indeterminacy of translation.

377 In fact the theory does not demand the existence of the referent and does not make any ontological commitments.35 Another semantic notion, the notion of truth, is used in the theory but the theory itself does not state which sentences are true. The interpreter regards speaker’s sentence as true when the corresponding sentence of his own idiolect expresses his own belief. It does not mean that there is no relation between system of beliefs expressed by the sentences and the reality but this relation is not semantic. According to Davidson, the notion of truth as correspondence is based on causal relation between utterances of some sentences and the events in the reality. The Davidson’s notion of truth is not only purely correspondential but coherential as well. The truth conditions are just the beliefs that are coherent with the sentence in question and help us to ascribe truth values to our sentences and the sentences of a speaker as well.36 The majority of our beliefs must be true in this mixed sense though we cannot be sure which in fact are true. This is one of many formulations of Davidson’s Principle of Charity.37 In the theory of interpretation the main drawbacks of Ajdukiewicz’s directival theory and Quine’s theory of meaning in terms of the manual of translation can be eliminated: 1. 2.

35

36

37

All expressions including sentences have meanings ascribed them by the theory of truth. The theory of interpretation is compositional.

The semantic terms like “object”, “sequence”, “satisfaction”, and “truth” are expressions of the interpreter’s idiolect. If we ascribe truth value to an occasional sentence It is a dog we take into account other occasional sentences accepted in the same circumstances, e.g. It is an animal, It has fur, etc. These sentences are regarded here as the truth conditions for the first one. In consequence, the semantics of Donald Davidson should not be called truthconditional as many authors call it, but a truth-theoretic one. This fact has been recognized by Lepore and Ludwig in the title of their book on the philosophy of Donald Davidson (see Lepore and Ludwig 2007).

378 3.

4.

The fact of communication is explained by the notion of interpretation that consists of the meaning ascription and the belief ascription. The change of the beliefs does not necessarily change the meanings.

In spite of the obvious differences between Davidson’s and Ajdukiewicz’s theories, they have one significant common feature. In case of the interpreter of his own beliefs expressed in his own idiolect, the theory of truth becomes trivial.38 In consequence, the theory of truth cannot ascribe meanings to our own words. In this case, the meanings of the linguistic expressions depend, like in Ajdukiewicz’s theory, on the role of expressions in the totality of the beliefs of a speaker. As Davidson’s theory is compositional, the meanings of all expressions of our own idiolect depend on the position of the simple expressions in our beliefs. An open problem is how to represent them in a reasonable way or even if such a representation is possible. The two dimensions of the process of interpretation and the extensionality of the theory enables one to solve various puzzles39 but the real challenge for the theory is the intensionality of natural language. Davidson proposed to solve the problem in Russellian style, maintaining that natural language in its deep structure is extensional. The interesting but very controversial tool of revealing the extensional structure of natural language is the paratactic analysis. This analysis was applied to quotations (Davidson 1979a), sentences about propositional attitudes (Davidson 1978), and speech acts (Davidson 1979b).

38

39

It is trivial because the relevant T-sentences are trivial, e.g. My sentence “Snow is white” is true if and only if snow is white. In case of interpreting other speaker words T-sentences are not trivial, e.g. The speaker’s sentence “Snow is white” is true if and only if snow is white. They are not trivial because they are the result of translating speaker’s words into interpreter’s words and ascription of beliefs to the speaker. Saul Kripke formulated his famous puzzle of Pierre’s beliefs about London and Londres (see Kripke 1979) as an argument against Mill’s approach to proper names. I propose the solution of the puzzle in Maciaszek (2008).

379 The idea of the paratactic analysis consists in reducing a sentence with non-extensional context (or rather the utterance of it) to the pair of utterances of extensional sentences. Both sentences have their truth conditions. In case of a performative sentence, e.g. an order, the first one is an utterance of an occasional or indexical sentence This is an order which indicates the utterance of the second one (treated as an event). The utterance of the second sentence gives the content of the order. The truth conditions of both sentences in the sense indicated above are independent, so the first sentence is true if the second one is an order or false if it is not.40 The second utterance is true if the order (or false order) was obeyed or not. Let us consider an example of an order given by an officer to the soldier: I order you to crawl to that tree In Davidson’s approach this is not an utterance of a single sentence but of two sentences: The next utterance is an order. You crawl to the next tree The next utterance is an order is an indexical sentence like This is a dog that has its own truth conditions, e.g. if is pronounced by an officer to the soldier in certain circumstances, etc. In some circumstances it can be false (the case of a false order) – e.g. pronounced not by an officer, or pronounced by the officer to the general). The utterance You crawl to the next tree has its own truth-conditions, e.g. the soldier moved towards the tree, he approached the tree, etc. In consequence we have four possibilities: – true order obeyed – true order but not obeyed – false order but obeyed – false order not obeyed 40

There is an analogy to typical occasional sentences. The utterance of This is a dog is true it the object in question in fact is a dog.

380 The same approach was proposed to analyze utterances expressing propositional attitudes. Let us come back to the previous example of two sentences: John believes that Tully was Roman orator and John does not believe that Cicero was Roman orator. In Davidson’s approach the analysis would be the following: John believes that. Tully was a Roman orator. John does not believe that. Cicero was a Roman orator. Though the interpreter treats Tully and Cicero as synonyms that are interchangeable in all his beliefs it does not prevent him from noticing that they are not synonyms in John’s idiolect. The program of constructing an extensional and non-referential semantics of natural language in Davidson’s style is far from being ended (see Ludwig and Lepore 2007). In my opinion, the main drawback of Davidson’s original approach is the overestimation of the theory of truth and underestimation of beliefs. I tried to amend it in this section. But in spite of various difficulties Davidson’s program could enable a proper treatment of many phenomena in natural language like tense, propositional attitudes or modalities. The Davidson’s approach can also be applied to the interpretation of metaphor (see Davidson 1978) and other figures of speech (metonymy, irony and allusion). In spite of the contention that metaphors do not have special metaphorical meanings, the popular Gricean idea of speaker’s meaning can be conceived as the ascription of beliefs to the speaker. 8. Final remarks One of the last attempts to save referential semantics of natural language was the causal or historical theory of reference for proper names and natural species names (see Kripke 1980 and Lycan 2000: 50-71 for further references, standard objections to the theory, and responses of its advocates). The idea is that a name is meaningful because one can trace (at least theoretically) the causal chain of the acts of its use up to the first act of “baptism” of an object or the representative of a species. The

381 arguments I presented in this paper were directed against all referential semantics including this theory. The act of first baptism can be considered neither sufficient nor necessary condition of the meaningfulness of the name, though the picture drawn by the advocates of the theory is true in prevailing number of cases. I maintain only that the theory has no explanatory power. If we accept semantics without reference, it does not mean that we do not believe in external world. It does not even mean that we do not believe that expressions can refer the objects of the external world. It only means that for a particular meaningful expression we do not demand that it refers something. The expressions of natural language can refer and many of them do refer, but this relation is not to be assumed in semantics. The relation of reference is the problem of philosophy, and not of semantics. Davidson states that the utterances of sentences are just events. Some events in the environment result in utterances of occasional sentences which express our empirical beliefs. The nonreferential semantics I have sketched in this article does not lose the relation with reality. This relation holds because the majority of our beliefs must be true. This contention is justified by the fact that our mutual communication is possible and usually successful as based on the causal relation of utterances and events in the reality. References Ajdukiewicz, Kazimierz 1931. On the Meaning of Expressions. In: Kazimierz Ajdukiewicz, The Scientific World-Perspective and Other Essays. Jerzy Giedymin (ed.) 1978. Dordrecht: Reidel Publishing Company, 1-35. Ajdukiewicz, Kazimierz 1934a. Language and Meaning. The Scientific WorldPerspective and Other Essays. Jerzy Giedymin (ed.) 1978. Dordrecht: Reidel Publishing Company, 34-66. Ajdukiewicz, Kazimierz 1934b. The World-Picture and the Conceptual Apparatus. In: Kazimierz Ajdukiewicz, The Scientific World-Perspective and Other Essays. Jerzy Giedymin (ed.) 1978. Dordrecht: Reidel Publishing Company, 67-89. Carnap, Rudolf 1947. Meaning and Necessity. Chicago: University of Chicago Press.

382 Davidson, Donald 1966. Theories of Meaning and Learnable Languages. In: Donald Davidson 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press, 3-15. Davidson, Donald 1967. Truth and Meaning. In: Donald Davidson 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press, 17-36. Davidson, Donald 1968. On Saying That. In: Donald Davidson 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press, 93-108. Davidson, Donald 1978. What Metaphors Mean. In: Donald Davidson 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press, 245-64. Davidson, Donald 1979a. Quotation. In: Donald Davidson 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press, 79-92. Davidson, Donald 1979b. Moods and Performance. In: Donald Davidson 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press, 109-21. Frege, Gottlob 1892. On Sense and Reference. In: Peter Geach and Max Black (eds.), Translations from Philosophical Writings of Gottlob Frege. Oxford: Blackwell, 56-78. Hylton, Peter 2003. The Theory of Descriptions. In: Nicholas Griffin (ed.), The Cambridge Companion to Bertrand Russell. Cambridge: Cambridge University Press, 202-240. Ingarden, Roman 1947/1948. Spór o istnienie świata t. 1/2 [Controversy over the Existence of the World, vol. 1- 2]. Warszawa: PWN. Kripke, Saul 1979. A Puzzle about Belief. In: A. P. Martinich, (ed.) 2001. The Philosophy of Language. Oxford: Oxford University Press, 405-432. Kripke, Saul 1980. Naming and Necessity. Oxford. Blackwell. Lepore, Ernie and Kirk Ludwig 2005. Donald Davidson. Meaning, Truth, Language, and Reality. Oxford: Clarendon Press. Lepore, Ernie and Kirk Ludwig 2007. Donald Davidson’s Truth-Theoretic Semantics. Oxford: Clarendon Press. Lycan, William 2000. Philosophy of Language. A Contemporary Introduction. Routledge. Maciaszek, Janusz 2008. Znaczenie, prawda, przekonania. Problematyka znaczenia w filozofii języka [Meaning, Truth, Beliefs. Problems of Meaning in Philosophy of Language]. Łódź: Wydawnictwo UŁ. Modrak, Deborah 2006. Philosophy of Language. In: Mary Louise Gill and Pierre Pellegrin (eds.), A Companion to Ancient Philosophy. Oxford: Blackwell, 640663. Quine, Willard Van Orman 1960. Word and Object. Cambridge, MA.: The MIT Press. Quine, Willard Van Orman 1961. Two Dogmas of Empiricism. In: W. V. O. Quine From the Logical Point of View. Cambridge, MA.: Harvard University Press, 20-46.

383 Russell, Bertrand 1905. On Denoting. In: A. P. Martinich, (ed.) 2001. The Philosophy of Language. Oxford: Oxford University Press, 212-220. Shramko, Yaroslav and Heinrich Wansing 2009. The Slingshot Argument and Sentential Identity. Studia Logica 91, 429-455. Strawson, Peter 1950. On Referring. In: A. P. Martinich, (ed.) 2001. The Philosophy of Language. Oxford: Oxford University Press, 228-242. Tarski, Alfred 1933. The Concept of Truth in Formalized Languages. In: Alfred Tarski (ed.), 1956. Logic, Semantics, Metamathematics. Oxford: Oxford University Press. Wittgenstein, Ludwig 1953. Philosophical Investigations. Oxford: Blackwell Publishing.

Andrei Mărăşoiu Georgia State University [email protected]

Are Propositions Facts? Abstract: This paper explores whether Jeffrey King’s theory of propositions is committed to an obscure metaphysics which identifies propositions with certain kinds of facts. Section 1 presents the problem King tries to provide a solution to, the problem of unity of the proposition. Section 2 presents King’s doubtful identification of propositions with certain existentially generalized facts. Section 3 sketches a host of objections to that identification read as a substantive metaphysical claim. Given the failure of such a reading, Section 4 argues in favor of a deflationist approach: we can better understand propositions by attempting to ramsify them – to provide substitutes that do all the explanatory work propositions do, but are not metaphysically dubious. I argue that King’s claim to identify propositions with facts is better interpreted as an example of such a ramsification project, and not in a metaphysically substantive way. Key words: Jeffrey King, structured propositions, possible worlds, facts, logical form, Ramsey-sentence, Bertrand Russell, unity of propositions, context of utterance

1. Propositional roles and the problem of the unity of propositions1 Propositions play important roles in linguistic and philosophical theorizing. Some of these roles are the following:2 (i) what is truth-apt; (ii) the semantic information uttered sentences encode; (iii) what translation preserves; (iv) what uttered sentences conversationally imply; (v) what speakers producing those uttered sentences mean; (vi) the 1

2

I would like to thank Nora Grigore, Alexandru Rădulescu, Petrişor Ivan, Tadeusz Ciecierski, Mircea Flonta and Andrei Moldovan for comments on earlier drafts. Roughly the same list of roles propositions are needed to play is present in King (2001: Introduction) and Soames (2010: 2-4). This list of roles presupposes Salmon’s (1986: 58) distinction between information semantically encoded in a sentence and information pragmatically imparted by uttering a sentence.

386 objects of propositional attitudes such as belief, desire, hope, fear, doubt; (vii) what is susceptible of having modal properties such as being necessary, possible, contingent; (viii) the designations of expressions such as “logicism”, “Goldbach’s conjecture” etc. I take the claim that these semantic roles are played by propositions as a prima facie plausible claim. Fictionalists about objects of propositional attitudes, or adepts of Russell’s multiple relation theory of judgment, will take issue with affirming the existence of propositions merely on the basis of the claim that propositions play these semantic roles. However, this dispute is beyond the scope of this essay. I assume from the very start that propositions exist and that they play (most of) the roles ascribed to them. These several roles assigned to propositions give us no clue as to what the kind or kinds of objects filling in those roles are. An inquiry into what kind of objects propositions are is an inquiry into the nature of propositions, and it involves explaining what binds propositional constituents into one and the same unity, the proposition. One way of starting such an inquiry is exhibited by Bertrand Russell, who says: A proposition, in fact, is essentially a unity, and when analysis has destroyed the unity, no enumeration of constituents will restore the proposition. The verb, when used as a verb, embodies the unity of the proposition, and is thus distinguishable from the verb considered as a term, though I do not know how to give a clear account of the precise nature of the distinction. (Russell 1903: 50) Owing to the way in which the verb actually relates the terms of a proposition, every proposition has a unity which renders it distinct from the sum of its constituents. All these points lead to logical problems. (Russell 1903: 52)

This is a dense passage, and many of Russell’s concerns are not strictly semantic. What are the points that are of specific semantic interest in the text? One answer could be the following. As Jeffrey King (2009: 257) points out, there are at least three questions underlying Russell’s metaphysical investigation into the unity of propositions, which may be formulated by taking an example, say, the proposition that Dara swims:

387 Unity Question 1 (UQ1) What holds the constituents Dara and the property of swimming together and imposes structure on them in the proposition that Dara swims? Unity Question 2 (UQ2) How does the “structured complex” that is the proposition that Dara swims manage to have truth conditions and so represent Dara as possessing the property of swimming? Unity Question 3 (UQ3) Why does it at least seem as though some constituents can be combined to form a proposition (Dara and the property of swimming), whereas others cannot be (George W. Bush and Dick Cheney)? (UQ1) concerns what holds propositions constituents together in a proposition; (UQ2) concerns how a proposition can have both a propositional structure and truth conditions which determine its representational content; and (UQ3) concerns what makes some structures propositional structures while others not, and what makes some groupings of constituents propositions and others not. It is not entirely clear what each of (UQ1)-(UQ3) aims at. It seems that (UQ1) is concerned with propositional structure, (UQ2) with truth conditions and representation, and (UQ3) with the arguments of which that propositional structure is a function. But, with respect to (UQ2), King is hasty in assuming that propositions “have truth conditions and so represent”. Scott Soames (2010: 102) would rather have it that propositions represent, and so have truth conditions. Although both points of view assume there is an essential relation between truth conditions and representation, neither author fully spells out why this is supposed to be so.3 Furthermore, the distinction between (UQ1) and 3

King believes propositions represent because they have truth conditions, and their representational character is nothing over and above these truth conditions. By way of contrast, two options may be formulated. For Soames (2010: 102

388 (UQ3) is not completely obvious, since we inquire into a structure and its arguments: and finding out what the structure is (UQ1) already constrains what the arguments could be (UQ3), although perhaps not sufficiently. (UQ1)-(UQ3) articulate Russell’s problem concerning the unity of the proposition. For the rest of this section, I will present one solution which has been formulated in response to this problem, Jeffrey King’s account of propositions. The rest of this paper will be concerned with a discussion of this solution to the problem. According to King, the propositional relation binding the propositional constituents is the sentential relation of the sentence expressing it, composed with the relation of semantic satisfaction. Since semantic satisfaction is assumed not to add any structure to the propositional relation, the structure of the proposition and of the sentence expressing it will be identical. For example, the structure of the proposition that Fa is the same as the structure of the sentence “Fa”; the difference between the two is that the relation of semantic satisfaction is added (“F” designates F, “a” designates a), the lexical items being replaced by their semantic values. In tree-form: (1)

R “F” S F-ness

R'

“a” S a

R: sentential relation R’: propositional relation S: semantic satisfaction “F”, “a”: words F, a: properties and objects

-103), propositions have truth conditions because they represent; propositions are cognitive event-types and their representational character (which determines their truth conditions) is inherited from the cognitive event-tokens they typify. A completely different stance is taken by McGlone (forthcoming, fn.8), for whom the fact that propositions have truth conditions (hence represent, just like for King) is a brute fact; there is no sensible question to be asked (as Soames would like to ask) about what it is that makes propositions have the truth conditions that they do. An assessment of these approaches goes beyond the scope of this paper.

389 A view along the lines of (1) has the resources needed to answer the questions (UQ1)-(UQ3). Below, I briefly sketch how it is able to do so. (UQ1) asks how it is possible for propositional constituents to be bound in a proposition together, and what role is played by the structure of the proposition. On King’s account, the propositional relation and the sentential relation have the very same structure, and so what structure the proposition will have will be determined by what structure the sentence expressing it will have. Jeffrey King (2007: 28, 66) identifies the sentential relation with what Robert May (1985/1997: 281-315) calls the logical form, or LF representation, of a sentence. According to May, the logical form of a sentence is obtained by transformations from the surface-structure of that sentence (e.g., by quantifier raising, which makes the scope relations between the expressions composing the sentence explicit).4 Viewing the propositional relation on the model of the sentence relation (or LF) will amount to substituting the words bound at LF with their semantic values, thereby obtaining the proposition expressed by the sentence whose LF was used in the substitution. This explains how propositional constituents, which are the semantic values of the words in the sentence, are bound in the proposition expressed by the sentence. (UQ2) asks how a proposition can represent the world and have truth conditions. The question concerns semantic values. For King, words receive semantic values in agreement with possible-worlds semantics. In particular, King (2007: 6-7) seems to favor a direct-referentialist semantics. Given that propositions are structured entities (with their structure given by the LF of the sentences expressing them), a proposition will not be identical to the set of possible worlds at which it 4

King’s presentations of his theory of propositions do not always distinguish between LF and surface-structure. King focuses on simple examples such as “Rebecca swims”, “Dara swims” etc. which do not exhibit scope ambiguities or other phenomena which could discriminate between surface-structure and LF. Although it is sometimes ambiguous which of the two King refers to when he uses the expression “sentential relation”, I take him to be referring to LF, in agreement with other works of his (King and Stanley 2005: 130, 149), as well as with the explicit statement: “I am ultimately interested in the relations words stand in at LF.” (King 2007: fn. 46)

390 is true. However, for each proposition, there will be such a set, and membership of the world of evaluation in the truth-set of the proposition will constitute the condition under which the proposition will be true. This view is meant to account for how a proposition receives its truth conditions. The same view is to account for how a proposition manages to represent the world, given King’s (2009: 257) claim that propositions “have truth conditions and so represent” (see fn. 3 above). (UQ3) asks how it is that some propositional constituents can be put together into a proposition, while others cannot. King’s answer to this is twofold. First, some objects which are semantic values, e.g., Bush for “Bush” and Cheney for “Cheney” cannot be bound together in a proposition because there would be no propositional relation binding them, and this would occur because there would be no grammatical sentence (and hence no LF) formed by concatenating just the proper names “Bush” and “Cheney” in that order, to use King’s (2009: 257) example. Consider, as a second case, the following: neither would “Numbers sing terribly” literally express a proposition, in spite of the fact that the sentence is grammatical, and one can build its LF.5 What prevents “Numbers sing terribly” from expressing a proposition is that the word “number” is specified in the lexicon with the feature: nonanimate, while the verb “sing” as requiring an animate agent. Not satisfying this selectional rule is what prevents “Numbers sing terribly” from expressing a proposition. One thing is worth emphasizing, since it has consequences in the division of labor: if propositional structure is sentence structure, then which structure it turns out to be is no longer a metaphysical question, but one to be settled by linguists rather than philosophers. 2. Identifying propositions with certain facts King’s main book, The Nature and Structure of Content (2007), complements the account of the structure of propositions sketched above 5

This approach to “Numbers sing terribly” and other absurd but syntactically wellformed sentences seems theoretically economical and in line with King’s general theorizing, but I have been unable to locate where King himself discusses them.

391 with an account of the nature of propositions. (King 2007: 25-26) aims at a naturalistic account, according to which we should have good reasons to believe propositions exist, independently of any semantic theorizing. To this end, he identifies a proposition “Fa” with the following fact: there is a language L, lexical items “F” and “a”, a context of utterance, a speaker, a time and a place, such that: F is the semantic value of “F”, a is the semantic value of “a”, “F” is the leftmost node of the sentential relation of “Fa” and “a” is the rightmost node of the sentential relation of “Fa”. To illustrate what he means, King says: The following fact is the proposition that Dara swims, where we include as part of the fact/proposition that the propositional relation in it encodes ascription: there is a language L, a context c and lexical items a and b of L such that a and b occur at the left and right terminal nodes (respectively) of the sentential relation R that in L encodes ascription and Dara is the semantic value of a in c and the property of swimming is the semantic value of b in c. (King 2009: 270)

Here is one way of cashing out what King says about Dara’s swimming. Consider the sentence “Dara swims”. This sentence expresses either the proposition that Dara swims at the time of the utterance or the proposition that she can swim. For simplicity, let us assume it expressed the proposition that Dara swims at the time of the utterance. What sort of fact would this proposition be identified with? It would be the following fact: – there is a language L (here English), such that L includes the lexical items a1 (“Dara”), and a2 (“swims”) and there is a context c, speaker s, time t and place p such that: (i) the semantic value of a1 relative to c, s, t and p is Dara; (ii) the semantic value of a2 relative to c, s, t and p is the property of swimming (iii) the ntuple satisfies the sentential relation [[ _ ]NP [ _ ]VP]S, and the result is the sentence [[ a1 ]NP [ a2 ]VP]S (iv) speaker s could have uttered the sentence obtained by concatenating a1 and a2 in context c, at time t and place p.

392 The first existential quantifier “there is” takes scope over all the clauses (i)-(iv) mentioned above. The fact that there exists a language with certain words (which have certain contents across possible worlds) which are conjoined and uttered comes to be seen, for King, as being identical to the proposition expressed by the sentence thus formed. 3. Which facts are propositions? This section will explore several problems with the metaphysical reading of King’s association of propositions with certain existential facts. Failure of the indiscernibility of identicals. In identifying propositions with facts, King (2007: 26-27) draws on the work of Russell and Wittgenstein. But then perhaps all three authors are subject to the following objection: a simple and basic intuition is that propositions aren’t any kind of facts (Speaks 2009: fn. 26). Propositions differ from facts because there are properties facts have and propositions don’t (facts obtain or not, propositions cannot be said either to obtain or not to obtain); and there are properties propositions have but facts don’t (propositions represent the world, and have truth conditions, while facts simply are or are not, they don’t represent what there is, nor do they have truth conditions). Failure of substitution of the words “proposition” and “fact”, regardless of how “fact” may be further modified. Where do these powerful commonsensical intuitions that propositions aren’t facts come from? One plausible answer is that it is a matter of language: more often than not, “proposition” and “fact” cannot be substituted one for the other salva veritate:6 (2a) It is a fact that France is not a monarchy. (2b) ? It is a proposition that France is not a monarchy. (3a) John was unaware of the fact that Mary had left. (3b) * John was unaware of the proposition that Mary had left. 6

The examples given in the text parallel examples given by Frederike Moltmann (2003: 82-84). I use “*“ to indicate ungrammaticality, and “?” to indicate questionable grammaticality.

393 (4a) The district attorney needs hard facts to prove her case. (4b) * The district attorney needs hard propositions to prove her case. (5a) Carnap denied the existence of facts. (5b) Carnap denied the existence of propositions. These examples share a common pattern: all (a) sentences are grammatical, a fact which competent speakers of English could testify for, while (b) sentences are either ungrammatical, viz. (3)-(5), or are of doubtful grammaticality, viz. (2). Substitution of “fact” for “proposition” fails in idiomatic expressions such as “it is a fact that” (1a) or “needs hard facts” (4a), in pleonastic expressions such as “unaware of the fact that” instead of “unaware that” in (3a), and in belief contexts (Carnap was free to extend or not his ontological suspicions from facts to propositions, this is not something a semantics for English could decide). The examples could be multiplied, cf. Moltmann (2003: 88-94) for additional important cases. Avoiding objections merely by denying ordinary language intuitions the role of data. Both the commonsensical objection that propositions are not facts (because they have different properties than facts) and the linguistic intuitions that “proposition” and “fact” have different distributional contexts are nothing more than ordinary language platitudes; they do not amount to a fully worked out argument against, nor an alternative theory to, the theories of Russell, Wittgenstein and King, who claim propositions are facts. Speaking from the standpoint of such theories, King disposes of our pretheoretical intuitions about propositions, saying they have no bearing on semantic theorizing. The notion of proposition expressed by a sentence is a theoretical one. But that means we just don’t have pretheoretical intuitions about which sentences express the same propositions. (King 2007: 101) Thus there is no reason to think that the ordinary use of ‘proposition’ outside philosophy […] tracks the theoretical notion in philosophy. (King 2007: fn. 82)

Nevertheless, if these platitudes can be given satisfaction by being kept as part of an overall theory of language, methodological considerations

394 of conservatism suggest they should be so kept. Rejecting the platitudes would be coherent, but it would take semantic theorizing further away from the semantic intuitions which serve as its main source of evidence. Saying that the theory works would no longer seem satisfactory from the layman’s viewpoint; one should also prove no better theory, one incorporating the platitudes, is available. Outright rejection of pretheoretical intuitions is a non-reply King gives to the platitude that propositions aren’t facts (because they have different properties). A second attempted reply King has against the platitude is a curious one: King (2007: 52) seems to accept the literal falsity of the identity he upholds: he says that when we view the same things one way, we view them as propositions, and when we view those same things another way, we view them as facts (about languages, contexts and speakers). I believe this reply is incoherent. The only rephrasing I can fathom would have it that propositions are mysterious in nature, so it would perhaps be theoretically parsimonious if we could have some kinds of facts play all the roles we take propositions to play. But I will argue that the sort of facts King would like to be surrogates of propositions are themselves ontologically obscure, and of little explanatory value. Compound facts. I would like to begin by addressing an ontological worry: how can an existential generalization be or be made true by a fact? In particular, how can the existential generalization over languages, words, contexts and speakers which King proposes to identify propositions with count as a fact? Do truth-functional compounds denote truth-functional facts? Consider (6) and (7): (6) (E) (A1x o A2x o A3x o …. o Anx), where A1,…, An are all the words defined in the enlarged Webster dictionary. (7) (Ex1, x2, …, xn)(A1x1 o A2x2 o … o Anxn), where A1,…, An are any complex predicates. Do (6) and (7) designate facts? If the answer is yes, then, in reply, the complexity of the examples could be increased ad libitum. As Mulligan and Correia (2007: §1) point out, this bloated ontology of facts is

395 “simply incredible”. Only a thoroughgoing realist about facts will accept the challenge. But the thoroughgoing realist will have to answer the following objections. What is the fact in (7) about? It predicates A1 of x1 or … or An of xn. How could this be spelled out more clearly? Wouldn’t we say there is a disjunction, in which either A1 is predicated of x1, or A2 is predicated of x2 or...; is there a separate disjunctive predication, over and above these atomic ones? Which would the predication be? It can’t be A1 o A2 o A3 … predicated of , because A2 is only predicated of x2, not of x1 or x3. An elegant solution out of these qualms is offered by Russell (1918/2007: 191-192). He accepts that Anxn is a fact, but thinks atomic facts are sufficient for the purposes of connecting language and reality. For Russell, complex facts are not needed; is he wrong in believing compound facts are not needed? If so, why, and how could the remaining difficulties connected with highly disjunctive predication be solved? The ontological issue is at least twofold. First, there is the issue of the ontological status of composed facts: King would simply have it that (6) and (7) express truth-functional facts over and above the facts being compound by means of truth-functions. Here I have suggested Russell differs from King. Second, there an ontological problem which surfaces even at the level of atomic facts, because the facts identical to the propositions expressed by (6) and (7) would be highly disjunctive: not only over A1, …, An, but also over all spoken languages spoken on Earth, all speakers, all contexts, all space-time regions. One sentence, infinitely many facts. In order to focus on variation in contexts and speakers, consider “Snow is white” and pretend it did not have any faithful translations in any natural language. Moreover, suppose it did not have any suitable paraphrase in English, so that the only way of expressing the proposition that snow is white by means of an English sentence would be by uttering “Snow is white”. Given that the language and the lexical items are fixed, the proposition that snow is white would then boil down, on King’s view, to an existential generalization over contexts, speakers, places and times; and if this existential generalization were not properly speaking a fact, but a generalization over atomic facts, what would those atomic facts be?

396 There are (denumerably) infinitely many possible speakers of English, and a continuum of space-time points to make up a continuum of possible contexts of uttering “Snow is white”. Are we to understand King as claiming that a proposition is identical to an existential generalization over a continuum of facts about contexts? Then we have a dilemma. If the proposition in question is highly sensitive to a particular context, then there will be continuum-many other contexts which will be superfluously quantified over. If the proposition is context-insensitive, then why bother to quantify over contexts at all, in addition to quantification over languages and their lexical items? Explanatory value? If facts abound and obtain for cheap, as (6)-(7) and examples of increasing complexity might suggest, then one may wonder what the explanatory value of these facts is. In what context could (6) or (7) be usefully appealed to as explanations? The genuine concern is that propositions could not be the facts that King identifies them with, precisely because propositions have explanatory value. In Section 1 above, I have mentioned some of the important theoretical roles propositions play. If propositions could not fulfill most of these roles, their theoretical utility would be doubtful; this is why the theoretician of propositions is committed to saying propositions are notions which have explanatory value, and why there is an embarrassment in identifying explanatory propositions with nonexplanatory facts, as King does. This suggests another dilemma for King’s view of propositions. On the one hand, if the proposition expressed by the sentence “Amelia talks” is identified with a specific fact about a specific language, lexical items and context of utterance, then that same proposition could not be expressed using different words or on different occasions. But this is precisely why propositions were mostly needed in semantics: to be communicable pieces of information which can remain unchanged even when languages and contexts of utterance are changed (Salmon 1986: 11-18). If, on the other hand, the proposition expressed by “Amelia talks” is identified with an existential generalization over facts about specific languages, contexts and speakers, then there is the ontological worry that an existential generalization is not a fact, and the explanatory worry that such an existential generalization will be of little value in

397 semantic explanation, whereas the proposition it is identified with has a theoretical existence accepted precisely because of the fruitfulness of propositions in semantic explanations. On both horns of the dilemma, there are problems with identifying propositions with kinds of facts, and the natural conclusion seems to be that propositions aren't any kind of facts. 4. Ramsifying propositions away Section 3 presented a case against associating a metaphysics of facts about language to propositions. The objections mentioned in Section 3 may be debated, and some of them may be finally rejected. But an adequate theory of propositions does not only have to withstand objections (and it is difficult to see how it can withstand all objections), but also to be explanatorily satisfactory. The identification of propositions with existentially generalized facts seems to bring more metaphysical obscurity than clarification of the “nature and structure” of propositions. Why not renounce King’s main account? Other than this identification, King’s account of propositions seems to be a good candidate for explaining propositions, a candidate worth keeping. It tries to build a compromise between the tradition of structured propositions and the tradition identifying a proposition with the set of possible worlds at which it is true. The Russellian tradition of structured propositions has the advantage of higher psychological plausibility: it is not a set of worlds, or a set of situations, that humans represent when they think. Provided human thoughts could be represented as events which instantiate attitude types captured by relations such as “knows that”, “believes that” and “desires that”, and provided the second argument of these relations is always a proposition, while the first argument is the agent having the attitude, then it is plausible to assume that cognitive representation of a proposition involves, inter alia, grasp of a certain propositional structure, giving weight to Russellian considerations. On the other hand, the possible-worlds account of propositions well explains how propositions have truth conditions and representational content,

398 while the Russellians will have a hard time explaining failure of substitution in belief contexts, or empty names. Any attempt to compromise between the two, and to extract their benefits while avoiding, if possible, their shortcomings, is welcome. King’s account of propositions seems to do this. On the one hand, given that propositional structure is identified with sentential structure, Russellian concerns are appeased. Furthermore, the identification of propositional structure with sentence structure (i.e. with logical form) has the advantage of bringing theories of natural language syntax to bear on the abstruse topic of the metaphysics of propositions. On the other hand, letting the propositional constituents (the semantic values of the expressions making up the sentence that expresses the proposition) be the possible-worlds contents associated with those expressions has the advantage of accounting for representation conditions. Furthermore, although a proposition is not identified with a set of possible worlds, each proposition (understood à la King) can be uniquely association with a set of possible worlds, thus accounting for truth conditions as well. In sum, it seems that requirements (UQ1)-(UQ3) are satisfied by this account in a better way than they are satisfied by either Russellian propositions or sets of possible worlds. What does it mean to identify propositions with facts? Notice that King’s identification of propositions with facts plays no role in the argument made in the preceding paragraph in favor of it. The purpose such a fact-proposition identification is supposed to achieve is ontological and explanatory clarification, and it fails at that, as section 3 tries to show. The issue seems to be complicated by the fact that King seems to be pushing a metaphysical point: he seems to wish to account for the “nature and structure of content”, and not for an adequately explanatory substitute for propositions. In what follows, I propose to renounce this claim to metaphysical truth or insight into the nature of propositions. How can this be achieved? Section 1 has presented a list of explanatory roles propositions play in a theory of language. These explanatory roles themselves seem to be accounted for whatever abstract entities satisfy (UQ1)-(UQ3), i.e. by whatever abstract have structure, truth conditions, and representationality (though perhaps, on elaboration, additional

399 restrictions may need to be imposed). The deflationist route is the following: find whatever does the job you wish propositions to do, and label those as the adequate ersatz propositions your theory of language will work with. The most convenient way to express this formally is a modified Ramsey-sentence for propositions (Ramsey 1929, Lewis 1970): (8a) (E1x)(c1(x) a c2(x) a … a cn(x)), or, more explicitly, (8b) (Ex)(c1(x) a c2(x) a … a cn(x) a (Ay)((c1(y) a c2(y) a … a cn(y)) → y = x)) In (8a-8b), c1,…, cn are the constraints the entities have to satisfy in order to count as surrogate propositions. In the simplified version above, there are only three constraints (structure, truth conditions, representationality). The unique existence claim is meant to impose the further constraint that, once you choose a set of entities that play the role of propositions in your theory, you should consistently use that set of entities throughout. Is there any philosophical justification for these technicalities? Yes, an obvious philosophical justification is clarifying one’s ontology. This is precisely the deflationist line taken in Ramsey’s “Facts and Propositions” (1927: 157-159). For Ramsey, propositions, true propositions, facts, facts that propositions are not false, etc. all come down to the same thing, namely, interpreting as true a sentential formula. But there is a second philosophical justification, apt to appeal philosophers not moved by ontological parsimony or sweeping pronouncements of eliminating entities. In his Philosophy of Logical Atomism (1918: 187-188, 192), Russell’s reasoning seems to be the following: if the only reason why we believe propositions as abstract entities exist is because we have posited them to fill certain semantic roles, then there is no loss of either information or function in constructing the Ramsey-sentence for propositions, and then saying there may be values of the variable quantified over which are not, intuitively, what we had thought propositions are. Our intuitive conception of propositions was mistaken because it assumed entities

400 which were not only redundant (this much was clear from Ramsey’s parallel point highlighted above), but obscure as well. If the only thing which made us think propositions exist were the functions they fulfilled, then having some set of entities fulfill those functions is quite enough for having what we wanted from propositions. Saying that propositions are thereby eliminated or that they are thereby identified with the new entities is a terminological issue. What I want to claim is that King’s attempt to identify propositions with existentially generalized facts is an unfortunate result of good intentions. The result is unfortunate because it brings metaphysical and explanatory obscurity, as suggested in Section 3. But the two good intentions are the following. First, to ramsify propositions away. King may not have actually had this intention, since this hardly explains his insistence on revealing the (true) “nature” of propositions, and this can hardly be done by replacing entities with more kosher entities to play their semantic roles. But King can be reinterpreted as advocating a ramsification of propositions. The theory producing the constraints c1, …, cn in (8a) would be a theory of language, broadly understood, so as to comprise both semantics and pragmatics. The theoretical term to be ramsified (eliminated in favor of an existentially bound variable in (8a)) would be the term “proposition”. The term which could count as providing an instance which would make (8a) true would be “LF in which the words have been replaced by their semantic values”, where “semantic value” would further be cashed out in possible-worlds terminology. King’s theory could then be seen as an attempt to show that the term “LF in which the words have been replaced by their semantic values” satisfies the constraints c1, …, cn. The word “proposition” in its traditional, metaphysical understanding, would then be clearly explained and would, in principle, be eliminable in favor of its LF explanans. Hence there would be no need to seek metaphysical identifications of propositions with facts of one sort or another. The second good intention lying behind King’s metaphysical identification of propositions with facts is this: the fact identified by him with a proposition also generalizes over contexts, speakers, times and places. The term that should replace (or explain) the term “proposition” is then not, strictly speaking, the one mentioned above, but rather “LF in

401 which the words have been replaced by their semantic values relative to a context of utterance” (cf. King 2007: fn.35). This is evidence of the fact that King is sensitive to criticism coming from pragmatic quarters, and suggesting that which proposition a given uttered sentence expresses may largely be determined by context of utterance. King’s own take on the issue is presented at length in (King and Stanley 2005), but it is important to see that, at least in a general fashion, context enters into what a proposition is. One could then further add a constraint in (8a) to the effect that there is a speaker that could utter in context (and at a given place and time) the sentence which would then express the proposition.7 King’s appeal to existentially generalized facts, although metaphysically obscure, is seen in retrospect to have had a reasonable explanation, namely, that there is a context relative to which semantic values are assigned to words and replace them in the LF to produce the proposition. The existential quantification over contexts is also needed in order to allow for the possibility of one and the same proposition being expressed in two different contexts. My own appeal to a Ramseysentence for the term “proposition” is meant to dispel the impression that there is any theoretical or explanatory role whatsoever played here by (existential) facts, metaphysically understood. A feature of the view under consideration is that it leaves it open what a context is as well as what a speaker is. One could, for instance, argue that a future computer passing the Turing test would qualify as a speaker in spite of not grasping any proposition. But perhaps a proposition would be expressed by the sentence produced by the computer in spite of the computer’s incapacity to grasp that proposition. What a context is may also be controversial. Are contexts the formal constructions presented in Montague, Kaplan or Stalnaker, or do the contexts belong to the actual external environment of the speaker and audience? King’s account of propositions, as amended by the ramsification of “proposition” suggested above, remains neutral on all these important questions. On 7

Notice that this constraint does not limit the number of propositions to those expressed so far in the conversational history of mankind. All propositions are allowed into existence as long as they could have been expressed by an uttered sentence at one point or another.

402 the one hand, this may seem unwelcome, as it clearly leaves much work to be done. On the other hand, it may prove beneficial, as it tries to separate which parts in semantic and pragmatic theorizing are relevant to a general characterization of what all propositions are (and this issue is partly addressed), and which parts are relevant to finding out which particular proposition is expressed in which particular context (the suggestion being that such an enterprise lies beyond a general theory of propositions). 5. Conclusion The topic of this essay has been Jeffrey King’s theory of propositions. In Section 1, I have presented Russell’s problem of the unity of proposition, and how this problem ramifies into accounting for why propositions have structure, truth conditions, and representationality. I have further presented King’s theory of propositions as an attempt to answer Russell’s problem. In Section 2, I have formulated King’s claim that propositions are identical to certain kinds of facts, existentially generalized ones, where what is being generalized over are languages, words, speakers and contexts of utterance. In Section 3, I have argued that the identificatory claim presented in Section 2 is metaphysically obscure, and that it fails to satisfy the explanatory demands placed on propositions. As a way of rescuing King’s attempt to compromise between structured propositions and possible-worlds semantics, I have suggested, in Section 4, how one may avoid King’s metaphysical commitments by a ramsification of the theoretical term “proposition”, i.e. a second-order claim placing restrictions on what entities may count as propositions. King’s own candidate ersatz propositions are the logical forms of sentences, in which the words have been replaced by their semantic values relative to context. Identifying these entities with surrogate propositions makes no appeal to facts and is metaphysically neutral.

403 References King, Jeffrey and Jason Stanley 2005. Semantics, Pragmatics, and Semantic Content. In: Zoltán Gendler Szabó (ed.), Semantics versus Pragmatics. Oxford: Clarendon Press, 111-164. King, Jeffrey 2007. The Nature and Structure of Content. New York & Oxford: Oxford University Press. King, Jeffrey 2009. Questions of Unity. Proceedings of the Aristotelian Society 109, 257-277. Lewis, David 1970. How to define theoretical terms. The Journal of Philosophy 67, 427-446. May, Robert 1985/1997. Logical Form as a Level of Linguistic Representation. In: Peter Ludlow (ed.), Readings in the Philosophy of Language. Cambridge, Mass.: MIT Press, 281-315. McGlone, Michael forthcoming. Propositional Structure and Truth Conditions. http://www.acsu.buffalo.edu/~mmcglone/Propositional Structure and Truth Conditions.pdf Moltmann, Frederike 2003. Propositional Attitudes Without Propositions. Synthese 135, 77-118. Mulligan, Keith and Fabrice Correia 2007. In: Edward Zalta (ed.), The Stanford Encyclopedia of Philosophy (Fall 2008 Edition). http://plato.stanford.edu/ archives/fall2008/entries/facts Ramsey, Frank P. and George E. Moore 1927. Facts and Propositions. Proceedings of the Aristotelian Society 7, 153-206. Ramsey, Frank P. 1929/1931. Theories. In: Richard Braithwaite (ed.), The Foundations of Mathematics and Other Essays. London: Routledge and Kegan Paul. Russell, Bertrand 1903. Proper Names, Adjectives and Verbs. In: The Principles of Mathematics. Cambridge: Cambridge University Press, 42-52. Russell, Bertrand 1918/2007. The Philosophy of Logical Atomism. In: Robert Marsh (ed.), Logic and Knowledge: Essays 1901-1950 (2nd ed.). London: George Allen and Unwin, 175-283. Salmon, Nathan 1986. Frege’s Puzzle. Cambridge, Mass.: MIT Press. Soames, Scott 2010. What is Meaning? Princeton: Princeton University Press. Speaks, Jeff 2009. Facts, Properties, and the Nature of the Proposition. Manuscript. http://www.nd.edu/~jspeaks/papers/facts-properties-propositions.pdf

Alexander Miller University of Birmingham [email protected]

Judgement-Dependence, Tacit Knowledge and Linguistic Understanding* Abstract: In his widely discussed 1989 paper “Wittgenstein’s Rule-Following Considerations and the Central Project of Theoretical Linguistics”, Crispin Wright argued that a judgement-dependent conception of meaning may be in some tension with the project of giving a cognitive-psychological explanation of speakers’ abilities to understand novel utterances. In this paper, by broadening Wright’s bipartite distinction between judgement-independent and judgement-dependent properties into a tripartite distinction between judgement-independent, strongly judgement-dependent and weakly judgement-dependent properties, I explore the possibility that there is a version of the idea that meaning is judgement-dependent that can be squared with a cognitive-psychological explanation of speakers’ abilities to understand previously unheard sentences. I explore the consequences for this idea of the so-called “Doctrine of Essential Linguistic Structure”. Key words: judgement-dependence, rule-following, semantic creativity, tacit knowledge, Wittgenstein, Crispin Wright

0. Introduction In this paper, I will return to the main issue broached in Crispin Wright’s (1989a). In that paper, Wright outlines what he takes to be a fundamental commitment of the “Central Project of Theoretical Linguistics”:

*

I’m grateful to audiences in Oslo, Lodz and Vienna for useful discussion. For very helpful comments and discussion I’m grateful to Helen Beebee, Paul Broadbent, Florian Demont, Martin Kusch, Nicholas Melville, Ali Saboohi and Crispin Wright.

406 What is characteristic of the [Central] Project [of theoretical linguistics] is the thought that … it is appropriate to seek an explanation [of our ability to understand novel utterances] in, broadly, cognitive-psychological terms: it is an ability which we have because we are appropriately related to a finite body of information which may be inferentially manipulated in such a way as to entail, for each novel string on which we can successfully exercise our ‘linguistic-creative’ power, appropriate theorems concerning its grammaticalness and content. (Wright 1989a: 170-171)

Wright then identifies this commitment as one that is put under pressure by the later Wittgenstein’s rule-following considerations: [W]hether Wittgenstein’s later discussions should be seen as issuing a challenge to the factuality of meaning and rule-following, they undeniably feature a recurrent, explicit discomfort with a certain conception of the autonomy of rules, the image of a rule as a rail laid to infinity, tracing out a proper course for a practice quite independently of any judgement of the practitioners. [I]t is hard to avoid finding this conception of autonomy in the picture of language and linguistic competence which drives the Central Project: the picture of language as a kind of syntactico-semantic mechanism, our largely unconscious knowledge of which enables us to compute the content which, independently of any response of ours, it bestows on each ingredient sentence. [T]he mechanism does the generating, and the competent adult merely keeps track of what (and how) it generates. (Wright 1989a: 178-179)

Again: [T]he principal negative point of Wittgenstein’s discussion of rulefollowing was precisely that the ability to make acceptable rule-informed judgements allows of no coherent construal as a tracking ability, in the fashion which Platonism requires. (Wright 1989a: 210)

Does this mean that if we accept the anti-Platonist upshot of the rulefollowing considerations we have to give up on the Central Project?1 1

Although the main focus of Wright (1989a) is on theoretical linguistics, and specifically the work of the early Chomsky, it is clear from Wright’s other writings that he thinks that there is potential for a similar clash between the rulefollowing considerations and the idea that tacit knowledge of a semantic theory

407 Possibly, but also possibly not: Wright floats the idea that we can perhaps adjust our conception of the Central Project so that it can survive the rejection of the idea that optimal semantic judgements track the facts about the meanings of novel utterances: [T]he cognitive-psychological project is properly seen as directed not at the description of the conditions for a certain kind of tracking accomplishment – that conception has to fall with Platonism – but at the detailed elaboration of the C-conditions whose realisation ensures that a subject’s judgement of the content of a particular novel utterance will be best. (Wright 1989a: 212)

Whether or not this allows space for the Central Project then depends on the answer to the following question: [Will] the … notion of best opinion [in the judgement-dependent conception of meaning] contain components whose proper description will require – or, less, allow – invoking the apparatus of theoretical linguistics? (Wright 1989a: 213)

Wright’s thought – that the Central Project might survive the repudiation of a tracking epistemology for knowledge of meaning and in some way figure in the outline of what makes for optimality in the case of semantic judgements – is certainly worth exploring, although linguists may wonder how well it captures what is central to their day-to-day theoretical enterprise. In this paper, however, I want to explore the possibility of a more conservative reconciliation of the Central Project and the rule-following considerations: specifically, whether there is a way of combining the judgement-dependent conception of meaning with the idea that optimal semantic judgement tracks the facts about meaning. In other words, I want to explore the possibility of a “compatibilist” judgement-dependent position that doesn’t give up the idea that optimal semantic judgements can play a tracking role with respect to facts about meaning and which therefore leaves room for the kind of cognitive-

– whether construed along broadly Davidsonian or Dummettian lines – can play a role in explaining semantic creativity. See in particular Wright (1986).

408 psychological explanation of semantic creativity essayed by the Central Project.2 1. The defeat of the Missing-Explanation Argument In order to open up the possibility of this sort of “compatibilist” position, I will look first at the demise of Mark Johnston’s “Missing-Explanation Argument” (see e.g. Johnston 1993). This argument attempts to show that a judgement-dependent or more broadly response-dependent account of e.g. redness is inconsistent with the idea that redness can appear in empirical explanations of optimal colour judgement.3 In other words that (a) It is a priori that: x is red if and only if x is disposed to look red to normal perceivers in standard conditions is incompatible with (b) x is disposed to look red to normal observers in standard conditions because x is red where the “because” is that of empirical explanation. Johnston’s argument proceeds via the following substitution principle: substitution of a priori equivalents preserves “possible explanatoriness”. So, for example, given (i) It is a priori that: S* if and only if S 2

3

This paper thus develops (and hopefully improves upon) a suggestion I ventured in Miller (2007a). There are important differences between the accounts of judgement- and response-dependence developed by Wright, Pettit and Johnston: Wright, for example would not accept (a) as an adequate formulation of a responsedependent account of colour. However, these differences are generally not germane to the central line of argument of this paper, so I am not fastidious about matters that would call for more precision in a different context. (For some of the relevant detail, see the appendix to Chapter 3 of Wright 1992.)

409 we could not also have (ii) S because S* where the “because” is that of empirical explanation. (Since substituting into (ii) on the basis of the a priori equivalence asserted in (i) yields the explanatory solecism (iii) S because S it follows from Johnston’s substitution principle that (ii) does not record a possible empirical explanation either: the empirical explanation (ii) “goes missing” in the presence of the a priori equivalence expressed by (i)). Similarly, it appears that the kind of response-dependent view of colour expressed by (a) rules out the possibility of an empirical explanation of an object’s appearing red in terms of its redness. Given the a priori equivalence asserted by (a), we can substitute into (b) to get the explanatory solecism (c) x is disposed to look red to normal perceivers in standard conditions because x is disposed to look red to normal perceivers in standard conditions. By Johnstone’s substitution principle this shows that (b) too is an explanatory solecism, since substitution of a priori equivalents preserves possible explanatoriness. To the extent that we don’t wish to give up on the possibility of redness figuring in empirical explanations, we are pressured to give up a response-dependent view of red, at least when construed as a descriptive thesis about our ordinary concept of the colour. Peter Menzies and Philip Pettit undermine the Missing-Explanation Argument by showing that when (a) and (b) are properly understood, Johnston’s substitution principle cannot be applied:

410 The first claim [(a)] says that there is an a priori linkage between possession of the disposition by something and it’s being red: what it is to be red, as it were, inherently involves the realization of that disposition. The second claim [(b)] says, not that the object possesses the disposition because it is red, but that when the disposition is manifested – when something looks red to normal observers in normal conditions – it is manifested because the object is red. (Menzies and Pettit 1993: 106)

Menzies and Pettit thus argue that (a) should be read as (d) It is a priori that: (if normality reigns, then x looks red) if and only if (x is red) while (b) should be read as (e) If normality reigns and x looks red, then ((x looks red) because (x is red)). Given the form of these statements, there is simply no way to use Johnstone’s substitution principle to rule out the possibility of (e)’s recording a genuine empirical explanation. Independently, Crispin Wright suggests a similar move in defence of a judgement-dependent account of colour: [A judgement-dependent account of colour is] quite consistent with the belief that, as it happens, there are interesting physical characteristics in common among the objects which best opinions determine as, say, red; characteristics which are involved in the aetiology of, inter alia, those very opinions, and which, by – as it happens, being common to and distinctive of red things, have a case to be regarded as physically constitutive of redness. (Wright 1992: 132)

The judgement-dependent view of colour holds that: (f) It is a priori that: if C, then x is red if and only if a suitable subject S judges that x is red

411 where the specification of the conditions C and suitable subject S satisfy the various conditions laid down by Wright (see the appendix to Chapter 3 of Wright 1992). Wright’s point in the passage above is that (f) is consistent with (g) A suitable subject S in conditions C judges that x is red because x is red. If best opinion does as a matter of fact hit on a causally efficacious kind we can read (g) along similar lines to (e) as (h) If conditions C obtain and S judges that x is red then ((S judges that x is red) because (x is red)), so that we have an explanation of S’s judgement that x is red in terms of the characteristics that are physically constitutive of redness. As Wright puts it: [(g)] can be saved for [a judgement-dependent account of redness] if it so happens that best opinion about [redness] hits on a causally efficacious kind. (Wright 1992: 132 n.38)

Importantly for my purposes, since the “because” in (h) is that of empirical explanation, we can retain the idea that optimal judgement plays a tracking role with respect to the facts about redness.4 2. Three grades of judgement dependence Wright’s distinction between judgement-independence and judgementdependence is usually thought of as a binary distinction. Where we can construct a provisional biconditional (1) If C then (S judges that x is P iff x is P) 4

See also Wright (1991: 42-43): “[T]he linguistic data in the vicinity are better accounted for by viewing the explananda not as items’ possession of certain … dispositions but as episodes of those dispositions’ manifestation”.

412 with conditions C and suitable subject S spelled out so that (1) is a priori, substantial, and satisfies Wright’s independence and extremal conditions, we can say that P is judgement-dependent and that C-conditioned judgement partially5 determines the extension of P. In cases where (1) is at most an a posteriori truth P is judgementindependent and C-conditioned judgement merely tracks the independently determined extension of P. I want to suggest that one moral of Section 1 above is that this binary distinction should be replaced by a tripartite distinction, as follows. Judgement-Independence: as before, (1) is at most a posteriori true, and best (C-conditioned) judgement merely tracks the independently determined facts about the instantiation of P. Optimal judgement plays merely an extension-reflecting role with respect to P. An example of a judgement-independent property would be squareness. Optimal judgement merely tracks the facts about squareness: the whole truth about the relation between best judgement and squareness is captured in claims like “It is because the box is square that suitable subjects in optimal conditions judge that the box is square”. So where “becauseE” is the “because” of empirical explanation we have (2) Suitable subjects S in optimal conditions C judge that x is square becauseE x is square. The second category in the new tripartite distinction is Strong Judgement-Dependence: in this case (1) is a priori and satisfies the other conditions on judgement-dependence, but the objects that optimal judgement deems to be P have no underlying physical characteristics in common that would be capable of explaining why suitable subjects in optimal conditions judge that such and such is P. An example of this sort of case is perhaps provided by Pettit’s example of U-ness and the Sloane Rangers 5

See Wright (1992: 119-120). I omit the qualification in what follows.

413 (Pettit 1991: 611-615): the whole truth about the relation between best opinion and U-ness is captured in the claim (3) x is U becauseC suitable subjects in optimal conditions judge that x is U (where “becauseC” is the “because” of conceptual determination). The third (and for our present purposes most interesting) category is Weak Judgement-Dependence: in this case (1) is a priori and satisfies the other conditions on judgement-dependence, and the objects that optimal judgement deems to be P do have underlying physical characteristics in common that could figure in explanations of why suitable subjects in optimal conditions judge that such and such is P. So these are cases where we have judgement-dependence, but in addition, due to the failure of the Missing Explanation Argument, empirical explanations along the lines of (e), (g) and (h) in Section 1 above. Here, best opinions play an extension-determining role and an extension-reflecting role, insofar as the extension-determining role is consistent with facts about redness figuring in empirical explanations of optimal colour judgements. One difference between this case and the case of judgement-independence is that were it to turn out that there is a chaotic physical heterogeneity among red things, we would not adopt an error theory with respect to ascriptions of redness, whereas in the case of squareness chaotic heterogeneity at the geometrical level among objects deemed by visually determined optimal judgement to be square would force us in the direction of an error theory.6 So in this kind of case we can have both

6

So although in both cases we have genuine empirical explanations of optimal judgements, it is only in the case of judgement-independence that the availability of such an explanation is demanded by the semantics of the relevant predicate. See Wright (1992: 132). This is one reason why we can have optical illusions of shapes but not colours. See Wright (1989a: 198).

414 (4) Suitable subjects in optimal conditions judge that x is red becauseE x is red and (5) x is red becauseC suitable subjects in optimal conditions judge that x is red.7

7

Although the possibility of combining (4) and (5) in the way envisaged is suggested by some remarks of Wright’s, in discussion he has voiced some concerns about the plausibility of the manoeuvre. Suppose that as a matter of fact the physical characteristics common to and distinctive of the items deemed red by optimal colour judgement are P1, …, Pn. Then, what we appear to have left room for is the idea that suitable subjects in ideal conditions judge that x is red becauseE x is P1 or…or Pn. In this sort of situation, we can say that optimal judgement tracks some physical properties that are correlated with redness, or maybe even that optimal judgement tracks the extension of redness, but we seem to have lost the idea that optimal judgement tracks redness itself, and it is the analogue of the latter that we would require in the case of meaning in order for the central argument of this paper to be sustained. In response to this, I think we can invoke the notion of “program explanation”, and the concomitant distinction between causal relevance and causal efficacy, developed in a number of papers by Frank Jackson and Philip Pettit (see e.g. Jackson and Pettit 1990). To see this, suppose that conditions C in fact obtain and that a suitable subject S judges that a particular object y is red. Suppose too that in this case physical characteristic P1 from the complex disjunction mentioned above is causally efficacious (in the Jackson and Pettit sense) in the production of S’s judgement. What do we need for the truth of: (1) y’s redness is causally relevant to S’s judging that y is red? What we need is something like: in possible worlds close to the actual world in which y is red but not P1, S still forms the judgement that y is red when he views it in optimal conditions. This doesn’t strike me as implausible: in such worlds, y, although it doesn’t have P1, will have one of P2 … Pn, and since the causal laws will remain the same, S will still judge that y is red. Since we have (1) we can have a perfectly respectable version of (2) If (conditions C obtain and S judges that x is red) then (S judges that x is red because x is red)

415 3. Knowledge of meaning and weak judgement-dependence It is this third possibility that opens up the prospect of a “compatibilist” position. According to Wright Platonism, is, precisely, the view that the correctness of a rule-informed judgement is a matter quite independent of any opinion of ours. (Wright 1989a: 210)

and on a Platonist view of meaning [T]he mechanism does the generating, and the competent adult merely keeps track of what (and how) it generates. (Wright 1989a: 178-179)

Since Wright takes the rule-following considerations to undermine any view of meaning as judgement-independent, he also takes them to undermine any view on which optimal judgements track the facts about meaning: hence the need for the (arguably implausible) kind of rapprochement between the rule-following considerations and the Central Project ventured by Wright mentioned in the Introduction above. However, the tripartite distinction outlined in Section 2 above allows us to explore a more conservative approach: rather than looking for a way to square the cognitive-psychological project of theoretical linguistics with the repudiation of a tracking epistemology, we note that weak judgement-dependence (in virtue of the likes of (4)) still allows a notion of tracking to be in play. So we can have a tracking epistemology even though we reject Platonism (the idea that meaning is judgementindependent). On this kind of view we would have (as a modification of the passage from Wright quoted immediately above): The mechanism does the generating, and the competent adult keeps track (but doesn’t merely keep track) of what (and how) it generates. so we can say that courtesy of the empirical facts optimal judgements do track the facts about redness, in the sense that the latter are causally relevant (in the Jackson and Pettit sense) to the former (though not, of course, causally efficacious). So despite the fact that redness is assumed to be judgementdependent, there is still a sense in which optimal judgement tracks redness itself.

416 What are the prospects for a weak judgement-dependent view of meaning? In order to explore this question I’ll consider a toy language L that consists of 5 names (“Amber”, “Archie”, “Ruby”, “Dino” and “Trudy”) and 5 predicates (“black”, “white”, “tortoiseshell”, “canine”, and “feline”) and which (since there are no sentential connectives) has only 25 well-formed sentences. Now suppose that as a matter of empirical fact a speaker Jones, through training and exposure, acquires competence with “Amber is feline”, “Archie is canine”, “Ruby is tortoiseshell”, “Dino is white”, and “Trudy is black”, and on that basis alone acquires competence with the remaining 20 sentences of L. Let us suppose that a formal theory of meaning for a language is a truth-theory in the usual sense and also that such theories should be constructed in accordance with what Martin Davies calls the “Mirror Constraint”: Mirror Constraint: If, and only if, the operative states implicated in the causal explanation of a speaker’s beliefs about the meanings of sentences S1, …, Sn are jointly sufficient for a causal explanation of his belief about the meaning of S, should the semantic resources sufficient for a canonical derivation of truth-conditions specifications for S1, …, Sn be sufficient for the canonical derivation of a truthcondition specification for S.8 The empirical fact mentioned above about Jones’s acquisition of competence with the sentences of L suggests that the causal states implicated in his beliefs concerning the meanings of “Amber is feline”, “Archie is canine”, “Ruby is tortoiseshell”, “Dino is white”, and “Trudy is black” are jointly sufficient for a causal explanation of his beliefs about the meanings of all of the remaining sentences of L. It follows from the Mirror Constraint that in constructing a theory of meaning for L, we ought to ensure that the semantic resources sufficient for deriving specifications of the truth-conditions of the five sentences just 8

A fuller outline of the Mirror Constraint would need to mention facts about semantic decay and revision in addition to facts about acquisition. For details, see Miller (1997).

417 mentioned are also sufficient for deriving specifications of the truthconditions of the remainder of L’s sentences. In fact, the Mirror Constraint is satisfied by the following very simple theory T: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.

“Amber” denotes Amber “Archie” denotes Archie “Ruby” denotes Ruby “Dino” denotes Dino “Trudy” denotes Trudy “black” applies to an object o iff o is black “white” applies to an object o iff o is white “tortoiseshell” applies to an object o iff o is tortoiseshell “canine” applies to an object o iff o is canine “feline” applies to an object o iff o is feline A subject-predicate sentence is true iff the predicate applies to the object denoted by the name

Where the Mirror Constraint is satisfied by a theory of meaning with respect to a given speaker we can think of tacit knowledge of that theory as a matter of possessing causal-informational states corresponding to the axioms for each name and each predicate of the language, and this ascription of tacit knowledge is capable of figuring in a cognitivepsychological explanation of a speaker’s ability to know the meaning of a novel utterance. Consider Jones, and his ability to understand the novel sentence “Amber is black” on the basis of his prior training and exposure. Ascribing tacit knowledge of T to Jones enables us to explain how he is able to understand “Amber is black” in the sense that the causal route from the causal informational states corresponding to the name and the predicate are mirrored by the derivational route in T from the relevant axioms to the relevant truth-condition specifying theorem: using axioms 1, 6 and 11 we can derive the theorem 12. “Amber is black” is true iff Amber is black More generally, ascription of tacit knowledge of T is ascription of a linguistic ability the causal structure of which matches the derivational

418 structure of T. On this picture, Jones tacitly knows T in the sense that he has a set of causal informational subdoxastic states corresponding to the names and predicates of L. Collectively, these states are potentially explanatory of exercises of his ability to understand novel utterances: the causal route from tacit knowledge of the meanings of constituents of a novel utterance to tacit knowledge of its truth condition is mirrored by the derivational route in the theory from axioms to theorems. The question for us is: can this picture of tacit knowledge and linguistic competence sit alongside the idea that meaning is weakly judgement-dependent in the sense of Section 2? In order to pursue the possibility of an analogy between colour and meaning we’ll look at a slightly different version of the point about colour made by Wright (see Section 2). In the case of linguistic understanding we are dealing with a single speaker, so in considering the possibility of an analogy with the colour case we shall consider not a range of different objects with the same colour, but a single object that is the subject of colour judgement over a period of time. Suppose, then, that throughout a period of time t1 … tn an object x is judged to be red by suitable subjects in optimal conditions. Suppose also, that as a matter of fact these judgements are the causal upshot of a determinate range of underlying physical states P1 … Pn. As outlined in Section 1, we can give an empirical explanation of a suitable subject’s judgement that x is red in terms of the redness of the object, notwithstanding the fact that redness is – we are supposing – judgement-dependent. However, if we subsequently discovered that the underlying physical states of x implicated in the etiology of optimal colour judgement were wildly heterogeneous we would not revise our judgement that the object was red throughout t1 … tn: in other words, we would not retrospectively adopt an error theory with respect to our previous judgements about the redness of the object. Now suppose that throughout a period of time t1 … tm Jones – our speaker in the example above – is stably disposed to judge that “Amber is black” is true iff Amber is black. Let’s call the casually structured linguistic ability of Jones mentioned above A. Suppose that as a matter of empirical fact each of Jones’s optimal judgements about the meaning of the sentence “Amber is black” involves the exercise of A. In this

419 situation, parallel to the colour case just considered, we can give an empirical explanation of Jones’s optimal judgements in terms of his tacit knowledge of theory T. Since one of the theorems of T is given by 12. we can have 13. Jones judges (in optimal conditions) that “Amber is black” is true iff Amber is black becauseE “Amber is black” is true iff Amber is black This allows us to view Jones’s optimal judgement about the meaning of the sentence as tracking the content of the sentence as he understands it. Our question is whether this is consistent with a judgementdependent account of meaning according to which 14. “Amber is black” is true iff Amber is black becauseC Jones judges (in optimal conditions) that “Amber is black” is true iff Amber is black If we subsequently found that Jones’s optimal judgements about the meaning of “Amber is black” were actually the causal upshots of a set of linguistic abilities with chaotically heterogeneous underlying causal structures, would we revise our judgement about what Jones meant by the sentence throughout the relevant time period? In other words, would we retrospectively adopt an error theory with respect to Jones’s optimal semantic judgements? If not, parallel to the colour case, we could thus have a genuine empirical dependence of optimal judgements on facts about meaning and understanding without this being required by the semantics of ascriptions of meaning. If the parallel with colour can be sustained, via accommodating empirical claims like 13. alongside conceptual claims like 14, we would have the resources for a judgementdependent view of meaning capable of co-existing with an orthodox cognitive-psychological explanation of semantic creativity.

420 4. The challenge from the Doctrine of Essential Linguistic Structure9 But can the parallel with colour judgement be sustained? On the picture outlined above, the semantic structure of a language is the structure projected into the language by an adequate theory of meaning for the language. If the Mirror Constraint is imposed this is effectively the causal structure of a competent speaker’s overall capacity to correctly use the sentences of the language. As we saw, the analogy with colour appears to imply that just as the ascription of colour to a single object over time would survive the discovery of wild cross-temporal heterogeneity among the physical causes of optimal colour judgement, the ascription of truth-conditions to the sentences of L made in theory of meaning T would survive the discovery that the causal structure of Jones’s ability varied significantly throughout the relevant time period. In the limiting case, at some times in the period Jones could have the same overall ability as above but consisting of 25 causally independent sub-abilities, one for each sentence of L. Accepting this as a genuine possibility would appear to imply that the semantic structure of L is one of its contingent features, and we would have a clash with the Doctrine of Essential Linguistic Structure (DELS, see e.g. Fricker 1982), according to which the semantic structure of a language is one of its necessary features. In the face of this, one option that could be pursued in defence of the “compatibilist” position suggested above would be simply to deny the DELS. Suppose we think of sentential understanding as a kind of recognitional ability, perhaps, as Wright puts it “an overall ability to suit one's use of the expression to the obtaining of factors which can be appreciated by oneself and others to render one's use apt” (Wright 1989c: 54). Then we might note that, in general, the causal structure of a recognitional ability is not one of its essential features. As Mark Sainsbury puts it: It is not generally true that the causal structure of a recognitional ability is essential to it. (Sainsbury 1980: 130) 9

In this section I hopefully go some way towards honouring a promissory note I issued in Miller (1997: 173, n.13).

421 As an instance of a recognitional ability with a contingent causal structure, Sainsbury gives the example of chicken-sexing: Let us suppose that there are chicken sexers who associate young birds with a number from 0 to 1, taken to represent the probability of that bird being male, and that they cannot articulate the features which cue their response. They might share a common ability: over a given range of inputs (these being visual or tactile properties) they assign the same number. One learnt slowly, requiring at least one lesson from an old hand for each type of input, and the sub-abilities which make up the overall ability are causally independent: it is causally possible that the ability to associate any one set of observable properties with a number fail, yet the other property-number associations remain untouched. Another sexer learnt quickly, and extrapolated to associating a number with hitherto unperceived combinations of properties along lines reflected in the theory of probability, and there are causal dependencies among the sub-abilities: there are pairs of sets of properties such that it is causally impossible, as things are with him, that his ability to associate the other set of the pair with the appropriate number survive. Here we have a clear case in which the same ability is realized in distinct causal structures. (Sainsbury1980: 131)

Although Sainsbury’s remarks about the chicken-sexing example seem plausible, and although, as Sainsbury suggests, it may well be the case that in general the causal structure of a recognitional ability is not one of its essential features, there are reasons to think that for a proponent of a judgement-dependent view of meaning, the case of linguistic understanding is an exception to the rule. This is suggested by the reflection that a judgement-dependent account of meaning is an instance of a general stance in the philosophy of language and mind that Alex Byrne (1998) has dubbed “Interpretivism”: It is an a priori truth that there is no gap between our best judgements of a subject’s beliefs and desires and the truth about the subject's beliefs and desires. (Byrne 1998: 199)

Indeed, Byrne goes so far as to say Interpretivism in general is the thesis that mental content is judgementdependent: the facts about what propositional attitudes someone has are

422 exactly captured by the (potential) judgements of some Ideal Interpreter (or Interpreters). (Byrne 1998: 205)

and we can take this to extend to matters of what meaning or content a speaker attaches to an expression of their public language or mental repertoire. He also explicitly characterizes Wright’s judgementdependent account of intention – held by Wright to be analogous to the cases of meaning and understanding – as a first-personal form of Interpretivism “where the Interpreter is taken to be the subject of interpretation” (Byrne 1998: 205). Suppose – as seems reasonable – that Byrne’s classification is correct. This is relevant to our current concerns given Elizabeth Fricker’s claim that: Insofar as interpretation is determinate, the primary feature ensuring this is the fact that the assignment of a meaning to one sentence constrains and is constrained by the assignments made to other sentences ... Some principle for discerning structure in sentences is essential to interpreting a language, and the correctness of an interpretation cannot be considered to be independent of facts about their structure. (Fricker 1982: 59)

Fricker’s point is that when we are in the process of interpreting the utterances of speakers of a language, it is in virtue of the fact that a sentence contains expressions which systematically contribute to the meanings of those and other sentences that the interpretation of any given sentence is constrained: and without recurring contributions of this nature, at the level of sentences, determinate interpretation will not be possible (see Miller 2007b: 267 for more detail). This apparently undermines the idea ventured above, that in the limiting case L’s sentences could have the meanings assigned by the truth-theorems of T, but in virtue of a speaker’s competence consisting in a set of 25 causally independent sub-abilities, in effect have no semantic structure at all. It also suggests that the putative possibility – necessary for the analogy between colour and meaning to be sustained – that the meanings attached by a speaker to a set of sentences could “float free” of the underlying causal structure of his overall linguistic competence, is in fact not a genuine possibility at all.

423 These reflections suggest that Sainsbury’s comments on the general case notwithstanding, a judgement-dependent account of meaning does not sit well with the rejection of the Doctrine of Essential Linguistic Structure. But the rejection of the Doctrine appears to be a requirement for sustaining the idea that meaning is, like colour, plausibly viewed as an instance of Weak Judgement-Dependence.10 In the remainder of this section, I will examine whether there is a way in which the Weak Judgement-Dependent view of meaning can instead seek to accommodate the Doctrine of Essential Linguistic Structure. A hint as to how such an accommodation might be achieved can be gleaned from an exchange between Colin McGinn (1984) and Crispin Wright (1989b) on the proper reading of Wittgenstein’s view of the relationship between the mental and the physical (interestingly, in the context of a discussion of Wittgenstein’s antipathy towards viewing understanding as a state of mind). In a notorious passage in Zettel, he writes: It is … perfectly possible that certain psychological phenomena cannot be investigated physiologically, because physiologically nothing corresponds to them. … Why should there not be a psychological regularity to which no physiological regularity corresponds? (Wittgenstein 1981: Section 609-610)

A consequence of this view, according to McGinn (1984: 113) is that there could be “physical events that have no physical explanation, viz. those that issue from psychological differences which correspond to no physiological differences”. He finds this consequence intolerable, and suggests that Wittgenstein needn’t have committed himself to it in claiming that use is the ultimate criterion for linguistic understanding. According to McGinn, if we think of use as the ultimate criterion for understanding, we can say that in the case of a man whose use of an expression suggested a genuine grasp of its meaning but who was 10

Fricker’s argument takes place in a context in which a non-reductionist view of meaning (“semanticalism” in Fricker’s terminology) has been set aside. It may be that this makes a material difference to the argument about interpretation given that the judgement-dependent view of meaning is self-consciously nonreductionist.

424 subsequently discovered to have a head full of sawdust, we would not withdraw our judgement that he was an understander. Crucially, McGinn argues, this does not commit us to the claim “that someone really could have a head full of sawdust or a chaotic nervous system and yet understand words” (McGinn 1984: 114). The reason is that a.

If it turned out that p, it would be rational (or even true) to say such-and-such

does not imply b.

It is metaphysically possible that p

since in uttering a. we might be considering what would be the case on an impossible supposition.11 In the case at hand [W]hen we admit that if it were to turn out that other people had heads full of sawdust we would still say that they understood, we are not thereby committed to allowing that this is a real metaphysical possibility. (McGinn1984: 114)

Commenting on this, Wright suggests that McGinn has not taken the correct measure of Wittgenstein’s view. According to Wright, Wittgenstein’s remarks in the passages from Zettel do not imply that it is genuinely metaphysically possible for someone with a head full of sawdust to be an understander. Rather: What follows is only that, for all the concept of understanding has to say about the matter, there could be behavioural, ergo physical events whose explanation was not to be found in the internal physical state of the behaving subject … Wittgenstein’s claim is one about what is compatible with our concepts of the psychological, not about what is possible tout court. (Wright 1989b: 153-154, Wright’s emphases)

11

Citing Kripke, McGinn notes that although “it might turn out that water is not H2O, … (given that water is H2O) it is necessary that water is H2O” (McGinn 1984: 114).

425 Let Jones* be a speaker with a grasp of the sentences of L but with a set of 25 causally independent sub-abilities, one corresponding to each sentence. Our worry was that the construal of meaning and understanding as Weakly Judgement-Dependent appeared to imply that Jones* is a genuine metaphysical possibility, a possibility apparently at odds with the Doctrine of Essential Linguistic Structure. However, the remarks from McGinn and Wright rehearsed above suggest that in adopting the Weak Judgement-Dependent view of meaning we do not have to admit that Jones* is possible. Rather, acceptance of Weak Judgement-Dependence commits us to the view that for all the concept of understanding has to say, Jones* is possible: and this falls short of accepting that Jones* is possible tout court. Alternatively, we can say that if it turned out that Jones had an ability with a Jones*-like structure, we would not revise our view of what meanings he attaches to the sentences of L. This does not commit us to viewing Jones* as a genuine metaphysical possibility.12 So, one can adopt the Weak JudgementDependent view of meaning and also accommodate the Doctrine of Essential Linguistic Structure.13 12

13

Note that this need not embroil us in any worries about counterfactuals with impossible antecedents. We can make the point in terms of what we would judge in the counterfactual situation in which we believe that a speaker has an ability with a Jones* like structure. In terms of the taxonomy in Footnote 13 below, the difference between judgement-independence and Weak1 Judgement-Dependence would be that in the case of the former, but not the latter, were we to believe that apparent instances of P were wildly heterogeneous at the relevant underlying or theoretical level, we would adopt an error theory with respect to P-judgement. The reflections in this section suggest that we may have to broaden the tripartite distinction outlined in Section 2 into a four-way distinction. Judgement-Independence: As before, optimal judgement-plays at most an extension-reflecting role. As a matter of conceptual necessity, there has to be a set of underlying (theoretically specified) characteristics common to and distinctive of P-things: if we discovered that P-things were in fact radically heterogeneous at the relevant underlying or theoretical level, we would embrace an error theory. However, which underlying or theoretical properties are identified with being P is an a posteriori matter. Strong Judgement-Dependence: As before, we have a substantially specified and a priori relationship between best judgement and fact. There is no

426 5. Conclusion We can venture the following tentative conclusion. If the judgementdependent account of meaning and the project of giving a cognitive psychological explanation of our capacity to understand novel utterances are both otherwise unobjectionable, since we don’t have to give up the idea that competent speakers are capable of achieving a tracking accomplishment in comprehending a novel utterance, if the Doctrine of Essential Linguistic Structure can be rejected or accommodated in the manner suggested, there is nothing to prevent us from yoking them together in a way that is consistent with the orthodox construal of the Central Project of Theoretical Linguistics.14 conceptually necessary requirement of homogeneity at the relevant underlying or theoretical level, and as a matter of fact there is no such homogeneity. Weak1 Judgement-Dependence: We have a substantially specified and a priori relationship between best judgement and fact. There is no conceptually necessary requirement of homogeneity at the relevant underlying or theoretical level. However, as a matter of metaphysical necessity there is homogeneity at the theoretical level, and which underlying properties are identified with being P is an a posteriori matter. Weak2 Judgement-Dependence: We have a substantially specified and a priori relationship between best judgement and fact. There is no conceptually necessary requirement of homogeneity at the relevant underlying or theoretical level, and it is not the case that as a matter of metaphysical necessity there is homogeneity at the theoretical level (although there is as a matter of empirical fact).

14

The thought adumbrated in this section is that by viewing meaning and understanding as an instance of Weak1 Judgement-Dependence we can view meaning as judgement-dependent whilst accommodating the Doctrine of Essential Linguistic Structure. It is an interesting question where colour should appear in this broader taxonomy. Another way of squaring the Weak Judgement-Dependent view with the DELS might involve distinguishing between stronger and weaker versions of the latter. According to the weaker version: if L is a language then necessarily (∃S)(S is a structure and L has S). According to the stronger version: if L is a language and L has structure S, then necessarily: L has structure S. Now, arguably the weaker version is sufficient to avoid indeterminacy at the level of sentence meaning: in

427 References Byrne, Alex 1998. Interpretivism. European Review of Philosophy 3, 199-223. Davies, Martin 1987. Tacit Knowledge and Semantic Theory: Can a 5% Difference Matter? Mind 96, 441-62. Evans, Gareth 1981. Semantic Theory and Tacit Knowledge. Reprinted in: Collected Papers. Oxford: Clarendon Press 1985. Fricker, Elizabeth 1982. Semantic Structure and Speakers’ Understanding. Proceedings of the Aristotelian Society 83, 49-66. Jackson, Frank and Philip Pettit 1990. Program Explanation: A General Perspective. Analysis 50, 107-117. Johnston, Mark 1993. Objectivity Refigured: Pragmatism Without Verification. In: Crispin Wright and John Haldane (eds.), Reality, Representation and Projection. Oxford: Oxford University Press, 85-130. McGinn, Colin 1984. Wittgenstein on Meaning. Oxford: Blackwell. Menzies, Peter and Philip Pettit 1993. Found: the Missing Explanation. Analysis 53, 100-109. Miller, Alexander 1997. Tacit Knowledge. In: Bob Hale and Crispin Wright (eds.), A Companion to the Philosophy of Language. Oxford: Blackwell, 146-174. Miller, Alexander 2007a. Critical Notice of Crispin Wright Rails to Infinity. International Journal of Philosophical Studies 15, 125-40. Miller, Alexander 2007b. Philosophy of Language (2nd edition). London: Routledge. Pettit, Philip 1991. Realism and Response-Dependence. Mind 100, 587-626. Sainsbury, Mark 1980. Understanding and Theories of Meaning. Proceedings of the Aristotelian Society 80, 127-144. Wittgenstein, Ludwig. 1981. Zettel. Oxford: Blackwell. Wright, Crispin 1986. Theories of Meaning and Speakers’ Knowledge. In: Realism, Meaning and Truth. Oxford: Blackwell, 204-238. order to avoid indeterminacy it is necessary that L have some structure or other. Since the weak DELS rules out the possibility of a language with no semantic structure, it rules out the possibility of the “limiting case” mentioned in the text. Can a Weak Judgement-Dependent view of meaning accept the weaker version of DELS? Reflection on the analogue in the colour case suggests that it can. The judgement-dependent view of colour can presumably accept that if x is red, then necessarily x is in some physical state or other. It needn’t accept that if red object x is in physical state P, then necessarily: it is in physical state P. So in the case of meaning, we could embrace the weak version of DELS and rule out the “limiting case” without having to say that a language has the particular structure that it has as a matter of metaphysical necessity (as in Weak1 Judgement-Dependence). I hope to pursue this line of thought in further work.

428 Wright, Crispin 1989a. Wittgenstein’s Rule-Following Considerations and the Central Project of Theoretical Linguistics. Reprinted in: Rails to Infinity. Cambridge: Harvard University Press 2001, 170-213. Wright, Crispin 1989b. Critical Notice of McGinn Wittgenstein on Meaning. Reprinted in: Rails to Infinity, 143-169. Wright, Crispin 1989c. Misconstruals Made Manifest. Midwest Studies in Philosophy 14, 48-67. Wright, Crispin 1991. Order of Determination, Response-Dependence and the Euthyphro Contrast (MS). Wright, Crispin 1992. Truth and Objectivity. Cambridge, Mass.: Harvard University Press.

Ulrich Reichard University of Durham [email protected]

Making Events Redundant: Adnominal Modification and Phases∗ Abstract: In the last two decades, Davidson’s event-argument hypothesis has become very popular in natural language semantics. This article questions that event-based analyses actually add something to our understanding of the respective phenomena: I argue that they already find their explanation in independently motivated grammatical assumptions and principles which apply to all kinds of modification. Apart from a short discussion of Davidson’s original arguments in favour of his hypothesis, I address Larson’s event-based account of the distinctions between stage-level vs. individual-level modification and adverbial vs. adjectival modification in the nominal domain. I argue that his analysis of the former reduces straightforwardly to the grammatical structure of the nominal phase. As for the latter, I provide reasons which motivate a re-description of the phenomenon in terms of sensitivity to descriptive content rather than events. I argue that, so described, the phenomenon can be explained in terms of interface conditions, given a phasal architecture of grammar. Key words: events, entailment, stage-level vs. individual-level predicates, adverbial vs. adjectival modification, ontological commitments, Davidson, Larson

0. Introduction Davidson’s (1967b) hypothesis that certain verbs have an additional argument-position for event-arguments has proven a very fruitful field of research in natural language semantics. The proposal has been extended to all kinds of verbs (Parsons 1990), to deverbal nominals, and even to expressions which, initially, do not seem to have much to do with events ∗

Many thanks to Wolfram Hinzen, Alex Drummond and Andrew Woodard for useful suggestions and conversations on the topic of this paper, and to Wolfram Hinzen for detailed comments on an earlier draft.

430 (Parsons 2000). This paper questions whether an analysis with the help of event-arguments does actually add much to our understanding of the respective phenomena by arguing that they are predicted given the pure grammatical structure of the relevant expressions. It further points out some problems which seem to be due to an event-based analysis rather than the explanada themselves. In the first section, some initial motivations for assuming eventarguments are shortly reviewed. The second section delineates in what sense many phenomena which have been taken to support the assumption of event-arguments seem to follow from independently motivated assumptions about grammatical structure-building and thus do not provide evidence for the event-argument hypothesis. The general account of modification developed there provides the background for the argument in Section 4. In Section 3, I turn to certain kinds of ambiguities concerning adnominal modification which have been analyzed with the help of event-arguments (Larson and Segal 1995; Larson 1998). I point out some reasons for preferring this analysis over its Montagovian competitor. However, in Section 4, I argue that Larson’s distinction between adjectival and adverbial modifiers should be generalized to a distinction in sensitivity to descriptive content in which event-arguments don’t play a role. With the help of recent theories of grammatical phases, the distinction, so described, can furthermore be explained in terms of interface conditions. In respect to Larson’s analysis of the stage/individual-level distinction (which goes back to Chierchia 1995), I claim that it follows directly from the organization of the nominal phase as described in Section 2. Section 5 concludes with some remarks about the relation between natural language, metaphysics and logic. 1. The event-argument hypothesis Davidson (1967b) famously argued that action verbs do not only have two argument positions, as traditionally assumed, but that they have a third argument position for an event-argument. For Davidson, the rationale for this move mainly consists initially in entailment relations between expressions which include adverbial modifiers. Consider (1-4):

431 (1) (2) (3) (4)

Shem boiled the soup at nine o’clock in London. Shem boiled the soup at nine o’clock. Shem boiled the soup in London. Shem boiled the soup.

(1) entails (2-4), (2) and (3) independently entail (4), but the conjunction of (2) and (3) does not entail (1): We can imagine that Shem boiled the soup at six o’clock in London and again at nine a clock in Leeds. In this case both (2) and (3) are true, but (1) is not. Traditional first order predicate logic cannot explain this phenomenon. However, there is a phenomenon reminiscent of (1-4) which first order predicate logic seems to account for. Consider (5-8) (5) (6) (7) (8)

There is a blue house in London. There is a blue house. There is a house in London. There is a house.

The entailment relations between these sentences exactly parallel those in (1-4): (5) entails (6-8), (6) and (7) entail (8), but their conjunction does not entail (5). Grammatically speaking, blue and in London modify house in these sentences. However, the simplest way of capturing the facts about entailment in first order predicate logic is to abstain from these grammatical facts and to assume that blue, house and in London are all predicates of the same variable which is then existentially closed. If, in addition, all the predicates are combined by conjunction, the implication is a mere matter of conjunction reduction. This assumption also provides a reason for why the conjunction of (6) and (7) does not entail (5): The variables are in the scope of different existential quantifiers.1 The logical form of (5) would thus be (9): (9) ∃x [blue(x) a house(x) a in London(x)]

1

A problem of this approach is that there are a number of adnominal modifiers for which some or all of the implications mentioned do not hold (see Section 3).

432 Davidson suggested that the same strategy could be applied to adverbial modification: If we treat the adverbial modifiers at nine o’clock and in London as somehow on a pair with the main sentence Shem boiled the soup in that we assume that all of them are predicates of the same variable and are combined with the help of conjunction, we could explain the implication with first order predicate logic and we could account for the intuition that the two cases are exactly parallel. In (1), however, the variables cannot possibly range over objects traditionally perceived, since there does not seem to be a way in which Shem boiled the soup describes an object. We rather seem to be inclined to say that this is a certain kind of event. Furthermore, we can say about events that they happen at a certain time, for example at 9 o’clock, and that they happen at some place, for example in London. Thus we could theorize that, despite grammatical differences, Shem boiled the soup, at nine o’clock and in London are all predicates of a variable ranging over events. And we can think of sentences (or at least certain kinds of sentences) as existential quantifications over events. The logical form of (1) would thus be (10). (10) Ee [boil(Shem, the soup, e) a in London(e) a at 9 o’clock(e)] In prose: There is an event e, such that e is a boiling of the soup by Shem, e happened in London and e happened at 9 o’clock. There is some further motivation for splitting up the three-place predicate Shem boiled the soup, since there are implications which our current treatment does not capture: (4) entails both (11) and (12). (11) Shem boiled. (12) The soup boiled. It has been proposed that this implication can be captured if we make use of thematic roles, a notion originally developed in syntactic theory (Baker 1988). The logical form of (4) is then (13): (13) Ee [boiling(e) a AGENT(e, Shem) a THEME(e, the soup)]

433 Read: ‘there is an event e such that e is a cooking and the Agent of e is Shem and the Theme of e is the soup’. Agent and Theme are then taken to be two possible roles which participants of an event can play in an event. This analysis captures our intuition that, even though Shem in (11) and the soup in (12) both grammatically occupy the subject position, they play different roles in the respective sentences: Shem is the person who boils something in (11), in most cases he will keep his original temperature (as long as we are not speaking metaphorically, at least), but the soup in (12) needs to be very hot in order for this sentence to be true (again at least as long as we are not in a fictional context in which soups boil something else and thus take the Agent role). Splitting up the arguments of a verb into arguments of thematic predicates which are united by an event-argument has been dubbed a ‘neo-Davidsonian’ analysis by Dowty (1989). The idea goes back to a proposal by Castaneda, who commented on Davidson’s paper when it was first delivered. Davidson retains Quine’s (1953) idea that an analysis which involves existential quantification creates an ontological commitment to whatever is quantified over. For him, the event-argument hypothesis thus has metaphysical implications. In retrospect, he writes: ‘If an ontology of events were the only way to give a satisfactory semantic analysis of these sentences and the relations between such sentences, it would, in my opinion, provide a very strong argument for the claim that there are events’ (Davidson 1993: 42). Davidson thinks of events as analogous to objects. Events are thus taken to be spatio-temporal particulars (Davidson 1969, 1970, 1971, 1985a). In his earlier work, he nonetheless takes events to be a different kind of entity than objects (Davidson 1967a), but in later work, he adopts Quine’s (1960) conjecture that the difference is gradual rather than absolute (Davidson 1985b). In the semantic literature, the event-argument hypothesis is seen as an empirical hypothesis in the sense that it has to be justified in terms of its explanatory power (Parsons 1990). In the introduction to a volume on event-semantics, Rothstein (1998: 2-3), for example, admits that ‘since the event arguments never appear as such in the sentences, the evidence [for their existence] can only be indirect. [...] The strongest evidence would be a construction that cannot be reasonably explained without

434 positing an event argument, and a weaker argument would be a construction that can be analysed much more simply if we posit an event argument.’ The implication relations and the parallelism between adjectival and adverbial modification which have been delineated in the first section are two of Davidson’s three original linguistic arguments in favour of the existence of event-arguments in language.2 The third one is that we actually seem to refer to events with the help of nominalisations. Thus, in addition to sentences like (14), where for Davidson there is hidden quantification over events, we also have expressions like (15) and (16). Natural languages, thus, seem to enable us to refer to events in a similar way as they allow reference to objects. (14) Brutus murdered Caesar. (15) Brutus’ murdering Caesar (16) Brutus’ murder of Caesar Further arguments have been added by a number of linguists and philosophers. For example, (14) could be followed by This happened in 44 BCE. It has been argued that only events can happen – the anaphoric reference of this must thus be to an event. Moreover, a number of aspectual phenomena like the distinction between telicity and atelicity have been analyzed as involving event-arguments. In addition, Schein (1993) argues that it is not possible to analyze the relation between distributivity and cumulativity in certain sentences without neoDavidsonian thematic separation and an event-argument which ensures the unity of the event. Finally, there have been attempts to analyze some phenomena of adnominal modification with the help of event-arguments (Larson 1998). For many of these phenomena, alternative explanations have been developed in a Montagovian framework where events are not taken to be primitives but properties of times. If it is indeed correct that these 2

Davidson also provides metaphysical arguments for the existence of events as spatio-temporal entities, which I shall ignore here. He stresses that his main argument for the existence of events always remained the case from language (Davidson 1985b).

435 phenomena can be modelled without the assumption of an eventargument, the assumption of the event-argument is justified to the extent to which the ontological costs of assuming it outweigh the processual costs of assuming a relatively complicated mechanism involving higher order logic. It is not clear what criteria could settle this dispute, since the losses and gains are of very different natures in the two cases. Hinzen and Reichard (2011) thus try to avoid this debate by questioning the explanatory value added by a semantic event-analysis – and the very same question could be raised in respect to its Montagovian competitor. This paper takes a similar approach towards the question. It addresses in particular aspects of adnominal modification (Sections 3 and 4), which has been left unaddressed in Hinzen and Reichard (2011). In Section 2, I shall briefly motivate a general approach of modification, thereby addressing some of the other phenomena that have been argued to back up event-arguments. The general picture also provides a background for the more specific discussion which follows it. 2. Events and the organization of grammatical phases It is widely recognized that there is a difference between arguments and other modifiers in natural languages. In general, it is not possible to drop arguments without causing ungrammaticality but other modifiers are mostly not obligatory. It has thus been assumed that these other modifiers are generated as adjuncts rather than as specifiers or heads.3 A related difference is that in many languages arguments receive case whereas modifiers do not. Furthermore, adjuncts in general are islands in 3

In recent theory, the adjunctive status of these modifiers has often been questioned. As a result of Kayne’s (1994) antisymmetry-thesis which assimilates specifiers and adjuncts, Cinque (1999), for example, argues that the rigid order in which adverbial modifiers occur crosslinguistically can only be accounted for if we assume that they are generated as specifiers of functional projections (for a similar point about adnominal modification see Scott 2002). This, however, does not change the general observation that arguments have to be present (and in most cases overt), whereas other modifiers don’t need to. Cinque’s functional projections can be left unspecified. And whether they are specified or not, does in most cases not change anything categorically.

436 respect to movement. But most importantly, the relevant modifiers do not change anything categorically – syntactically a modified object retains whatever properties it has had before modification. Moreover, it is striking that, from a grammatical point of view, the modified expression contains the un-modified expression as a proper part. The last two features suggest that also semantically the unmodified expression is a proper part of the modified one – any specification of the unmodified expression will be a specification of the modified expression as well; and this seems to sufficiently explain why in general the modified expression implies the unmodified expression.4 Where the facts are exactly as we expect them to be, no additional theory is needed – especially in light of the fact that the event-argument hypothesis does not come without further complications. Note that the present considerations are not restricted to adverbial modification and equally apply to adnominal modification. They may, thus, be seen as the source of the parallel nature of the two domains of modification. An analysis with the help of modern logic, and a set-theoretic interpretation thereof, in a certain way obscures these results. In settheory, the primitive is the individual which we can either refer to or quantify over. Predicates then are analyzed as sets of individuals, intensions are analyzed as the extensions of the relevant entities in possible worlds, etc. However, in line with the remarks above, it seems to be the case that the grammar of natural languages is organized exactly the other way round. Let’s consider first the nominal domain. Borer (2005a) argues that (17) has the grammatical structure given in (18): (17) these three lambs (18) DP[these #P[three Classifier Phrase[-s NP[lamb]]]] As Hinzen (2010; cf. also Hinzen 2007) explains, the most inner constituent lamb is a ‘pure lamb space’ which is, if it is not divided or 4

This is obviously only the case in non-negated environments. An odd number of negations changes an environment from downward to upward entailing and thus reverses the possible inferences. Ludlow (2002) shows that it is in principle possible to use the insights of the medieval logicians to provide a syntactic account of upward and downward entailing environments.

437 restricted with the help of quantification, understood as a mass term, like in I ate lamb. The grammatically simplest expression thus receives a generic interpretation which is rather complicated to describe in settheoretic terms. The classifier, which features as a separate word in some languages, for example in Chinese, divides the ‘pure lamb space’ into individual units. The numeral three orders these units in triples and the demonstrative determiner picks out one of these triples. It thus seems that reference to individuals is not a lexical phenomenon, but that reference to (or quantification over) individuals, which is set-theoretically primitive, is grammatically the last of several steps which the computational system of human languages can (but does not have to) take in the nominal domain. Referentiality, thus, seems to be an aspect of the meaning which grammar as opposed to the lexicon contributes. As Longobardi (1994, 2005) argues, also proper names are grammatically not primitives but involve movement of the lexical item to the determiner position. A similar pattern can be found in the verbal domain (Borer 2005b). As noted above, Davidson thinks of events as spatio-temporal particulars; and this is indeed what is suggested by the event-argument hypothesis which utilizes predicate logic or set-theory, since in these domains the primitives are individuals. However, the grammatically simplest eventive expressions are generic. In contrast, to get a fully individuated event, the maximal structure available in the verbal domain is necessary. Thus, again, reference to individuals seems to be part of grammatical rather than lexical meaning. The expression coming home in (19), for example, is generic and not specified for tense. Also, destroying Syracuse in (20) is something which, in principle, many people can do at many different times; and the destruction of Syracuse in (21) can be used to pick out any such event. What all these expressions have in common, according to current syntactic theory, is that they are specified for Aspect but lack a sufficiently specified tense-projection T: the derivation of the eventive expression stops before tense is sufficiently specified. In contrast, in (22) the derivation proceeds to the T-projection and the result is a fully individuated event (for a more detailed discussion see Hinzen and Reichard 2011. They conclude from

438 a consideration of these examples that grammar does not provide evidence for Davidson’s thesis that events are temporal particulars). (19) Coming home is always wonderful. (20) Destroying Syracuse is something Caesar did. (21) The destruction of Syracuse is Caesar’s deed. (22) Caesar destroyed Syracuse. The same pattern is again found within vP: telic expressions have to be grammatically more complex than atelic expressions. For example, (12) lacks a sufficiently specified v-head, which is responsible for introducing a place for the Agent (Chomsky 1995: Ch. 4). As a result, the structure in (12) cannot be telic. On the other hand, such a head is available in (4) and, since the Theme is also specified and sufficiently specific, a telic reading is possible there. The distinction between telicity and atelicity in the verbal domain is often seen as mirroring the distinction between mass and count in the nominal domain. In sum, considering only the few projections just discussed, the syntactic structure of (22) may be (23). The more structure is added to the left, the more specific the eventive structure becomes. In parallel to the nominal domain, quantification over fully individuated events is only the last of several steps which the grammatical derivation can (but does not have to) take. In (23), the clitic tense head attracts the verb and thus triggers movement of the verb. Caesar, first generated in the specifier position of v, is moved to the subject position. (23) TP[Caesar destroy-Ted vP[Caesar v VP[destroy DP[Syracuse]]]] Since the expression gets the more specific the more functional structure is added to its left, it is expected that a grammatical structure implies its proper part. Thus, if I ate the lamb, I ate lamb. If I boiled the soup, the soup boiled and if Caesar destroyed Syracuse, Syracuse was destroyed. This is expected, given the grammatical structure, and a neoDavidsonian analysis like (13) does not add to our understanding of it. Rather, the use of predicate logic and set-theory obscure this result, as it requires reference to individuals as a primitive, although, from a

439 grammatical point of view, individuality is never a primitive or lexical phenomenon. Furthermore, it seems that reference (or quantification) itself is a phenomenon which is not primitive but only occurs at the end of the nominal or verbal phase. What is referential is the whole phrase these three lambs, not the embedded expression lamb (see Section 4 for further discussion). 3. Adnominal modifiers and events As has long been noted, the implicational behaviour of modifiers is not as straightforward as suggested in the first section, at least not in the nominal domain. Summarizing previous literature, Kamp and Partee (1995) distinguish between four different types of adjectives according to their inferential behaviour. First, some adjectives are intersective. A blue house, for example, is an object which is both blue and a house. If, ignoring the problems discussed in the last section, we assume that both blue and house have a set as their extension, the extension of blue house can, in line with (9), be described as the intersection of these two sets. Since the intersection of the two sets is a subset of both sets, the meaning of the combination of noun and adjective implies both the meaning of the noun and that of the adjective. Second, there is a kind of adjective which is similar to the first in that the combination of adjective and noun implies the noun, but that differs from the first one in that the combination of noun and adjective does not imply the adjective. A skilful surgeon, for example, may be a violinist without being a skilful violinist. Set-theoretically, the denotation of the combination can be described as a subset of the denotation of the noun. These adjectives are therefore called ‘subsective’. Third, some adjectives, like alleged, do not even allow this kind of inference. An alleged murderer may not be a murderer at all. Adjectives which behave in this way are called ‘nonsubsective’. Finally, some adjectives seem to negate the denotation of the noun. Thus, in some relevant sense, a fake gun is not a gun. These adjectives are called ‘privative’. However, Partee (2007), in part criticising her own earlier work, has made a case for the thesis that these adjectives can also be classified as subsective.

440 Ignoring the rather rare case of non-subsective adjectives, Reichard (2011) argues that the distinction between intersective and subsective adjectives should not be treated as a lexical but rather as a grammatical phenomenon, since both intersective and subsective readings are systematically available for most adjectives. Furthermore, in many languages, the different readings correspond to grammatical differences. This has recently been taken as an argument for the thesis that the semantic difference corresponds to a difference in syntactic structure (Alexiadou et al. 2007; Cinque 2010; Sproat and Shih 1988). Two relevant ambiguities, first described in Bolinger (1967), are the distinction between individual-level and stage-level readings, which is available if the modifying adjective is deverbal (this distinction, arguably, finds a correlate for non-deverbal adjectives which give rise to kind-level and individual-level readings), and the ambiguity between adjectival and adverbial modification, which is available if the noun modified has a deverbal origin. Concerning the first ambiguity, consider (24) and (25). If visible occurs postnominally like in (24), in English, there is only one reading available: the stars which are currently visible. This use has been called ‘stage-level’ modification. The same reading is also available if visible occurs prenominally, like in (25). However, in this case, there is also an additional reading according to which those stars are referred to that are visible in general. Understood in this sense, (25) would be true, even when uttered at day time such that you couldn’t currently see any star. This use has been called ‘individual-level’ modification (cf. Carlson 1980). (24) The stars visible include Cappella. (25) The visible stars include Cappella. Concerning the distinction between adjectival and adverbial modification, consider (26) and (27). Again, if the modifier occurs postnominally like in (26), in English there is only one reading available. In this case beautiful modifies a person; it is thus an adjectival modifier. However, (27) is ambiguous between this reading and a reading according to which beautiful modifies not the person but the person’s dancing.

441 (26) The dancer more beautiful than her instructor (27) The beautiful dancer (Cinque 2010: Ch. 2) The stage-level and the adjectival modification are usually taken to be intersective, whereas the individual-level and the adverbial modification are not: As noted, if visible is understood in the individual-level reading, the truth of (25) is compatible with Cappella not being (currently) visible, and if beautiful is understood in the adverbial sense, a beautiful dancer may not be so beautiful after all. The Montagovian tradition takes intensionality to be the reason for the non-intersectivity in these cases (Clark 1970; Kamp 1975; Montague 1970; Parsons 1970; Siegel 1976). In general, adjectives are thus taken not to modify the extension but the intension of their modifee. They are, therefore, property modifiers in the sense of ‘property’ developed by Montague, that is, they are functions from intensions of noun phrases to such intensions of noun phrases. However, McConnell-Ginet (1982) has provided arguments which suggest that the failure of intersectivity does not need to be due to intensionality but can also be caused by a different (hidden) relationality. For example, even if we assume that everyone who cooks also eats and vice versa, Mary may cook fish without eating fish. Intuitively the failure of this substitution is not due to what happens in other possible worlds, but to the fact that you can cook things which you don’t eat and eat things that you haven’t cooked yourself in the actual world, even if you both cook and eat. Therefore, ‘cooking fish and eating fish can be distinguished in a model that does not distinguish cooking and eating, with no appeal to alternative situations’ (McConnell-Ginet 1982: 163). Furthermore, as McConnell-Ginet notes, in the case of adverbial modification, an appeal to possible worlds does intuitively not seem to be correct. Let’s assume that the singers and dancers are coextensive. Even if this was necessarily the case, it does not follow from the fact that Mary dances beautifully that she also sings beautifully. The reason for this, therefore, cannot lie in properties which Mary has in different possible worlds, but in properties she has in the actual world.

442 The explanation lies not in the existence of an alternative situation (where individuals have different properties), but simply in the possibility of a different sorting of the individuals, given a refinement of the sorting principles. (McConnell-Ginet 1982: 163)

Larson (2002: 249; 1998) argues that in the case of adverbial modification, the reason for substitutivity failure consists in hidden relations to events: ‘even if the same people sing and dance, there is a performance. And even if the same people dance and sing, the performances are still different. And one might be beautiful, and the other not.’ Larson and Segal (1995) and Larson (1998) therefore suggest that the difference may be captured by assuming that expressions like dancer do not only have an argument to be taken by an object, but also an event-argument. The ambiguity between adjectival and adverbial modification can then be captured by assuming that (certain) adjectives can modify either of the two arguments. If it modifies the eventargument, we get the adverbial reading, and if it modifies the objectargument, we get the adjectival reading. Refining an analysis from his (1998), Larson (1999) suggests the logical form of the adverbial reading of (28) may be (29). (28) Olga is a beautiful dancer. (29) Γe[[Con(e, Olga)] [Ee'[overlap(e,e') a dancing(e') a Agent(Olga,e')]] [beautiful(e, C)]] (30) Γe[[Con(e, Olga)] [Ee'[overlap(e,e') a dancing(e') a Agent(Olga,e')]] [beautiful(Olga, C)]]

Γ is a generic quantifier over events (Chierchia 1995). Con(e, Olga) is the contextually determined restrictor of the generic event-quantification which ensures that Olga does not always have to dance in order to count as a dancer, but only on certain occasions. The overlap relation ensures that even on the given occasion – let’s say when Olga is in clubs – she does not necessarily always dance; she only is asserted to dance at some times during these events. That Olga is the Agent of such events of dancing which overlap with a generic event makes Olga a dancer. What is important for our purposes is that beautiful now is a predicate of the

443 generic event (relative to the context C) – it does not modify Olga directly. In contrast, if beautiful receives an adjectival interpretation, it is a predicate of Olga, as illustrated in (30). According to this proposal the reason for the ambiguity does now not lie in the adjective but in the noun. Note that according to this theory, both readings are extensional. Event semantics can also be used for explaining the difference between the stage-level and the individual-level readings. As Chierchia (1995) notes, individual-level modifiers are inherently generics while stage-level modifiers are not. In Larson’s theory, the difference between them thus turns out to be a difference in the quantifier over eventvariables. Whereas individual-level predicates are in the scope of the generic quantifier of the event they modify, stage-level predicates are not.5 As in the case of the adjectival/adverbial distinction, this move has some intuitive plausibility, since, again, the alternative doesn’t: it does not seem to be the case that the stage-/individual-level distinction is a phenomenon which has something to do with intensionality. It is not even clear that there is an inference failure between visible star (where visible has an individual-level reading) and visible – a visible (in the individual-level reading) star is a subset of things which are generally visible. Thus, as Bolinger (1967) originally proposed, individual-level and stage-level predicates seem to designate different kinds of properties, and an object can be visible in one sense and fail to be visible in the other. This has nothing to do with properties which the object has in different possible worlds, but with different properties the object has in this world. A further strength of Larson’s analysis is that it offers different treatments for the stage-level/individual-level ambiguity and for the adjectival/adverbial ambiguity. That different treatments are necessary is suggested by two facts about their distribution. First, whereas the 5

It seems that this analysis could be extended to the dichotomy between kind-level and individual-level modifiers discussed in Reichard (2011), which similarly seems to be a difference between genericity and individuality – the only difference then is that the relevant quantifiers will in these cases scope over objects, not events.

444 adverbial reading cannot occur in predicate position, the individual-level reading can.6 Thus, there is no adverbial reading available in (31), but there is an individual-level reading in (32). (31) The dancer is beautiful. (32) The Danube is navigable. Furthermore, it seems that the two phenomena can in principle co-occur. Thus, arguably, there are four possible readings for (33). First, responsible may be used as an adverbial modifier. In this case, the dancer, let’s call her A, is not necessarily a responsible person in everyday life, but she is responsible for the dance – she may, for example, be leading the dance. At the same time, responsible may be used as an individual-level or as a stage-level predicate. On the adverbial individual-level reading, our dancer is usually responsible for leading the dance, but may currently be on holiday and not be dancing at all. During this time, a second dancer, let’s call her B, may take over A’s job. For B, being responsible is an adverbial stage-level predicate. However, responsible may also be used adjectival. In that case, the dancer, C, may or may not be responsible for the dancing, but is generally behaving in a responsible way – C helps children and elderly people crossing the street, cycles to work rather than using a car, and does whatever you think responsible people do. This in turn may be a stage-level or an individual-level predicate: A dancer may behave responsibly in general, but perhaps not at the moment of your assertion, or be usually irresponsible, but doing a great job at the moment regarding responsibility. On the other hand, if responsible appears post-nominally as in (34), only the adjectival, stage-level reading is available. The independence of the two phenomena is supported by the fact that when responsible appears twice prenominally, the first can, for example, have an adjectival individual-level reading: a person who is generally responsible; and the second an adverbial stage-level reading: someone responsible for tonight’s dancing. Thus, neither of the two occurrences 6

Judgments about the latter seem to differ – ironically, Larson (1999) claims that individual-level readings cannot occur in predicate position.

445 of responsible is fully intersective in the sense of Kamp and Partee (1995). To get a fully intersective reading in this sense, the adjective has to have both, a stage-level and an adjectival reading. (33) A responsible dancer (34) A dancer responsible for the tickets (35) I need a responsible responsible dancer for tonight’s show. 4. Adnominal modification, grammatical organization, and the interface In this section, I first argue that the analysis which Larson provides for the stage-/individual-level distinction reduces to the grammatical structure of the DP, described in Section 2. Then I turn again to the adjectival/adverbial modification distinction. Here, I argue that even though Larson’s theory seems to capture the phenomenon in some respects better than the Montagovian alternative, there seem to remain some problems. First, there is the intuition that even when someone asserts (28) in the adverbial reading, it is still Olga who is said to be beautiful, not her dancing. Second, there are problems created by describing the phenomenon with the help of event-arguments, as this description does not capture central aspects of the linguistic phenomena. Third, given a more suitable description of the phenomenon, it seems to follow from the phasal organisation of language. Larson analyzes the difference between stage-level and individuallevel predicates in terms of a difference between generic and existential quantification over event-arguments. He notes that crosslinguistically the individual-level modifier is always closer to the noun than the stagelevel modifier. Thus if, in English, visible occurs twice in front of a noun, the first (counted from left to right) has the stage-level and the second the individual-level reading like in (36). These facts are captured if the generic quantifier, which binds all the open event-arguments in its scope, has a relatively narrow scope, as exemplified in (37) (based on Larson 1998, example 24).

446 (36) The visible visible stars include Cappella. (37) [ AP [Γe [ AP N]7] AP ] stage-level individual-level stage-level It has to be noted, though, that this is mainly a redescription of the phenomena in logical terms which independently follows from the grammatical structure of the nominal phase. As discussed in Section 2, a bare nominal will have a generic interpretation. Structure and individuality is gained only with the help of a number of functional heads. A modifier will modify whatever it is attached to. And if, as Larson shows, the individual-level modifier is closer to the noun than the stage-level modifier, it is expected that the latter will exhibit more individuality than the latter (at least as long as there are relevant intermitting functional projections). Let’s turn to Larson’s explanation of the adjectival/adverbial modification. Larson (1998; 1999) provides two paraphrases for the adverbial reading of (28) given in (38) and (39). (38) Olga is beautiful as a dancer. (39) Olga dances beautifully. The two paraphrases seem to differ in meaning only slightly but in an important way. In (38), we predicate something of Olga – as noted above (cf. 31), the adverbial reading is generally accepted not to be available in predicate position. Beautiful in (38) should thus not be a predicate of an event but a predicate of an object in Larson’s terms. In contrast, in (39) beautifully is a proper adverb and modifies a verb. According to Davidson’s event-argument hypothesis, it thus is a predicate of an event. Larson’s analysis of the phenomenon presupposes (39), understood as a generic sentence, as the correct paraphrase of (28) when he analyzes adverbially used adjectives as generic event modifiers. However, (38) seems to be a much better paraphrase of (28) than (39). Even if we assert (28) using beautiful in the adverbial meaning, we do not say that Olga’s 7

This is the place for postnominal individual-level modifiers which we find, for example, in Romance languages (cf. Cinque 2010).

447 dancing is beautiful, but that Olga herself is when she dances. Further evidence for this claim comes from examples like (40) which, according to Larson’s theory, should have (41) as a paraphrase. However, the pure adverbial reading in this sense does not seem to be readily available here. Nonetheless, when someone asserts (40), we would naturally think of Olga as singing beautifully as well. Indeed, if we knew that she did not, (40) would in most circumstances sound ironic. (40) Olga is a beautiful singer. (41) Olga sings beautifully. This suggests that the adverbial reading of an adjective, after all, does not seem to modify an underlying event.8 The difference rather seems to be that an adverbial adnominal modifier is sensitive to the descriptive content of its modifee in a way in which the adjectival modifier is not. Whereas the adverbial modifier modifies its host under a description, the adjectival modifier modifies the referent of the nominal phrase it attaches to irrespective of its descriptive content. According to this picture, the difference between the adverbial and the adjectival readings thus turns out to be a difference in sensitivity to the descriptive content of the modifee.9

8

9

A further argument towards this conclusion is provided by Szabó (2001). He argues that events on their own are not sufficient for a proper analysis of adverbial modification. He asks us to imagine a possible world where dances are only performed in order to persuade the gods to let it rain. It seems that in such a world the rain-making events would be the same events as the dancing-events. Nonetheless, someone may be a good dancer but not a good rainmaker if she has perfected her dancing technically to a high degree but is nonetheless not very successful in making rain. Szabó’s solution is to assume a contextual variable which specifies the respect in which something is called good; the same strategy could be used for other adjectives like beautiful. However, Szabó’s challenge is also met by the proposal informally discussed in the remainder of this section. This proposal, furthermore, avoids postulating contextual variables which solve the problem by stipulation rather than explanation. For anyone who does not accept this analysis, there is, of course, a fallback position which simply syntactizises the semantic account. It might thus be argued

448 A virtue of this analysis is that it can accommodate some examples which have been taken to be problematic for Larson. Arguably, the event-argument is due to the verbal origin of the noun. Thus, (42) owes its adverbial reading to the fact that we could also say that the referent rules justly. But what about (43)? There does not seem to be a verb to king. As Larson (1999) admits, there are a number of examples like this, and it does not seem to be convincing to assume that there are eventarguments hidden in many of the most common nouns which do not seem to have anything to do with events.10 (42) The just ruler (43) The just king (Vendler 1968: 91) To recapitulate, we have seen that the adverbial reading is unavailable in predicative position (cf. 31). In English, it is also unavailable in postnominal position (cf. 24). In prenominal position, both the adverbial and the adjectival readings are available. However, even in prenominal position, there is a difference in that if the same adjective occurs in both the adjectival and the adverbial reading, the adjective with the adverbial reading is closer to the noun than the one with adjectival reading: In (44), the first occurrence of heavy (from left to right) can be understood as heavy in the sense of weight and the second as meaning that John smokes a lot, but not vice versa. (44) John is a heavy heavy smoker. If the phenomenon is correctly described in terms of sensitivity to the descriptive content of the modifee, then it seems that descriptive content is only available relatively close to the noun but unavailable further away. And indeed, the notion of a grammatical phase seems to provide an explanation of the unavailability of descriptive content at least for the predicate position and (in Germanic languages) the postnominal

10

that the availability of the adverbial modification is due to the verbal origin of the noun. Examples like (43) could then be taken as cases of reconstruction. Vendler (1968: 92) mentions fast horse, slow car, careful scientist, good poet and good father as further examples of the same phenomenon.

449 position: Chomsky (2001, 2008) proposes that grammatical structure is derived in units which he calls ‘phases’. Grammatical evidence for their existence is mainly provided by island conditions, that is, grammatical structure becomes unavailable for the further derivation at some point. The hypothesis is that at these points, the complex syntactic object is spelled out and receives an interpretation. A rationale for this hypothesis is that a derivation in phases reduces the computational burden – when the phase is finished, the complex structure is gone from the derivational workspace and only the non-complex referent derived in the phase is left. If this is correct, it follows that the further derivation is insensitive to the descriptive content of the referents generated so far. Our redescription of the difference between adverbial and adjectival adjectives seems to perfectly match this prediction of the phasal architecture of language: To the extent to which we do not cross phaseboundaries in cases where descriptive content is available and do cross phase-boundaries where it is not, grammatical phases can be taken to explain the difference in meaning between adverbial and adjectival adnominal modifiers. Indeed, if DP is a phase,11 it follows that its descriptive content is unavailable in the predicate position. It also seems that relevant postnominal modification in English, which has to be phrasal in order to be licensed, attaches in a separate phase: more beautiful than her instructor in (26) is even phonologically detached from its modifee. In respect to the prenominal position, it seems that the adverbial reading is the most salient one (so much so that Bolinger 1967 argued that the adjectival reading is not available in this position). And this is predicted, given the fact that the modifier modifies an object under a certain description; that is, given the fact that there is still grammatical structure available at the time of modification. In contrast, the event-analysis does not predict this preference here. However, structure can also be ignored, hence the availability of the more paratactic adjectival reading even in this position. Note that a purely adjectival reading is rather marginal: in fact, it only seems to be available in cases involving movement-phenomena like focus. Thus, 11

Chomsky originally proposed that only CP and vP are phases. However, DP has been added to the list in many more recent proposals (cf. Chomsky 2007).

450 apart from ironic statements, we may use the beautiful dancer to refer to a person who dances dreadfully, only in contrast to someone who, in addition, looks ugly too.12 5. Conclusion In sum, it has been argued that the phenomena which have been taken to support the event-argument hypothesis find their explanation in mainly grammatical terms. Thus the fact that, ceteris paribus, an expression modified adverbially entails its non-modified counterpart is expected given the general structure in which grammatical meaning is computed and the fact that the verb-phrase of the latter is a proper part of the verbphrase of the former. The parallel nature of adnominal and adverbial modification finds a natural explanation in the grammatical fact that modification seems to work in similar ways across all grammatical domains. And the phenomena which originally motivated the neoDavidsonian approach can be accounted for with a standard syntactic verb-shell analysis (for a more detailed discussion of these and other phenomena see Hinzen and Reichard 2011). Also the stage-/individuallevel distinction, when analyzed as a phenomenon about genericity, 12

The latter observation may suggest that also in the prenominal position there is a grammatical difference between the two readings. However, it is not clear whether the phasal explanation can get away without ‘undoing’ structure, since, even though the focused modifiers occupy higher nodes in the grammatical structure than unfocused ones, they are still taken to belong to the same phase. There is an alternative explanation for the ambiguity of prenominal adjectives explored in Reichard (2011). The readings of adjectives which Kamp and Partee (1995) classify as intersective (and this includes adjectival readings) have recently been argued to originate in a reduced relative clause whilst the readings of adjectives classified as subsective are taken to have no clausal origin (Alexiadou et al. 2007; Cinque 2010). Given the assumption that CPs are phases in the relevant sense, the descriptive content of the adjective cannot be modified relative to the modifee in the case of the clausal origin, whereas this possibility is open for non-clausal modifiers. A problem for this proposal in respect to the adverbial/adjectival modifier distinction is that it does not seem to be the descriptive content of the modifier the availability of which seems to matter, but the availability of the descriptive content of the modifee.

451 seems to directly follow from the structure of the grammatical phase. Finally, it has been argued that Larson’s analysis of the adjectival/adverbial modification distinction, even though it captures some phenomena more adequately than the traditional Montagovian account, should be refined: rather than assuming the nominal to have a position for an event-argument in addition to the one for the objective argument, the phenomenon should be described in terms of sensitivity to the descriptive content of the modifee. The adverbial reading then is due to such sensitivity. A phasal account of grammar predicts that this reading is unavailable in predicative and, in the case of English, in postnominal position. We can conclude from this that the assumption of event-arguments does not add anything to our understanding of the relevant phenomena, given that they find their explanation in independently motivated grammatical principles. This does not mean that the hypothesis is incompatible with the phenomena described. But rather than adding to their explanation, an account based on eventarguments seems to add further problems which have to be cared of. As noted, Davidson thought of his event-argument hypothesis as the core evidence for a metaphysical thesis concerning the existence and particularity of events in a metaphysical sense. However, if the position defended here is correct, the event-argument hypothesis does not find independent support as far as the linguistic phenomena are concerned, the metaphysical theses thus remains in want of support from natural language. The fact that grammatical principles seem to provide an explanation for why certain kinds of expressions have a certain meaning does furthermore suggest that grammar is not only a way of encoding meaning, but is part of the creation of meaning; we can, thus, speak of something like ‘grammatical meaning’ as opposed to different aspects of meaning (for example lexical meaning). Whether something is picked out as an event, object or proposition, whether it is described as exhibiting generality or individuality, then, seems to be part of this grammatical meaning. This provides another argument against the validity of using ontological commitments in respect to natural language to back up metaphysical theses. The ontological categories we find in natural languages seem to be part of how we refer to the world using

452 grammatical structures rather than categories of the world which are independent of grammar. Finally, the event-argument hypothesis seems to be a case in point where additional problems are introduced by the analysis of the phenomena. As noted in the second section, this problem is not restricted to event-semantics but extends to other uses of logic in the analysis of natural languages, since the structure of its common set-theoretical interpretation at least seems to differ radically from the organisation of natural language. This is not surprising, given that logical systems were initially developed for very different purposes. But we should, therefore, be very sensitive to whether, and in what sense, they provide an adequate analysis of natural language, when using them for this purpose. References Alexiadou, Artemis, Liliane Haegeman and Melita Stavrou 2007. Nouns Phrase in the Generative Perspective. Berlin: Mouton de Gruyter. Baker, Mark 1988. Incorporation. Chicago: University of Chicago Press. Bolinger, Dwight 1967. Adjectives in English: Attribution and Predication. Lingua 18, 1-34. Borer, Hagit 2005a. In Name Only. Structuring Sense, Volume I. Oxford: Oxford University Press. Borer, Hagit 2005b. The Normal Course of Events. Oxford: Oxford University Press. Carlson, Greg 1980. Reference to Kinds in English. New York: Garland. Chierchia, Gennaro 1995. Individual-Level Predicates as Inherent Generics. In: Greg Carlson and Francis Jeffry Pelletier (eds.), The Generic Book. Chicago: University of Chicago Press, 176-223. Chomsky, Noam 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Chomsky, Noam 2001. Derivation by Phrase. In: Michael Kenstowicz (ed.), Ken Hale. A Life in Language. Cambridge, Mass.: MIT Press, 1-52. Chomsky, Noam 2007. Approaching UG from below. In: Uli Sauerland and HansMartin Gärtner (eds.), Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics. Berlin: Mouton de Gruyter, 1-29. Chomsky, Noam 2008. On Phases. In: Robert Freidin, Carlos P. Otero and Maria Luisa Zubizarreta (eds.), Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud. Cambridge, Mass.: MIT Press, 133-166.

453 Cinque, Guglielmo 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Cinque, Guglielmo 2010. The Syntax of Adjectives: A Comparative Study. Cambridge, Mass.: MIT Press. Clark, Romane 1970. Concerning the Logic of Predicate Modifiers. Nous 4, 311335. Davidson, Donald 1967a. Causal Relations. In: Essays on Actions and Events. Oxford: Oxford University Press (2001), 149-162. Davidson, Donald 1967b. The Logical Form of Action Sentences. In: Essays on Actions and Events. Oxford: Oxford University Press (2001), 105-122. Davidson, Donald 1969. The Individuation of Events. In: Essays on Actions and Events. Oxford: Oxford University Press (2001), 163-180. Davidson, Donald 1970. Events as Particulars. In: Essays on Actions and Events. Oxford: Oxford University Press (2001), 181-188. Davidson, Donald 1971. Eternal vs. Ephemeral Events. In: Essays on Actions and Events. Oxford: Oxford University Press (2001), 189-203. Davidson, Donald 1985a. Adverbs of Actions. In: Essays on Actions and Events. Oxford: Oxford University Press (2001), 293-304. Davidson, Donald 1985b. Reply to Quine on Events. Appendix B. In: Essays on Actions and Events. Oxford: Oxford University Press (2001), 305-311. Davidson, Donald 1993. Method and Metaphysics. In: Truth, Language, and History. Oxford: Oxford University Press (2005), 39-45. Dowty, David R. 1989. On the Semantic Content of the Notion of ‘Thematic Role’. In: Gennaro Chierchia, Barbara H. Partee and Raymond Turner (eds.), Properties, Types, and Meaning. Vol. II: Semantic Issues. Dordrecht: Kluwer, 69-129. Hinzen, Wolfram 2007. An Essay on Names and Truth. New York: Oxford University Press. Hinzen, Wolfram 2010. An Argument Against Compositionality. Manuscript. Hinzen, Wolfram and Ulrich Reichard 2011. The Event-Argument Hypothesis: In Search of Evidence. Manuscript. Kamp, Hans 1975. Two Theories about Adjectives. In: Edward Louis Keenan (ed.), Formal Semantics of Natural Language. Cambridge: Cambridge University Press, 123-155. Kamp, Hans and Barbara H. Partee 1995. Prototype Theory and Compositionality. Cognition 57, 129-191. Kayne, Richard 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Larson, Richard K. 1998. Events and Modification in Nominals. In: Devon Stolovitch and Aaron Lawson (eds.), Semantics and Linguistic Theory VIII (SALT8). Ithaka, NY: Cornell University, 145-168.

454 Larson, Richard K. 1999. The Semantics of Adjectival Modification. Delivered at the Duch National Graduate School (LOT), Amsterdam. http://semlab5.sbs.sunysb.edu/~rlarson/LOT(99)/Contents.htmld/index.html. Larson, Richard K. 2002. The Grammar of Intensionality. In: Gerhard Preyer and Georg Peter (eds.), Logical Form and Language. Oxford: Clarendon, 228-262. Larson, Richard K. and Gabriel Segal 1995. Knowledge of Meaning: An Introduction to Semantic Theory. Cambridge, Mass.: MIT Press. Longobardi, Giuseppe 1994. Reference and Proper Names. A Theory of N-Movement in Syntax and Logical Form. Linguistic Inquiry 25, 609-665. Longobardi, Giuseppe 2005. Towards a Unified Grammar of Reference. Zeitschrift für Sprachwissenschaft 24, 5-44. Ludlow, Peter 2002. LF and Natural Logic. In: Gerhard Preyer and Georg Peter (eds.), Logical Form and Language. Oxford: Clarendon, 132-168. McConnell-Ginet, Sally 1982. Adverbs and Logical Form: A Linguistically Realistic Theory. Language 58, 144-184. Montague, Richard 1970. English as a Formal Language. In: Richmond H. Thomason (ed.), Formal Philosophy: Selected Papers of Richard Montague. New Haven: Yale University Press (1974), 189-224. Parsons, Terence 1970. Some Problems Concerning the Logic of Grammatical Modifiers. Synthese 21, 320-334. Parsons, Terence 1990. Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge, Mass.: MIT Press. Parsons, Terence 2000. Underlying States and Time Travel. In: James Higginbotham, Fabio Pianesi and Achille C. Varzi (eds.), Speaking of Events. Oxford: Oxford University Press, 81-93. Partee, Barbara H. 2007. Compositionality and Coercion in Semantics: The Dynamics of Adjective Meaning. In: Gerlof Bouma, Irene Krämer and Joost Zwarts (eds.), Cognitive Foundations of Interpretation. Amsterdam: Royal Netherlands Academy of Arts and Sciences, 145-161. Quine, Willard van Orman 1953. On What There Is. In: From a Logical Point of View: Logico-Philosophical Essays. New York: Harper & Row (1963), 1-19. Quine, Willard van Orman 1960. Word and Object. Cambridge, Mass.: MIT Press (1993). Reichard, Ulrich 2011. Inference and Grammar: Intersectivity, Subsectivity and Phases. To appear in Proceedings of the Irish Network in Formal Linguistics Conference 2011. Belfast Working Papers in Linguistics. Rothstein, Susan 1998. Introduction. In: Susan Rothstein (ed.), Events and Grammar. Dordrecht: Kluwer, 6-11. Schein, Barry 1993. Plurals and Events. Cambridge Mass.: MIT Press.

455 Scott, Gary-John 2002. Stacked Adjectival Modification and the Structure of the Nominal Phrases. In: Guglielmo Cinque (ed.), Functional Structure in DP and IP. New York: Oxford University Press, 91-120. Siegel, Muffy E. A. 1976. Capturing the Adjective. University of Massachusetts, Amherst, Mass. Sproat, Richard and Chinlin Shih 1988. Prenominal Adjectival Ordering in English and Mandarin. In: James Blevins and Julie Carter (eds.), Proceedings of NELS 18. Amherst: GSLA, 465-489. Szabó, Zoltán Gendler 2001. Adjectives in Context. In: István Kenesei and Robert M. Harnish (eds.), Perspectives on Semantics, Pragmatics, and Discourse. Amsterdam: John Benjamins, 119-146. Vendler, Zeno 1968. Adjectives and Nominalisations. The Hague: Mouton.

Mª Uxía Rivas Monroy University of Santiago de Compostela [email protected]

A Philosophical Reflection on Language: Is Ontology Needed in Semantics?∗ Abstract: Frege was concerned about the ontological status of senses, which he located in the third realm, opening the way to reifying similar notions as meaning. Late Wittgenstein, Quine and Davidson didn’t accept such reification and they introduced in semantics pragmatic features to explain meaning. This text focuses on the justification of the ontological status of senses provided by Frege. For Frege ontology is needed in semantics in order to establish the objectivity of knowledge, and to avoid psychologism. Possibly, emphasis on the ontological status of Fregean senses is not the best solution to account for the objectivity of knowledge, but emphasising only pragmatic features is not enough to get a good explanation of how language gives us information about the world. My point of view is that pragmatic accounts of meaning are possible just because semantic proposals, such as Frege’s, could explain the cognitive content of language. Key words: cognitive content, Fregean sense, meaning, ontological compromise, ontology, semantics, third realm, use

1. Preliminaries: setting the question In a certain way the question we are reflecting about, namely, if ontology is needed in semantics, has a clear affirmative answer. In order that the words apply or refer to reality it is necessary to give a structure or an ontology to that reality, and in that sense ontology is needed in semantics. Nevertheless, this part of the problem, difficult as it is, will be taken for granted in what follows. The question we are dealing with in this paper has to do with some of the main concepts used in semantics, ∗

This paper was supported by the FFI2009-08828 of Spanish Ministry of Science and Innovation.

458 such as meaning or sense and its ontological status. In relation with this part of the problem we will consider first Frege’s point of view about senses. Frege distinguished two main semantic relations, which he called sense and reference (Sinn and Bedeutung were the German words), and which gave rise to a traditional distinction, in Analytical Philosophy, between intensional and extensional semantics – following Carnap’s terminology. Sense was the semantic notion that connected world and language, and explained the cognitive content of language, and which is considered the predecessor of the notion of meaning.1 Ontological compromises in semantics are always difficult to take, and Frege took them in order to avoid worse consequences as psychologism or solipsism. This way of thinking was also connected with the importance of science and the view that science is the better means to obtain knowledge about the world. Similarly, the notion of truth was the relevant one. Frege put senses in a third realm, a realm that has to be recognised, because neither the outer world of spatio-temporal objects neither the inner world of images, subjective representations, sensations or emotions were appropriate to house senses. Russell’s logical atomism, Wittgensteins’s Tractatus, and Logical Positivism followed the stream of the semantic doctrine opened by Frege, sometimes to criticize it as Russell did, or sometimes to adjust it to a new framework as it was in Carnap’s distinction of intension and extension. Nevertheless, a new way of understanding language and its relation with reality began to grow around 1950. The contribution of Wittgenstein’s Philosophical Investigations (1953) with his statement that the meaning of a word is its use in language was crucial to the establishment of this new approach to language. The Fregean way of understanding senses have contributed to the reification of similar notions used to explain the informative content of expressions; among them the most usual was the wide and general term of “meaning”. After publication of Wittgenstein’s work, philosophers of language linked to the Analytical tradition, such as Quine or Davidson, refused to consider 1

The identification of Fregean sense with meaning began with Russell’s paper “On Denoting” (1905), where he translated “Sinn” as “meaning”.

459 meaning as an entity, and in their work on semantics they didn’t accept explicitly the ontological status of meaning or akin notions; therefore, they introduced into semantics pragmatic features such as use and holism to explain meaning. For example, Quine’s proposal was to change the myth of the museum – where meaning is like a piece on a shelf to which different names of various languages are attached – by introducing a more pragmatic and behavioural concept, as that of stimulus meaning. In addition, Quine and Davidson explained the meaning of words as an abstraction, where the systematic contribution of expressions to the meaning of sentences is balanced with the contribution of each sentence in which a word appears to the meaning of that word. So, holism is here the solution that is to explain meaning without any ontological compromise. In Fregean semantics senses could be treated as special kind of objects, given the way Frege characterizes them; on the contrary, the assumption that meaning is use does not need to take any ontological compromise. Putting the question in such terms, and taking into account Occam’s razor, it seems reasonable to favour the idea that meaning is use, and that it is related to human practices of language such as speaking and understanding; but it is also necessary to ask if this account of meaning is powerful enough to explain the cognitive content of statements. In my opinion, the account of meaning as use has in its favour to have opened the narrow view that the main use of language is constatative or representative, and, therefore, to emphasize that language not only give us information about the world, but it is intertwined with many other social activities and it cannot be separated from them. But, in my view, the doctrine of meaning as use is not good enough to provide an explanation of the cognitive content of linguistic expressions, that is, the fact that through language we acquire knowledge about the world. Concepts such as use, practice, rules and similar ones explain the relevant role of speakers and their intentions. They explain also the social character of language, and how language is a powerful instrument to create and construct social and particular identities; but when an explanation of the representational function of language is at stake, pragmatic accounts are not enough. What we are looking for are explanatory devices or theoretical notions to get a better understanding

460 of the complexities of the language. Neither of the two accounts have given a complete explanation of these complexities. Both are needed, and my claim is that pragmatic accounts of meaning are possible just because semantic approaches, such as Fregean semantics, have done the job of explaining the cognitive value of language as a means of acquiring some knowledge about the world. 2. In favour of ontology: the justification of Fregean senses To begin with, I will quote a text by Burge, from his book Truth, Though, Reason. Essays on Frege, which highlights the importance of Fregean sense in a reflection about meaning: It seems to me that philosophy of language is even now so focused on conventional linguistic meaning that it has not adequately exploited Frege’s notion of sense. (…) In philosophy the idea that meaning is a common denominator shared among normally competent speakers of a communal language continues to dominate reflection on language. Such a view features the aspects of language and language use that are common to different speakers and common to different occasions of use. Thus the rules governing indexicals, and what communicators have in common (…) are the center of investigation. Ways in which think about the entities referred to on particular occasions that are not easily shared or are not incorporated in such rules tend to be neglected. (…) These approaches take actual use, or functional relations, or conventional understanding to be determinative of linguistic meaning. But, (…) certain aspects of ‘meaning’ that Frege was responding to in his theoretical notion of sense escape these strictures. Here too Frege’s conception promises to supplement our philosophical understanding of language as well as thought. (Burge 2005: 58-59)

Sharing this interpretation of Burge about the powerful explanatory capacity of Frege’s notion of sense to understand language and thought, I would like to emphasize how Fregean semantics – and particularly, the notion of sense and thought – could explain both the cognitive or representative character of language, that is, how language hooks onto the world, and the compositionality of meaning. Any semantic theory has to account for this latter feature of language in order to explain the creativity and the capacity of learning a language. And Fregean

461 semantics is a suitable semantics to fulfil those aims. My claim is that even theories of meaning, which are far away from ontological compromises about meaning, like Davidson’s theory of meaning, are in debt to Fregean semantics in some of its most basic principles such as that of compositionality. Taking into account that one of the most questionable points of Frege’s semantics is its ontology of senses and thoughts, that is, its compromise with the third realm of abstract entities, we will describe briefly Frege’s ontology and semantics to get a better understanding of his reasons to assume such compromise. At first sight, Frege’s ontology is very simple; it has only two kinds of entities: objects and functions. At the same time, the criterion to decide to which kind an entity belongs is equally simple: everything which is not a function is an object. The important issue is, then, to identify a function, and for doing that Frege uses a mathematical resource: a function is that kind of entity which needs completion, because it is an unsaturated entity. In order to distinguish functions from objects, Frege follows the syntactic and grammatical structure of language. The main unit of meaning is for Frege the sentence or, better said, the statement, which expresses a complete sense – called thought by Frege.2 For Frege only thoughts can be true or false. Normally, statements are composed of a proper name and a predicative expression. Names and sentences are complete expressions, and predicative or relational expressions are incomplete and need to be completed by names. Frege was a realist about the external world, because there are only spatio-temporal objects in it, independent of human beings. Proper names refer to those objects. But proper names not only refer to objects, they also express a sense, which contains the way the objects are given. This way of understanding sense, apparently very simple, is a powerful theoretical device to explain the cognitive content of language and to grant the objectivity of knowledge. In a metaphorical way, the Fregean notion of sense works as an intermediate 2

Frege explained that what he calls thought has not to be equated with the internal activity of thinking. Thoughts are not subjective entities; on the contrary, they are objective, sharable, and communicable, cf. Frege (1984: 351-372).

462 link between world and language, and gives an answer to the question of how language can represent, or speak about, or hook onto the world. All the information we get through language is due to the sense of the words, and this information is not arbitrary, because the sense of a proper name contains an objective property of the object, namely, the way the object is given, that is, its mode of presentation. The property that is ontologically possessed only by an object, for example, that of being the president of Spain in 2010, is transferred to the language through the sense of the name “the president of Spain in 2010”. Fregean semantics has many subtleties, but a very important role in it is played by thoughts. Thoughts are senses of complete statements, and when they are true, Frege calls them facts. In the external world there are no facts but spatio-temporal objects. Through the language we apprehend true thoughts or facts. These entities, such as senses, thoughts, and facts, belong to the third realm, and, as we have already said, they share the property of being objective, but not spatio-temporal. A very important reason for Frege to say that senses and thoughts are not dependent on human beings is this: if it were the case they should be subjective, and Frege shows strong anti-psychologism in the foundations of both logic and knowledge. He claims that a third realm has to be recognised, because the ontological properties of senses do not let us classify them either as external entities or as internal or mental ones. The role of language is only to be a vehicle, which makes senses accessible to human beings. Therefore, the ontological compromise between senses and thoughts is a very relevant element in Frege’s explanation of why language conveys knowledge about the world, and why this knowledge constitutes facts, namely, true, eternal and immutable thoughts. The independence of senses and thoughts could be connected with the realistic thesis of the independence of objects, which says that objects are as they are without our intervention. In Fregean semantics, then, senses and thoughts are abstract entities, which are objective and also independent from human beings, as spatio-temporal objects are, even when they do not share the same reality as that of the external world. Their reality is not the same as that of the external objects, because they do not have the property of being causally effective, in the sense of having the causal power proper of external objects. Therefore,

463 Frege characterises the third realm as objective and “non-actual” (das Nichtwirkliche),3 choosing objectivity over reality. As we have said, the sense of proper names gives us information about the way objects are, because it contains this way of being the objects referred by the name. Both ways of being, that of the objects and that contained in the senses of proper names are objective and independent of human beings. The knowledge we obtain by the thoughts expressed through statements is not invented, created, imagined or produced by us; on the contrary, we discover what was already there, namely, the facts (the true thoughts), and we unveiled them by means of language. We know about the world and its objects, because we apprehend thoughts, which in any case are not our creation. In a way, this objectivity of thoughts and senses is related to the objectivity of external things, which also are objective and independent of us, because what the thought expresses, if it is true, is precisely how things are. Theoretically, thoughts exist even when we do not have epistemic access to them. As already mentioned, thoughts are characterised by Frege as eternal, immutable, objective, and independent of human beings; for that 3

“Actual” is the translation from the German word “wirklich” in the partial English translation of the Grundgesetze der Arithmetik, and it has to be understood in the sense of lacking causal effectiveness, therefore a convenient synonym for it could be “not causally effective”, following Burge (2005: 301). Frege names this feature of the third realm in the following excerpt from Grundgesetze der Arithmetik: Ich erkenne ein Gebiet des Objectiven, Nichtwirklichen an, während die psychologischen Logiker das Nichtwirkliche ohne weiteres für subjective halten. Und doch ist gar nicht einzusehen, warum das, was einen vom Urtheilenden unabhängigen Bestand hat, wirklich sein, d.h. doch wohl fähig sein müsse, unmittelbar oder mittelbar auf die Sinne zu wirken. (Frege 1966: xviii).

The English translation of this text clarifies in a footnote why “wirklich” is translated as “actual”: [F]or me there is a domain of what is objective, which is distinct from what is actual, whereas the psychological logicians without ado take what is not actual to be subjective. And yet it is quite impossible to understand why something that has a status independent of the judging subject has to be actual, i.e., has to be capable of acting directly or indirectly on the senses*. (Frege 1964: 15-16).

*The translation reproduces a play upon ‘wirklich’ and ‘auf die Sinne wirken’.

464 reason science has to discover true thoughts, or what amounts to the same, the task of science is to discover facts and not to create them. Then, in Frege’s view thoughts are the guarantee of scientific knowledge. They open the possibility of science, because science discovers facts, and as it was said, facts are true thoughts. In relation to that topic, Frege gives special attention to the argument that thoughts are not subjective ideas (Vorstellungen); if that were the case, science would be impossible: each person would defend her idea of what she thought the world was, and the right way of speaking in that case would be for example “my subjective idea of the moon is …” and nobody could argue against that, everybody would have his or her truth; there would be then only an interior and private world. True thoughts are thus the warrant of objectivity and, therefore, the task of science is to discover of them. Facts, as independent of human minds, are the “firm foundation for science” (Frege 1984: 368). Frege also emphasizes two features of thoughts: (i) that of being entities directly related with truth,4 and (ii) that of being independent of the particular thoughts of individuals,5 these are, then, the main basis to justify that true thoughts become the guarantee of scientific knowledge. Not only the kind of scientific considerations we have considered led Frege to the acknowledgment of the third realm; semantic reflections were also important, because for him, thoughts are not only the sense of statements but also the references of sentences when sentences are found in indirect speech. In that case, Fregean semantics is appropriate to explain how language works in different contexts of speech (ordinary, direct and indirect), making use of only two semantic relations of words or sentences, namely, sense and reference. But not only that, Fregean semantics has done the great job of formulating the interchange of expressions salva veritate, the compositional principle for sense and reference, and the context principle. With these technical principles Frege could explain many linguistic phenomena as the informational 4

5

Frege (1984: 368): “For what I have called thoughts stand in the closest connection with truth”. Frege (1984: 368): “That someone thinks it has nothing to do with the truth of a thought”.

465 value of identity statements, the systematicity of language, propositional attitudes, etc. One of the most important ways by which Fregean semantics has been fruitful, with an influence that continues nowadays, is through the connection it establishes between thoughts and truth. This relation is grounded on the cognitive value of the notion of sense, that we have already presented, and allows us to understand thoughts as the truth conditions of statements. This close relation between thought – later identified with proposition6 – and truth introduced a main line of inquiry, which was followed, among others, by Wittgenstein in his Tractatus,7 developed in an empiricist way by the Neo-positivists in the well known principle of verification, and elaborated in modern semantics by Davidson through the form of a theory of meaning, which can be condensed in the dictum saying that “the meaning of a sentence is its truth conditions”.8 This understanding of meaning as the truth conditions of a statement offers an explanation of the meaning of the statement in terms of its cognitive content. Therefore, we have got an explanation of meaning that is not far away from that notion of thought given by Frege. Nevertheless, there is a very noticeable difference between these views, namely, the ontological compromise. While Davidson can easily get rid of meaning as an entity, by not recognising any explanatory power to such reification, Frege needs to assume such a compromise to reinforce his arguments about the objectivity of knowledge and science. But the theory of meaning proposed by Davidson makes use of important 6

7

8

Once again, Russell translated the Fregean notion of thought by proposition in his correspondence with Frege in the letter of 12.12.1902. See Frege (1980: 150). Especially in proposition 4.021: “The proposition is a picture of reality, for I know the state of affaires presented by it, if I understand the proposition”. And also in proposition 4.024: “To understand a proposition means to know what is the case, if it is true. (One can therefore understand it without knowing whether it is true or not.)” (Wittgenstein 1955: 47). The exact formulation of this idea can be found in several places of Davidson’s works, e.g. in his article of 1967, “Truth and Meaning”, where he said: “[..] to give truth conditions is a way of giving the meaning of a sentence” (Davidson 1984: 24); and also in his later paper of 1986, “A Coherence Theory of Truth and Knowledge”, where he stated: “We can accept objective truth conditions as the key to meaning” (Davidson 1986: 307).

466 principles of Fregean semantics, such as the compositionality principle and the context principle. 3. Against meaning reification (I): Quine’s behavioural semantics Prefering a behavioural semantics,9 and arguing against mentalist or intensionalist semantics, Quine did not accept meaning as entity. Among other reasons, because in his opinion meaning lacks identity criteria – following his slogan of “no entity without identity” – and because of understanding meaning as an entity which does not bring about any good explanation of language.10 Therefore, instead of speaking of linguistic meaning Quine preferred to use a more scientific and behaviourist notion, such as stimulus meaning – introduced in his book Word and Object (1960), which in Pursuit of Truth (1990) he changed to stimulus range. On the contrary, for Frege this notion would be a psychological one. But for Quine, his new approach to meaning would make of semantics a science and not a speculative philosophy. For Quine stimulus meaning is related to an objective situation with objective elements, such as the speaker, the hearer and the words expressed in a particular situation, nevertheless stimulus meaning is a private event and only the expected behaviour of the hearer in the situation gives an index that the words were correctly understood. For Quine stimulus meaning is not equivalent to linguistic meaning, on the contrary it is a kind of empirical meaning. The beginning of mastering a language, both in a mother language or in a foreign language, goes through observational 9

10

“In psychology one may or may not be a behaviorist, but in linguistics one has no choice. [...] There is nothing in linguistic meaning beyond what is to be gleaned from overt behaviour in observable circumstances” (Quine 1990: 38). “To question the old notion of meanings of words and sentences is not to repudiate semantics. Much good work has been done regarding the manner, circumstances, and development of the use of words. Lexicography is its conspicuous manifestation. But I would not seek a scientific rehabilitation of something like the old notion of separate and distinct meanings; that notion is better seen as a stumbling block for philosophers than for scientific linguistics, who, understandably, have simply found it not technically useful” (Quine 1990: 56).

467 sentences, to which stimulus meaning is attached. After that first step sentences got their linguistic meaning through the relation established among them, which helps to get abstractions and generalizations,11 without any ontological compromise. What is common to language and world is the stimulus situation which affects our neural receptors, and Quine prefers these terms instead of senses and meanings; he works on a naturalized epistemology for explaining the relation between language and world; and we can agree on that, but is it all what we need in order to get an explanation of how language give us informative content about the world, and how can we use compositionality for building new sentences in a language, or for knowing that if we change two coextensive expressions, the truth value of the sentence holds, but its sense changes? Quine explains the meaning of a sentence as that which the sentence shares with its translations in another language. Therefore, there is not any entity which corresponds to meaning, only a sentence in another language would do the job, and the beginning of a translation goes through the vague but objective observational sentences. Notwithstanding, Quine’s doctrine of the stimulus meaning is linked to the indeterminacy or inscrutability of reference and the indeterminacy of translation, two consequences that Frege would not accept, due to the fact that senses are objective and allow to translation. The indeterminacy of translation is a thesis, which Quine formulated in his book Word & Object, and in the context of the radical translation12 as follows: “manuals for translating one language into another can be set up in divergent ways, all compatible with the totality of speech dispositions, yet incompatible with one another” (Quine 1960: 26) and the inscrutability of reference is a thesis stated in “Ontological Relativity”:13

11

12

13

Quine called these kind of sentences “observational categorical” or “focal observation categorial”. See Quine (1990: 10-11). Quine understands by “radical” translation a translation between unrelated languages and unshared cultures, more precisely, using Quine’s words, a “translation of the language of a hitherto untouched people” (Quine 1960: 28). “Ontological Relativity” was first published as an article in the Journal of Philosophy (1968) and one year later it appeared as the second chapter of

468 “The terms ‘rabbit,’ ‘undetached rabbit,’ and ‘rabbit stage’ differ not only in meaning; they are true of different things. Reference itself proves behaviourally inscrutable” (Quine 1969: 35). The indeterminacy of reference supports the indeterminacy of translation, but the latter is, according to Quine, a holophrastic and stronger thesis than the former, because the indeterminacy thesis applies to terms but the indeterminacy of translation applies to whole sentences, and specially to observational sentences. For Quine, an observational sentence, seen holophrastically as a whole, is conditioned by stimulatory situations, and thus theory-free, while when the sentence is seen analytically, word by word, it is theoryladen, and thus relative to a language, a theory or a manual of translation. The reference of terms is inscrutable because the stimulation of a particular situation is not enough to differentiate between an ontology of objects, parts of objects, or events, and for that reason the reference of terms depends on the theory or the language. Therefore, a flaw of Fregean semantics is the reification of senses, but a flaw of the Quinean doctrine is the indeterminacy of translation. 4. Against meaning reification (II): Davidson’s theory of meaning Davidson recognises that in some semantic theories, such as that of Frege, the consideration of meaning as an entity may play a role, particularly, when senses or thoughts are considered as references, as happens in indirect speech. But meaning as entity does not work in a theory of meaning14 under the conditions that Davidson demands for it to be a satisfactory theory.15 These conditions are related to both formal and empirical conditions; the former are met within the Tarskian

14

15

Quine’s book Ontological Relativity and Other Essays (1969), “corrected and extended” – in Quine’s words in the Preface to that book. Precisely, Davidson (1984: 20) states this idea in a very visual and figurative form: “the one thing meanings do not seem to do is oil the wheels of a theory of meaning”. These two demands, in Davidson’s words, are: “it would provide an interpretation of all utterances, actual and potential, of a speaker or group of speakers; and it would be verifiable without knowledge of the detailed propositional attitudes of the speaker” (Davidson 1984: xiii).

469 framework of the semantic definition of truth, particularly, Tarski’s Convention T; and the latter within the Quinean semantic framework of the radical translation, which Davidson modified into the radical interpretation. Similarly to Quine, Davidson’s main reason to reject meanings as entities is that “they have no demonstrated use”. Therefore, his objection is grounded more on operative, practical and functional reasons than on philosophical issues concerning the nature and identity of meanings, as being abstract entities or the fact that their identity conditions are obscure, as Quine had also previously stated. As a consequence of this account of meaning and of its dispensability Davidson maintains that we can speak about meaning in a harmless and “ontologically neutral sense” as the systematic contribution that a word makes to the meaning of the sentences in which it occurs. Nevertheless, he prefers the holistic point of view that understands the meaning of a word as “an abstraction” made of the contribution that each word makes to the totality of sentences in which it appears. As Davidson recognises, this insight has its roots in Frege’s context principle, which he extended to the whole language: “only in the context of the language does a sentence (and therefore a word) have meaning” (Davidson 1984: 22). But this Davidsonian account of meaning as ontologically neutral or of meaning as an abstraction did not explain from where meaning obtains its cognitive value. Davidson’s proposal of a theory of meaning explains many logical and semantic features of language in a comprehensive and systematic way, as well as the relations among language, truth, meaning, knowledge, belief, action. He has included in the theory of meaning fundamental aspects such as the important role played by the syntactic structure of sentences considering the meaning of the sentence, or the empirical proof of the theory of meaning through the methodological and theoretical device of the radical interpretation. But the most basic insight of how language hooks onto the world is taken for granted in his approach to build a satisfactory theory of meaning and there is not a plausible, or even creative, explanation for it.

470 5. Conclusion Pragmatic or behavioural approaches to language show us that some aspects of language can be explained without assuming that “meaning” has to be considered an entity. But semantics is the study of meaning in its more comprehensive dimension, namely, sense, reference and force.16 Reference has to do with the relation between words and objects; usually, objects are external elements to language, and we do not question ontology in that case. Force has to do with the pragmatic element of the modality in which a sentence is expressed, that is, assertive, interrogative, desiderative, etc., and it is related to the speaker’s intentions. But sense, and, of course, linguistic meaning are, on the one hand, an internal feature of language, because in order to explain the meaning of an expression we use other expressions, but on the other hand, what language expresses is in close connection with the way the world is. The notion of sense, at least as it is introduced by Frege, explains how this is possible: sense is not only a semantic notion, which determines the referent of proper names, but also a cognitive one, which contains the mode of presentation of the referent. Therefore, Fregean sense explains how language is informative: even when senses are apprehended through language, they are independent from it – language is only a means – and they contain “the way the referent is given” or its mode of presentation. This mode of presentation is a feature of objects but, according to Frege, it is contained in the sense of proper names; thus, senses are “ways things are presented to a thinker,” or ways of thinking about the referent.17 Moreover, the sense of the statement – the thought it expresses – is also associated with the truth conditions of the statement, being that another way of reinforcing that the notion of sense is linked with the cognitive or informative content of the statement. The latter dimension of language is a very important one, and it represents the essential character of language as a system of signs, that is, to give information about the world when the world is not right in front of us. 16 17

See Dummett (1993: 40). Burge (2005: 243).

471 In “Two Dogmas of Empiricism” Quine suggested that in order to get a better explanation of how the world is, scientific theories have to expand ontology, because an inflated ontology simplify the explanation of the phenomena we try to explain.18 We can paraphrase Quine’ words and say that to get a better explanation of semantics it could be also necessary to extend the ontology; or to get a good explanation of the cognitive content of expressions in a language it could sometimes be preferable and appropriate to inflate the ontology, because it simplifies the explanation and makes clear the point that the doctrine or the theory is trying to clarify. Fregean semantics could work as a good example for it. It is true that it has a strong ontological character, supported by a very extreme scientific tendency, where facts are completely objective and independent of human beings; but even within this ontological Platonism, it is a coherent and well-elaborated explanation of the informative content of language and of how language hooks onto the world. Pragmatic or behavioural proposals such as these of Quine or Davidson need not take ontological compromises – in the way Frege was obliged to recognise a third realm of abstract and objective entities, such as thoughts and senses – but in general they work along the main semantic principles of Fregean semantics, such as that of the interchangeability of co-denoting expressions, the compositional principle and the context principle,19 especially when they have to 18 19

See Quine (1961: 45). These principles are very important in Fregean semantics and they play a relevant role in explaining difficult semantic phenomena such as the failure of interchangeability salva veritate in indirect speech, or the lack of truth value in sentences where proper names do not refer. The compositionality principle has two versions, one when it applies to senses and the other when it applies to references. In general, it states that the sense, or reference, of a complex expression is functionally dependent on the senses, or references, of the relevant component expressions. What is known as the context principle consists in underlining the importance of the sentential context in order to determinate the sense or reference of an expression; more precisely it says: “never ask for the meaning (Bedeutung) [later, sense/reference] of a word in isolation, but only in the context of the sentence.” (Frege 1968: x). The context principle is found in Frege’s Foundations of Arithmetic (1884), before his distinction between sense

472 explain the preservation of truth in ordinary and indirect speech. In ordinary speech the truth value of the sentence is preserved if expressions with the same reference are interchanged, and, similarly, the sense of the sentence is preserved if expressions with the same sense are interchanged. In indirect speech the preservation of truth of the whole sentence does not depend on the truth value of the subordinate sentence, but of the thought it expresses, which in Fregean semantics is the reference of the subordinate sentence. In order to understand why the reference of the subordinate sentence is a thought (the sense of a sentence in ordinary speech) and not a truth value, it is necessary to take into account the compositionality principle, because in indirect speech the references of the component expressions are senses. Pragmatic and behavioural approaches are not always successful in explaining the informative content of languages. Fregean semantics not only took ontological commitments and strange reification of senses, as negative as it can be seen by philosophers of language as Quine or Davidson, but it brought also some semantic principles, such as the ones already mentioned, which are accepted by almost all linguists and philosophers of language. Quine or Davidson, for example, extended Frege’s context principle to cover not only sentences – as Frege stated it – but the whole system of language, defending a semantic holism. These Fregean principles explain the learning of language and the philosophical issue of the cognitive value of sentences, and linguists or philosophers of language do not get rid of them easily. Through this short presentation I did not try to answer the enormous, wide, complex, and difficult question of what meaning is; this is a challenging and extremely complicated enterprise, which has been the aim of many philosophers – and philosophers of language – during the whole history of philosophy without having yet a clear and unanimous answer, even in contemporary semantics. On the contrary, my aim was very modest and restricted; I only wanted to stress the great value, that in and reference was introduced, and, therefore, when it was formulated Frege used the term “Bedeutung” without the technical meaning of “reference”, which it would have later in “On Sense and Reference” (1892). If the principle is understood taking into account the distinction, then it can apply to references as well as to senses.

473 my point of view, Frege’s senses have, especially when they are used to explain the cognitive and informative content of some expressions. And as I have emphasised, even when their explanatory force is burdened by the strong Platonism involved in the postulation of the third realm. Nevertheless, according to what recent approaches about meaning have shown, it might be also possible to weaken Frege’s ontological compromise about senses to the extent that the third realm became only a realm of abstract entities, not completely independent from human beings. But Fregean semantics is coherent and systematic in explaining many other linguistic phenomena, which can not be completely taken apart from the constitutive nature of senses, thoughts and facts. In Fregean semantics the principles of context, compositionality and interchangeability salva veritate are used to explain the different cognitive value of analytic and synthetic identity sentences, the peculiarities of oblique contexts, the creativity and learnability of languages, etc. Other semantic proposals, which have turned to pragmatics, have still taken into account these basic principles formulated by Frege. We think through language, and we do many activities which imply language. Fregean semantics teach us many insights about thinking and the cognitive value of meaning; pragmatic semantics teach us many insights about how communication works, paying attention to the speaker’s role and intentions. If within this third realm we avoid Frege’s strong ontological Platonism (eternity, total independence from human beings, language as mere vehicle of thoughts, etc.), but retain the main traits that explain how language gives us information about the world in a way that is not arbitrary to our will, there will be then some theoretical elements in meaning, similar to those of Fregean senses, which will be supported by some kind of harmless ontology. I have begun with a quotation by Burge, and I will finish this paper with another one by him, which highlights once again that ontological compromises in semantics need not be so strong as Frege has taken them, and that in spite of them Fregean senses retain their power to explain the cognitive character of language, thought and knowledge: The key point about Fregean sense is its explanatory fruitfulness. I believe that one can see, rather easily, that such fruitfulness depends on

474 commitment to abstract entities. But one can make considerable progress by making use of Frege’s insights about sense and thought contents without deciding the exact metaphysical nature of the abstractions to which one appeals. This is the analog of the situation with respect to mathematical entities in all sciences. (Burge 2005: 28)

References Burge, Tyler 2005. Truth, Thought, Reason. Essays on Frege. Oxford: Oxford University Press. Davidson, Donald 1967. Truth and Meaning. Reprinted in: Donald Davidson, 1984, 17-36. Davidson, Donald 1984. Inquiries into Truth and Interpretation. Oxford: Oxford University Press. Davidson, Donald 1986. A Coherence Theory of Truth and Knowledge. In: Ernest Lepore (ed.), Truth and Interpretation. Oxford: Basil Blackwell, 307-319. Dummett, Michael 1981. Frege. Philosophy of Language. London: Duckworth (2nd edition). Dummett, Michael 1993. What is a Theory of Meaning (II). In: Michael Dummett, The Seas of Language. Oxford: Clarendon Press, 34-93. Frege, Gottlob 1964. The Basic Laws of Arithmetic. Exposition of the System. Berkeley and Los Angeles: University of California Press. Frege, Gottlob [1893/1903] 1966. Grundgesetze der Arithmetik. Hildesheim: Georg Olms Verlagsbuchhandlung. Frege, Gottlob [1884] 1968. The Foundations of Arithmetic. Evanston, Ill.: Northwestern University Press. Frege, Gottlob 1980. Philosophical and Mathematical Correspondence. Oxford: Basil Blackwell. Frege, Gottlob 1984. Thoughts. In: Brian McGuinness (ed.), Gottlob Frege: Collected Papers on Mathematics, Logic and Philosophy. Oxford: Blackwell, 351-372. Frege, Gottlob 1992. Philosophical Writings of Gottlob Frege (3rd edition). Oxford: Blackwell. Quine, Willard Van Orman [1960] 2001. Word and Object. Cambridge, Mass.: Harvard University Press. Quine, Willard Van Orman 1969. Relativity and Other Essays. New York: Columbia University Press. Quine, Willard Van Orman 1961. Two Dogmas of Empiricism. In: Willard Van Orman Quine, From a Logical Point of View (2nd edition). Cambridge, Mass.: Harvard University Press, 20-46.

475 Quine, Willard Van Orman 1990. Pursuit of Truth, Cambridge, Mass.: Harvard University Press. Wittgenstein, Ludwig [1922] 1955. Tractatus Logico-Philosophicus (translated by C. K. Ogden). London: Routledge & Kegan Paul. Wittgenstein, Ludwig [1953] 1978. Philosophical Investigations. Oxford: Basil Blackwell.

Ali Saboohi Institute for Research in Fundamental Sciences, Tehran [email protected]

In Defense of Implicit Knowledge in a Full-Blooded Theory of Meaning∗ Abstract: I defend Dummett’s “full-blooded” theory of meaning against McDowell’s objections. In particular, I argue that McDowell’s objections against Dummett’s invocation of the notion of implicit knowledge – the central notion in a full-blooded theory – cannot be sustained. Key words: theory of meaning, modesty, full-blooded, psychologism, implicit knowledge, rule-following considerations, judgement-dependence

0. Introduction The central task of the philosophy of language is to illuminate or explain the concept of meaning (sense).1 There is consensus among the key figures in philosophy of language – namely Dummett, Davidson and McDowell – that the best way to achieve this goal is to construct a theory of meaning. “Meaning,” on this account, is what a theory of meaning is a theory of.2 These philosophers diverge, however, when it comes to the issue of how a theory of meaning should be constructed. Dummett suggests that an appropriate theory of meaning would be a “full-blooded” one, meaning that a theory of meaning is not allowed to ∗

1

2

I would like to thank Alex Miller, Bill Brewer and Hamid Vahid for their comments on an earlier version of this paper The distinction between sense and reference is Frege’s most famous thesis in the philosophy of language. In Frege’s conception, the sense of an expression is its “cognitive value” and is what is grasped by the speaker when she understands an expression; so the meaning of an expression can be equated with its sense. Throughout the paper I use “sense” and “meaning” interchangeably. See McDowell (1998: 3) and Dummett (1993: 1).

478 “take as already given any notions a grasp of which is possible only for a language speaker” (Dummett 1991: 13). In other words, meaning must be approached from “outside the content” and language. McDowell, in contrast, believes that the only possible way to form a theory of meaning is “a modest” one, from within the language. There has been ongoing debate between the proponents of fullbloodedness and modesty.3 The aim of this paper is to defend fullbloodedness in a theory of meaning against some objections raised by McDowell. In particular, I argue that McDowell’s worry that a fullblooded theory is impossible to construct is misplaced. I will state McDowell’s objections against the notion of implicit knowledge in a full-blooded theory and argue that they cannot be sustained. 1. What is a full-blooded theory of meaning? Dummett sees the requirement of full-bloodedness in a theory of meaning as being grounded in Frege’s work in the philosophy of language, particularly in Frege’s criticism of psychologism in logic and language. Frege robustly believed that meaning is not to be associated with mental images, pictures and ideas (or, in general, psychological items). Frege’s hostility to psychologism lies in the fact that psychological items are subjective and private to the holder, whereas meaning is objective and communicable. Frege saw this as telling against views which tend to identify meaning with psychological items.4 In his several inspiring works on the interpretation of Frege’s philosophy of language, Dummett suggests that anti-psychologism – as the most fundamental principle of analytic philosophy – implies that language must take priority over thought in terms of explanation: 3

4

Dummett holds that the debate between modesty and the full-bloodedness has a significant and decisive impact on the realism/anti-realism debate in metaphysics. For further discussion, see “What is a Theory of Meaning? (II)” in Dummett (1993). The identification of the meaning of an expression with mental items such as ideas or images is associated with the British Empiricists, especially Locke. See Locke (1975: Book III).

479 Historically, the common principle uniting all the very diverse versions of analytical philosophy has been the priority of language over thought in the order of explanation. The fundamental ground of this priority thesis lies in Frege’s doctrine …that a thought, expressible by sentences, and the senses of the component words, which are constituents of that thought, are not content of consciousness. (Dummett 1991a: 315)

But, one might wonder, if the meaning of expressions and sentences are not to be characterized by mental items, how then can they be explained at all? Frege’s suggestion in his famous paper “Thought” (1918) is to associate the meaning (senses) of assertoric sentences with what he calls “thought”. In Frege’s conception, thoughts are neither things in the external world nor ideas in the mind of speakers. Rather, they are inhabitants of a mysterious third realm. This account, however, is silent as to how the speaker grasps the meaning of expressions as inhabitants of a third realm. Dummett, though, suggests that Fregean meaning should be taken as “an ability to determine the truth-value of sentences” (Dummett 1991: 278). When someone is credited with the knowledge of the meaning of a sentence she must be able to say whether the sentence is true or false.5 Accordingly, a theory of meaning must set itself the task of describing the speaker’s outward linguistic behaviour. Thus a full-blooded theory describes the ability in which competence with a concept – in this case, square – consists as follows: At the very least, it is to be able to discriminate between things that are square and those that are not. Such an ability can be ascribed only to one who will, on occasion, treat square things differently from things that are not square; one way, among many other possible ways, of doing this is to apply the word “square” to square things and not to others. (Dummett 1993: 98)

5

Thus Dummett disagrees with the “Descriptive” theory of meaning which has it that the meaning of an expression is associated with a definite description. In fact, according to Dummett, Frege never gave the details of a theory of meaning (sense). Rather, he showed the meaning of expressions by saying what the referent is. See Dummett (1973: 227).

480 How can a full-blooded theory of meaning manage to avoid psychologism? A full-blooded account, in fact, aims to explain what it is to grasp a concept without using that concept within the scope of a content-specifying “that-” clause. More precisely, in a full-blooded theory of meaning the grasp of a concept is not articulated by clauses such as “know that”.6 And as a result no relation is presupposed between the speaker and concepts. McDowell, for reasons we will explore below, claims that a fullblooded theory is not possible at all. Before we deal with McDowell’s objections to full-bloodedness, however, we need to look at some other aspects of a full-blooded theory. 2. Full-blooded theory, implicit knowledge and the manifestation It was remarked above that a full-blooded theory aims to characterize meaning “as from outside the content” by describing the speaker’s linguistic behaviour. But one might object that the full-blooded theory does not incorporate the role of mind in a theory of meaning. If this is so then the threat of behaviourism – an account that aims only to systematically predict the speaker’s linguistic behaviour and regularities – will loom large. Dummett, however, has clearly eschewed behaviourism in philosophy of language: I have many times remarked that a theory of meaning is not to be assessed as a scientific systematisation of regularities in complex phenomena. (Dummett 1987: 260)

A theory of meaning, according to Dummett, must describe the use of language as a rational activity and this is not achieved unless a theory of meaning is viewed as a theory of understanding: 6

A modest theory, in contrast, describes the grasp of a concept in a way that involves using that concept within the scope of a content-specifying that-clause. Recalling that psychologism equals the priority of thought over language, the relation between psychologism and modesty is evident; a modest theory of meaning presupposes the speaker’s grasp of a concept and thereby tries to explain linguistic meaning on the basis of the content of that concept.

481 I will repeat what I have maintained elsewhere, that a theory of meaning is a theory of understanding; that is, what a theory of meaning has to give an account of is what it is that someone knows when he knows the language, that is when he knows the meanings of the expressions and sentences of the language. (Dummett 1993: 3)

So the primary aim of a theory of meaning is not to give predictions about the speaker’s linguistic behaviour; rather, it is to capture the speaker’s knowledge of language. In Dummett’s conception, however, knowledge of language is implicit. Dummett remarks that if the speaker had explicit knowledge of a theory of meaning of her language, then she would be able, in principle, to manifest that knowledge. But a verbal manifestation presupposes mastery of language; and to state the content of knowledge of language one needs to be a master of that language, which is obviously circular. Dummett thus proposes that the knowledge of language must be viewed as implicit knowledge: the speaker’s knowledge of language, according to Dummett, is implicit knowledge and cannot be manifested explicitly.7 There might be a temptation to object that the invocation of the notion of implicit knowledge in a full-blooded theory might bring psychologism back; if the speaker is not able to express her knowledge of language verbally, then the communicability of meaning will be violated. To circumvent this problem Dummett adds what he calls the 7

Evans, however, has argued that knowledge of language cannot be construed as a genuine propositional attitude or intentional state. Evans argues that a genuine propositional attitude is necessarily “at the service of many distinct projects,” meaning that a propositional attitude can interact with others of one’s beliefs and desires to produce new beliefs and desires. But implicit states of speakers, according to Evans, violate this constraint; “[implicit states are] exclusively manifested in speaking and understanding a language; the information is not even potentially at the service of any other project of the agent, nor can it interact with any other beliefs of the agent … to yield further beliefs” (Evans 1981: 133). Nevertheless, as Wright (1986: 227-228) points out, Evans’ objection does not apply to implicit states of theorems which govern the use of a whole sentence, but only to states of axioms of a theory which govern the use of words. For an interesting attempt to show that implicit states of words can be taken as intentional states, see Weiss (2004).

482 “manifestation principle”: according to this principle, the speaker must manifest her implicit knowledge of language in use. The meaning of … a statement cannot be, or contain as an ingredient, anything which is not manifested in the use made of it, lying solely in the mind of individual who apprehends that meaning. (Dummett 1978: 216)

The manifestation principle will block psychologism because the speaker will manifest her implicit knowledge in behaviour. In sum, the invocation of implicit knowledge enables Dummett to steer between the twin evils of psychologism and behaviourism. A full-blooded description of outward behaviour, therefore, should not be taken as a mere description of the behaviour of the speaker. Rather, it is a manifestation of implicit knowledge of language. On the other hand, the speaker’s implicit knowledge will not remain hidden from view, because the speaker will manifest it in her behaviour. McDowell beautifully summarizes the bulk of the full-bloodedness account as follows: Psychologism is avoided because the implicit knowledge is (in, so to speak, its outward aspect) manifested in behaviour; nothing is hidden from view. But there is no lapsing into mere behaviourism, because the implicit knowledge is (in, so to speak, its inner aspect) capable of acknowledgment by speakers as what guides their practice. (McDowell 1987: 95)

3. McDowell’s objections against full-bloodedness McDowell has raised some objections, on the basis of Wittgenstein’s rule-following considerations, against a full-blooded theory of meaning. The key feature of the objections concerns the insufficiency of the behavioural manifestations in determining the content of the speaker’s knowledge as guiding her linguistic behaviour. Wittgenstein proceeds by considering cases in which someone is to extend a series of numbers.8 Suppose you teach a pupil to expand the series of natural numbers. He masters it and now can expand a series, 8

Wittgenstein (1974: §185).

483 say “+2,” as follows; 2, 4, 6, 8 … 1000. Now we get him to continue the series beyond 1000. Because you have set him to follow the rule of addition “+,” it is quite obvious that you expect him to come up with 1002, 1004. Strangely, he writes down 1004, 1008, 1012. At first you might think that he has made a mistake in his operation, i.e. he has forgotten what he had been doing before reaching 1000 and so went on in a different way. Thus you might ask him to look back and reflect on how he started the series. To your surprise, he returns the answer: “But I went on in the same way”. The way he has expanded the series, he goes on, is exactly, what he initially meant. In fact, though, he has gone on in a radically different way from the one we intended. So, any rule which we set someone to follow may be applied by him at some stage in manner both consistent with his past application of it and other than that which we intended. So no pattern of behaviour can be taken as the manifestation of the knowledge of a specific rule.9 McDowell writes: Now the point from Wittgenstein is that for any pattern to which a stretch of linguistic behaviour can be made out to conform, as described in terms that function below the level of explicitly specifying meanings as such, there are always other patterns that the same stretch of behaviour can equally be made out to exemplify. (McDowell 1998: 115)

The lesson of the rule-following considerations, according to McDowell, is that adopting a perspective outside language and constructing a fullblooded theory is not possible.10 Does McDowell’s Wittgensteinian observation undermine the notion of full-bloodedness? One Dummettian answer to the rule-following objection might take issue with the speaker’s ability to “acknowledge” 9

10

It is worth noting that McDowell does not take the Wittgensteinian point to undermine Dummett’s idea that meaning is open to view in linguistic behaviour. McDowell embraces Dummett’s rejection of psychologism. Nevertheless, he holds that it is modesty and not full-bloodedness which satisfies the requirement of openness of view. McDowell suggests that the rule-following issue implies modesty because the only possible way to make sense of the speaker’s behaviour as obeying the instruction to “add 2” is to be a member of a linguistic community. See McDowell (1998: 115).

484 the correctness of the formulation of a theory of meaning. Dummett remarks that although one might not be able to formulate for herself the knowledge on the basis of which she acts, she might be willing to acknowledge the correctness of a statement of that knowledge when is represented to her.11 What makes this suggestion plausible is that a subject has a special authority about her own mental states. Normally, a subject’s knowledge of her own mental states is groundless and noninferential: beliefs about one’s own mental states are not usually inferred on the basis of other evidence; it does not make sense to ask for reasons for a subject’s belief about her own mental states. Nevertheless, there is another feature of understanding which seems to be incompatible with a first-person epistemology: an ascription of understanding to a speaker has to answer to how that speaker would go on to apply that expression. In other words, an ascription of meaning is “disposition-like.” These two features of understanding do not seem to be reconcilable. Wright, however, in his discussion of the rule-following considerations,12 has outlined a plausible account of how the two features sketched above can be reconciled. Wright suggests that our linguistic judgement, under cognitively best conditions (C-conditions), does determine the extension of a concept and does not merely track it. Wright names this account the judgement-dependent account of meaning. To determine whether a given class of disputed concepts is judgement-dependent, Wright introduces a provisional equation: (x) (C→(A suitable subject S judges that Px iff Px) If the provisional equation for a property, say P, satisfies the following conditions, then P will turn out as judgement-dependent: (1) The a Prioricity Condition: The provisional equation must be a priori true.

11 12

See Dummett (1993: 96). Wright (1989). Page references are to the reprint in Wright (2001).

485 (2) The Substantiality Condition: The C-condition must be specifiable non-trivially. (3) The Independence Condition: The question as to whether the C-conditions obtain in a given instance must be logically independent of the class of truths for which we are attempting to give an extension-determining account. (4) The Extremal Condition: There must be no better way of accounting for a priori covariance of best opinions and facts. Colour concepts straightforwardly satisfy the conditions (1)-(4) and thus they can be viewed as judgement-dependent concepts. Shape, however, cannot be taken as judgement-dependent as the Independence Condition is violated (since we must presuppose that no change in the shape of an object occurs during the observations). Can meaning be taken as judgement-dependent? If there is an a priori and non-trivial true provisional equation for meaning, which can also meet the Independence and Extremal Conditions, then meaning could be taken to be judgement-dependent.13 Crispin Wright has suggested that the judgement-dependent account of meaning constitutes satisfactory ground for answering the rule-following problem. In particular, Wright asserts that a judgement-dependent account of meaning can accommodate both the features of first-person epistemology of understanding and its likeness to dispositions. The first-person authority 13

The satisfaction of the Substantiality Condition (2) and the Independence Condition (3) in the case of meaning are controversial: in order to form the best judgement about her understanding, a subject must be free of self-deception. But including a clause among the C-conditions stipulating that the subject is not selfdeceived is just the sort of insubstantial, whatever-it-takes formulation which condition (2) was meant to exclude. However, Wright suggests that no selfdeception is positive presumptive: “such is the ‘grammar’ of ascription of [understanding], one is entitled to assume that a subject is not materially selfdeceived … unless one possesses no positive evidence to the contrary” (Wright 1989: 201). For an interesting criticism of this treatment, see Miller (2008). See also Boghossian (1989) for a criticism of the Independence Condition.

486 of subject about her own understanding is explained by the fact that her best judgement on the matter conceptually determines the facts about her understanding, while the “disposition-like” feature is explained by the fact that her best judgement conceptually determines whether a current piece of behaviour implements an understanding. What is the bearing of the judgement-dependent account on the fullbloodedness of meaning? The judgement-dependent account of meaning is a full-blooded one because its independence condition states that whether C-conditions obtain in a particular case must be logically independent of facts about the subject matter in question. So we can say that the Independence Condition in the case of a meaning property prohibits making assumptions about semantic facts in determining whether the C-conditions obtain (so it could be viewed as from outside the content). Furthermore, the judgement-dependent account of meaning is not a reductive account since the concept in question – in this case, meaning – appears on both sides of the conditional. In sum, the judgement-dependent account of meaning can answer the rule-following problem “as from outside of the content,” without assuming that meaning is reducible to behavioural facts. We should now move on to address McDowell’s second objection against acknowledgment. McDowell’s second objection to the speaker’s ability to “acknowledge” the correct formulation of the theory of meaning concerns an alleged circularity in the full-blooded explication of possessing the concept square in Dummett’s example. McDowell says that the word “square” is used in Dummett’s explanation of grasping a concept. But if the speaker is going to acknowledge the correctness of the formulation of the theory of meaning, then she needs to understand the theory, and this means that a grasp of the concept in question is presupposed by Dummett: Comprehension of this object of acknowledgment would be an exercise of the very capacity we were trying to see as guided by the implicit knowledge that the acknowledgment supposedly reveals. (Dummett 1987: 95)

This objection seems to be confusing two different sorts of circularity: epistemic circularity and analytic circularity. To spell this out, it is

487 helpful to have a brief look at the Gricean account of meaning and the alleged charge of circularity against it. Grice (1989), in contrast to Dummett, suggests that thought should be considered as prior to language in the project of constructing a theory of meaning. The notion of linguistic meaning (or “sentence–meaning,” in Gricean terminology), according to Grice, should be analysed in terms of mental content (“speaker’s-meaning,” in Gricean terminology). So, according to Grice, to construct a theory of meaning we should proceed by explaining speaker’s meaning in terms of her intentions and then move on to explain sentence meaning in terms of the speaker’s meaning. Grice says, however, that there are some conditions that a speaker’s intentions must fulfil in order to convey meaning by an utterance: first, the utterer must intend to induce the belief that p in the hearer by uttering ‘p’. Second, she must intend the hearer to recognize the intention behind her utterance, and finally the hearer’s recognition of the utterer’s intention must play a part in the explanation of why he forms the belief that p. Once these conditions are met, we can say that the speaker meant something by p. Some writers, however, have objected that Grice’s account of speaker-meaning suffers from circularity. Mark Platts remarks that to understand what intention a person has we must know what expressions of his language mean. But this, Platts says, renders the Gricean account circular: Any explanation of how such intentions are recognized will inevitably rely upon the audience’s recognition of the literal meaning of the sentence; that meaning is the route to speaker’s intentions, the reverse journey usually being implausible. (Platts 1979: 91)

Platts’ point is that one cannot know what intentions the speaker has without knowing the literal meaning of her expressions. So Grice’s attempt to explain sentence meaning in terms of the speaker’s meaning is, in fact, circular. But, as several writers have pointed out, there is a confusion in Platts’ remarks: the circularity Platts detected in the Gricean account is, in fact, an epistemic circularity and not an analytic circularity. To flesh this out we can borrow two notions from Martin

488 Davies: analytic priority versus epistemic priority. Davies (1996) defines analytic priority as follows: To say that the notion of X is analytically prior to the notion of Y is to say that Y can be analysed or elucidated in terms of X, while the analysis or elucidation of X itself does not have to advert to Y. (Davies 1996: 96)

And epistemic priority is characterized as follows: To say that X is epistemologically prior to Y is to say that it is possible to find out about X without having to proceed via knowledge about Y, where finding out about Y has to go by way finding out about X. (Davies 1996: 96)

What is the connection between analytic priority and epistemic priority? As Miller remarks, “analytic priority is one thing, epistemological priority another: the two notions are logically independent” (Miller 1997: 260). Miller makes this clear by an example. He notes that although the notion of personal identity should be analysed in terms of the notion of physical continuity, it does not follow from this that when I meet you in the street I need to know these physical facts in order to know you are the same person I was talking to yesterday. In this light, the kind of priority Grice has adverted to is an analytic priority, and this does not bring commitment to an epistemic priority. Platts’ confusion results from failing to realize this point. What is the bearing of this point on McDowell’s circularity objection? In the same way, McDowell’s confusion seems strikingly similar to Platts’ misunderstanding: McDowell fails to see that what is definitive in Dummett’s account is analytic priority: to define meaning without adverting to prior understanding of concepts. But this does not necessarily make Dummett committed to the epistemic priority of language over thought which says that we should find out the theory of meaning of a speaker without having to proceed via the speaker’s understanding and acknowledgment of a theory.

489 4. Concluding remarks I have argued that a full-blooded theory of meaning is not undermined by McDowell’s attacks on the notion of implicit knowledge. It is possible, contrary to what McDowell says, to form a non-behaviouristic theory of meaning as from outside the content. Wright’s judgementdependent account of meaning outlines this. I also argued that the usage of a word in a full-blooded description of meaning does not give rise to circularity; a full-blooded theory is only committed to the priority of language over thought and this is different from the epistemic priority which might bring with it a problem of circularity. References Beaney, Michael (ed.) 1997. The Frege Reader. Oxford: Blackwell Publishers. Boghossian, Paul 1989. The Rule-following Considerations. Reprinted in: Alexander Miller and Crispin Wright (eds.), Rule-Following and Meaning. Chesham: Acumen, 2002, 141-187. Davies, Martin 1996. Philosophy of Language. In: Nicholas Bunnin and Eric P. Tsui-James (eds.), The Blackwell Companion to Philosophy. Oxford: Blackwell Publishers, 90-146. Dummett, Michael 1973. Frege: Philosophy of Language. London: Duckworth. Dummett, Michael 1975. What is a Theory of Meaning? (I). Reprinted in Dummett 1993, 1-33. Dummett, Michael 1976. What is a Theory of Meaning? (II). Reprinted in Dummett 1993, 34-83. Dummett, Michael 1978. What Do I Know When I Know a Language? Reprinted in Dummett 1993, 94-105. Dummett, Michael 1991a. Frege and other Philosophers. Oxford: Oxford University Press. Dummett, Michael 1991b. The Logical Basis of Metaphysics. London: Duckworth. Dummett, Michael 1993. The Seas of Language. Oxford: Oxford University Press. Evans, Gareth 1981. Semantic Theory and Tacit Knowledge. In: Steven Holtzmann and Christopher Leich (eds.), Wittgenstein: To Follow a Rule. London: Routledge, 118-137. Grice, Paul 1989. Meaning. In: Studies in the Ways of Words. Cambridge, Mass.: Harvard University Press, 213-223. Locke, John 1975. An Essay Concerning Human Understanding, edited by P. Nidditch. Oxford: Oxford University Press.

490 McDowell, John 1984. Wittgenstein on Following a Rule. Reprinted in McDowell 1998b, 221-261. McDowell, John 1987. In Defense of Modesty. Reprinted in McDowell 1998a, 87107. McDowell, John 1998. Meaning, Knowledge and Reality. Cambridge, Mass.: Harvard University Press. Miller, Alexander 2007. Philosophy of Language, 2nd ed. London: Routledge. Miller, Alexander 2008. Primary Qualities, Secondary Qualities and the Truth about Intention. Synthese 171, 433-442. Platts, Mark 1979. Ways of Meaning. London: Routledge and Kegan Paul. Weiss, Bernhard 2004. Knowledge of Meaning. Proceedings of the Aristotelian Society 104, 75-94. Wittgenstein, Ludwig 1974. Philosophical Investigations. Oxford: Basil Blackwell. Wright, Crispin 1986. Theories of Meaning and Speakers’ Knowledge. In: Wright Realism, Meaning and Truth. Oxford: Blackwell, 204-238. Wright, Crispin 1989. Wittgenstein’s Rule-Following Considerations and the Central Project of Theoretical Linguistics. In: Alexander George (ed.), Reflections on Chomsky. Oxford: Blackwell Publishers, 233-264. Wright, Crispin 2001. Rails to Infinity. Cambridge, Mass: Harvard University Press.

Arthur Sullivan Memorial University of Newfoundland [email protected]

On Pragmatic Regularities Abstract: The aim of this essay is to help illuminate a relatively neglected aspect of a Gricean picture of the semantics-pragmatics interface: namely, the notion of a pragmatic regularity. (Another term which Bach (2004, among other places) has used to designate this phenomenon is “standardized non-literality”.) The guiding idea is that better charting the notion of a pragmatic regularity will have some significant effects on some ongoing debates. Key words: semantics, pragmatics, Grice, Bach, Devitt

0. Introduction A pragmatic regularity is a kind of pragmatic implication (i.e., carried by the act of using a particular expression in a particular context, as opposed to being constitutive of what is semantically expressed) which is nonetheless systematic and regular. In being systematic and regular, pragmatic regularities are clearly distinct from one-off, thoroughly context-dependent particularized conversational implicatures (PCIs). They are also distinct from such similar, more familiar phenomena as conventional implicatures,1 idioms and dead metaphors, and indirect speech acts. Rather, they are just simply common patterns of usage which regularly convey a certain implicature, independent of what is semantically expressed. Paradigmatic examples of pragmatic regularities include Gricean generalized conversational implicatures (GCIs) (e.g., ‘X is meeting a woman this evening’), rhetorical uses of interrogatives (e.g., ‘Who do you think you are?’) and certain kinds of standardized irony (e.g., ‘Way 1

Bach (1999) has questioned whether CIs exist at all, but for present purposes I will follow Potts (2005) in reading Bach as arguing for certain theses about CIs.

492 to go, Einstein!’). Many take these kinds of phenomena to be precisely the sorts of considerations which motivate alternative, non-Gricean approaches to the semantics-pragmatics interface (such as relevance theory, or some variety of semantic contextualism). However, this essay is crafted as a development within the Gricean framework. While, in this context, I cannot engage in a global battle against relevance theory or contextualism, in the first section I will recapitulate an important contribution of Grice’s, which is still entirely relevant to many active contemporary debates at the semantics-pragmatics interface. In the second section I work through a few different examples of pragmatic regularities, toward the aim of sketching a tenable and significant account of the category. The third section is focused on one open debate in which the notion of a pragmatic regularity plays a key role – namely, the debate between Bach (2004, 2007) and Devitt (2004, 2007) over whether the phenomenon of referential uses of descriptions is amenable to a pragmatic explanation. In the final section, I draw out some general morals. 1. Framing the issues (within a neo-Gricean framework) Grice’s systematic factoring of what is communicated with a use of a sentence (in context) into what is semantically expressed and what is pragmatically implicated was an epochal development in the philosophy of language, which has also helped to shape related areas of linguistics and psychology.2 By ‘neo-Gricean’, I mean to designate the variety of 2

Here I depart from Grice’s terminology, in employing the following terms to characterize Grice’s factorization: WHAT IS COMMUNICATED with a use of a sentence (in context) is a function of two factors – (i) WHAT IS SEMANTICALLY EXPRESSED, and (ii) WHAT IS PRAGMATICALLY IMPLICATED. Along with many others nowadays, I find ‘WHAT IS SAID’ to be hopelessly vague and ambiguous – while Grice was exceedingly consistent in his usage of the expression, unfortunately one finds it subsequently used in different ways by different theorists. I will also make no use of the notion of ‘conventional meaning’, as my main focus here is on a kind of pragmatic implication that is systematic and regular, and so ‘conventional’ in some senses of the term.

493 views which accord a central theoretical place to (some version of) Grice’s factorization.3 Neo-Griceans hold that semantics and pragmatics are distinct and discrete (though intimately interrelated) channels involved in linguistic communication. To cite one paradigmatic example, Grice’s (1975) famous letter of reference semantically expresses that Mr. X has an excellent command of English and a good attendance record, while pragmatically implicating that Mr. X is no good at philosophy. Neo-Griceans, thus understood, may depart from the letter of Grice’s views on many specific points; and the varieties of neo-Gricean views are rather heterogeneous, in several respects. Some examples of points of contention within the neo-Gricean ranks include whether conventional implicatures should be counted as truth-conditionally irrelevant, whether metaphors can or should be understood as PCIs, or how much stock should be put in the classic Gricean indicators for pragmatic implications (e.g., cancelability, calculability). (There is further discussion of the status of these Gricean indicators below.) Nonetheless, the term ‘neo-Gricean’ is not so broad as to include all parties to the debates at the semantics-pragmatics interface. In particular, the centrality of Grice’s factorization has been challenged by relevance theorists and semantic contextualists. One core objection is that the entire factorization itself is a drastic oversimplification and an untenable relic from a bygone era. (As Borg (2007: 340) puts it, non-neo-Griceans hold that “pragmatic effects are endemic throughout the literal, truthevaluable content expressed by sentences”, or that “pragmatics infects semantic content in a substantial way.”) Some good reasons have emerged to question whether semantics and pragmatics are, in general, discretely separable.4 Notwithstanding all that, for several reasons, this essay is crafted as a development within the neo-Gricean framework. For one thing, the case 3

4

Examples of neo-Griceans, thus understood, include Kripke (1977), Neale (1990, 2007), Levinson (2000), and Horn (2004). Of particular relevance to Section 3 below is that neo-Griceanism is part of the commonly conceded ground between Bach (2004, 2007) and Devitt (2004, 2007). Recanati (2004) is a good recent survey of them.

494 study in which I am interested in Section 3 takes place within that framework. More deeply, though, I think that Grice’s factorization is a great leap forward in our understanding of linguistic communication, and that many moves made by relevance theorists and contextualists run afoul of the core, basic insights on which it is founded. Part my aim here is to further develop (and hence, ultimately, to shore up) the neo-Gricean stance. The general picture is tailored to account for semantic regularities and pragmatic happenstance; but what of pragmatic regularities? How would such phenomena fit in? What theoretical work could or should this notion do? To round off this opening section, I should develop at some length what I mean by the epochal impact of Grice’s factorization for the philosophy of language. It immediately follows from Grice’s factorization that not all differences in what is communicated by two expressions E1 and E2 (in context) should be counted as, or traced to, a difference in what E1 and E2 semantically express. (A key sub-case is where E1 = E2, in which case the question is whether the relevant expression should be understood as ambiguous.) This was a devastating blow to some strands within the ordinary language tradition in philosophy, which fed on case-by-case cataloguing of subtle distinctions between what is communicated via different uses of certain terms.5 More generally, though, Grice’s factorization has a drastic and across-theboard effect on how philosophers can go about motivating conclusions about meanings. I’ll call this ‘Grice’s challenge’ (i.e., differences in what is communicated do not suffice to show differences in what is semantically expressed). It is my view that many contemporary strands within 5

A case in point is Strawson’s (1952) argument that ‘and’ is ambiguous, based on the consideration that ‘P and Q’ (in context) sometimes communicates temporal order, sometimes communicates a causal relation, sometimes communicates neither such relation, etc. In effect, Grice’s factorization demonstrates why it is that this kind of argument is hopelessly incomplete – one has to address the alternative of a difference in pragmatic implications, in order to shore up the gap between premises and conclusion. Other instances of what Grice does here for the case of ‘and’ include Grice’s (1975) discussion of ‘a’ (cf. Section 2 below), and Kripke’s (1977) work on ‘the’ (cf. Section 3 below).

495 relevance theory and semantic contextualism run afoul of Grice’s challenge; and, as a consequence, are guilty of positing implausibly many distinct meanings. (That is, relatives of Strawson’s (1952) ill-fated arguments concerning ‘and’ live on in the guise of referential meanings, pragmatic meanings, metaphorical meanings, etc.) It is of course controversial exactly how much weight should be afforded to such general methodological principles as Ockham’s semantic razor, but still some fundamental considerations about the learnability and systematicity of meanings (among other things) impose considerable pressures against their proliferation. (Another consequence of Grice’s factorization is that sameness in what is communicated by E1 and E2 (in context) does not suffice to show synonymy (i.e., sameness in what is semantically expressed). I will call this ‘the reverse-Grice point’. While this latter point lacks for the same kind of historically significant splash as the former, it is my view that certain positions on the contemporary landscape run afoul of it. In this latter case, the charge is the opposite of the above, too-manymeanings charge; it is rather that important semantic distinctions get conflated.6) 6

Consider, for example, a line of argument against the Russellian approach to definite descriptions, developed by Schiffer (1995, 2005), and dubbed by Neale (2004) ‘the argument from psychological parity’. The idea here is that, in context, ‘The woman is insane’ and ‘She is insane’ (say) communicate exactly the same content; and hence that ‘the woman’ has as strong a claim to conventional referring status as does ‘she’. A second case concerns some of King’s (2001, 2008) arguments in favor of a quantificational analysis of complex demonstratives – specifically, what King calls ‘NDNS’ uses (short for ‘no demonstration, no speaker reference’) such as “That student who scored 100% must be a genius”. Here, again: the premise that such uses of demonstratives are communicatively equivalent to definite descriptions is taken to be a key reason in favor of their semantic equivalence. At least prima facie, both moves might fall prey to the reverse-Grice point. Just as difference in what is communicated does not entail a failure of synonymy, sameness in what is communicated does not suffice to establish synonymy. In both cases, all things considered, the notion of a pragmatic regularity may afford the best explanation of the data in question (in conjunction with the claims that, contra Schiffer, pronouns are referring expressions but definite descriptions are

496 To sum up, then: In the wake of Grice’s factorization, it is more difficult to draw any conclusions, back or forth, between what is communicated and what is semantically expressed. One cannot establish synonymy merely on the basis of sameness in what is communicated, nor can one refute a claim to synonymy by pointing to a difference in what is communicated. One would have to address the alternative of a pragmatic explanation, in order to justify any such conclusions. The aim of this paper is to work towards illuminating one aspect of this broadly Gricean picture: namely, the notion of a pragmatic regularity. While this notion plays at least an implicit role in some ongoing debates at the semantics-pragmatics interface, it deserves a more extended treatment than it has yet been given (at least, to my knowledge). The guiding idea is that charting and further illuminating the notion of a pragmatic regularity might have significant affects on some ongoing debates. 2. What is a pragmatic regularity? Some work towards a theoretical account In this section I work through a few different examples of pragmatic regularities, with a view to sketching the general contours of the category. First off, I take the phenomenon of generalized conversational implicatures (GCIs) as a paradigm case of a pragmatic regularity. Here are two constructions which Grice (1975) uses to illustrate this notion: (1) X is meeting a woman this evening. (2) X went into a house. (1) would normally implicate that “the person to be met was someone other than X’s wife, mother, sister, or perhaps even close Platonic friend”, and (2) would normally implicate that the house is other than X’s own. Based on such cases, it might be said that a lack of familiarity, not, and that, contra King, complex demonstratives are referring expressions but definite descriptions are not).

497 ownership, or otherwise (context-relative) close relation is commonly communicated by the use of the indefinite article ‘a’. However, this is no part of what is semantically expressed by the indefinite article, as the following cases suffice to illustrate: (3) X broke a finger. (4) X has been sitting in a car all morning. This claim rests only on the following three (fairly safe) assumptions: (i) that (3) would normally implicate that X broke X’s own finger, (ii) that (4) is completely neutral as to whether or not the car was X’s own, and (iii) that there is nothing remotely non-literal or non-standard about these latter uses of ‘a’. Hence what (1) and (2) instance here is, though systematic and regular, a pragmatic matter (i.e., carried by the speaker’s use of the indefinite article in this particular context, as opposed to a matter of what is semantically expressed by ‘a’). Not surprisingly, we find these GCIs to carry such classic pragmatic hall-marks as calculability and cancelability.7 So, this phenomenon instanced by (1) and (2) is a case wherein something communicated (with a use of an expression, in context) seems to be pragmatically implicated, not semantically expressed. However, it clearly differs from one-off, thoroughly context-dependent PCIs in being 7

That is, it might be decidedly odd, but is no contradiction, to say: “I am meeting a woman this evening, and she is my mother.” Similarly, it is easy to imagine scenarios in which the GCI associated with (2) is canceled. Hence, GCIs are cancelable. As for calculability, the idea here is that pragmatic implicatures can be represented as the conclusion of a certain chain of reasoning, sometimes called a ‘Gricean derivation’. The key point is that semantic competence is necessary, but not remotely close to sufficient, for working out pragmatic implicatures. Extra-semantic calculation, in some form or other, is also required. (It is important not to take calculability as a claim about psycho-linguistic processing. As Bach (among others) has pointed out, “Grice did not intend his account of how implicatures are recognized as a psychological theory or even as a cognitive model. He intended it as a rational reconstruction” (2005: 8). Some objections to the Gricean picture may well be attacks on a straw target, for want of attention to this distinction.)

498 relatively systematic and regular. Hence, it should be understood as a pragmatic regularity. Now, what else might belong in such a category? The case of rhetorical questions provides a second candidate. Consider the following: (5) Why are you so lazy? What a use of (5) (in context) semantically expresses is a question about the cause of, or explanation for, the addressee’s laziness. Literal uses of (5) are thus relatively rare (e.g. a therapist might literally ask this of her patient). What is more common is that speakers use (5) rhetorically – i.e., to pragmatically implicate that the addressee is lazy. In general, and as is typical of rhetorical uses of questions, while what (5) semantically expresses is interrogative, what it pragmatically implicates (by asking that question in that context) is assertoric. So, prima facie, we have here a second case wherein something that is communicated (with a use of a sentence, in context) is at once relatively systematic and regular, while at the same time independent of what is semantically expressed. It is systematic and regular in that some questions (like (5)) especially lend themselves to this kind of usage, even to the extent that literal, interrogative uses of them are hard to motivate. Another such case is: (6) Who do you think you are? So, this phenomenon, too, seems rather distinct from a PCI – i.e., it would be hard to deny that there is a context-independent communicative regularity to use (5), (6), and etc. rhetorically. Again, there are good reasons to think of that regularity as pragmatic. Intuitively, it is using that question in that context that generates the implication in question; and, again, the implication is both cancellable and calculable. Certainly, one who wanted to argue that (5), (6), and etc. are semantically ambiguous between an interrogative meaning and an assertoric meaning would be subject to pretty rough handling. There would be serious questions both pertaining to the theoretical motivation for these putative assertoric meanings, and pertaining to the boundaries of the category.

499 To the contrary, regardless of how common the association between (5), (6), or etc. and its implication has become in the minds of speakers, still the implication is carried by the act of using that expression (in a specific context), not simply by what is semantically expressed. Here again, as in the case of ‘and’ and ‘a’, the neo-Gricean stance seems vastly preferable to the proliferation of meanings. Rhetorically-used questions seem to fall naturally into the category of pragmatic regularities, along with GCIs. I will briefly discuss a third type of example, not because I think that it exhausts the category of pragmatic regularity, but rather to illustrate exactly the opposite: i.e., the notion of a pragmatic regularity is bound to be quite widely applicable, should its legitimacy and interest be conceded. The following brief discussion is tentative because it involves irony, and the question of the correct theoretical explanation of irony (and of how irony intersects with the semantics-pragmatics interface) is a complex and thorny matter, which (as far as I know) has not received much in the way of sustained philosophical treatment. Nonetheless, despite all that, this next case does help to illustrate some of the general confines of the category of pragmatic regularity, and hence helps to ward off the suggestion that the phenomenon instanced by the above cases is isolated and circumscribed. Consider an utterance of the following, in a context in which a mistake or accident has just occurred: (7) Way to go, Einstein! Arguably, this instances two distinct kinds of pragmatic regularity. First, there is the ironic use of the compliment, to indirectly assert its opposite – compare: ‘Nice job, Einstein!’, ‘Good work, Einstein!’. Second, there is the non-literal, mocking use of a famous name – compare: ‘Check out Elvis over there’, ‘Finally, Socrates deigned to enlighten us with his brilliant opinions’.8 8

Note that this is a distinct phenomenon from what is often called an ‘inverted commas’ use of a name, such as when one humors a delusional man who thinks he is Napoleon by so-referring to him. Both kinds of use fit fairly well within the category of what Kripke (1977) calls ‘speaker reference’ (provided that we explicitly reject Kripke’s occasional suggestion that it is essential to the cases of

500 In any case, again, clearly we have here, as in the above cases, a communicative regularity. This is why I call this ‘standardized’ irony – as an empirical matter, non-literal, mocking uses of (7) are epidemic (in at least some contemporary dialects of English). But, also, I take it that it is just as clear that this should hardly be taken to show that either the compliment or the proper name is semantically ambiguous. Surely one does not want to say that the currency of these kinds of ironic, mocking uses of ‘Einstein’ (and ‘Elvis’, ‘Socrates’, etc.) has engendered the terms with a new semantic meaning. Apart from its sheer implausibility, that option would also bring in its train a host of difficult questions, concerning the precise boundaries of this distinctive kind of ambiguity. To the contrary, this kind of communicative regularity too is, on balance, best explained in terms of pragmatic implicature – i.e., the message is carried by the use of that sentence in that context, as opposed to by what (7) semantically expresses. All things considered, the claim that these names are semantically univocal, but that they are subject to a pragmatic regularity, seems much more satisfactory. So, the neo-Gricean has a straightforward explanation which covers many such cases. The distinction between what is semantically expressed and what is pragmatically implicated has long been evident; and it suffices to ground a simple, comprehensive account with potential to cover an indefinitely broad range. The fact that speakers often use questions (e.g., ‘Why are you so lazy?’) rhetorically, to make statements, does not mean that the interrogative form is semantically ambiguous; the fact that speakers often use compliments (e.g., ‘Way to go!’) sarcastically, to insult, does not show that the relevant sentences are semantically ambiguous; and the fact that certain famous names have come to be used in a certain mocking way does not show that they have become semantically ambiguous. This ties in with some aforementioned instances of Grice’s challenge (i.e., differences in what is communicated do not suffice to show differences in what is semantically expressed). In speaker/semantic reference divergence that the speaker and audience share a factual misconception as to the semantic reference). However, the inverted commas uses are more explicitly meta-linguistic.

501 each case, the literal semantic meaning is operative, and plays a role in the further post-literal speech act. The regularities in question are pragmatic ones, and so should not be understood as semantically encoded or determined. We are now in a position to see how it is that pragmatic regularities differ from such similar phenomena as conventional implicatures (CIs), idioms and dead metaphors (IDMs), and indirect speech acts (ISAs). For one thing, the distinctive communicative properties of CIs and IDMs are more clearly a function of the semantics of the expressions in question, as opposed to properly characterized as pragmatic phenomena. Relatedly, as distinct from CIs or IDMs, pragmatic regularities need not be lexically encoded, or traceable to (the semantics or pragmatics of) one distinguishable sub-component of the expression in question.9 (For example, while it makes sense to isolate questions about the semantics and pragmatics of (say) pejoratives or idioms – considered as independent sub-sentential lexical items – no such point holds, in general, across the range of pragmatic regularities.) Further, as distinct from ISAs, there need not be two items of information (or propositions) involved, let alone need there exist any regular sort of relation between the communicatively relevant bits of information. So, while many multiplicities of use are cases of semantic ambiguity, crucially, not all are. Some should be understood as pragmatic regularities; i.e., cases in which one expression with its univocal semantic properties has come to be used in diverse ways. Pragmatic regularities, then, may well constitute an interesting and useful pragmatic natural kind, which merits further investigation.10 9

10

The ‘need not be’ qualification is important – for some putative pragmatic regularities (e.g., those associated with ‘a’ discussed above and ‘the’ discussed below) might be seen as lexically encoded or triggered. The key point is that, even if that is so, that per se does not suffice to show that those putative pragmatic regularities should instead be classified as a semantic phenomenon, akin to CIs or IDMs. As mentioned above, Bach’s sometimes refers to this phenomenon as “standardized non-literality”. Cf. Bach (1995, 1998) for development, and (2004, 2007) for applications. It is crucial for Bach that no Gricean derivation (let alone a deliberate, conscious inferential process) is required in order to arrive at what is

502 3. A case study: One contemporary debate in which the notion of a pragmatic regularity plays a key role The notion of a pragmatic regularity plays a role in ongoing debates between Bach (2004, 2007) and Devitt (2004, 2007) over whether referential uses of definite descriptions admit of a satisfactory pragmatic explanation. I will use the term ‘RDs’ as an abbreviation for ‘referential uses of definite descriptions (principally, among some other quantified noun phrases)’. One of the points which are conceded in common by both Bach and Devitt (in addition to the commitment to the neo-Gricean framework, as already mentioned) is that while the phenomenon in question is most prevalent for the case of definite descriptions, it is also – to varying degrees – instanced by other quantified noun phrases. Next, I will sketch what I will call ‘the Russellian orthodoxy on RDs’. Second, I will describe some of Devitt’s (2004, 2007) criticisms of the Russellian orthodoxy. After that, I will describe Bach’s (2004, 2007) replies to Devitt on behalf of the Russellian orthodoxy, which appeal to the notion of a pragmatic regularity. The Russellian orthodoxy on RDs is the view that ‘the’ has one univocal meaning, i.e., roughly, the quantificational meaning articulated by Russell (1905), though as modified by subsequent developments in syntactic and semantic theory.11 (Especially important here are some developments within quantification theory, and the notion of a restricted quantifier.) The Russellian orthodoxy concedes that there are common

11

pragmatically implicated in these cases. This is one of the important differences between pragmatic regularities and PCIs. The Russellian orthodoxy is sketched by Grice (1969: 143) – as Grice puts it, despite admitting of both ‘identificatory’ and ‘non-identificatory’ uses, “descriptive phrases have no relevant systematic duplicity of meaning; their meaning is given by a Russellian account”. Kripke (1977) then develops this line of thought into a powerful counter-argument against anti-Russellians – in his (1977: 263) words, the phenomenon of RDs “shows nothing against a Russellian or other unitary account”, and so treating referential uses via “a general theory of speech acts” is, all things considered, preferable to positing a “semantic ambiguity” in the definite article. Neale (1990) and Bach (2004) add important amplifications and further developments. See Neale (2004) for more thorough discussions of the development of the view.

503 and significant non-quantificational (i.e., referential) uses of definite descriptions (as well as of various other kinds of quantified noun phrases). However, it holds that, all things considered, these should be understood as a pragmatic phenomenon (i.e., a kind of speech act), not by positing a distinct referential meaning for ‘the’. Basically, the idea is that certain of the putative criticisms of Russell (1905) raised by Strawson (1950) and Donnellan (1966) fall prey to Grice’s challenge – i.e., that what Grice does for ‘and’ and ‘a’ also applies to ‘the’. While it can scarcely be denied that there exist non-trivial differences between what sentences of the form ‘The F is G’ (in context) can be used to communicate, conceding this still falls well short of establishing anything along the lines of a semantic ambiguity. Second, as for Devitt’s views: while Devitt is hardly the first or only critic of the Russellian orthodoxy,12 he has helped to push the confines of the debate in a way that is relevant to the central interests of this paper. For one thing, he is one of the few to take a bold, categorical step beyond Donnellan (1966) – i.e., on Devitt’s (2004, 2007) view, ‘the’ is absolutely and unequivocally semantically ambiguous between quantificational and referential senses (like ‘bank’ is ambiguous between financial and geographical senses). Hence, there is a distinctively clear split between Devitt and the Russellians. Furthermore, Devitt has also contributed to the development of what many take to be a strong line of argument against the Russellian orthodoxy, which Neale (2004) dubs ‘the argument from convention’. The core idea here is that RDs are so “common, standard, regular, systematic, and cross-linguistic” (Neale 2004: 173), that it seems somewhat ad hoc and theoretically-driven to deny that they are a significant semantic phenomenon. (After all, what are semantic properties, if not ‘common, standard, regular, systematic’ properties of linguistic communication?) Devitt (2004, 2007) has been stubbornly demanding, when it comes, specifically, to the question of exactly how the Russellian orthodoxy’s pragmatic explanation of RDs is supposed to go. He has helped to force the recognition that Russellians are generally quite vague on the 12

In particular, many of the points discussed with reference to Devitt below were anticipated or developed by Reimer (1998).

504 mechanics. Is this supposed to be a PCI? A GCI? Some other sort of implication? If so, what, exactly? In addition, Devitt has helped to further cloud the issue of how we might determine whether or not RDs can or should be classified as a pragmatic phenomenon. For example, one upshot of his dead-metaphor argument (which, again, follows Reimer 1998) might be taken to be that calculability buys you nothing, because anything can be described as calculable. (It is a fairly trivial matter to sketch a Gricean derivation from more or less any P to any Q; and so the availability of such a calculation can hardly do much theoretical work.) The challenge for the Russellian orthodoxy, then, is to explain exactly how it is that, in the case of RDs, one specific singular proposition gets pragmatically implicated, while a distinct general, quantificational proposition is semantically expressed. Devitt is surely right, too, that there is not much in the way of specific directions in the classical Russellian orthodox corpus (cf. note 11) for how to meet this challenge. So, third, then: onto Bach’s (2004, 2007) responses to Devitt, in which the notion of a pragmatic regularity is employed in order to bolster the Russellian orthodoxy. One thing to observe right off the top is that, as mentioned above, proponents of the Russellian orthodoxy typically take this issue to be a battle in the global wars over the status and tenability of the neo-Gricean framework. A core idea motivating the Russellian view is that Strawson (1950), Donnellan (1966), and much of the subsequent work inspired by them, fall prey to Grice’s challenge. Thus, there is something idiosyncratic about Devitt’s conceding the neoGricean framework in general, but then arguing in favor of referential meanings for RDs. At the very least, many of his neo-Gricean allies will take Devitt to be putting himself in with some unsavory elements, when it comes to the unnecessary and unprincipled proliferation of meanings. Of course, according to Devitt, there is nothing unprincipled about this particular case of RDs, because of the force of such considerations as the argument from convention. So, one of Bach’s (2004) main contributions to this dialogue has been to counter that argument, from a Russellian perspective. Bach concedes that, to be sure, there is a much stronger case for classifying definite descriptions as referring

505 expressions, than there is for any other kind of quantificational expression. Definite descriptions especially lend themselves to referential uses, precisely because they semantically single out exactly one individual. When one wants to express object-dependent information about an unnamed individual, or about an individual whose name is not mutually known among present interlocutors, definite descriptions are often the best means available. Still, there is nothing here that a Russellian must downplay or whitewash. Given the Russellian claim that definite descriptions are quantificational expressions whose compositional semantics single out exactly one individual, then it is not at all surprising (let alone inconvenient or embarrassing) that referential uses of them are ‘common, standard, regular, systematic, and cross-linguistic’. The definite description’s one univocal, quantificational semantics is still evident and operative on referential uses.13 So, contra the argument from convention, definite descriptions stand out among quantified noun phrases as the best tailored for referential uses, but it is far from clear that this does anything to undermine the Russellian orthodox view.14 Bach (2004) explicitly concedes that RDs do not involve any sort of PCI, and – in response to Devitt’s badgering – Bach (2007) also concedes that they are not GCIs. The question then becomes whether there is an alternative pragmatic explanation of how a singular proposition can be communicated while a distinct general proposition is semantically expressed. 13

14

For example, even if the description is used referentially – and even if the relevant descriptive information does not in fact apply to the intended designatum – still the context-independent semantic properties of, say, ‘the man drinking the martini’ play an ineliminable role in characterizing what is communicated with an use of a sentence containing it. This highlights why it is that I want to avoid the term ‘conventional meaning’ (cf. note (2). To concede a communicative regularity is to concede that there is something conventional going on, in at least some senses of the term ‘convention’. However, provided that there is a significant and coherent notion of pragmatic regularity, conceding a communicative regularity is not yet to concede anything having to do with meaning. Hence, if there are pragmatic regularities, then Grice’s challenge shows that we ought to be wary of the idea of conventional meaning.

506 Bach (2004, 2007) clearly thinks that there is – here see especially (2007: 38-39) – and this is where he appeals to the notion of ‘pragmatic regularity’ (or, alternatively, ‘standardized non-literality’). He calls RDs “akin to GCIs”. The idea is that the pragmatic natural kind sketched above in Section 2 might provide the best means for understanding the phenomenon of RDs. The claim is not that RDs are rhetorical questions, or GCIs, or etc. – they are a distinct kind of speech act – but nonetheless they are enough like both of them that this notion of a pragmatic regularity may be exactly what the Russellian ought to reach for, at this stage in the dialogue. Now, one reason why Devitt remains unmoved by this is that Bach (2007) is still (though perhaps not inappropriately) vague on the mechanics. That is, for example, concerning the paradigm case cited at the outset of Section 1, Grice (1975) provides an explicit, stepwise calculation for how semantically expressing that Mr. X has an excellent command of English and a good attendance record thereby pragmatically implicates that Mr. X is no good at philosophy. Perhaps Devitt would give up, if a Russellian could offer him something parallel and solidly plausible, for the case of RDs; but, as far as I can tell, Bach does not even attempt to meet Devitt’s challenge in this direct way. In support of Bach, and against Devitt, I think it is important to bear in mind that, once we get beyond straightforward paradigm cases of conversational implicature (e.g., Grice’s 1975 letter of reference), crisp, clean Gricean derivations are exceedingly hard to come by. Further, this factor is predicted by the Gricean picture, rather than a decisive knock against it (cf. the discussion of “indeterminacy” in the last paragraph of Grice 1975). Further exculpating Bach’s lack of explicit precision is the fact that we are running up against massive open questions about quantifier domain restriction here. I will elaborate on this theme for a bit, before returning to the main issue. To illustrate the problem of quantifier domain restriction: We typically take sentences such as the following to make significant, and possibly true, assertions: (8) Everyone is here now. (9) There is no coffee left.

507 Hence the quantifiers are not taken to range over the entire universe – otherwise neither (8) nor (9) would, as a matter of fact, ever semantically express truths. So, such constructions in natural language are generally taken to be restricted quantifiers – quantifiers which range over some limited (more or less vague) contextually salient domain. (The fact that there is coffee on the island of Sumatra does not contradict my utterance of (9), at 7 am in my kitchen, because Sumatra is clearly outside of the relevant contextually salient domain.) What, then, determines the exact truth-conditions for such cases? Again, this is a massive and difficult open question in the philosophy of language.15 The important present point is that, to the extent that definite descriptions have a quantificational semantics, the same complications will pertain to ‘The book is on the table’ or ‘The car broke down’. The complexity of these issues, then, surely goes some way to exculpating Bach for not directly meeting Devitt’s challenge. If the challenge for the Russellian orthodoxy is to explain exactly how it is that, in the case of an RD, one specific singular proposition gets communicated, while a distinct general, quantificational proposition is semantically expressed, then it seems that this challenge cannot be comprehensively addressed until we have a satisfactory account of quantifier domain restriction.16 It seems that the same challenge could be pressed for (8) or (9), and (8) and (9) are completely independent of any controversial Russellian theses about descriptions. To sum up, then: all parties to these debates concede that RDs constitute a communicative regularity (i.e., definite descriptions are often used with the intention to communicate a singular proposition). Both Bach and Devitt concede that this phenomenon cannot easily be 15

16

For a good programmatic discussion see Stanley and Szabó (2000), and the responses by Bach (2000) and Neale (2000). To run a parallel case against Devitt, imagine someone trying to argue that ‘everyone’ cannot plausibly be given a univocal, quantification semantics because it is sometimes used to communicate information about my logic class, other times used to communicate information about a certain soccer team, other times used to communicate information about the participants at a certain conference, etc. I take it that this argument is very weak, exactly what Grice’s challenge was designed to eradicate.

508 accounted for on the model of Gricean PCIs or GCIs. Devitt then concludes that this phenomenon should be given a semantic explanation; while Bach holds that the notion of a pragmatic regularity provides the best means of theoretically incorporating RDs into the neo-Gricean picture. So, while it is beyond doubt that RDs constitute a communicative regularity, it is far from clear how to settle the question of whether that phenomenon is best amenable to a semantic or pragmatic explanation. Indeed, the question of whether any phenomenon admits of a pragmatic explanation has become more complicated, as more theoretical positions get peopled at the semantics-pragmatics interface. Grice (1975) did bequeath some tentative criteria (i.e., calculability, cancelability, etc.); but, as we have seen, one hardly needs to be a radical iconoclast to be skeptical as to the theoretical worth of some of these. (More on this ongoing theme immediately below.) Not only does the question of where semantics ends and pragmatics begins receive rather different sorts of answers from neo-Griceans vs. relevance theorists vs. contextualists vs. etc.; even further, the idea that the question should receive different answers for different parts of the lexicon has even been floated (cf. Taylor 2007). Some are skeptical as to whether there is such a thing as a pre-theoretical semantics-pragmatics interface, which it is the job of linguists and philosophers to limn and chart (cf. Cappelan 2007). To illustrate the murkiness of this semantics-pragmatics interface a bit further: We have seen reasons not to rest too much on the notion of calculability, but what of cancelability? Well, if RDs were cancelable, then surely that would help Bach against Devitt, in suggesting that a pragmatic explanation is most appropriate. However, the status of that antecedent is rather murky. Recall (cf. note 7) that Gricean PCIs and GCIs provide paradigm instances of cancelability – as well as illustrating why cancelability has been thought to be relevant to the pragmatics/semantics interface. For example, if Grice’s (1975) aforementioned letter of reference had continued, after mentioning Mr. X’s handwriting and attendance record, “Furthermore, Mr. X is also the most brilliant philosophical mind of his generation”, the result would be an odd letter of reference, but it would

509 not carry any implication that Mr. X is no good at philosophy. The original PCI gets canceled by the added material; and that it does so is a reason to think that the original PCI is pragmatically implicated, rather than semantically expressed. However, the relation between RDs and cancelability is not straightforward. RDs turn out to pattern with irony, in something of a halfway-house between clearly cancelable pragmatic implications and clearly uncancelable semantic meanings. Consider first the following: (10) Way to go, Einstein! Though, I do not mean to mock you. A speaker of (10) has not semantically expressed a contradiction; and hence the strong negative message that is communicated by the ironic use of the first sentence uttered is distinct from what is semantically expressed. However, clearly, there is an explicit and drastic contradiction between what is pragmatically implicated and what is subsequently semantically expressed. Irony is thus less cancellable than paradigmatic pragmatic implications (i.e., PCIs or GCIs); but still it is merely an odd speech act, as opposed to a contradiction, to cancel it. Turning now to RDs, we find the very same phenomenon. Consider a referential use of the following: (11) The man drinking the martini looks interesting. Though, he [demonstrating the referent of the previous RD] doesn’t.” Again, (at least to me) this falls short of semantically expressing a contradiction – i.e., ‘P and ~P’ has not been semantically expressed.17 As in the case of irony, though, there is here an explicit and drastic contradiction between what is pragmatically implicated and what is subsequently semantically expressed. When it comes to cancelability, then, RDs seem to fall into this halfway-house of less cancelable than paradigmatic pragmatic implications, though more cancelable than semantic meanings. 17

Perhaps it is possible to run a pro-Bach, anti-Devitt argument on this intuition; though I will make no attempt to do so here.

510 Of course, the point was already made by Camp (2006) that the fact that irony does not seem to be cancelable (in the same way as Grice’s PCIs) does not show that irony is semantic, but rather shows that we have to be careful not to put too much stock in cancelability as an indicator of semantic vs. pragmatic content. In any case, the primary aim of this present work is not to draw a firm, comprehensive semantics-pragmatics divide, and then to prove that RDs belong on the pragmatics side of the divide. It is rather to further the study of pragmatic regularities, for the neo-Gricean movement. 4. Conclusions and morals The most general point I wish to urge here is that the notion of a pragmatic regularity merits further investigation. It has some promise to be quite a useful notion for neo-Griceans, in the course of global battles against the non-neo-Gricean forces. A related upshot is to criticize the methodology behind Devitt’s line of argument against the Russellian orthodoxy. While, again, I think that Devitt has done us all (including Russellians) a valuable service in obstinately refusing to rest content with vague gestures, as opposed to more specific explanations, still he is being unimaginative in carrying on as if one could refute the Russellian orthodoxy by merely showing that RDs are not paradigm cases of either PCIs or GCIs. Provided that there is a coherent notion of a pragmatic regularity, that line of argument is no more likely to successfully attain its goal than was Strawson’s (1952) case for the ambiguity of ‘and’. References Bach, Kent 1995. Standardization vs. Conventionalization. Linguistics and Philosophy 18, 677-686. Bach, Kent 1998. Standardization Revisited. In: Asa Kasher (ed.), Pragmatics: Critical Concepts. Vol. IV. London: Routledge, 712-722. Bach, Kent 1999. The Myth of Conventional Implicature. Linguistics and Philosophy 22, 322-386.

511 Bach, Kent 2000. Quantification, Qualification, and Context. Mind and Language 15, 262-283. Bach, Kent 2004. Descriptions: Points of Reference. In: Anne Bezuidenhuit and Marga Reimer (eds.), Descriptions and Beyond. Oxford: Oxford University Press, 189-229. Bach, Kent 2005. Top Ten Misconceptions about Implicature. In: Betty Birner and Gregory Ward (eds.), Festschrift for Larry Horn. London: Benjamins. Bach, Kent 2007. Referentially Used Descriptions. European Journal of Analytic Philosophy 3, 33-48. Borg, Emma 2007. Minimalism vs. Contextualism in Semantics. In: Gerhard Preyer and Georg Peter (eds.), Context Sensitivity and Semantic Minimalism. Oxford: Oxford University Press, 346-371. Cappelen, Herman 2007. Semantics and Pragmatics: Some Central Issues. In: Gerhard Preyer and Georg Peter (eds.), Context Sensitivity and Semantic Minimalism. Oxford: Oxford University Press, 3-22. Camp, Elizabeth 2006. Why Isn’t Sarcasm Semantic Anyway? Unpublished MS. Devitt, Michael 2004. The Case for Referential Descriptions. In: Anne Bezuidenhuit and Marga Reimer (eds.), Descriptions and Beyond. Oxford: Oxford University Press, 280-305. Devitt, Michael 2007. Referential Descriptions and Conversational Implicatures. European Journal of Analytic Philosophy 3, 7-32. Donnellan, Keith 1966. Reference and Definite Descriptions. Philosophical Review 77, 281-304. Grice, H. P. 1969. Vacuous Names. In: Donald Davidson and Jaakko Hintikka (eds.), Words and Objections. Dordrecht: D. Reidel, 118-145. Grice, H. P. 1975. Logic and Conversation. Syntax and Semantics 3, 31-48. Horn. Laurence 2004. Implicature. In Laurence Horn (ed.), Handbook of Pragmatics. London: Blackwell, 3-28. King, Jeffrey 2001. Complex Demonstratives: A Quantificational Account. Cambridge: MIT Press. King, Jeffrey 2008. Complex Demonstratives, QI Uses, and Direct Reference. Philosophical Review 117, 99-117. Kripke, Saul 1977. Speaker’s Reference and Semantic Reference. In: Peter A. French, Theodore E. Uehling Jr. and Howard K. Wettstein (eds.), Contemporary Perspectives in the Philosophy of Language. Minneapolis: University of Minnesota Press, 255-277. Levinson, Steven 2000. Presumptive Meanings. Cambridge, Mass.: MIT Press. Neale, Stephen 1990. Descriptions. Cambridge, Mass.: MIT Press. Neale, Stephen 2000. On Being Explicit. Mind and Language 15, 284-294.

512 Neale, Stephen 2004. This, That, and the Other. In: Anne Bezuidenhuit and Marga Reimer (eds.), Descriptions and Beyond. Oxford: Oxford University Press, 68-182. Neale, Stephen 2007. Heavy Hands, Magic, and Scene-Reading Traps. European Journal of Analytic Philosophy 3, 77-132. Potts, Chris 2005. The Logic of Conventional Implicature. Oxford: Oxford University Press. Recanati, François 2004. Literal Meaning. Cambridge: Cambridge University Press. Reimer, Marga 1998. The Wettstein/Salmon Debate. Pacific Philosophical Quarterly 79, 130-151. Russell, Bertrand 1905. On Denoting. Mind 14, 479-493. Stanley, Jason, and Zoltán Szabó 2000. On Quantifier Domain Restriction. Mind and Language 15, 219-261. Strawson, Peter 1950. On Referring. Mind 59, 320-344. Strawson, Peter 1952. Introduction to Logical Theory. London: Metheun. Taylor, Ken 2007. A Little Sensitivity Goes a Long Way. In: Gerhard Preyer and Georg Peter (eds.), Context Sensitivity and Semantic Minimalism. Oxford: Oxford University Press, 23-49.

Mieszko Tałasiewicz University of Warsaw [email protected]

A New Perspective for the Theory of Intentionality Abstract: The aim of the paper is to initiate a naturalistic approach to intentionality. According to the presented account intentionality in the primary sense is a property of some physical, perceptible objects such as arrows and sticks (but also gestures or glances (gaze)), relative to subjects that perceive these objects – much as colors are properties of some physical objects relative to perceptual abilities of subjects. Human beings are naturally equipped with ability to perceive some objects as directed somewhere, or intentional. This ability, as psycholinguists argue, is prior to and required by first language acquisition. I will argue that intentionality of linguistic expressions is derived from this natural primitive intentionality (or more exactly: human ability to perceive some material entities – collections of ink molecules or sounds – as meaningful expressions is derived from ability to perceive glances and gestures as directed). Key words: intentionality, Husserl, first language acquisition, perception, evolution of language, speaker-hearer distinction

0. Introduction The concept of intentionality in modern philosophy features in many explanatory contexts (or shall we say, in many problem contexts), including contexts which are quite remote from language issues (e.g. analysis of mental states connected with perception). The purpose of the paper is to present an outline of a uniform naturalistic account of the notion of intentionality that is capable of being applied within many of these contexts, yet the intentionality of language is the primary focus. I assume that the reader will keep observing the difference between ‘intentionality’ as ‘directedness’, which is what we are after here, and ‘intentionality’ as ‘having an intention/wish/purpose etc’. Within the

514 following account intentionality is particularly far away from intention, so any uncertainty about that would be devastating. In order to understand better the intentionality of language expressions, let us look first at directedness of other things, such as arrows, gestures or glances. 1. Intentionality of material things An arrow in itself is an ordinary piece of wood or metal. Yet everyone who sees an arrow sees it as indicating a certain direction. Whether the arrow was made and mounted or not, in other words, whether it has an ‘encoder’ and whether the encoder had intended (wished) to point it at something or not, is of no direct relevance. The same goes for whether there is actually anything the arrow is pointing at. An arrow can be an arbitrary arrangement of sticks whipped into formation by the wind, it does, however, direct us in a certain way; it appears to us as pointing at something. This is all the more true in the case of some gestures and glances. We interpret certain gestures as directional before the person making the gesture explains to us that he or she actually wants to turn our attention that way. The order of interpretation is such that it is the gesture that betrays the intention (wish), not the other way round. We read the intention (wish) from the gesture, not the meaning of the gesture from the declaration of intention. Sometimes in such situations we misinterpret the gesture, as in when the gesture is an accidental hand wave (while we follow the outstretched finger). Other people’s gaze is also interpreted as directed at something. We follow the gaze to see what that person is looking at before he or she tells us that they had intended to look at something in the first place. A gaze is always directional, and the subtle but perceptible accommodation clues (which an external observer can pick up) suggest the distance in this direction of the object being looked at. A glance tells us about the intention of the person who casts it – sometimes misleading us, as when the person is deep in thought and just stares around blankly. Extrinsic knowledge about the circumstances in which the object (an arrow, a gesture, a glance) was made, in particular a person’s declaration

515 about his or her intention to use the object as an indicator, may to some extent correct our interpretation. The maker of the gesture may say that he or she wanted to point at this not that, or that they have stumbled and thrown up their arm. We will then go on and correct our interpretation. The person who gazes at something may say that he was looking not at this but at that, or that he was not looking at anything in particular, just staring into space and thinking about yesterday’s supper. We will likewise change our interpretation of the way he was looking. What is important is that, even though knowledge about the intentions of the encoder can change our interpretation of the object he produced, it does not determine it: we do not need any declarations to trigger an interpretation, correct or incorrect. In fact we may even know that something is not consciously directed, yet perceive it as directed and use its directedness: we may say to someone asking about the right way: ‘follow the direction of that branch over there’; or we may know that someone’s finger is outstretched accidentally and yet say to some third person: ‘look, his finger points at some sleeping guy!’ The impression which the recipient forms in his mind that the gestures, glances or arrows made of sticks are directed towards something is primary and independent of the intentions of the encoder (or the encoder’s existence: as mentioned, certain stick formations we find on our path can be interpreted as directed at something, even though they are the result of wind or rain). Neither does it matter if the thing at which these objects are pointing exists or not. Certain gestures are directional, even if they do not point at anything. We may summarize these remarks in a following proto-definition: (DEF) Intentionality is a relative quality of certain perceived objects, in particular arrows and sticks, but also other people’s gestures and glances, which exhibits itself in a relationship with a certain ability of the subject perceiving those objects. The result is that the subject perceives those objects as being directed towards something. As can be seen, we take for our starting point a position in which intentionality is expressed in a simple, naturalistic way, and is similar to

516 such relative qualities as, say, colors. Compare J. J. C. Smart’s definition for the latter: “Colors [are] dispositions of physical objects to evoke characteristic patterns of discriminatory color behavior by normal human percipients in normal circumstance” (Smart 1997: 1).1 Intentionality is not a relation between the mind and what the mind is directed at – contrary to the standard post-Brentanian account. If we do choose to express intentionality in relational terms (which we are allowed to do in so far as intentionality as a relative quality is founded in a certain relation), then we can talk about a relation between an object of a certain kind and the subject who perceives this object as being directed. Intentionality is a relational property of a material sign (in relation with a perceiver of it). This approach changes the conventional meaning of the expression ‘intentional object’. This expression is typically used to describe the alleged object towards which something is directed. Our conception proposes that the intentional object is an object which evokes the impression of being directed at, thus the sign: an arrow, a glance or a gesture (as opposed to a ball, ocean or pig’s tail, which in normal circumstances do not evoke the impression of being directed and which are thus not intentional). The objects which are being pointed at can best be termed ‘intended objects’. First of all however, under this approach the question: whose acts are intentional acts (acts of perceiving something as being directed), is highlighted. Husserl, as his predecessors, Brentano and Twardowski, who studied intentionality, speaks in general terms about the subject’s acts. His perspective does not differentiate between the subject as an ‘encoder’ and the subject as a ‘percipient’ of the intentional object (sign). When we come to consider this problem in ‘solitary life’ (Husserl 1

The analogy is far-reaching: just as the perception of intentionality of a certain object can be corrected by reference to a certain knowledge about the circumstances in which it was made, so a certain extrinsic knowledge can further correct our perception of color. When we are looking at Renoir’s La Promenade, we perceive directly gray-green, dirty, shapeless trousers which the man is wearing. Yet thanks to our being accustomed to how dark foliage casts shadows we are predisposed to think that he is actually wearing clean, snow-white, elegant trousers.

517 2001, Vol. 1: 190), and this kind of perspective seems to prevail in intentionality studies, this difference does not stand out clearly and can be easily missed: the subject is in a sense both the encoder and the percipient. Nevertheless, it is a very important difference and failure to spot it can lead to serious difficulties in understanding the problem of intentionality correctly. If this distinction is slipping out of our control, then we automatically assume that it is an act of the encoder what constitutes the intentionality of a sign. Often, in such cases, intentional acts merge with the communicative intentions of the encoder, while it should be borne in mind, somewhat ahead of introducing the term ‘expression’, that the communicative intention of the encoder of an expression (e.g. his conscious intention to convey some message to the hearer) should be contrasted sharply with the intentionality of the expression, i.e. an expression being directed towards a certain object. For although: all expressions in communicative speech function as indications [and] they serve the hearer as signs of the ‘thoughts’ of the speaker [...] [-] this function of verbal expressions we shall call their intimating function [-] (Husserl 2001, Vol. 1: 189) an expression’s meaning [...] cannot coincide with its feats of intimation. (Husserl 2001, Vol. 1: 190)

Hence, even if intentionality were in fact founded in the encoder’s acts, it would still be different from his communicative intention.2 As our approach shows, however, intentionality is not founded in the acts of the encoder (who may not exist if the intentional object is not really a sign, but only appears to be one), but of the percipient.3

2

The communicative intention of speaker’s utterance that ‘It is raining’ is directed at a situation wherein the hearer forms an idea that [the encoder wants to (give the impression that he wants to) communicate that] it is raining. The intentionality of this sentence, on the other hand, consists in it being directed towards the fact that it is raining, a situation which is independent of the interlocutors.

518 Intentionality can be imputed to conscious intentions of sign encoders, but only in a secondary sense, in that person A, aware that a certain object C (in particular: an expression) will be interpreted by person B as directed towards object D (which means that C is intentional and D is intended in the primary sense), produces this object deliberately and presents it to person B in order to, while relying on this 3

In Husserl’s system, far from what his critics say, the question is not settled. In many places the description of an intentional act essentially takes the perceiver’s perspective: What is involved in the descriptive difference between the physical signphenomenon and the meaning-intention which makes it into an expression, becomes most clear when we turn our attention to the sign qua sign, e.g. to the printed word as such. If we do this, we have an external percept [...] just like any other, whose object loses its verbal character. If this object again functions as a word, its presentation is wholly altered in character [...]. Our interest, our intention, our thought [...] point exclusively to the thing meant in the sense-giving act [...]. [I]ntuitive presentation, in which the physical appearance of the word* is constituted, undergoes an essential phenomenal modification when its object begins to count as an expression. While what constitutes the object’s appearing remains unchanged, the intentional character of the experience alters. There is constituted [...] an act of meaning which finds support in the verbal presentation’s intuitive content, but which differs in essence from the intuitive intention directed upon the word itself. (Husserl 2001, Vol. 1: 193-194) *

In (2001) edition there is ‘world’ in this place, which is an obvious proof-error.

All objects and relations among objects only are what they are for us, through acts of thought essentially different from them, in which they become present to us, in which they stand before us as unitary items that we mean. (Husserl 2001, Vol. 1:194) The meaning of the assertion [...][-] we continue to recognize its identity of intention in evident acts of reflection: we do not arbitrarily attribute it to our assertions, but discover it in them. (Husserl 2001, Vol. 1: 196) The soliloquizing thinker ‘understands’ his words, and this understanding is simply his act of meaning them. (Husserl 2001, Vol. 1: 321 – note 5 to page 213)

That Husserl was vague on this point was stressed by Kotarbińska (1957). She herself adopted the perspective of the percipient, as I am doing in the present paper. However, her general theory of intentionality goes in a different direction than mine. A particularly important difference is that she chose to characterize intentionality in terms of semantic concepts, such as denoting, which in my view is putting the cart before the horse.

519 intentionality, attract the attention of person B to object D.4 It should be borne in mind that intentionality of one’s intention must always adjust to natural intentionality, which is not determined by intentions but which is simply perceptible. We can, for instance, use arrows in a game of chalk chase to show others the direction in which they should be going. However, in order for our endeavors to be successful, we have to mark the arrows with the head pointing in the direction of the chase, that is, in accordance with their natural intentionality. If we mark the arrows side-on to the direction of the chase, we can be sure that the players (percipients) will lose their way and will not accept our explanation that the intention all along was to indicate the direction of the chase using the side rather than the head of the arrow. During a boring conference, we can indicate with our eyes to a neighbor a sleeping professor, but only on condition that we use the natural intentionality of our gaze and look meaningfully in the direction of the professor. We can’t just look through the window and get huffy because the addressee of our visual communication will not cotton on that by means of this glance we are trying to show him an object which is perpendicular to our line of sight. (In a group, we can of course establish a convention according to which what we want to indicate is not along the line of natural intentionality of indicators but, e.g. perpendicular to it. Yet such a convention is itself based on natural intentionality, only somewhat modified in a certain systematic way.) Primary intentionality is always an intersubjectively perceived quality of material objects, not a person’s intention. (As we have said, intentionality is a relative property – a property which is exhibited by some objects when they are perceived by someone. In this sense it is not an objective property. But it is not subjective, either. If we assume that cognitive apparatus of different people reacts similarly to the same stimulus, then intentionality of the same object will be similar for these people, thus intersubjective. If two persons may agree that the light at

4

Again, such a procedure is described also in Kotarbińska (1957).

520 the street crossing was red, they equally may agree which of them is pointed at by the finger of the policeman who is trying to hold them.)5 2. Intentionality of expressions The intentionality of language expressions works in an analogous way. Moreover, it is not just a structural but a genetic analogy: intentionality of language comes from the intentionality of glances and gestures. Psycholinguists point out that the ability to intentionally interpret another person’s gaze is fundamental in order to learn a language: Joint Attention Comes First [...] In a successful conversation, the two participants must agree on what is being talked about. One way to ensure this is to start with the same locus of attention. But how does one make a one- or two-year-old systematically attend to what one is saying? [...] [B]y age one, infants have become quite good themselves at checking on the adult’s gaze, stance, and physical orientation. (Clark 2003: 32)

I do not claim that the ability to interpret language expressions is exactly the same as the ability to recognize the intentionality of gestures or a person’s gaze. The best proof that it is not so is that one-year old children have the latter ability without yet having the former. I maintain though that the two abilities are of the same kind and moreover, one is functionally dependent on the other, even if materially they may be rooted in different brain structures. Sensitivity to the directionality of a person’s gaze may well be a congenital trait (at any rate, functionally it is treated as such: no cognitive ability that is more fundamental has yet been found which would determine the acquisition of the ability to recognize a direction of a gaze or gesture). Sensitivity to the intentionality of expressions is an acquired trait. However, the nature of ‘being directed towards’ or ‘pointing at’ is in both cases the same. 5

Such an agreement would be much harder to result from cooperation with a creature characterized by an essentially different cognitive apparatus: it sometimes gets frustrating, when we are trying to direct a dog’s attention to something, for instance. On the other hand, a talk about colors with a dog seldom is very fruitful, either.

521 As with arrows and a person’s gaze, expressions are also interpreted as referring to something before the speaker declares that he is actually going to refer to it. The use of an expression alone is reason enough for us to think that the person is talking about something, even if the sentences we just happened to read are an arbitrary result of a monkey bashing the keyboard. If we are made aware that the gripping yarn we are just reading is the handiwork of some monkey, then our attitude to what we are reading may change to some extent. But we start with an attitude and change it later. Just as arrows or human gaze are directional even when they actually do not point at anything, so expressions look as if they refer to something even if they do not. Finally, just as arrows and gestures can be manipulated to indicate this or that to a person, using only their natural intentionality, so expressions can be manipulated in the process of speech production in order to put together a message others can understand, but only on condition that we use their ‘natural intentionality’ encoded in the meaning (which we, as the recipients, learn from others). The latter shows that we are not at risk of slipping into a HumptyDumpty-like attitude towards meaning. Humpty-Dumpty, talking to Alice in Through the Looking-Glass, maintains that she cannot know the meaning of a word he has used (‘glory’) until he communicates his intentions in the matter. Acceptance of such attitudes is rightly believed to be absurd – if it were to take place, the impact on any intersubjective communication would be disastrous. A danger of this kind would indeed be present if the meaning of expressions were to form in the acts of the encoder.6 According to our approach, however, meaning is formed in the acts of the recipient, when he or she is perceiving the expression as directed at something. And such a quality of this perception may be 6

Michael Dummett claims that for this reason ‘It is difficult to acquit Husserl of maintaining a view of the matter like Humpty-Dumpty’s’ (Dummett 1993: 44). However, as we have shown, Husserl’s views about whether acts constituting meaning are the acts of the subject as an encoder or as a recipient are ambiguous. Hence, it is not clear to what extent Dummett’s claim is justified. The uncertainty concerns equally the other critical remarks which Dummett made about Husserl. They can be justified if we accept that Husserl was speaking of the acts of encoders, but are not if he meant the recipient.

522 induced by those, whom the recipient is learning the language from. In certain cases, the meaning can be literally shown to us by our teachers’ gestures, which guarantees its intersubjectivity. HumptyDumpty imparts meaning to the word ‘glory’ not when he wants to subject Alice to ‘a nice knock-down argument’ but when he himself learns the word for the first time. And if he uses it against Alice, then, if Alice was taught this word in the same way as Humpty-Dumpty, then her acts will constitute the same meaning in her head. None of HumptyDumpty’s fantasies about what he would like the word to mean are of any relevance here. The same aspect of our theory helps it to stand up to another, although related, charge against intentional Husserl-like conceptions of language, namely that the construction of meaning in such conceptions starts with individual mental acts and that social practice of language plays no part in it. Within such conceptions the general meaning is unintelligible.7 Within ours, on the contrary, the matter is clear: if meaning is constituted by individual acts, it is so only in so far as the impression of color is constituted by individual perception. Perhaps some qualia in it are private – they are not important though. What is important, in Smart’s terminology, is ‘a pattern of discriminatory color behavior,’ which is intersubjective. Likewise in the acquisition of meaning – there are some private qualia connected with our perception of expressions as intentional objects, but they are not important. The important thing, which is intersubjective, is a ‘pattern of discriminatory verbal behavior’, ultimately passed on to children in the process of language learning in the social practice of using language. And so we have come to the question of meaning. The intentionality of expressions alone, that is, the fact that they are perceived as directed towards something, or indicating something, does not yet constitute their meaning. Meaning is founded in intentionality, but it covers not only the indicating, but also a broadly understood, way in which this indicating is done. This question is very important, but it exceeds the scope of this paper, calling in fact for a separate one. What are ways of indicating something, how do the semantic relationships of expressions work; what 7

Again, such a charge has been raised against Husserl by Dummett (1993: 67).

523 is a proposition and what are the truth conditions of propositions – these are just some of the questions which come to mind and which must be postponed till another occasion. 3. Conclusion For the purposes of this paper it is important that a quality of the human mind whereby certain objects (including expressions) are perceived as being directed towards something can be understood completely naturalistically, as something that originates in the evolution of our cognitive apparatus. It is not hard to imagine that being in possession of such a quality was a very significant evolutionary gain, increasing the fitness of its bearer. Humanoids, who possessed this quality, were able to locate food in a dense foliage when the true finder gave away its location by gazing at it. They were also able to turn around quickly and defend themselves against a predator when they saw the shock in the eyes of another individual staring into space behind their back. The humanoids who contemplated the look of a friend instead of turning around to see what had scared him are unlikely to have passed their genes onto us. Speaking of intentionality in the sense just outlined, I do not mean any mystery – available only via an idealizing Anschauung, but the natural ability of the human mind. Natural yet fundamental: I believe that intentionality so understood is constitutive for the humans’ ability to reason about the world in general and for any higher cognitive process, although we have to remember that it is our attitude as recipients that is the origin of intentionality. If our thoughts are intentional,8 then they are intentional in the sense that we perceive them as being directed towards something, which manifests itself in acts of a kind of meta-reflection when we are thinking about our own thoughts.9 Thinking of this kind is 8

9

I do not attempt to resolve whether all thoughts are intentional, viz. whether all thinking is thinking about something, in particular whether it is articulated, language-shaped thinking. Prima facie, I do not think this is the case. It does not mean, however, that intentionality manifests itself only in sentences expressing propositional attitudes, or more generally, in intensional sentences. This view can be found in literature (see Smith and McIntyre 1982), but is not, in

524 not possible without prior acquisition of language. A well-known and poignant fact is that if a person was not exposed to human speech at a certain stage in his or her childhood, they will have missed the chance ever to learn to speak. Such a person is not able to think articulated thoughts, engage in any ‘inner dialogue’, for if he or she were, they would be able to learn to translate that kind of dialogue into any intelligible spoken language at any time in their life, just as an anthropologist at any time is able to learn to speak the language of any tribe he studies. This shows that the encoder’s perspective, which asserts itself as the dominant one when we think casually about our language behavior (including inner dialogue) is in fact secondary to the recipient’s perspective and cannot function without it. References Brentano, Franz 1874. Psychologie vom empirischen Standpunkt. Leipzig [Psychology from an Empirical Standpoint, London 1973]. Clark, Eve Vivienne 2003. First Language Acquisition. Cambridge: Cambridge University Press. Dummett, Michael 1993. Origins of Analytical Philosophy. Cambridge, Mass.: Harvard University Press. Husserl, Edmund 2001. Logical Investigations. London: Routlege. Kotarbińska, Janina 1957. Pojęcie znaku [The Concept of Sign]. Studia Logica VI, 57-143 [main text is in Polish, with a large English summary on pages 140143]. Smart, John J. C. 1997. On Some Criticisms of a Physicalist Theory of Colors. In: Alex Byrne and David R. Hilbert (eds.), Readings on Color, vol. 1: The Philosophy of Color, Cambridge, MA: The MIT Press, 1-10 [first published in C. Cheng (ed.) (1975), Philosophical Aspects of the Mind-Body Problem, Honolulu: University Press of Hawaii]. Smith, David Woodruff and Ronald McIntyre 1982. Husserl and Intentionality. Dordrecht: D.Reidel Publishing.

my opinion, justified – neither in the conception proposed here nor in Husserl’s original writings. Every sentence is intentional: including simple and fully extensional sentences of the kind ‘The cat is on the mat’.

525 Twardowski, Kazimierz 1894. Zur Lehre vom Inhalt und Gegenstand der Vorstellungen. Wien (On the content and object of presentations. The Hague: Martinus Nijhoff 1977).

527 Tomoo Ueda Ruhr-Universität Bochum, University of Tokyo [email protected]

A Conceptual Role Semantics for Attitude Reports Abstract: Reports of propositional attitudes have two components. One is a frame and the other is an inset. This paper is focusing on the relation between them and aims at individuating the content expressed by insets. First, examples from the English and Japanese languages are examined. Second, it shall be argued that there are two sorts of opacity. One is based on Frege’s discussion of modes of presentation, the other arises if we evaluate attitude reports in the discursive context and the second opacity is treated in the same way as deixis. Finally, both sorts of opacity are explained in terms of one-factor conceptual role semantics. Key words: propositional attitudes, attitude reports, conceptual role semantics, deixis, English, Japanese

0. Introduction A reporter reports to her listener about a propositional attitude of someone else which the reporter apprehends in another context. The reporter shall typically use the following sentences in English and Japanese respectively: (1) Betty believes that Willy Brandt was born in Lübeck (2) Betty-wa Willy Brandt-ga Lübeck-de UMARETA to shinjiru Betty-T Willy Brandt-NOM Lübeck-in be born PAST QT believe1 [Intended: Betty believes that Willy Brandt was born in Lübeck] 1

Note for abbreviations: ACC (accusative case), DAT (dative case), IP (interactional particle), NOM (nominative case), NON-PAST (non-past tense), PAST (past tense), QT (quotative marker), T (theme marker). Ungrammatical sentences are indicated by putting ‘*’ in front of the sentence. ‘#’ indicates a following sentence is semantically or pragmatically anomalous.

528 Attitude reports, such as (1) and (2), are assertions and hence, they have truth-values. Grammatically, they belong to a class of indirect quotations and they consist of two sub-sentential parts. Following Sternberg’s terminology (1982: 108), I shall call the part written in small capitals inset and the other frame.2 This paper concentrates on insets of attitude reports. Especially, the discussion is directed toward how to individuate their contents. Since Frege’s seminal paper (1892), this has been widely discussed in connection with the opaque occurrences of singular terms in insets. This paper shall argue that a holistic theory of meaning is a plausible candidate for individuating contents expressed by insets of attitude reports. Here is the structure of this paper. First, the role which deixis plays in attitude reports and its relevance to truth-evaluations will be identified. This discussion implies that attitude reports are discursively evaluated by nature and this will be examined by introducing a second sort of opacity. Soames (2002) provides a theoretical account which seems to capture the second sort of opacity but a closer look shall provide the reason why this is not the case. Finally, a provisional account in terms of a holistic theory of meaning shall be provided. 1. Deixis and reporter’s commitment Our discussion starts with a passage from Frege’s “Über Sinn und Bedeutung” (1892). In the following passage, Frege discusses sentences containing content clauses: The case of an abstract noun clause, introduced by ‘that [dass]’, includes the case of indirect speech, in which we have seen the words to have their 2

It is not addressed here whether the English connective that and the Japanese quotative marker to belong to the inset. There are actually two ways of interpretation. The classic discussions following Frege propose that insets are treated as identical to that-clauses (Sternberg 1982: 111), while Davidson (1968), Maynard (1996), and Recanati (2000) claim that the insets contain only the sentences quoted. I am in favor of the latter line, but this is an independent point and I will not give any defense of this position.

529 indirect Bedeutung, coincident with what is customarily their sense. In this case, then, the subordinate clauses has for its Bedeutung a thought, not a truth-value, and for its sense not a thought, but the sense of the words ‘the thought that … [der Gedanke, dass…]’, which is only part of the thought in the entire complex sentence. This happens after ‘say [sagen]’, ‘hear [hören]’, ‘be of the opinion [meinen]’, ‘be convinced [überzeugt sein]’, ‘conclude [schließen]’, and similar words. (Frege 1892: 37)3

Note that the first sentence contains three claims: a. b. c.

attitude reports are treated in a parallel way to indirect speech; dass S is a noun phrase (in the German language); taking a dass-clause as complement is a distinctive feature of indirect speech compared to direct speech (in the German language).

These claims characterize attitude reports linguistically. Furthermore, it seems that Frege argues from these linguistic characterizations for what we shall call later the Eigenname view: The subordinate clause could be regarded as a noun, indeed one could say: as a proper name [Eigenname] of that thought, that command, etc., as such a proper name it [the subordinate clause] comes into the context of the compound sentence. (Frege 1892: 39)4

Note that Frege’s use of the term Eigenname includes definite descriptions and hence, it is equivalent to what we mean by “singular term” rather than by “proper name”. The dass-clause in an attitude report functions as a singular term which denotes believer’s thought since “one talks about the sense, e.g., of another person’s remarks” in an indirect speech (Frege 1892: 28).5 3

4 5

The English translation of this text is based on the version of Max Black, which is edited by Michael Beany (1997: 151-171). My modifications are indicated in footnotes. Page numbers are referred according to the original German text and original German words are inserted by using square brackets. The latter half of the translation is mine. As a consequence of this, Frege has to introduce the doubly indirect sense of expressions since he has introduced the notion of indirect sense. We shall not get

530 This section examines this argument. Although (b) and (c) are syntactic claims about the German language, Frege’s Eigenname view has a general nature. That is to say, equivalent claims to (b) and (c) should apply to other natural languages as well. In the following, some counterevidence against the generality of (b) and (c) will be introduced and the notion of deixis is proposed as a more plausible feature characterizing the indirectness of speech reports. This discussion also implies that the Eigenname view does not hold. 1.1. Indirectness and deixis In the first half of the section, I shall present some evidence from the English and Japanese languages in order to examine whether the following are valid: be cj

ce

that S is a noun phrase (NP) in the English language; taking to-clause as a complement is a distinctive feature of indirect speech compared to direct speech in the Japanese language; taking that-clause as a complement is a distinctive feature of indirect speech compared to direct speech in the English language

If (be) is true, then it is assumed that content clauses can be appear wherever noun phrases can. However, Huddleston (2002a: §8.3) provides some linguistic evidence against the NP interpretation of content clauses following that. According to him, the content clauses cannot be understood in terms of the classic classification of noun, adjectival and adverbial clause. Especially, the parallelism between objective noun phrases and objective content clauses in indirect speech does not hold. He provides three observations. First, there are verbs which license a content clause but not an NP. For example, observe into the discussion here what Frege should mean by “the sense of the words ‘the thought that … [der Gedanke, dass…]’, which is only part of the thought in the entire complex sentence” (Frege 1892: 37).

531 marvel and vouch. Second, compare (3a) and (3b). NP objects generally come immediately after the verb otherwise the sentence is unacceptable. But content clauses do not have to follow the verb directly. Observe (3c): (3) a. *He opened slowly the door b. He opened the door slowly c. He believes generally that IT IS GOOD FOR HIS HEALTH Finally, content clauses cannot occur as the complement governed by a prepositional verb such as insist on. Observe the following: (4) a. He insisted on an adjournment b. He insisted that we adjourn c. *He insisted on that we adjourn. Therefore, content clauses behave differently from noun phrases and (b) is not supported in the English language. (4c) is also generally not valid among languages. Frege differentiates indirect speech from direct speech based on their syntactic features. Namely, a compound sentence belongs to the class of indirect speech if it contains an attitude verb followed by a dass-connective, while direct speech takes a sentential complement with quotation marks (in writing).6 However, the indirect quotation markers are neither sufficient nor necessary for characterizing indirectness of the speech report. Against (ce), existence of the indirect quotation makers is unnecessary because (1) remains acceptable and grammatical even if the connective that is omitted (Huddleston 2002a: 955). Observe the following: (5) Betty believes WILLY BRANDT WAS BORN IN LÜBECK

6

This interpretation can be disputed because there is a possible tension among Frege’s text. The interpretation here can be supported by his later text: Gedankengefüge (Frege 1923: 36).

532 Against (cj), the existence of the quotative markers to of the Japanese language is insufficient to characterize a speech report as indirect because it is used in both direct and indirect speech reports (see sentence (7); Kamada 2000: Fig. 1 [96]; Nihongo Kizyutsu Bumpô Kenkyûkai 2008: §2.6). Now, we have to find an alternative way of discerning the indirect speech reports from direct speech reports. The features we shall focus on are called deixis. For example, terms like here, now or you are deictic terms. In order to interpret deixis, some contextual information must be provided. The notion of deixis is introduced as follows: Deixis is the name given to those formal properties of utterances which are determined by, and which are interpreted by knowing, certain aspects of the communication act in which the utterances in question can play a role. (Fillmore 1997: 61)

These include (i) person deixis; (ii) place deixis; (iii) time deixis; (iv) discourse deixis and (v) social deixis. The deixis characterizes indirectness of speech and attitude reports in general ways since they are applicable to both Japanese and English (Kamada 2000: 95-97; Huddleston 2002a: §9.1). Observe the following three reports: (6) a. Joe said to Betty(i): “You(=i) like climbing” b. Betty: Joe said to me(i) that I(=i) LIKED CLIMBING c. #Betty: Joe said to me(i) that YOU(≠i) LIKED CLIMBING First, the tense of the inset is back-shifted to preterit in (6c) and (6b). The back-shifted preterit can occur when the frame is past or the time of the frame is past (Huddleston 2002b: 153). This is because the English language uses absolute tense in insets. Second, (6b) but not (6c) says the same as (6a) because of the assignments for personal pronouns. The second point is observed in the Japanese language too. Observe that (7a) and (7c) but not (7b) say the same (after Kamada 2000: 107):

533 (7) (Context: Megumi said to Koichi: “I will marry you”) a. Megumi-wa Koichi(i)-ni ANATA(=i)- TO KEKKON-SHIMASU to itta Megumi-T Koichi-DAT you-with marry (polite NON-PAST) QT say PAST [Intendend: Megimi said to Koichi(i): “[I] will marry you(=i)”] b. #Megumi-wa Koichi(i)-ni ANATA(≠i)- TO KEKKON-SURU to itta Megumi-T Koichi-DAT you-with marry (NON-PAST) QT say PAST [Intendend: Megimi said to Koichi(i) that [she] would marry you(≠i)”] KARE(i)- TO KEKKON-SURU to itta c. Megumi-wa Koichi(i)-ni Megumi-T Koichi-DAT he-with marry (NON-PAST) QT say PAST [Intendend: Megimi said to Koichi(i) that [she] would marry him(i)”]

The issue about the tense is different here because the Japanese language uses the relative tense in insets (Kamada 2000: 24). The time of inset is set relative to the time of the frame and using non-past tense form indicates that the time of the inset is the same as that of the frame. Instead, compare kekkon-shimasu and kekkon-suru in (7). Note that shimasu used in (7a) is a polite form of ‘suru’ used in (7b) and (7c). If polite forms like shimasu are used in insets, the speech reports can only be interpreted as direct speech reports (Kamada 2000: §1.2.3). We have seen how tense, pronouns and politeness markers behave in insets of indirect speech reports. We can characterize the indirectness of speech reports as follows: A speech report is indirect if (and only if) the deixis used in its inset is determined or interpreted relative to the discursive context, in which the speech report is uttered. This way of characterizing indirectness of speech reports is more generally applicable than the characterizations Frege introduced. 1.2. Reporter’s commitment Recall Frege’s Eigenname view introduced above. If we assume that a belief predicates express binary relation between believers and thoughts which content clauses name, then it implies the following (d): d.

An attitude report is false if its inset conveys any additional information which is only relevant for the reporter.

534 If there were such additional information, it would have to be expressed in the inset of an attitude report because of the logical form of belief predicates. However, the additional information cannot be a part of the thought the believer has. If it were conveyed in the inset, then the inset would no longer be the singular term of thought. Hence, (d) must be the case. In contrast, our characterization of indirectness permits reports to convey some additional information which is not part of beliefs believers actually have in mind. Features of social deixis used in insets of Japanese attitude reports provide some data for the discussion. Observe the following attitude reports: (8) (Context: Peter reports Paul’s thought to Mary(i)) a. ANATA(=i)-GA KOKO-NI KURU you-NOM here come NON-PAST [Intended: Paul thinks that you(=i) come here] b. #ANATA-GA KOKO-NI KI MASU you-NOM here come polite NON-PAST [Intended: Paul thinks that you come here] c. ANATA(=i)-GA KOKO-NI IRASSHARU you-NOM here come respectful NON-PAST [Intended: Paul thinks that you(=i) come here] d. #ANATA-GA KOKO-NI IRASSHAI MASU you-NOM here come respectful polite NON-PAST [Intended: Paul thinks that you come here]

to Paul-wa omou QT Paul-T think to Paul-wa omou QT Paul-T think to Paul-wa omou QT Paul-T think to Paul-wa omou QT Paul-T think

We have discussed the difference between (8a) and (8b) already. In insets of indirect speech reports including belief reports, the polite forms do not occur in insets. Therefore, it is anomalous to use the polite form in thought ascriptions as indicated in (8b). Observe now (8c) and (8d): irassharu in (8c) is a respectful form of kuru used in (8a). The speaker uses respectful forms in the Japanese language in order to honor individuals and/or their activities and this form is used when these individuals are socially higher than the speaker (see Tsujimura 2007: 429). The use of the respectful form does not make attitude reports unacceptable. If we add the politeness form to the inset of (8c), (8d) contains polite forms of the honorific form, where it is

535 unacceptable or at least anomalous in the context at hand. These data suggest that the honorific feature can only be interpreted relative to the discursive context. It is, hence, the reporter not the believer, Paul, who honors the person referred to by the second person pronoun. In order to demonstrate the point, let us suppose Peter utters (8c) to Mary. Suppose further Paul is socially superior to Mary and Mary to Peter. In this case, the original utterance of Paul will be (9a) rather than (9b): (9) (Context: Paul informs Peter about Mary’s coming) a. Mary wa soko ni kuru Mary T there come [Intended: Mary comes to you] # b. Mary wa soko ni irassharu Mary T there come honorific [Intended: Mary comes to you] Note that the social relation between Paul and Peter is not expressed in either case of (9). Hence, (8c) contains some additional information to the content expressed by (9). The observations so far can be interpreted as follows: Reporters of indirect speech do not simply represent and forward some contents which they apprehend. Rather, the insets of attitude reports also convey reporters’ point of view shared with their audience in each discursive context. This contradicts (d) and, hence, the Eigenname view. The remaining question is whether the deixis expressed in the insets is relevant to the truth-values of whole attitude reports. The answer is yes. The person deixis is clearly relevant for truth-evaluations. Observe (8) again. (8b) is anomalous because the second person pronoun cannot refer, and hence it is false. Failures in modifying social deixis are related to the truth-values in more complicated ways. The interpretation of social deixis is necessary for assigning values to personal pronouns. Observe (8) again. The difference between kuru and kimasu provides a context of interpretation for the second person pronoun anata contained in each inset. In

536 accordance with this difference, the insets of (8a) and (8b) express different truth-evaluable contents and, therefore, they may differ in truth-values. In this example, lack of the politeness form is a reliable indication for the indirectness of speech report and, hence, it is a characteristic feature concerning the anomaly of the thought report. The time deixis used in the English attitude reports can be analyzed in a parallel manner. It is clear from the characteristics of the honorific form that (8c) would become anomalous if the reporter were socially superior to the addressee. This anomaly should be treated as relevant to truthevaluations if the other social deictic features are relevant. In this section, we have seen that the reporters convey some additional information by modifying the deixis and hence, Frege’s Eigenname view is argued against. The deixis used in the insets of attitude reports is relevant for the truth-value of attitude reports as a whole. 2. Indirectness and discursive opacity In this section, it is shown that opacity can be treated in the same framework. Frege’s argument introduced above is also seen as a demonstration for the opacity of attitude reports if it is put together with his notion of modes of presentation. Suppose Betty has two different modes of presentation associated with the term ‘Willy Brandt’ and ‘Herbert Frahm’. Then, she may accept (10) while denying (11): (10) Willy Brandt was born in Lübeck (11) Herbert Frahm was born in Lübeck This difference gives reason to assume that singular terms may occur in insets opaquely; the different modes of presentation Betty attaches to both of the names make the following sentences different in truth-values: (12) Betty believes that Willy Brandt was born in Lübeck (13) Betty believes that Herbert Frahm was born in Lübeck

537 That is to say, singular terms occur in the insets of attitude reports opaquely because of different modes of presentations the believer attaches to the singular terms. In the last section, we have shown that the reporter may convey some additional information to the information which believers have in mind. The example we have discussed was some of the social deictic features. If the relevance of social deixis to truth-evaluation is of a general character and the content individuations of insets have essentially a discursive nature, then there are normally two relevant contexts for evaluating the truth of an attitude report. One is the context in which a reporter apprehends what a believer thinks or believes, e.g. as Rudolf and Betty are talking about the believer’s former neighbor, in the following example: (14) (Context A: Rudolf and Betty talks about Betty’s former neighbor) Betty: Herbert Frahm was a child in my neighborhood and he was not born in Lübeck. The context A is the context which Frege’s discussion is relevant to. Namely, suppose Betty has two different modes of presentation attached to ‘Herbert Frahm’ and ‘Willy Brandt’. Then, she could claim something contradictory to (14) if the occurrence of ‘Herbert Frahm’ in (14) is substituted with ‘Willy Brandt’. It is the discursive context in which Rudolf reports Betty’s attitude to Lucy and the discursive context is only relevant context for evaluating attitude reports. For example, Betty’s attitude which Rudolf apprehended through (13) can be reported as follows: (15) (Context B: Talking about history of German politics and Lucy is not a specialist about the issue) a. Rudolf: Betty believes that Willy Brandt was not born in Lübeck b. Rudolf: Betty believes that Herbert Frahm was not born in Lübeck

538 (15a) and (15b) have different truth-values if the discursive context requires evaluating the attitude reports opaquely. Furthermore, the similar case can be built in virtue of some modes of presentation which the participants of the context B associate with both of the names. Suppose Rudolf knows who should be assigned to both of the names in question but Lucy does not know who ‘Herbert Frahm’ is. In this case, she cannot attach any modes of presentation to the name. The name ‘Willy Brandt’ in (15a) cannot be substituted with ‘Herbert Frahm’ in the discursive context since Lucy does not accept (15b) regardless of which of the names Betty actually used. The case of Lucy’s different mode of presentation can well be regarded as another variant of opacity. Namely, this case is regarded as relevant for truth-evaluations. The case with Lucy’s modes of presentation has a structural similarity to the discussion about the person deixis (Section 1.2). The assignment of values to proper names like ‘Willy Brandt’ and ‘Herbert Frahm’ will be determined in the discursive context as personal deixis and the opacity is caused by some failures to assign or evaluate such proper names in the discursive context. Therefore, ‘Willy Brandt’ occurring in (15a) is opaque but it is not due to the believer’s modes of presentation. Rather, it should be called “discursive opacity”. A hypothesis to explain the discursive opacity will be this: occurrences of proper names in insets of attitude reports contain some hidden deictic features which are made explicit if a demonstrative like ‘that’ occurs before any proper name and it demonstrates the objects that are determined in the discursive context. This section provided some arguments for postulating a second sort of opacity. Some occurrences of singular terms are opaque, and substituting such occurrences with some other co-referring singular terms may affect the truth-values of the whole belief reports in the same way as the classic opacity-case. Nonetheless, discursive opacity arises not due to any modes of presentations believers have. Rather, this opacity is due to some ignorance or discursive presuppositions which reporters and their audience share. Furthermore, we have proposed that discursive opacity must be explained in a parallel way to the person deixis.

539 3. An implementation Now, we have three sorts of occurrences of singular terms in the insets: i. transparent occurrences ii. opaque occurrences (classic) iii. opaque occurrences (discursive) (ii) is the opacity which is caused by the cases Frege observed, and (iii) is the opacity which is observed in (15). How to implement these factors in the model of content individuations? Soames (2002: Chapter 8) analyzes a similar example as we have discussed in the context of (14) and (15). In this section, we shall examine his analysis in order to make the characteristic features of our example clear. According to him, opacity is explained by some salient common background information shared by the believer and her audience. This information is descriptive and it constitutes a piece of additional information to the singular Russellian proposition. Here is how Soames formulates the situation: (i) says that there is some descriptive information and (ii) says that this descriptive information is a part of belief content: An assertive utterance in a context C of a propositional attitude ascription, α believes that Fn, containing an ordinary proper name n, results in the assertion of the proposition (semantically) expressed by α believes that F[the x: (Dx & x=n)] in C if i.

it is part of the common background information shared by speakers and hearers in C that the name n is associated by them with the description the x: (Dx & x=n), and as a result of this, an assertive utterance of Fn in C would result in an assertion of the proposition (semantically) expressed by F[the x: (Dx & x=n)] in C; and

ii.

the common background information shared by speakers and hearers in C is such that given it, conversational participants in C will readily assume that if the speaker’s assertive utterance is true in the context, then the proposition (semantically) expressed by α believes that F[the x: (Dx & x=n)] is true; moreover, each knows this about the

540 other. (Soames 2002: 221; original italics and bold face; my underline)

Note that the descriptive contents are determined solely in the discursive context C shared by the reporter and her audience. This is certainly a positive characterization which is harmonious to the discursive nature of attitude report which we have been discussing. Although this principle seems to capture the cases of our interest, there is a small problem of analysis. The analysis of Soames takes the cases as a standard in which not only the reporter and her audience share the additional information but also the believer shares it as well. This assumption is not generally valid and Soames is well aware of it. If the believer does not share any discursive context, she has different information than the reporter and her audience. According to Soames (2002: 222-224), this situation can be explained by weakening the shared descriptive background information. In other words, the believer, the reporter and the listener still share some very thin descriptive information even for the cases where the believer and the listener do not share any context. The question we face is this: how to find any common background information shared among all the players in (14) and (15)? Recall that Betty and Lucy do not share any contexts (it can be even the case that Betty and Lucy live in two completely different ages); they use two distinct proper names; and these proper names are associated with two distinct clusters of information so that the transitions between contexts are only possible because Rudolf keeps track of referring relations between the two names and the person. This transition does not have to keep any descriptive information attached to each of the names. It is not the right strategy here to weaken the descriptive common information in order to explain the opacity. Although Soames treats the cases like those we have in (14) and (15) as exceptional to his theory, they are more general in nature, what we have observed in the last section. Hence, his failure in explaining our more general cases undermines his theoretical account.

541 4. Capturing what others think Shared common information among three participants did not explain the discursive opacity. In this section, a version of conceptual role semantics (CRS) will be examined in order to judge whether it can explain both the classic and discursive opacities. Furthermore, it will be sketched how both types of opacities relate to each other. 4.1. Conceptual role semantics CRS is a semantic and pragmatic theory determining truth-evaluable contents of each utterance. According to CRS, contents of one’s utterance, say Betty’s utterance of (1), are determined by their conceptual role, such as inferential connections concerning the utterance in question. Inferential connections are connections among speech acts (Brandom 2000) and attitudes (Cummins 1992) rather than among their contents. A few things are worth noticing here. First, CRS is wrong as a theory for the representational contents of propositional attitudes. Although content individuations are dependent on the speech acts and attitudes one has, the individuated contents are contents about states of affairs external to the individuals. This is also indicated by the fact that “it is possible to describe two computational systems that have all the same attitudes, but which differ in how those attitudes are realized” (Cummins 1992: 111). Second, due to its use of inferential connections, CRS is holistic. Each content must be located in a whole inferential nexus. Holism seems to have a problem with compositionality because an inferential role of complex expression, say ‘brown cow’, depends “not only on the inferential role of ‘brown’ and the inferential role of ‘cow’, but also on what you happen to believe about brown cow” (Fodor and LePore 1991: 334, original italics). However, as Block (1993) argues, there is a holistic notion of compositionality. Basically, a semantic theory is compositional if and only if there is a functional relation between a semantic value of a sentence or a report and semantic values of the part expressions contained by the sentence or report. This notion of

542 compositionality is held if we explain semantic values of part expressions in terms of semantic values of sentences which contain each part expression in question. Block calls this kind of compositionality “hyper-compositionality”. Third, the holism used here can be an individualistic theory as well as a social one. Block (1987) introduces the distinction between narrow and wide contents. The narrow content is determined solely by the inferential nexus of each person and hence, it supervenes on her physical constitution. Contrastingly, wide contents are determined also by individuals outside of one’s mind which are referred to. There are two parties of wide content theorists. I follow Block (1987) and call the one party “two-factor theorists” and the other “one-factor theorists”. A version of CRS which captures the wide contents is a two-factor theory (Field 1977, 1978; Loar 1982; Block 1987, 1993). It postulates two factors which are relevant for content individuation. The first factor captures the conceptual roles which are ascribed to one’s mind and the other is “an external component that has to do with the relations between the representations in the head (with their internal conceptual roles) and the referents and/or truth conditions of these representations in the world” (Block 1987: 627). Harman (1982 and 1999) proposes a different version of CRS for determining the wide contents.7 According to him (1999: 221), the content of a thought is “a matter of how mental states are related to each other, to things in the external world, and to things in a context understood as a normal context”. Against two-factor theories, Harman (1999: 224-225) argues that there is no clear border between the internal conceptual roles (narrow contents) and the external contexts. Furthermore, he points out that the internal conceptual roles are partly specified in the external contexts, i.e. the concepts are partly individuated in relation to the things in the external world.

7

According to Harman (1999: 207), “the basic use of symbols is taken to be in calculation, not in communication”. This conception of content looks not harmonious to our communicative approach towards attitude reports. However, I think this point is well separable from his claims about content individuations.

543 4.2. Application to the attitude reports My claim will be this: A version of CRS individuates the truth-evaluable contents expressed by insets of attitude reports most properly. This claim has two parts: First, CRS can capture an inferential nexus of others who are not members of the discursive context. Second, CRS explains opaque occurrences of singular terms. The idea of applying CRS to the contents expressed by insets is not remarkably new. Especially, there are some interesting applications by the narrow content theorists (Brandom 1994: 497; 2000: Chapter 5; Dresner 2010). However, narrow contents are individualistic and belong to someone. Since a believer has an inferential nexus different from the one a reporter has, it is not clear whose nexus is in charge of individuating the content in question. There are proposals which address this issue. For example, Brandom (2000: Chapter 5) claims that the reporter’s inferential nexus is used in the transparent report and the believer’s is used in the opaque report.8 But this does not work. According to his strategy, insets of opaque attitude reports would express contents individuated in an inferential nexus of the believer. However, the believer does not always share any discursive context with the audience of the report as we have discussed in connection with Soames above. It means that the believer’s inferential nexus is epistemologically inaccessible for the audience in normal contexts. This is problematic for Brandom because contents are grasped, and hence individuated, by the believer and her audience essentially in a discursive context. Furthermore, this strategy is not applicable if we take opacity as a feature of each occurrence of a singular term. There are reports containing more than one singular term such as (15a). It may very well be the cases for these reports that some of the singular terms occur transparently while others occur opaquely. This case cannot be explained in Brandom‘s way of treating opacity, which presupposes that all of the

8

Brandom prefers the terminology of de re/de dicto but this difference does not have any influence here.

544 occurrences of singular terms are either transparent or opaque in a report. One of the lessons learned from this consideration is this: if you apply CRS to the content expressed by the insets of attitude reports, you have to extend an inferential nexus of the reporter so that you can catch the inferential nexus of the believer. The opacity must be explained in the extended inferential nexus. This is where the wide content theories are needed. In order to extend the inferential/functional nexus, one has to merge the inferential/functional nexus of each participant in the communicative framework. This merged inferential/functional nexus cannot be individualistic; rather it is social. The two-factor theory is more ad hoc in applying this point than the one-factor theory. Because of the non-holistic character of the second factor, the two factor theory seems to have to introduce a third factor to capture the merger, which should characterize the inferential/functional nexus of the believer. By introducing the third factor, this proposal is not only ad hoc but it faces the same problem as Brandom does if an inset of an attitude report contains more than one singular term. Namely, the distinction between the opaque and transparent belief reports is determined in accordance with which inferential nexus is used to evaluate the reports. Contrastingly, the one-factor theory counts the merged inferential connections as a part of external connections which originally captures the functional connections of the outer world. This is to say, the inferential/ functional connection of the believer is characterized in terms of her verbal reaction in giving and asking for reasons. 4.3. Extending nexus Let us now try to characterize informally how the extension works. Recall the situation depicted in (14) and (15) (section 2). Suppose a fragment of Rudolf’s inferential/functional nexus consists of seven assertions, namely |ρ| = {P1, P2, P3, P4, P5, P6, P7} where:

545 P1 P2 P3 P4 P5 P6 P7

Willy Brandt was born in Lübeck Herbert Frahm was born in Lübeck Willy Brandt is Herbert Frahm Anybody who was born in Lübeck has seen the city hall Willy Brandt has seen the city hall Herbert Frahm has seen the city hall Anybody who was not born in Lübeck does not know what the Sieben Türmen are

All of them are located in Rudolf’s global inferential/functional nexus. Now, suppose a fragment of his inferential nexus contains the following inferences: ρ = {{P1, P3} ⊢ P2; {P2, P3} ⊢ P1; {P1, P4} ⊢ P5; {P2, P4} ⊢ P6; P7 ⊢}

The semantic content of, for example, P1 is defined as a subset of ρ which contains all the inferences including P1. Suppose further that a fragment of Betty’s inferential nexus consists of five assertions, namely |β| = {P7, P8, P9, P10, P11} where: P8 P9 P10 P11

Every kid in my neighborhood was not born in Lübeck Herbert Frahm was a child in my neighborhood Herbert Frahm was not born in Lübeck Herbert Frahm does not know what the Sieben Türmen are

Suppose further that a fragment of her inferential nexus contains following inferences: β = {{P8, P9} ⊢ P10; {P7, P10} ⊢ P11} As we have discussed, β is not epistemologically accessible to Rudolf and all he can get in the context A is Betty’s verbal assertion shown in (14), namely P9 and P10. Based on these speech acts, Rudolf adds the following to his inferential nexus β/ρ: First, P′9 and ⊢ P10 are introduced without justifying it:

546 ⊢ P′9; ⊢ P10; where: P′9 Herbert Frahm was a child in Betty’s neighborhood. Furthermore, if they are connected to the inferential/functional nexus of Rudolf, the following inferences should be also introduced: {P10, P1, P3} ⊢ ⊥; {P10, P2} ⊢ ⊥; {P10, P7} ⊢ P11 However, it is assumed that ⊥ does not belong to the inferential/functional nexus. Rudolf must modify the nexus in order to avoid it. There are two ways to do this: One way is to keep P3 and delete P1 and P2 from β/ρ so that the reporter does not assume that these are part of the inferential/functional nexus which is relevant for evaluating the truth of the attitude report. The other way is to keep P1 and delete P2, P3 from β/ρ. Let us call the former β/ρt and the latter β/ρo: |β/ρt| = {P3, P7, P′9, P10, P12, P13} β/ρt = {⊢ P′9; ⊢ P10; {P3, P10} ⊢ P12; {P7, P10} ⊢ P11; {P7, P12} ⊢ P13} |β/ρo| = { P1, P3, P4, P5, P6, P7, P′9, P10, P11}

β/ρo = {⊢ P′9; ⊢ P10; {P2, P4} ⊢ P6; {P7, P10} ⊢ P11; {P1, P10} ⊢ P3}

where: P12 Willy Brandt was not born in Lübeck P13 Willy Brandt does not know what Sieben Türmen are Now, here is my proposal: For every β/ρx, the attitude report (15a) is true in β/ρx just in case P12 is supported in β/ρx, i.e. P12 is a member of |β/ρx|. The question about which reading we should take is determined contextually. In order to explicate this thesis, we have to investigate how

547 to make sense of the notion of truth in such a holistic semantics. This is a major issue concerning CRS in general and has to be discussed separately. These two extensions correspond to the distinction between transparent opaque occurrences of singular terms in the context A. Recall that if you use CRS as a theory for content individuation then the compositionality in question is hyper-compositionality. That is to say, P3 has as its semantic content a set of pairs of sentences which differ in one occurrence of the singular terms. If P3 is introduced in the nexus, the substitution between ‘Willy Brandt’ and ‘Herbert Frahm’ is licensed. Not only the transparent understanding but also the opaque understanding of Rudolf is distinct from the original nexus of Betty. On the one hand, P9, P10 are introduced without any justifications in both of β/ρt and β/ρo. On the other hand, they are connected via P8 in β. Although they are different from the inferential nexus of the believer, both transparent and opaque understandings are still essentially images of it because they capture part of the original nexus, such as {P7, P10} ⊢ P11 Let us now turn to the discursive opacity. In the context B, Rudolf conveys some of the information contained in β/ρ to Lucy. As we have been discussing, this is where the discursive opacity plays a role. The discursive opacity is understood as Rudolf’s modification of β/ρ so that Lucy can apprehend the content Rudolf conveys. In our informal model, it is to understand that part of the inferential connections in the β/ρ will be canceled. In our example of (15), P3, P10 and hence, P11 will be canceled from β/ρt and hence, β/ρd looks as follows: β/ρd = {⊢ P′9; ⊢ P10; {P7, P12} ⊢ P13} The major difference between β/ρo and β/ρd is described as follows. On the one hand, the classic opacity and transparency are introduced in order to avoid contradiction in an inference nexus. On the other hand, the discursive opacity is introduced after the choice between β/ρt and β/ρo

548 is made. That is to say, Rudolf can introduce the discursive opacity based on β/ρo too. 4.4. Implications This way of characterizing the content individuation of insets has some consequences for the discussion about the nature of opacity. First and foremost, opacity is related to the frame rather than the inset. Recall that the opacity is explained in terms of the relations which are depicted as the inferential/functional nexus. A belief report is therefore true just in case the content expressed by the inset of the belief report is supported by the inferential and functional nexus. It is this supporting relation that the frame expresses. Second, this characterization does not require contents of insets to be descriptive. In our characterizations, it is indifferent for the opacity whether the contents expressed by insets are descriptive or not. This is the reason why the distinction between opacity and transparency is used in this paper rather than de dicto/de re distinction. Third, although the content of any inset is not necessarily descriptive, it is determined relative to each inferential nexus and hence, each occurrence of P7 in β/ρt, β/ρo and β/ρd expresses a different content. 5. Conclusion This paper discussed the following theses. First, attitude reports are discursive in nature if we accept the deixis as better indications for the indirectness of attitude reports than the classic indications which Frege assumes. Furthermore, the deictic features are relevant for the truth of attitude reports. Second, there are two sorts of opaque occurrences of singular terms if we evaluate attitude reports in their discursive contexts. This is because there are always two contexts involved for an attitude reports. The first sort of opacity is the classic opacity due to different modes of presentation believers have and the other is discursive opacity due to different modes of presentation the reporters or the listeners have.

549 Finally, it is argued that a version of one factor CRS explains these opacities. An attitude report is true just in case the content expressed by the inset of the attitude report is supported by the inferential/functional nexus. Our version of CRS postulates the inferential and functional nexus in the discursive context. Acknowledgements I would like to acknowledge Albert Newen, Shigeki Noya, Colin Allen, Leon de Bruin, Markus Eronen, Anika Fiebich, Lena Kästner, Christoph Michel, Ulrike Pompe, Raphael van Riel, Ryoji Sato, Tobias Starzak and Anna Welpinghus for helpful comments. Shuichi Kuchikata suggested many linguistics sources in the Japanese language. An earlier version of this paper was delivered at PhiLang2011 conference in Łódź, Poland. References Beaney, Michael (ed). 1997. The Frege Reader. Malden, MA: Blackwell. Block, Ned 1987. Advertisement for a Semantics for Psychology. Midwest Studies in Philosophy 10(1), 615-678. Block, Ned 1993. Holism, Hyper-analyticity and Hyper-compositionality. Mind and Language 8(1), 1-26. Brandom, Robert B. 1994. Making It Explicit: Reasoning, Representing, and Discursive Commitment. Cambridge, MA: Harvard University Press (Page citation based on the paperback edition in 1998). Brandom, Robert B. 2000. Articulating Reasons: An Introduction to Inferentialism. Cambridge, MA: Harvard University Press. Cummins, Robert E. 1992. Conceptual Role Semantics and the Explanatory Role of Content. Philosophical Studies 65(1-2), 103-127. Davidson, Donald 1968. On Saying That. Synthese 19(1), 130-146. Dresner, Eli 2010. Language and the Measure of Mind. Mind and Language 25(4), 418-439. Field, Hartry H. 1977. Logic, Meaning, and Conceptual Role. The Journal of Philosophy 74(7), 379-409. Field, Hartry H. 1978. Mental Representation. Erkenntnis 13, 9-61. Fillmore, Charles J. 1997. Lectures on Deixis. 65 of CSLI Lecture Notes Stanford, CA: CSLI Publications.

550 Fodor, Jerry and Ernest LePore 1991. Why Meaning (Probably) Isn’t Conceptual Role. Mind and Language 6(4), 328-343. Frege, Gottlob 1892. Über Sinn und Bedeutung. Zeitschrift für Philosophie und philosophische Kritik, NF 100, 25-50. Frege, Gottlob 1923. Logische Untersuchungen. Dritter Teil: Gedankengefüge Harman, Gilbert 1982. Conceptual Role Semantics. Notre Dame Journal of Formal Logic 23(2), 242-256. Harman, Gilbert 1999. (Non-solipsistic) Conceptual Role Semantics. In: Reasoning, Meaning and Mind. Oxford: Clarendon Press, 206-231. (Originally published in Ernest LePore (ed.), New Directions in Semantics, London: Academic Press, 1987). Huddleston, Rodney 2002a. Content Clauses and Reported Speech. In: Rodney Huddlesteon and Geoffrey K. Pullum (eds.), The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press, 947-1030. Huddleston, Rodney 2002b. The Verb. In: Rodney Huddlesteon and Geoffrey K. Pullum (eds.), The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press, 71-212. Kamada, Osamu 2000. Nihongo-no in’yo [Quotation of Japanese]. Tokyo: Histuji Shobo. Loar, Brian 1982. Conceptual Role and Truth-conditions: Comments on Harman’s Paper: “Conceptual role semantics”. Notre Dame Journal of Formal Logic 23(3), 272-283. Maynard, Senko K. 1996. Multivoicedness in Speech and Thought Representation: The Case of Self-quotation in Japanese. Journal of Pragmatics 25(2), 207-226. Nihongo Kizyutsu Bumpô Kenkyûkai (ed.) 2008. Gendai Nihongo Bumpo 6: Fukubun [Modern Japanese Grammar 6: Compound Sentence]. Tokyo: Kuroshio shuppan. Recanati, François 2000. Oratio Obliqua, Oratio Recta: An Essay on Metarepresentation. Cambridge, Mass.: MIT Press. Soames, Scott 2002. Beyond Rigidity: The Unfinished Semantic Agenda of “Naming and Necessity”. Oxford: Oxford University Press. Sternberg, Meir 1982. Proteus in Quotation-land: Mimesis and the Forms of Reported Discourse. Poetics Today 3(2), 107-156. Tsujimura, Natsuko 2007. An Introduction to Japanese Linguistics (2nd edition). Oxford: Blackwell.

551

Index of Names Ajdukiewicz, Kazimierz 1, 9, 85, 336, 368, 369, 370, 371, 375, 377, 378 Aristotle 37, 38, 40, 44, 45, 49, 50, 53, 358, 373 Bach, Kent 10, 492, 502, 504, 505, 506, 507, 508 Barsalou, Lawrence 296, 298, 299, 300, 301 Benacerraf, Paul 152, 153, 191 Boghossian, Paul 2, 3, 15, 16, 17, 18, 19, 20, 21, 22, 23, 25, 26, 27, 28, 29, 31, 34, 485 Buridan, Jean 3, 37, 38, 39, 40, 41, 42, 45, 46, 47, 48, 53 Carey, Susan 296, 303, 305, 306, 308, 309, 310 Carlson, Lauri 3, 57, 58, 59, 61, 64, 65, 66, 73, 440 Carnap, Rudolf 163, 357, 366, 393, 458 Chomsky, Noam 1, 4, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 154, 180, 238, 306, 406, 438, 449 Cmorej, Pavel 162, 163 Davidson, Donald 1, 9, 10, 59, 365, 366, 367, 371, 374, 375, 376, 377, 378, 379, 380, 381, 429, 430, 432, 433, 434, 437, 438, 446, 451, 457, 458, 459, 461, 465, 468, 469, 471, 472, 477, 528

Devitt, Michael 11, 128, 492, 502, 503, 504, 505, 506, 507, 508, 510 Doležel, Lubomir 4, 57, 58, 68, 73, 74 Duhem, Pierre 375 Dummett, Michael 10, 23, 153, 157, 202, 470, 477, 478, 479, 480, 481, 482, 484, 486, 487, 488, 521, 522 Fodor, Jerry 296, 298, 302, 303, 304, 306, 308, 310, 311, 313, 541 Frege, Gottlob 1, 10, 11, 24, 72, 246, 260, 357, 358, 359, 362, 364, 365, 366, 367, 368, 457, 458, 459, 460, 461, 462, 463, 464, 465, 466, 467, 468, 469, 470, 471, 472, 473, 478, 479, 527, 528, 529, 530, 531, 533, 536, 537, 539, 548 Gibbard, Allan 8, 276, 281, 315, 316, 317, 318, 320, 321, 322, 323, 326, 328, 329, 330 Glüer, Kathrin 15, 18, 27, 28, 33, 283 Grice, Paul 205, 253, 487, 488, 492, 493, 494, 495, 496, 500, 503, 504, 506, 508 Hintikka, Jaakko 3, 52, 57, 58, 59, 60, 61, 62, 63, 64, 66, 67, 68, 69, 70, 71, 72, 73, 74 Husserl, Edmund 359, 368, 516, 518, 521, 522, 524

552 Ingarden, Roman 64, 358, 359 Kaplan, David 6, 81, 197, 198, 200, 201, 202, 208, 248, 401 King, Jeffrey 6, 9, 83, 84, 177, 178, 179, 180, 181, 182, 183, 184, 186, 187, 188, 189, 190, 191, 192, 193, 194, 385, 386, 387, 388, 389, 390, 391, 392, 393, 394, 395, 396, 397, 398, 400, 401, 402, 496 Kripke, Saul 1, 5, 59, 127, 128, 129, 130, 133, 134, 135, 136, 137, 138, 139, 140, 142, 198, 246, 264, 266, 357, 378, 380, 424, 494, 499, 502 Larson, George 10, 430, 434, 442, 443, 445, 446, 447, 448, 451 McDowell 10, 477, 478, 480, 482, 483, 486, 488, 489 Mill, John Stuart 5, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 357, 364, 378 Montague, Richard 8, 69, 72, 147, 148, 151, 155, 334, 336, 354, 357, 401, 441 Nunberg, Geoffrey 241, 242, 243, 244, 245, 248, 251, 253, 255

Powers, Lawrence 3, 37, 44, 45, 46, 47, 48, 49, 50, 51, 52, 54 Predelli, Stefano 88 Quine, Willard Van Orman 1, 9, 10, 15, 16, 22, 151, 245, 308, 365, 367, 368, 374, 375, 376, 377, 433, 457, 458, 459, 466, 467, 468, 469, 471, 472 Ramsey, Frank 316, 399, 400, 401 Russell, Bertrand 1, 9, 198, 266, 360, 361, 362, 363, 364, 365, 386, 392, 393, 395, 399, 402, 458, 465, 502, 503 Stanley, Jason 6, 88, 197, 202, 203, 204, 205, 207, 208, 209, 401 Tarski, Alfred 1, 9, 59, 72, 93, 371, 372, 373, 374, 375, 376, 469 Travis, Charles 4, 87, 88, 89, 94, 103, 104, 105, 107, 108, 109, 208 Wittgenstein, Ludwig 1, 3, 10, 57, 59, 62, 63, 64, 68, 113, 357, 369, 392, 393, 406, 423, 424, 457, 458, 465, 482 Wright, Crispin 4, 9, 21, 30, 31, 113, 405, 406, 407, 408, 410, 411, 415, 418, 420, 422, 423, 424, 425, 481, 484, 485, 489

Index of Subjects abstract notion, 296, 299, 301, 306, 308, 310, 311, 312 adverbial vs. adjectival modification, 10, 429, 434, 445, 446, 451 ambiguity, 40, 41, 45, 47, 48, 49, 66, 87, 97, 152, 163, 274, 318, 320, 321, 322, 326, 329, 440, 442, 443, 450, 500, 501, 502, 503, 510 analyticity, 15 anaphora, 7, 216, 217, 227, 231, 234, 241, 246, 249, 250, 251, 254, 255, 256, 257, 258 attitude reports, 2, 6, 11, 177, 179, 185, 186, 189, 191, 195, 527, 528, 529, 532, 533, 534, 535, 536, 537, 538, 542, 543, 544, 548, 549 ballistic requirement, 88, 103, 104, 107, 108, 109 belief, 26, 27, 30, 31, 32, 33, 41, 43, 45, 67, 69, 70, 71, 75, 177, 250, 276, 279, 282, 286, 317, 318, 325, 328, 359, 360, 363, 368, 369, 370, 375, 376, 377, 378, 380, 381, 386, 393, 416, 421, 469, 484, 487, 534 biological internalism, 118, 119 calibration, 104, 105 cognitive content, 10, 170, 174, 457, 458, 459, 461, 465, 471 cognitive meaning, 163, 164, 165, 166, 167, 168, 169

compositionality, 7, 64, 72, 168, 170, 174, 235, 237, 359, 360, 365, 366, 375, 460, 461, 466, 467, 471, 472, 473, 541, 542, 547 concept empiricism, 295, 298 conceptual role semantics (CRS), 2, 11, 368, 527, 541, 542, 543, 544, 547, 549 conditional non-contradiction (CNC), 8, 315, 316, 318, 323 connotation, 129, 132, 133, 137, 138, 141, 142 consequence relation, 334, 344, 345, 347, 354 context, 6, 215, 216, 217, 218, 220, 221, 222, 224, 228, 230, 231, 232, 233, 234, 235, 237 context of utterance, 246, 249, 250, 255, 320, 391, 396, 400 context-sensitivity, 63, 107 contextual parameters, 171, 172, 173, 174 contextualism, 2, 168, 492, 495 continuity theory of concepts, 302 declarative unit, 334, 335, 338, 339, 340, 341, 345, 346, 347, 348, 353 deferred reference, 7, 241, 242, 243, 244, 245, 246, 249, 254, 255, 256, 257, 258

554 deixis, 11, 527, 528, 530, 532, 533, 534, 535, 536, 537, 538, 548 denotation, 130, 131, 133, 136, 138, 142, 181, 186, 358, 439 descriptive indexicals, 7, 241, 242 direct reference, 197, 198, 199, 201, 202, 243, 249 discourse games, 3, 57, 58, 64, 65, 73 discursive context, 11, 527, 533, 535, 537, 538, 540, 543, 549 Dynamic Syntax (DS), 2, 6, 8, 215, 221, 222, 224, 226, 227, 228, 234, 235, 238, 333, 334, 336, 354 entailment, 6, 28, 95, 96, 177, 185, 189, 190, 430, 431 equivocation, 3, 37, 41, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54 events, 10, 114, 150, 156, 301, 324, 377, 381, 397, 423, 424, 429, 432, 433, 434, 435, 437, 438, 439, 442, 443, 447, 448, 451, 468 fallacy, 2, 3, 37, 38, 39, 40, 41, 42, 43, 44, 45, 47, 48, 49, 50, 51, 52, 53, 54, 207 fallacy theory, 3, 44, 46, 49, 54 first language acquisition, 11, 238, 513 formal semantics, 58, 88, 89, 90, 92, 93, 94, 95, 97, 99, 101, 102, 103, 108, 146, 158 Fregean sense, 361, 457, 458, 460, 473

full-blooded theory of meaning, 10, 477, 478, 479, 480, 481, 482, 483, 489 Game-Theoretical Semantics (GTS), 2, 3, 57, 58, 59, 60, 61, 63, 64, 65, 69, 72, 73, 76, 77 general propositions, 197, 241, 244, 258 general terms, 5, 128, 129, 130, 133, 138, 140, 141, 142, 516 generalized quantifier, 247 holism, 63, 369, 459, 472, 541, 542 implicit definition, 2, 3, 15, 16, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 30, 31, 32, 33, 35 implicit knowledge, 2, 10, 477, 478, 480, 481, 482, 486, 489 indeterminacy of translation, 368, 376, 467, 468 Indexicalism, 208, 209 indexicals, 2, 7, 8, 167, 168, 174, 197, 198, 201, 202, 206, 208, 209, 210, 241, 242, 244, 245, 246, 247, 248, 250, 252, 253, 258, 264, 320, 321, 331, 460 indicative conditionals, 315, 316, 321 inscrutability of reference, 367, 368, 467 intensional logic (IL), 81, 148, 163 intentionality, 2, 11, 39, 149, 307, 308, 513, 514, 515, 516, 517, 518, 519, 520, 521, 522, 523 Japanese, 11, 183, 184, 527, 528, 530, 532, 533, 534, 549

555 judgement-dependence, 2, 9, 405, 411, 412, 413, 415, 423, 425, 426, 427 label, 193, 276, 287, 306, 335, 336, 337, 339, 341, 342, 345, 346, 399 labelled formula, 337, 338 lambda calculus, 230, 340, 343 Linear Dynamic Syntax, 2, 8, 333, 334, 335, 336, 349, 354 linguistic expression, 6, 11, 18, 88, 92, 99, 102, 103, 105, 161, 162, 163, 164, 166, 167, 168, 169, 170, 171, 172, 173, 174, 175, 334, 358, 359, 360, 370, 459, 513 linguistics, 1, 4, 6, 113, 116, 123, 125, 147, 161, 164, 165, 168, 190, 231, 324, 406, 415, 466, 492 logic, 1, 6, 8, 10, 16, 17, 21, 23, 29, 37, 48, 58, 61, 67, 69, 70, 81, 147, 155, 161, 224, 229, 238, 276, 333, 334, 335, 354, 360, 430, 431, 432, 435, 436, 437, 438, 452, 462, 478 logical form (LF), 6, 41, 45, 47, 48, 49, 53, 70, 177, 180, 181, 182, 184, 185, 188, 208, 209, 235, 389, 398, 402, 431, 432, 442, 534 logico-syntactic truth-conditions, 92, 93 mereological fallacy, 113, 118, 119, 120, 121, 122, 123, 125 methodological naturalism, 2, 4, 5, 113, 114, 115, 116, 117, 118, 122, 125

methodology, 4, 113, 116, 117, 120, 123, 125, 148, 150, 228, 510 Minimalism, 208 modal games, 2, 3, 57, 58, 66, 68, 69, 72, 73 modal properties, 6, 178, 197, 202, 203, 204, 386 modesty, 478, 480, 483 Montague grammar, 8, 333, 334, 354 Montague semantics, 148, 151, 154 narrative modalities, 4, 57, 58, 64, 74 natural kind terms, 2, 5, 28, 127, 128, 129, 133, 134, 135, 136, 137, 138, 140, 141, 142 natural kinds, 128, 134, 135, 136, 137, 138, 139, 140, 141, 142 natural language (NL), 1, 2, 3, 8, 9, 57, 58, 60, 61, 63, 65, 66, 70, 76, 77, 163, 168, 188, 208, 216, 218, 222, 226, 228, 229, 236, 237, 247, 324, 333, 336, 347, 357, 358, 359, 360, 361, 364, 365, 366, 367, 368, 369, 370, 374, 375, 378, 380, 381, 395, 398, 429, 430, 434, 435, 436, 451, 452, 507, 530 naturalism, 114, 118 non-referential semantics, 2, 8, 9, 357, 360, 361, 374, 380, 381 normativity of meaning, 2, 273, 274, 279, 281, 282, 283, 287, 289, 290, 291, 292 norm-relativity, 289

556 One Fallacy Theory, 3, 37, 44, 46, 48, 54 ontological commitments, 377, 451, 472 ontological compromise, 457, 458, 459, 461, 462, 465, 467, 471, 473 ontology, 2, 10, 145, 153, 156, 158, 179, 235, 358, 359, 360, 394, 399, 433, 457, 461, 468, 470, 471, 473 perception, 8, 295, 298, 299, 300, 301, 303, 513, 516, 521, 522 Polish notation, 337 possible worlds, 5, 57, 58, 59, 66, 67, 73, 74, 75, 81, 82, 83, 135, 139, 145, 147, 148, 149, 150, 152, 155, 156, 167, 190, 201, 202, 203, 318, 319, 389, 392, 397, 398, 414, 436, 441, 443 possible-worlds semantics, 57, 59, 65, 66, 67, 69, 316, 389, 402 pragmatic context, 161, 162, 167, 168, 169, 170, 171, 173, 174 pragmatics, 10, 63, 162, 168, 171, 191, 221, 222, 235, 400, 473, 491, 492, 493, 496, 499, 501, 508, 510 privileged access, 3, 15, 17, 19, 23, 24, 25, 28, 29, 32, 35 proper names, 2, 7, 28, 84, 127, 128, 133, 197, 201, 202, 204, 263, 264, 265, 266, 267, 268, 269, 270, 271, 364, 378, 380, 390, 437, 461, 463, 470, 471, 538, 540

propositional attitudes, 11, 67, 69, 250, 371, 378, 380, 386, 421, 465, 468, 541 propositional/sentential games, 58, 60, 61, 65, 73 propositions, 4, 6, 8, 9, 18, 19, 25, 49, 52, 53, 59, 61, 62, 65, 67, 69, 70, 72, 74, 81, 82, 83, 84, 85, 130, 148, 156, 167, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 197, 199, 200, 201, 202, 203, 204, 207, 210, 211, 212, 217, 276, 277, 278, 281, 315, 316, 318, 319, 320, 321, 322, 329, 330, 334, 385, 386, 387, 388, 389, 390, 391, 392, 393, 394, 395, 396, 397, 398, 399, 400, 401, 402, 501, 523 psychologism, 10, 457, 458, 478, 480, 481, 482, 483 Ramsey-sentence, 399, 401 rationality, 130, 141, 273, 274, 276, 292, 317, 319 reference, 9, 26, 70, 72, 84, 85, 127, 129, 133, 134, 136, 137, 138, 139, 140, 141, 142, 167, 168, 174, 180, 202, 203, 206, 245, 246, 249, 279, 282, 304, 307, 357, 358, 359, 360, 361, 365, 366, 367, 368, 373, 376, 380, 381, 434, 437, 438, 439, 458, 464, 467, 468, 470, 471, 472, 493 referential semantics, 357, 358, 359, 360, 361, 365, 366, 367, 368, 370, 371, 380, 381

557 relevance, 34, 52, 193, 249, 414, 492, 493, 494, 495, 508, 514, 522, 528, 537 rule-following, 63, 406, 407, 415, 482, 483, 484, 485, 486 semantic content, 264, 265, 266, 267, 268, 269, 338, 493, 545, 547 semantic creativity, 407, 408, 419 semantic realism, 279 semantics, 2, 4, 6, 7, 9, 10, 58, 59, 65, 70, 72, 73, 74, 77, 83, 95, 96, 148, 149, 151, 152, 154, 155, 157, 158, 161, 162, 165, 168, 171, 175, 177, 178, 185, 187, 189, 190, 191, 194, 220, 221, 225, 228, 229, 230, 235, 263, 264, 265, 266, 268, 275, 276, 316, 323, 327, 357, 358, 359, 360, 361, 366, 367, 368, 369, 374, 376, 377, 381, 389, 393, 396, 400, 413, 419, 429, 433, 443, 452, 457, 458, 459, 460, 461, 462, 464, 465, 466, 468, 470, 471, 472, 473, 491, 492, 493, 496, 499, 501, 505, 507, 508, 510, 547 semi-expression, 163, 167 semi-expressions, 162 set theory, 5, 81, 145, 147, 148, 149, 150, 151, 152, 153, 155, 156, 157, 158, 359, 436, 437, 438 singular propositions, 2, 6, 197, 198, 199, 200, 201, 202, 204 stage-level vs. individual-level modification, 10, 429, 440, 441, 443, 444, 445

structured propositions, 2, 4, 6, 81, 83, 85, 177, 178, 180, 183, 189, 192, 194, 397, 402 substantial truth-conditions, 92, 93, 94, 95, 103, 104 synonymy, 6, 161, 162, 164, 165, 166, 167, 171, 495, 496 syntax, 6, 83, 146, 180, 181, 182, 184, 185, 188, 192, 193, 194, 215, 217, 221, 222, 227, 228, 229, 230, 235, 264, 270, 398 tacit knowledge, 2, 9, 405, 406, 417, 418, 419 theory of meaning, 2, 8, 9, 66, 357, 359, 368, 369, 371, 374, 375, 377, 416, 417, 420, 461, 465, 468, 469, 477, 478, 479, 480, 481, 484, 486, 487, 488, 489, 528 theory of truth, 9, 357, 371, 372, 373, 374, 375, 376, 377, 378, 380, 465 third realm, 10, 457, 458, 461, 462, 463, 464, 471, 473, 479 translation, 5, 9, 123, 145, 146, 307, 354, 357, 365, 368, 369, 371, 372, 373, 374, 375, 376, 377, 385, 463, 467, 468, 469 translations, 148, 395 Travis cases, 4, 87, 88, 94, 103, 104, 105, 107, 108, 109 uniformity, 134, 263, 264, 265, 324 unity of propositions, 2, 385, 386 warrant transmission, 3, 15, 19, 32, 33

Linguistics&Philosophy

Piotr Stalmaszczyk (Ed.)

Philosophy of Language and Linguistics Papers gathered in the two volumes investigate the complex relations between philosophy of language and linguistics, viewed as independent, but mutually influencing one another, disciplines. They concentrate on the ‘formal’ and ‘philosophical’ turns in the philosophy of language, initiated by Gottlob Frege, with further developments associated with the work of Bertrand Russell, Ludwig Wittgenstein, Kazimierz Ajdukiewicz, W.O.V. Quine, Richard Montague, Pavel Tichý, Richard Rorty. The volumes bring together contributions by philosophers, logicians and linguists, representing different theoretical orientations but united in outlining the common ground, necessary for further research in philosophy of language and linguistics. The papers were submitted and, in most cases, presented at the first International Conference on Philosophy of Language and Linguistics, PhiLang2009, organized by the Chair of English and General Linguistics at the University of Łódź. About the editor Professor Piotr Stalmaszczyk holds the Chair of English and General Linguistics at the University of Łódź, and he is the Dean of the Philological Faculty. His research interests cover linguistic methodology, philosophy of language, generative grammar, and Celtic languages. His publications include Structural Predication in Generative Grammar (habilitationsschrift, Łódź 1999), two edited volumes on methodologies in language studies (in Polish), a co-edited volume on Cognitive Approaches to Language and Linguistic Data (Peter Lang 2009), and two books on Celtic languages and language contact in the British Isles.

ontos verlag

Frankfurt • Paris • Lancaster • New Brunswick 2010. 671pp. Format 14,8 x 21 cm Hardcover EUR 229,00 ISBN 978-3-86838-070-5

P.O. Box 1541 • D-63133 Heusenstamm bei Frankfurt www.ontosverlag.com • [email protected] Tel. ++49-6104-66 57 33 • Fax ++49-6104-66 57 34