Semantics and Beyond: Philosophical and Linguistic Inquiries 9783110362480, 9783110354386

Citation preview

Semantics and Beyond

Philosophische Analyse / Philosophical Analysis

Herausgegeben von/Edited by Herbert Hochberg, Rafael Hüntelmann, Christian Kanzian, Richard Schantz, Erwin Tegtmeier

Volume / Band 57

Semantics and Beyond Philosophical and Linguistic Inquiries Edited by Piotr Stalmaszczyk

ISBN 978-3-11-035438-6 e-ISBN 978-3-11-036248-0 ISSN 2198-2066 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.dnb.de. © 2014 Walter de Gruyter GmbH, Berlin/Boston Printing: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Contents Semantics and Beyond: Philosophical and Linguistic Inquiries. Preface Piotr Stalmaszczyk .................................................................................1 On the Normativity of Speech Acts Brian Ball ...............................................................................................9 Representations without Representa: Content and Illusion in Linguistic Theory John Collins .........................................................................................27 Kripke, Putnam, and the Description Theory Luis Fernández Moreno .......................................................................65 The Meaning of Formal Semantics Chris Fox .............................................................................................85 Frege’s Puzzle and the Direct Reference Theory Filip Kawczyski ............................................................................... 109 Situation Semantics, Time, and Descriptive Indexicals Katarzyna Kijania-Placek ................................................................ 127 The Prescription Argument Against the Normativity of Meaning Joanna Klimczyk ............................................................................... 149 The Sense of Finitude and the Finitude of Sense Paul Livingston ................................................................................. 161 Green Leaves Again! An Assessment of Kennedy and McNally’s Solution to Travis’ Challenge Ernesto Perini-Santos ....................................................................... 185 Borg’s Minimalism and the Problem of Paradox Mark Pinder ...................................................................................... 207 Lexical vs. Grammatical Meaning Revisited Tabea Reiner ..................................................................................... 231

vi

Contents

Sentence, Proposition, and Context. On the Idea of an Intermediate Level Stefan Riegelnik................................................................................. 241 On the Semantic Relevance of Romanovs Arthur Sullivan .................................................................................. 255 What Incompleteness Arguments Tell Us about Semantics and Linguistic Competence Massimiliano Vignolo ....................................................................... 271 On Underdetermination of Contextualism Marián Zouhar .................................................................................. 291 Index ..................................................................................................... 313

Piotr Stalmaszczyk University of Łód [email protected]

Semantics and Beyond: Philosophical and Linguistic Inquiries. Preface 0. Introduction Ruth Kempson, in her classical textbook, Semantic Theory, described semantics as a “bridge discipline between linguistics and philosophy” (Kempson 1977: ix). This metaphorical description is still very useful in discussing the scope and aims of semantics, considered as a vital part of both linguistics and philosophy. Some, especially older, studies identified philosophy of language with semantics, and pointed to meaning as the crucial issue in any approach to philosophy of language.1 On the other hand, numerous recent studies and collections of papers have been devoted to discussing the mutual relations between meaning and context, semantics and pragmatics,2 with some of them aiming at ‘pragmaticising meaning’, and even showing the “semanticist the way out of semantics” (Turner 2011: 14). Nevertheless, semantics and meaning have remained in the center of research carried out within contemporary linguistics and philosophy, especially philosophy of language.

1

2

Cf. Caton’s opening statement in his overview of semantics in philosophy: “The part of philosophy known as the philosophy of language, which includes and is sometimes identified with the part known as semantics” (Caton 1971: 3), see also Platts (1979). For a recent overview, see the comprehensive introduction by Davis and Gillon (2004). For some recent collections dealing with the semantics/pragmatics border, including studies from the linguistic and/or philosophical perspective, see the volumes edited by Baptista and Rast (2010); Bianchi (2004); von Heusinger and Turner (2006); Szabó (2005); Turner (2011).

2

Piotr Stalmaszczyk

1. Contents of the volume Brian Ball discusses illocutionary speech acts such as assertion which are subject to norms. He attempts to explain the normativity of these speech acts, focusing on the case of assertion itself. He is concerned not with whether speech acts are essentially conventional, but with whether they are constitutively governed by norms. Next, Ball suggests that one kind of attempt to explain the normativity of assertion by appeal to a reductive view of its essence fails; hence he sketches his own preferred explanation of this normativity which treats assertion as an irreducible kind. He also shows how the account he favors grounds the normativity of speech acts and their essences in more basic facts in a manner that is naturalistically acceptable – provided that the piecemeal conception of philosophical naturalism articulated there is correct. John Collins observes that contemporary linguistic theory seeks to explain linguistic phenomena by way of the attribution of representtations to speakers/hearers. A general philosophical problem arises concerning the apparent intentionality of representational states. A specific instance of this concern is how to understand the representation of a linguistic item when such an item does not exist independently of the representation itself. Collins proposes a resolution of this quandary by reviving a method of hyphenation originally due to Nelson Goodman (finessed and defended against a range of likely objections) and concludes that we can have representations without representa. The representations are states of a system that are typed in terms of the representa, which are the theoretical notions employed by the relevant theory. Luis Fernández Moreno proposes a version of the description theory of reference for natural kind terms that could constitute an alternative to Saul Kripke’s causal theory. This sort of description theory is grounded, on the one hand, in the extension to natural kind terms of some proposals by John Searle and Peter Strawson about proper names and, on the other hand, in the incorporation of a fundamental component of Hilary Putnam’s theory of reference for natural kind terms, that is, the division of linguistic labor. Chris Fox considers some of the foundational issues of formal semantics. These issues include the nature of the data that is to be accounted

Semantics and Beyond: Philosophical and Linguistic Inquiries. Preface

3

for, the nature of the formalism used to characterize that data, and the criteria for determining whether a particular formal account is adequate. If the relevant data is merely the arguments and truth conditions that subjects agree with, then the terms in which the theory is expressed may be irrelevant. But a case can be made that it is also legitimate to be concerned with intuitions that people have about how they reason with language, as well as technical and philosophical issues relating to the formalism chosen. The diversity of issues and objectives is highlighted by considering a selection of different problem domains, and by questioning the sometimes implicit assumptions about what counts as an appropriate formal framework. Ultimately the meaning of formal semantics depends upon assumptions about the nature of the data that is to be accounted for, and on that in what sense the formal theory itself needs to be sympathetic to that data. Filip Kawczyski discusses a possibility of defending the Direct Reference theory from its most dangerous threat which is the notorious Frege’s puzzle. Kawczyski discusses two possible ways of doing that. The first is based on Jeffrey King’s theory of propositions as facts; however, the tools provided by King’s theory are not enough to solve the puzzle. More promising is a method supported by Scott Soames’s new theory of propositions as cognitive event-types. The author tries to show that this framework allows developing a satisfying solution of the puzzle, one which focuses on the notion of cognitive value of sentence. Katarzyna Kijania-Placek argues that descriptive uses of indexicals pose a challenge to the situation semantics of Angelika Kratzer and Stephen Berman. Many of the proposed methods of analysis for descriptive indexicals are based on the situation semantics of Kratzer as supplemented by Berman’s notion of a minimal situation. As Kratzer observed, not all domains are countable, and especially not those that are overlapping, as is the case with some situations. Thus the notion of a minimal situation was initially intended to single out those situations which could serve as a counting domain for quantified sentences. Al-though minimality works well for timeless situations, Kijania-Placek argues that descriptive uses of indexicals require taking time into consideration, and that time, combined with static predicates, poses a challenge to minimality. She also suggests that the notion of a countable situation (intended by

4

Piotr Stalmaszczyk

Kratzer to replace the notion of a minimal situation) fails for cases that involve some static predicates. Joanna Klimczyk provides the prescription argument against the normativity of meaning. She observes that the main argument Kathrin Glüer and Åsa Wikforss build against the normativity of meaning and content relies on the assumption that the essential feature of normativity is guidance. However, they impose quite a strong condition on the normative guidance to be the case, namely they argue that if there is any normative guidance at all, it must take the form of prescription. Klimczyk shows that prescription conceived of as the paradigmatic form of normative guidance is not normative in the intuitive sense of normativity that they favor. That is so because reason-givingness is not an essential property of prescription. Paul Livingston observes that for Martin Heidegger in Being and Time, human existence (Dasein) is essentially finite in its directedness toward death as a final and unavoidable individuating possibility. In Kant and the Problem of Metaphysics, Heidegger, following Kant, further specifies the finitude of Dasein as the capacity to be affected by external objects. This gives rise to the temporal problem of the relationship between sensibility and understanding. There is also an idea of constitutive finitude that is decisive in analytic philosophy of language, and that this idea can usefully be compared with Heidegger’s. On this conception, language is an essentially finite system of terms and recursively applicable rules capable of infinite application to produce new sentences. One of the most developed applications of this picture is Davidson’s conception of the structure of a “theory of meaning” for a natural language. Livingston argues that certain structural aporias and paradoxes arising from this picture of language in the work of the late Wittgenstein, Cavell, and Turing/Gödel point to a different determination of the relationship of language and sense to the infinite. This points to a constitutive infinity of sense which is nevertheless not the theological or absolute infinite rejected by Heidegger. Ernesto Perini-Santos provides an assessment of Kennedy and McNally’s solution to Travis’ challenge. In one of the most discussed contextualist challenges to the minimalist framework, Travis argues that

Semantics and Beyond: Philosophical and Linguistic Inquiries. Preface

5

different uses of the same sentence containing color adjectives can talk about the same state of the world and have different truth-values, and none of usual ways to absorb contextual effects on what is conveyed by an utterance applies. Kennedy and McNally have proposed a solution to Travis’ challenge: color adjectives are ambiguous between gradable and nongradable meanings. Against this claim, Perini-Santos argues that, although Kennedy and McNally give perspicuous representations of the intended readings of the sentences in question, they fail to establish that color adjectives are ambiguous, instead of underdetermined. The different readings of color terms can be explained from a unitary meaning postulating their semantic flexibility. Perini-Santos tries to show that speakers can move from one reading of color adjective to another in a way that seems incompatible with the ambiguity claim. Mark Pinder focuses on Emma Borg’s version of minimalism and the problem of the liar paradox. According to Borg, minimalism is (roughly) the view that natural language sentences have truth conditions, and that these truth conditions are fully determined by syntactic structure and lexical content. A principal motivation for her brand of minimalism is that it coheres well with the popular view that semantic competence is underpinned by the cognition of a minimal semantic theory. Pinder argues that the liar paradox presents a serious problem for this principal motivation. He discusses two lines of response to the problem, and shows difficulties facing those responses. He concludes by issuing a challenge: to construe the principal motivation for Borg’s version of minimalism in such a way so as to avoid the problem of paradox. Tabea Reiner deals with the distinction between lexicon and grammar, particularly between lexical and grammatical meaning. Using the example of function words, she first reviews some relevant literature from Cognitive Linguistics, Formal Semantics as well as Modistic Theory, and next suggests an alternative account which provides clear-cut categories of lexical vs. grammatical meaning. Additionally, Reiner’s account is tentatively extended to syntax and beyond. Stefan Riegelnik observes that in contemporary theories of language it is common to appeal to propositions as expressed by utterances of sentences. Riegelnik’s aim is to question this idea. He argues that the relationship between sentences and propositions cannot be worked out in a

6

Piotr Stalmaszczyk

rewarding way. First, he explores reasons for introducing proposition, next he links them to contemporary theories of interpretation and, finally, he argues that these theories cannot explain the relationship between sentences and propositions. Arthur Sullivan works toward a refined conception of the distinction between what is semantically expressed and what is pragmatically implicated, for the case of proper names. The background context of this contribution is a recent exchange between Delia Graf Fara and Robin Jeshion. Sullivan starts with sketching some of the varieties of cases at issue in the Fara-Jeshion exchange. Next, he briefly sets up a neoGricean conception of the semantics/pragmatics interface. Finally, he explores what that framework has to say about some of these contested cases. Massimiliano Vignolo observes that philosophers and linguists disagree over the logical relation between semantics and linguistic competence. Some theorists maintain that semantic properties are psychologically real in the strong sense that they are represented or built in the psychological processes that are active in the production of linguistic behavior. Other theorists reject such view and hold that the logical relation between semantics and linguistic competence consists in the fact that it is constitutive of the very nature of linguistic competence to produce linguistic expressions which have semantic properties described by semantics. In this weak sense, semantic properties are psychologically real insofar as they are respected in the exercise of linguistic competence. Vignolo argues that there is a constraint that linguistic competence poses on semantics even in the weak sense of the psychological reality of semantic properties and that those semantic theories that violate the constraint are not acceptable. He addresses the issue by discussing Herman Cappelen and Ernie Lepore’s objection to Incompleteness Arguments for context sensitivity. Marián Zouhar concentrates in his contribution on underdetermination of contextualism. According to contextualism, the propositions expressed by utterances of certain kinds of sentences often involve constituents that are unarticulated at the level of syntactic representation. This claim is usually supported by a set of examples collected from everyday communication in which the utterances are taken as expressing

Semantics and Beyond: Philosophical and Linguistic Inquiries. Preface

7

richer contents than those determined solely by semantic conventions and compositionality. Zouhar tries to show that this kind of evidence cannot be used to uphold contextualism without further arguments. In other words, contextualism is underdetermined by this kind of evidence. It is argued that other approaches can accommodate the same set of examples and provide the same predictions. They can do so despite the fact that the purported unarticulated constituents are taken to be somehow articulated at the level of syntactic representation. These approaches are, thus, non-contextualist in the above sense. Consequently, to win their case, the contextualists have to demonstrate that all relevant competitors to their theory that provide the same predictions are defective for some reason or other. Acknowledgments The contributions collected in this volume are based on selected papers presented at the 3rd International Conference on Philosophy of Language and Linguistics, PhiLang 2013, held in Łód in May 2013, and organized by the Department of English and General Linguistics at the University of Łód. I want to thank all the participants for stimulating plenary lectures, talks, and discussions. I am grateful to Mr Ryszard Rasiski for comprehensive editorial assistance, and to Dr Piotr Duchnowicz who prepared the final manuscript. I wish to thank Dr Rafael Hüntelmann for advice and encouragement for the project, and especially warmly to Ms Olena Gainulina for her professional final formatting assistance.

References Baptista, Luca and Erich Rast (eds.) 2010. Meaning and Context (Lisbon Philosophical Studies. Vol. 2). Bern: Peter Lang. Bianchi, Claudia (ed.) 2004. The Semantics/Pragmatics Distinction (CSLI Lecture Notes). Stanford: CSLI. Caton, Charles 1971. Overview. In: D. Steinberg and L. Jakobovits (eds.), 3-13. Davis, Steven and Brendan S. Gillon 2004. Introduction. In: S. Davis and B. Gillon (eds.), 1-130. Davis, Steven and Brendan S. Gillon (eds.) 2004. Semantics. A Reader. Oxford: Oxford University Press.

8

Piotr Stalmaszczyk

von Heusinger, Klaus and Ken Turner (eds.). 2006. Where Semantics Meets Pragmatics (Current Research in the Semantics/Pragmatics Interface. Vol. 16). Oxford: Elsevier. Kempson, Ruth 1977. Semantic Theory (Cambridge Textbooks in Linguistics). Cambridge: Cambridge University Press. Platts, Mark 1979. Ways of Meaning. An Introduction to a Philosophy of Language. London: Routledge & Kegan Paul. Steinberg, Danny and Leon Jakobovits (eds.) 1971. Semantics. An Interdisciplinary Reader in Philosophy, Linguistics and Psychology. Cambridge: Cambridge University Press. Szabó, Zoltan G. (ed.) 2005. Semantics versus Pragmatics. Oxford: Clarendon Press. Turner, Ken 2011. Introduction: Preliminary ‘sketches of landscape’. In: K. Turner (ed.), 1-18. Turner, Ken (ed.) 2011. Making Semantics Pragmatic (Current Research in the Semantics/Pragmatics Interface. Vol. 24). Bingley: Emerald Group Publishing Limited.

Brian Ball

St Anne’s College, Oxford [email protected]

On the Normativity of Speech Acts Abstract: Illocutionary speech acts such as assertion are subject to norms. In this paper I describe how we might explain the normativity of such speech acts, focusing on the case of assertion itself. In the introduction I clarify what is at issue: we are concerned not with whether speech acts are essentially conventional, but with whether they are constitutively governed by norms; I am sympathetic to the view that they are, and wish to explain how this could be the case. Next, I suggest that one kind of attempt to explain the normativity of assertion by appeal to a reductive view of its essence fails; and I sketch my own preferred explanation of this normativity which treats assertion as an irreducible kind. Then, in the final section, I show how the account I favour grounds the normativity of speech acts and their essences in more basic facts in a manner that is naturalistically acceptable – provided that the piecemeal conception of philosophical naturalism articulated there is correct. Keywords: speech acts, normativity, naturalism, knowledge rule, constitutive rule, assertion

0. Introduction Speech acts are governed by norms: for instance, one ought not to assert a proposition p unless one knows p; and one ought not to ask a question q unless one wants to know q’s answer.1 But what is the status of such norms? Construed as moral or prudential norms they seem defeasible: it might, under certain circumstances, be morally acceptable to lie to 1

These are both norms requiring speakers to have attitudes which they present themselves as having by performing the speech acts in question; that is, they are what we might call norms of sincerity. It is not my present concern whether it is precisely these norms, rather than some variants (e.g. ask q only if you wonder what q’s answer is, assert p only if you have a justified true belief that p), that govern the acts in question; nor shall I address the question whether there are norms other than those of sincerity governing speech acts.

10

Brian Ball

someone, asserting something one knows to be false; or it might be prudentially advisable to flatter someone by asking them a question to which one does not seek an answer. Can these or similar norms be construed,2 however, in such a way that they are not defeasible, but apply universally and without qualification? What might the source of such normativity be? Unfortunately, early debates obscured these issues somewhat, focusing on the notion of convention and its role in the performance of speech acts. Thus, Austin argued that “the illocutionary act is a conventional act: an act done as conforming to a convention” (Austin 1962/75: 105); but Strawson objected that only some illocutionary speech acts are “essentially conventional” (Strawson 1964: 457), while others are not. It is important to be clear what Austin’s disputed claim amounts to – or, at the very least, what it does not amount to. Austin famously distinguished three types of speech act. The locutionary act is the act of saying something …in the full normal sense… which includes the utterance of certain noises, the utterance of certain words in a certain construction, and the utterance of them with a certain ‘meaning’ in the favorite philosophical sense of that word, i.e. with a certain sense and with a certain reference. (Austin 1962/1975: 94).

Performing an illocutionary act, however, requires more: in particular, it requires that one’s utterance have a certain communicative force, e.g. that of an assertion or a conjecture, a command or a request. Finally, perlocutionary acts are individuated by their non-communicative outcomes: for instance, in arguing something, one might convince one’s audience; if so, then one has not only been understood as adducing (perhaps conclusive) evidence in support of some proposition, thereby achieving a communicative effect – one has also achieved the result that one’s audience believes the proposition argued for. According to Austin, “the illocutionary act and even the locutionary act too involve conventions” (Austin 1962/1975: 107); but, by contrast, “perlocutionary acts are not conventional” (Austin 1962/1975: 121, ital2

See previous note.

On the Normativity of Speech Acts

11

ics original). It should be clear that the performance of a locutionary act does indeed require the existence of a convention, just as Austin claimed: since the sounds one utters might have been associated with other meanings, something must serve to effectuate the actual association between the two; and this something is, of course, a linguistic convention employed in the community to which the speaker belongs.3 Similarly, that the perlocutionary act is in no way conventional is also correct, though perhaps less obviously so. To convince someone, for instance, is to induce in them a belief by way of argument: and of course, to argue is, inter alia, to perform locutionary acts involving conventions; so it might be thought that conventions are necessary for the performance of perlocutionary act, which can therefore be regarded themselves as conventional. But it is better to think of the matter in the following way. Take two cases in which a speaker argues that p: in one her audience is convinced, and in the other not. Is there any convention, beyond those involved in the latter case that must be in place in the former case? Clearly, the answer is no: all that must happen is that the hearer must form the belief that p, and this is not a matter of any convention. Crucially, the claim that the illocutionary act is conventional is not merely the claim that it involves a locutionary act, and therefore involves a convention. Indeed, it is not even the (contentious) claim that illocutionary force has a conventional means of expression, just as sense and (perhaps) reference do. But then, what is required for Austin’s claim to be correct? Austin seemed to have in mind that there must exist something like a ritual or ceremonial practice in the community of language users from which the illocutionary act receives its significance, or force (Austin 1962/1975: 19, 70). But in what respects must the relevant practice be like rites and ceremonies? Must it, for instance, be arbitrary? Must it serve some broader function within the community? Austin did not address these questions, and accordingly left our central problems ill-formulated and vague. 3

I am not claiming here that convention is sufficient to effectuate the association of words with meanings required for the performance of a locutionary act (such as expressing the proposition that p) – a claim upon which doubt is cast by, e.g. Hawthorne (1990); rather, I am claiming that it is necessary.

12

Brian Ball

Searle made significant progress on this front, distinguishing three questions: First, are there conventions for languages? Second, must there be rules (realized somehow) in order for it to be possible to perform this or that illocutionary act? And third, are the conventions realizations of rules? (Searle 1969: 40)

He argued that the answer to all three questions is yes: indeed, even posing the third – which might equally be rephrased, “are the rules realized by the conventions?” – presupposes an affirmative answer to the first two. Clearly, the first of Searle’s questions receives an affirmative answer given, as we have seen, that the association between words and their meanings is (largely) arbitrary.4 The third question is one that, while interesting in its own right, will be set aside here.5 Crucially, however, by articulating the second question, and distinguishing it from the third, Searle allowed us to pull apart the question of conventionality from that which is of central interest to us. Searle’s insight was to recognize that a rule might be constitutive of an act, and even (necessarily) conventionally realized (if realized at all), though not itself a convention. Much more recently, Williamson has made it absolutely clear that constitutive rules (which may have normative force) are not conventional in nature. More specifically, he notes that [c]onstitutive rules are not conventions. If it is a convention that one must , then it is contingent that one must ; conventions are arbitrary and can be replaced by alternative conventions. In contrast, if it is a constitutive rule that one must , then it is necessary that one must . (Williamson 2000: 239)

4

5

This qualification is made by way of acknowledgement of the existence of onomatopoeia – though it must be recognized that even words such as “moo” and “baa”, while not arbitrary, involve an element of convention. It concerns the contentious claim mentioned above that there are conventional devices whose employment is necessary and sufficient for the performance of illocutionary acts.

On the Normativity of Speech Acts

13

Our question, then, is whether illocutionary speech acts are constitutively governed by rules where, as Williamson says, A rule… count[s] as constitutive of an act only if it is essential to that act: necessarily, the rule governs every performance of the act. (Williamson 2000: 239)

Clearly, our question is one of metaphysics: it concerns the essences of certain items, namely, speech act types such as assertion and command. In my (forthcoming b) paper, “Speech Acts: Natural or Normative Kinds?” I claimed, following Fine (1994), that essences in general are specified by way of (possibly non-reductive) definitions; and I suggested that we can distinguish two views of the essences of speech acts in the philosophical literature. According to one of these views, some normative term such as “must” or “may” will occur in the definition of any illocutionary act type; according to the other, only (naturalistically acceptable) descriptive terms will occur. On views of the latter kind, speech acts are natural kinds; on the former, they are normative kinds. Finally, I argued that despite appearances, it is possible to reconcile these views: speech acts are in fact both natural and normative kinds. I will describe below how this reconciliation is to be effectuated; but it will be helpful, in order to see more clearly what is at issue, to have a concrete proposal in front of us. Williamson claims that the speech act of assertion is constitutively governed by the knowledge rule (with which we began), according to which “[o]ne must: assert p only if one knows p” (Williamson 2000: 243). I find this claim plausible; but some theorists, such as Hindriks (2007) and Bach (2010), have not been convinced by Williamson’s (abductive) arguments in favour of it. Part of their resistance, I suggest, comes from general concerns about the metaphysics of speech acts it implies. In particular, these concerns are of two kinds. A first concern is that we should be able to say, in reductive terms, what it is to perform a speech act such as assertion; yet Williamson’s account of the essence of assertion is non-reductive – assertion is simply that speech act which is subject to the knowledge rule. A second concern is that normativity such as that involved in the knowledge rule ought to be explicable in more basic terms; yet Williamson’s claim that the

14

Brian Ball

knowledge rule is constitutive of assertion does not seem to leave room for such explanation – indeed, he says, It is pointless to ask why the knowledge rule is the rule of assertion. It could not have been otherwise. (Williamson 2000: 267)

In fact, however, these two concerns are intertwined: for if we were able to give a reductive definition of the essence of assertion, we might be able to show that it is subject to a norm similar to the one articulated by the knowledge rule; and Hindriks (2007) has attempted just this. In the first section below, I will describe a standard reductive naturalist account of the nature of assertion, and say briefly why I don’t think the kind of derivation of the norm of assertion based on it that Hindriks proposes works, before giving my own proposed explanation of the norm of assertion. In the second section, I will consider concerns about the naturalistic potential of my account. The two sections correspond roughly to the two kinds of concerns mentioned above, taken in reverse order. 1. Explaining the Normativity of Assertion Standard naturalist accounts of the essences of speech acts attempt to explain what it is to perform a given type of speech act by appeal to the kind of intention with which one makes an utterance, rather than by appeal to constitutive rules articulating norms governing their performance.6 Thus, Grice (1957) argued that an agent can mean something by performing an action with a certain sort of intention – a reflexive one whose fulfillment consists in part in its recognition. And Strawson (1964) suggested that we can understand some basic illocutionary forces, such as those of assertion and request, in terms of these Gricean intentions. More fully, Grice isolated, and attempted to characterize, a special kind of non-factive, “non-natural” meaning as being of interest to the 6

Searle (1969: 41) recognizes that there might be constitutive rules which are not normative.

On the Normativity of Speech Acts

15

study of communication. Strawson summarized Grice’s theory of this kind of meaning as follows: S non-naturally means something by an utterance x if S intends (i1) to produce by uttering x a certain response (r) in an audience A and intends (i2) that A shall recognize S’s intention (i1) and intends (i3) that this recognition on the part of A of S’s intention (i1) shall function as A’s reason, or a part of his reason, for his response r. (Strawson 1964: 446)

He claimed that …we must add to Grice’s conditions the further condition that S should have the further intention (i4) that A should recognize his intention (i2). (Strawson 1964: 447)

Strawson argued that the audience A has understood the speaker S provided that A recognizes S’s intentions (i2) and (i4), i.e. those aspects of S’s overall intention which make it reflexive, and so communicative.7 Accordingly, one might perform an illocutionary act provided that these sub-intentions are fulfilled, though one will only perform the corresponding perlocutionary act if the initial intention (i1) is fulfilled, and the response r is produced. On Strawson’s view, different speech act types can then be individuated, at least in part, by the different responses they are intended to bring about: belief in the case of an assertion; action in the case of a request. Bach and Harnish (1979), however, impressed perhaps by the thought (noted by Austin and stressed by Searle) that one can perform an illocutionary act without attempting to perform some perlocutionary act, abandoned the requirement that the speaker must intend the noncommunicative response r in order to perform an illocutionary act. They suggested instead that speech acts express attitudes, where roughly speaking, one expresses an attitude if and only if one reflexively intends one’s hearer to believe that one has that attitude (Bach and Harnish 1979: 15). And working in this tradition, Hindriks (2007) claims that 7

There is some debate over the relationship between nested intentions such as those described by Strawson and the reflexive intentions posited by Grice; but I leave it to one side here. For some comment, see Ball (forthcoming b).

16

Brian Ball [t]o assert that P is to utter a sentence that means that P and thereby express the belief that P. (Hindriks 2007:400)

He argues that we can derive, on this basis, a restricted version of the knowledge rule, having the force of a moral norm, and applying only to situations of normal trust: since such a norm would not govern every possible assertion (but only those made in situations of normal trust), it is not constitutive of that act given the necessary condition for this articulated by Williamson (above); but its existence, it might be thought, can account for the appearance of plausibility enjoyed by Williamson’s unrestricted, constitutive norm. Without going into too much detail, the idea is to appeal to the fact that knowledge is the norm of belief (Williamson 2000: 47), together with the fact that the essence of assertion is to express belief, to conclude that when one ought, morally speaking, to be sincere, one ought to assert a proposition only if one knows it. Although I am sympathetic to the claim that knowledge is the norm of belief as well as that of assertion, I do not think that Hindriks’ own derivation of the latter from the former, together with his belief-expression theory of the essence of assertion, succeeds. In my paper (Ball forthcoming a) “Deriving the Norm of Assertion”, I have argued this in some detail. The basic problem is that Hindriks equivocates between moral norms and epistemic norms, when what is ultimately needed is a distinctive kind of assertoric norm: the claim that knowledge is the norm of assertion is the claim that assertions ought, qua assertions, to be expressions of knowledge. Elsewhere (Ball forthcoming b) I have attempted my own explanation of why it should be that we engage in a practice – namely assertion – which is subject to the knowledge rule.8 The idea was to argue, first, that speakers normally assert only what they know, and second, that speakers always have (some, possibly overridden) reason to do what is normal. If successful, the two arguments together would show that speakers are subject to an unrestricted knowledge rule having normative force. Finally, I claimed that we can understand what it is for an assertion to be 8

Although Williamson does not countenance an explanation of why assertion is subject to the knowledge rule, he recognizes that it is legitimate “to ask why we have such a speech act as assertion in our repertoire” (Williamson 2000: 267).

On the Normativity of Speech Acts

17

normal in naturalistic terms. Since assertion was not taken to be special in any way amongst illocutionary acts, the upshot was taken to be that although speech acts are normative kinds, they are also natural kinds. My argument for the first claim is based on four premises, and runs as follows: (P1) Normally, speakers assert only if they intend to induce belief. (P2) Normally, speakers intend to induce belief only if hearers come to believe. (P3) Normally, hearers come to believe only if they thereby come to know. (P4) Normally, hearers come to know only if speakers know. Therefore, (C) Normally, speakers assert only if they know. My idea was that Grice, in effect, correctly described what Millikan (1984) would call the “normal” cases of assertion, viz., those which explain the continued existence of the practice. More fully (though still very roughly), Millikan thinks that a device type functions normally (or properly) if, and only if, it does what past tokens of it did which caused them to be copied, thereby causally explaining the fact that current tokens of the type exist. My first premise, then, claims that in all such normal cases of assertion, speakers reflexively intend to induce belief in their hearers, just as Grice and Strawson suggested. The second premise is supported by the thought that speakers would not continue to assert if hearers did not believe what they were told; accordingly, we have reason to believe that hearers do believe what they are told in these normal cases in which speakers (reflexively) intend them to do so. The third premise is supported by the thought that knowledge is the norm of belief, i.e. that one ought only to believe something if one thereby knows it: the idea is that there is no reason to think that hearers’ cognitive capacities should be malfunctioning in those cases of assertion in which it does what makes it advantageous to us; so we should expect that belief is perfectly normal, and hence constitutes knowledge, in these normal cases of

18

Brian Ball

assertion. Finally, the fourth premise is made plausible by reflecting on the fact that knowledge requires safety, and that this is likely to be secured in normal cases in which hearers accept what they are told only if speakers themselves believe what they say, and indeed do so safely – thus, likely, only if speakers assert only what they know. The conclusion then follows from the premises, by the transitivity of “only if”, together with the fact that “normally” serves as a universal quantifier over the same set of normal cases of assertion. Supposing this argument to be successful (i.e. sound), why think that its conclusion vindicates the thesis that assertion is constitutively governed by the knowledge rule? In particular, why think that what speakers normally do in respect of assertion is what they ought to do in this respect? One initial thought is that device types are propagated and reproduced because advantageous to those that employ them: accordingly, normal, knowledgeable assertions must provide some benefit to language users; and so, speakers themselves must, prima facie, secure some advantage by asserting only what they know, and therefore (prudentially) ought, prima facie, to do so. A second, slightly different argument was given in “Speech Acts”. This time the idea was that, since it is of the essence of assertion to normally produce hearer knowledge, and since speakers assert knowingly and intentionally (i.e. they are not ignorant of what they do when they assert), they in some sense intend hearer knowledge whenever they assert. But a means to the end of such hearer knowledge is asserting only what one knows; so speakers always have some pro tanto obligation of rationality to intend this means and assert only what they know. Neither of these lines of thought seems to capture the full normative force of the “must” occurring in the knowledge rule; yet each has the advantage of showing that the naturalistically acceptable conclusion (C) of the argument above articulates a constraint on assertion which induces a normative requirement on speakers, despite the fact that assertion is an irreducible kind. That is, we do not need a reductive definition of assertion to explain the normativity in the knowledge rule; we can get by with the thought that speakers possess reflexive intentions to induce hearer beliefs in normal cases of assertion. This allows us to recognize that abnormal assertions produced when such intentions are absent are asser-

On the Normativity of Speech Acts

19

tions nonetheless, and that no non-trivial necessary and sufficient conditions can be given for being an assertion; but that features of the normal cases are the source of the normativity involved. In the remainder of this section, I want to try to say some things that capture more adequately the normative force of the “must” occurring in the knowledge rule. The problem with each of the arguments given above is that they treat the “must” as having purely prudential force for the speaker: yet part of what is wrong with abnormal, unknowledgeable assertions concerns the position the hearer is put in by the speaker’s action; the abnormal asserter arguably does some wrong by the hearer. The normativity involved in the knowledge rule thus seems to have an intersubjective character, and not just the force of personal, prudential rationality. In particular, what I want to suggest is that, while the knowledge rule has the force of a pro tanto obligation which points beyond oneself, it can nevertheless be given a kind of prudential, or utilitarian explanation. My account here follows the work of Rawls (1955), who distinguished two ways of thinking of (normative) rules. On the first, “summary” conception, rules articulate generalizations which may have exceptions; they are “rules of thumb” which simply abbreviate a whole range of particular normative facts. Accordingly, whenever one is in the kind of situation with which the rule is concerned, the question of what one ought to do arises: the case may then be like the majority of other cases of its type, so that one ought to do as the rule says; or it may differ from them, so that one is under no obligation to do as the rule says. Crucially, on the summary conception, the rules in question can have no independent normative status themselves. By contrast, on what Rawls called the “practice” conception, Rules are pictured as defining a practice. Practices are set up for various reasons, but one of them is that in many areas of conduct each person’s deciding what to do on utilitarian grounds case by case leads to confusion, and that the attempt to coordinate behavior by trying to foresee how others will act is bound to fail. As an alternative one realizes that what is required is the establishment of a practice, the specification of a new form of activity; and from this one sees that a practice necessarily involves the abdication of full liberty to act on utilitarian and prudential grounds. (Rawls 1955: 24)

20

Brian Ball

Now, one of the cases that Rawls has in mind is the legal practice of punishing those convicted of a crime; and it is entirely possible (whether or not it is probable) that such a practice should have been deliberately set up by enlightened legislators recognizing the advantages it would offer. This, however, is inessential for our purposes: what is crucial is that Rawls thought that such rules can be given utilitarian defences; that is, it is because they are advantageous to us, and more so than any (nearby) alternative, that they are established and retained – whether or not we realize this. Clearly, Hindriks is operating with something like the summary conception of the knowledge rule, whereas on the Williamsonian view that I have been espousing, this rule is treated as defining a practice, namely, that of assertion. And I think the knowledge rule can be given a utilitarian defence of just the sort that Rawls suggests: in particular, it seems to me that normal assertions are advantageous to those involved (both speaker and hearer), and that it is because of this that there is a practice of assertion within human communities. But it is worth noticing that the utility alluded to is not that of the speaker in a given (perhaps abnormal) case: rather, it is that of the members of the community (or perhaps species) to which she belongs. It is in this way that the knowledge rule acquires its distinctive, sui generis normative force: but acquire it, nonetheless, it does; the normativity is not basic. That is, it is because having a practice of asserting is beneficial to us in general that we have it; and this means that in any particular case we have a pro tanto obligation, not just to ourselves, but one owed towards the members of our community, to assert only what we know. This account of the normative force of the knowledge rule is only possible given the non-reductive character of the essence of assertion. 2. Piecemeal Naturalism and the Essence of Assertion We have seen that the normativity involved in the knowledge rule of assertion – and, by analogy we might hope, that involved in the constitutive rules of other illocutionary speech acts – can be explained by the advocate of a non-reductive account of the essence of assertion, thus al-

On the Normativity of Speech Acts

21

leviating the second kind of worry mentioned at the end of the introduction. But what about the first kind of worry? Isn’t there something problematic about non-reductive accounts of high-level features of reality in general, and of the essences of speech acts in particular? The general worry is that without a reduction of some non-physical phenomenon P, it seems mysterious how P could be instantiated. After all, in the beginning was the world, with its wholly physical character. If, at some later time, phenomenon P arose, this cries out for explanation; yet given that P is not (reducibly) physical, it might seem no such explanation can be forthcoming. Must we suppose that God simply added P to the world at that stage? This seems problematically ad hoc and unexplanatory. In connection with speech acts in particular, we might worry that it is impossible to intend to perform an illocutionary act directly and immediately: assertion, for instance, doesn’t feel like a basic kind of action in the way that, say, moving one’s hand (or one’s lips) does. Surely, then, we must be able to say what it consists in, what we have to intentionally do if we are to assert. And this can seem to militate in favour of a reductive account of the essences of such speech acts as assertion; for given such an account we could readily say what we must do in order to assert, and so, for assertion to be instantiated. In order to address these worries, however, I want first to discuss an objection raised about my account at the PhiLang2013 conference by Maciej Witek. Witek noted that my account of assertion does not succeed in explaining the normativity of assertion in entirely natural terms, since I appeal to unexplained norms of rationality – and in particular, epistemic norms – in my argument that in normal cases, speakers assert only what they know. More specifically, Witek pointed out, quite correctly, that I appeal to Grice’s account of what it is to communicatively mean something in defending the first premise (P1), and that this involves something like Strawson’s intention “(i3) that th[e] recognition on the part of A of S’s intention (i1) shall function as A’s reason, or a part of his reason, for his response r” (Strawson 1964: 446). In short, the speaker must intend that the hearer treat his recognition of her perlocutionary intention to produce r as an epistemic reason to form that response – i.e. as subject to an epistemic norm.

22

Brian Ball

A cheap response is to note that the occurrence of the normative notion of an epistemic reason occurs here within the scope of an intention; so strictly speaking, I don’t need to recognize the antecedent existence of epistemic norms in appealing to Grice’s theory of meaning, but only that of our conception thereof. But this reply won’t do in general: Witek might as well have pointed to my defence of the third premise (P3) of my argument that normally speakers assert only what they know; for here I made direct use of the claim that knowledge is the norm of belief. Similarly, in arguing that the second premise (P2) is true, I appealed to norms of practical rationality: it would be practically irrational for speakers to intend to induce belief if they did not have a reasonable expectation that hearers would come to believe – an expectation that is best explained by its truth (in normal cases). My explanation of the normativity of the illocutionary act of assertion is run through with normative notions. The concern, then, is that I have not explained the normativity of the knowledge rule in entirely non-normative terms, and so I have not guaranteed that my account is naturalistically acceptable (by my own lights). Strictly speaking, this is not an objection to the fact that my account of assertion is non-reductive, since even if I had defined assertion in terms of Gricean intentions, as Bach and Harnish and Hindriks do, it would still apply; the problem is that the base properties appealed to are normative in character. Nonetheless, I think the current objection can be addressed in a way which also alleviates the concern that it is mysterious how assertion could be irreducible. Williams (2002) points out that there is, in general, a difficulty in characterizing philosophical naturalism. The problem is that we run the risk of either describing it as a position which is utterly trivial, or else as a view that is entirely implausible. In somewhat more detail, Williams’ thought is that naturalism about some phenomenon is, very roughly, the view that it is simply “part of nature”: but this is trivial if nature includes everything there is; and it is implausible if it means that we can reductively define that phenomenon in physical terms. Williams’ solution is to take a kind of piecemeal approach to naturalism.9 He says, 9

He himself calls it the “creeping barrage” (Williams 2002: 23) conception.

On the Normativity of Speech Acts

23

The question for naturalism is always: can we explain, by some appropriate and relevant criteria of explanation, the phenomenon in question in terms of the rest of nature? (Williams 2002: 23)

He thinks that there are open questions about “what we are prepared to regard… as an explanation” (Williams 2002: 23); and what the rest of nature is will depend on the target phenomenon. I am sympathetic to Williams’ view of naturalism. More specifically, my suggestion is that if we can specify sufficient conditions for the instantiation of a type in simpler terms, that will count, for the purposes of naturalism, as having explained it: for we will be able to answer the question how that type of phenomenon could have occurred employing only simpler terms, standing for (instances of) already instantiated types. (This will not count as a reduction, since the statement of necessary conditions employing only the simpler terms is also required for a reductive definition.) And when it comes to the target phenomenon of illocutionary act types and their normativity, my proposal is that the rest of nature includes the norms of practical and theoretical rationality by which agents are bound. Accordingly, if we can explain how, for instance, assertion could be constitutively governed by the knowledge rule appealing only to the existence of interacting agents subject to the norms of reason, as well as the (non-assertoric) intentions they form, we will have given a naturalistic account of the normativity of that speech act. And this, of course, is exactly what I take myself to have done.10 Thus, by adopting Williams’ piecemeal approach to naturalism we can see 10

In particular, on my view sufficient conditions for the performance of a speech act can be given in simpler terms – for instance, it is sufficient to assert that p that one utter a sentence which means that p, reflexively intending to thereby induce the belief that p in one’s hearer – and it is accordingly unmysterious how that speech act type might have arisen. The sufficient conditions specified, however, will not in general be necessary for the performance of the act in question, and so the account of is non-reductive in character: it does not purport to capture the whole essence of the act in the simpler terms of Gricean intentions. Of course, the question remains how we perform e.g. assertions in the remaining cases where the sufficient conditions are not met: but to say how we do something is not the same as saying what we thereby do; the means is no part of the essence of the act.

24

Brian Ball

both how the normativity of speech acts such as assertion can be recognized as naturalistically explicable, and how it is possible to perform such acts despite their not being (in some relevant sense) basic actions. 3. Conclusion I began this paper by remarking on the innocuous fact that illocutionary speech acts are subject to norms – for instance, that one ought only to assert a proposition if one knows it to be true. I then asked what the source of that normativity might be: in particular, I wondered whether the acts in question might be constitutively governed by such norms, taking care to distinguish this from the issue of whether they are essentially conventional; it emerged that our question concerned the metaphysics of speech acts. I then proposed to defend the view that assertion in particular, is constitutively governed by Williamson’s knowledge rule, and to show that this is compatible with philosophical naturalism. From there on the paper was in two parts. In the first of these I responded to the concern that only if speech acts can be reductively characterized in other terms can we explain their normativity, i.e. that otherwise we must, implausibly, treat it as basic. Appealing to Rawls’ distinction between the “summary” and “practice” conceptions of rules, I argued that it is not despite but because of its irreducible nature that the normativity of assertion can be adequately explained (rather than treated as primitive). In the final part of the paper, I responded to concerns that my approach was insufficiently naturalistic. More specifically, I showed how, appealing to Williams’ piecemeal conception of naturalism, we can accommodate the thought that some norms (those of rationality) are invoked in the explanation of the normativity of speech acts without compromising our naturalistic ambitions; and I argued that by giving sufficient conditions for the performance of a speech act such as assertion in simpler terms we can explain the possibility of its performance without providing a full reduction. In this way I have hoped to provide support for the view that speech acts are both natural and normative kinds.11 11

Thanks to Piotr Stalmaszczyk for inviting me to contribute this paper, and to Charlie Pelling and Maciej Witek for comments on an earlier draft.

On the Normativity of Speech Acts

25

References Austin, John 1962/1975. How to Do Things with Words. Edited by J. O. Urmson and Marina Sbisà, 2nd ed. Oxford: Clarendon Press. Bach, Kent 2010. Knowledge In and Out of context. In: J. K. Campbell, M. O’Rourke and H. Silverstein (eds.), 105-36. Bach, Kent and Robert Harnish 1979. Linguistic Communication and Speech Acts. Cambridge, MA: MIT Press. Ball, Brian (forthcoming a). Deriving the Norm of Assertion. The Journal of Philosophical Research. Ball, Brian (forthcoming b). Speech Acts: Natural or Normative Kinds? Mind and Language. Campbell, Joseph Keim, Michael O’Rourke, and Harry Silverstein (eds.) 2010. Knowledge and Skepticism. Cambridge, MA: MIT Press. Fine, Kit 1994. Essence and Modality. Philosophical Perspectives 8, 1-16. Grice, Paul 1957. Meaning. The Philosophical Review 66 (3), 377-388. Hawthorne, John 1990. A Note on “Languages and Language”. Australasian Journal of Philosophy 68 (1), 116-118. Hindriks, Frank 2007. The Status of the Knowledge Account of Assertion. Linguistics and Philosophy 30, 393-406. Millikan, Ruth 1984. Language, Thought, and Other Biological Categories. Cambridge, MA: MIT Press. Rawls, John 1955. Two Concepts of Rules. The Philosophical Review 64 (1), 3-32. Searle, John 1969. Speech Acts. Cambridge: Cambridge University Press. Strawson, Peter 1964. Intention and Convention in Speech Acts. The Philosophical Review 73 (4), 439-460. Williams, Bernard 2002. Truth and Truthfulness: an Essay in Genealogy. Princeton: Princeton University Press. Williamson, Timothy 2000. Knowledge and its Limits. Oxford: Oxford University Press.

John Collins

University of East Anglia [email protected]

Representations without Representa: Content and Illusion in Linguistic Theory Abstract: Contemporary linguistic theory seeks to explain linguistic phenomena by way of the attribution of representations to speakers/hearers. A general philosophical problem here arises concerning the apparent intentionality of representational states. A specific instance of this concern is how to understand the representation of a linguistic item when such an item does not exist independently of the representation itself. The paper proposes a resolution of this quandary by way of reviving a method of hyphenation originally due to Goodman, which will be finessed and defended against a range of likely objections. Keywords: representation, intentionality, intentional inexistent, Nelson Goodman Georges Rey

0. Introduction Theories in cognitive science characteristically attribute representations to the objects of the theories, both conscious and unconscious, and indifferently to human and non-human; indeed, many argue that representational theories are necessary for an adequate account of cognition (e.g., Chomsky 1980; Newell 1980; Marcus 2001; Gallistel and King 2009). Scientists and philosophers disagree over what is meant when a system is said to be representational; it might even be that there just is no uniform notion that holds across the relevant range of theories. Furthermore, it is an open question what theoretical attributions of representations have in common, if anything, with our quotidian representational attributions of beliefs, wishes, hopes, and other such attitudes. My present concern is not to compound or confound the apparent heterogeneity

28

John Collins

of representational talk in cognitive theory and our common parlance. My far narrower concern is for theoretical attributions of representations in the linguistic sciences, syntax in particular, where there is no corresponding object or property of the representation. Such are cases where the represented or representa in general do not exist to be represented, anyway, anyhow, anywhere; where, that is, nothing external to the representation constitutes the correctness or incorrectness of the representation. In the jargon of Burge (2010), such representations are ‘nonobjective’. A particularly striking example is offered by so-called empty categories, i.e., elements of syntax that do not have any phonological or morphological realisation.1 The present interest of such representations is that though they are essential to current explanations in linguistics, the items have no external signature at all, not even in speaker/hearers’ phenomenology.2 What is it, then, for a speaker/hearer to represent empty categories as they do according to modern syntactic theory? In fact, the same kind of question arises for non-empty linguistic categories, such as the humble noun, adjective, verb, and so on, insofar as these categories also lack any sensible signature, even though their instances do. One might, say, be able to recognise instances of dog when appropriately presented with them and also even be able to understand the tokens as being instances of the type noun. Yet at no stage can it be rightly said that one sees or hears nounhood, as it were. To be sure, one might be able to recognise a given noun (in English), but one cannot, ipso facto, do likewise when one hears a language unknown to one. Still, there would be little complaint, I think, against talk of our hearing or seeing nouns. Indeed, the sheer heterogeneity of the forms dog itself (or any other word, of course) may take both within and across modalities means that one isn’t even sensibly aware of dog-hood. In language, at least, it seems as if we see or hear tokens only via abstraction, which collects together the heterogeneous tokens in a way that has no essential signature in what is collected (just reflect on the differences between dog in 1

2

Empty categories, as understood in contemporary syntactic theory, were introduced in the mid-1970s (see Chomsky 1973, 1975; Chomsky and Lasnik 1977). This is not to suggest that empty categories have no effect on our linguistic phenomenology and processing. My point is merely that empty categories have no definite and uniform external feature associated with them.

Representations without Representa: Content and Illusion in Linguistic Theory

29

ASL, English brail, Times New Roman, and spoken in RP). The motive, therefore, for my focus on empty categories is that they make a general point most vivid: the representa of attributions in linguistic theory appear to be non-extant and so cannot constitute the correctness conditions for the representation attributed. Similar issues also arise with the same force wherever we have the attribution of a representation of x, where x is wholly abstract and has no definite conscious or external signature, e.g., Euclidean figures posited in visual processing or dead reckoning posited in insect navigation. In such cases, there is nothing external to the mind to which the representation is answerable, qua the particular representation it is. My attention, though, will be on the linguistic case and I forswear any ambition to give a general account of representation throughout cognitive science or beyond. In the following, I shall offer a picture of how to understand the linguist’s representational talk that renders representation as a non-relational or non-transitive notion. I shall call this the method of hyphenation for reasons that will become clear. The bare idea of my approach goes back to Goodman (1949, 1976) and Dennett (1969) via McGilvray (1999), who cites only Goodman. My motives, problems, and uses of the idea are quite distinct from Goodman’s and Dennett’s, and McGilvray fails to broach central problems his appeal to Goodman faces. The resolution of those problems will lead us to make a distinction between representations that are part of the technology of linguistic theory and representations as part of the object of the theory. The fundamental idea is that the theorist employs dyadic representational talk in order to frame generalisations about monadic, non-relational states.3 A crucial twist to all of this is that we may accept the underlying motive of the hyphenation approach without accepting the content of it, which, as we shall see, suffers from severe problems if presented as an analysis or depiction

3

The position thus has an obvious affinity with so-called measure-theoretic approaches to the propositional attitudes, as most recently and comprehensively argued for by Matthews (2007). Although I am very sympathetic to Matthews’s position, my arguments are broadly independent of his, not being specifically concerned with the propositional attitudes.

30

John Collins

of the logical form or content of representational talk. The hyphen will be one of those ladders we may kick away.4 If I am right in all of this, the linguist will not be lumbered with the problem of how to understand intentional inexistence in the sense that has hitherto concerned philosophers worrying about our thoughts of Zeus and Father Christmas.5 Zeus might well be, in some sense, as might empty categories construed as mind-independent objects; in the latter case, at least, no intentional relation featuring the categories as relata is presupposed or entailed by the theories that posit the categories; so, issues of non-existence don’t arise as a problem for the theories. This result, however, will not translate into a general solution of the problem of intentional inexistence, how, say, it can be true that little Johnny hopes that Farther Christmas leaves him a bike. The account on offer only bears on the theoretical attribution of representations. I shall turn to this notion first in order to distinguish my concern from the colloquial notion of representation (well, colloquial among philosophers). 1. Representations: Separating the Personal from the Theoretical Essential to understanding others and ourselves as thinking things is the attribution of representations. Indeed, it is widely assumed that a ‘theory’ of representations and how to attribute them is a fixed aspect of human (and potentially higher primate) cognition.6 Whatever the facts of developmental psychology for primates in general prove to be, it is undeniable that we possess what I shall call a personal-level system of representations, exhibited in our talk and our own awareness of thinking of 4

5

6

I thank Tim Crane for helping me see the aptness of this familiar philosophical analogy. A great deal has been written about the ontology and semantics of fiction and myth. See Everett and Hofweber (2000) for an overview of the debate and samples of many varied positions. The literature in this area is vast. See, for example, Perner (1991) for a seminal developmental account of the cognitive use of representations in humans, and Byrne and Whiten (1991), Whiten and Byrne (1997), Whiten (1991), and Carruthers and Smith (1996) for discussion of the varied accounts of representtational capacities in human and non-human primates.

Representations without Representa: Content and Illusion in Linguistic Theory

31

others and ourselves in terms of propositional attitudes of belief, desire, and the rest, and other intentional states such as seeking, imagining, chasing, etc. The account of representation that I seek to promote for linguistic theory is not meant to model this system, although having at least certain key features of personal representations in view will allow us to see theoretical representations in the right light. Let me, then, put forward five features characteristic of personal representations. (i) Answerability. Representations are answerable in the following sense: Answerability: If S represents [x], then S’s representation is assessable (answerable) to x.7 The modes of assessment may vary, but the basic idea is that representations are judged, and are taken to be judgeable by the representer (S), in terms of how things are for what is represented. The principle amounts to a disquotational property of representation, where to specify in the most minimal way what a representation is answerable to is just to state the content of the representation, what is represented as being thus and so. Crucially, this thin condition does not presuppose that there is any x at all, but only that representations are assessed as if there were an x. This, again, is akin to disquotation, which holds uniformly across all representations as a formal condition regardless of differences across the contents. Thus, if there is no x, then S might be immediately reckoned to have represented falsely, but still to have represented [x] in terms of which he is held to account. On the other hand, again where there is no x, maybe S is understood (by himself or others) to have represented things truly, because the representation [x] is assessable regardless of there being an x or not; this would be because whatever is relevant to the assessment of the representation is picked out in terms of x. Imagine that S holds that Holmes was a great detective. Foolishly, though, S thinks that the Conan Doyle stories are genuine reports from Victorian London, just as Conan Doyle intended them to appear. In which case, S is answerable to how things are for Holmes and is judged to have represented things falsely, for there is no Holmes. On the other hand, S’ also holds 7

‘[x]’ stands in for any content whether propositional or not.

32

John Collins

that Holmes was a great detective, and she is answerable to something beyond her representation too, but since she is not naive and knows the stories to be just stories, she is judged to have represented things truly, for being a great detective is exactly how Conan Doyle depicted Holmes to be. Note that what makes the representation true in this case – the content of Conan Doyle’s stories – is still rightly packaged in terms of [Holmes is a great detective]. I appreciate that much more needs to be said on this matter, but I only want to establish the plausibility, without prejudging the issue, that representations are assessable in light of how things are in the world, but a simple inventory of the world, as it were, will not determine the value of assessment of each token of a content. Answerability, in this sense, does not presuppose a content externalism, only that there is something packaged as the represented in terms of which representations are assessed. (i) Accessibility. Representations in the present personal sense are accessible to us. Minimally, this just means that we can metarepresent them in terms of holding ourselves answerable. Indeed, being able to meta-represent representations is constitutive of representations, in that the representations are contents of attitudes we adopt towards them. We do not, that is to say, just represent – punkt. (ii) Promiscuity. Representations are attitudinally promiscuous in that the same representations can fall under different attitudes for the one agent and be shared equally between agents. Obviously, this is essential for interference and inter-agentive actions, such as deception or disagreement. (iii) Internal structure. Representations have internal structure in that, again, they relate to each other in terms of their parts, and we can build up ever more complex representations out of constituent ones. Again, this feature is essential to inference. (iv) Generalisations. Representations are generalisable in that we can generalise over something/everything/most things, etc.

Representations without Representa: Content and Illusion in Linguistic Theory

33

an agent believes or desires and we can reason in terms of such generalisations. On the basis of these five features, the system of personal representations has the appearance of a symbolic system or code (a language), which has its syntactic and semantic properties independent of the attitudes we take towards elements of the system, where we, when adopting an attitude, inherit the answerability of the representation. For present purposes, it does not matter if we take this system to be an abstraction from natural language competence or an independent system of thought with language-like properties. As indicated above, I have outlined this notion of representation in order to put it to one side. My interest is in representations as they are feature in linguistic theory, where, as a theoretical statement, the linguist attributes a representation of syntactic structure or a principle to an agent as part of an explanation of their linguistic competence. It is necessary, though, to outline the personal notion of representation precisely because only if the theoretical notion is separated from the personal one may we get a clear view of the former, which, I shall contend, lacks the first three features (answerability, accessibility, and promiscuity), and so possesses the last two features in a distinctive way. Before embarking on the main business, two points bear emphasis. Firstly, we should reject an unthinking monism about representation, as if the use of representation by laity and theorists must mean exactly the same thing. I can see no a priori reason to favour such a blanket understanding; besides, it contradicts what I take to be a general view we should have towards any theoretical vocabulary, viz., theoretical terms are to be understood relative to the explanations in which they feature and the phenomena such explanations target. So, if representation is employed in theoretical linguistics, we should hardly expect it to be correlated with a propositional attitude such as belief, at least not without further ado. In general then, we should not a priori assume that there is a univocal notion of representation that holds within and across colloquial discourse and theoretical inquiry, such as linguistics. It is, surely, not so shocking to think that Bill’s dreaming of Vienna, the honey bee’s ‘representing the angular position of the sun relative to its terrain’ (Gallistel

34

John Collins

1990: 134), and Caravaggio’s Supper at Emmaus depicting the Risen Lord are not the same phenomenon. Secondly, I do not intend my characterisation of personal representations to map onto what is now normally referred to as the ‘personal level’ in distinction to the ‘sub-personal level’. This is so for two somewhat trivial reasons. Firstly, my concern here is just for representation not other kinds of mentality. For instance, Dennett (1969), who first drew the distinction, had pain chiefly in mind (sentience rather than sapience, one might say). Secondly, the distinction is used in many varied ways, and I see no compelling reason to think that there is an underlying uniform notion. For instance, following Davidson (1980) and Dennett (1978), it is often thought that the personal level is governed by ‘constitutive norms of rationality’. On the other hand, Stich (1981), Cherniak (1986), and numerous others, have resisted any such normative constraint. For my part, I take the five features of personal representation adumbrated to be neutral on this matter. One can, for instance, accept that answerability is constitutive of personal representtations, but then argue about the extent of the demand: does it involve the absence of systematic, albeit understandable, error? Similarly, Stich (1978) singled out the inferential promiscuity of belief at the personal level to distinguish it from sub-doxastic (sub-personal) states. Again, I am neutral on this view. My promiscuity (as it were) relates to contents, not attitudes. Moreover, while Stich is clearly right that promiscuity does pertain to the attitudes at the personal level, nothing immediately follows about the sub-personal level, either because one may seek to articulate a suitably promiscuous sub-personal state, or because one may resist all talk of sub-personal belief on other grounds (answerability, say).8 8

The putative lack of inferential promiscuity or integration of sub-doxastic states might lead one away from the traditional philosophical conception of a speaker/ hearer knowing or, at any rate, having a propositional attitudinal relation to their grammar. See, for example, Graves et al. (1973), Fodor (1983), and Knowles (2000) for the traditional view defended against the kind of threat Stich’s subdoxastic diagnosis poses. As it is, I think all talk of sub-personal propositional attitudes is wrong-headed for reasons independent of promiscuity in Stich’s sense. The chief problem is answerability. Also see Collins (2004) for some of the background and more specific issues pertaining to the interpretation of Chomsky.

Representations without Representa: Content and Illusion in Linguistic Theory

35

For good or ill, then, I am not intending to make a stand on the personal/sub-personal distinction in general. My concern is just to fix a set of features that I hope are uncontroversial as descriptors of our normal conception of ourselves and others, and to package them under the title of ‘personal representation’. We may then contrast this notion with another notion of representation designed for a specifically theoretical purpose. 2. A Minimal Theoretical Notion of Representation Notwithstanding my plea to distinguish theoretical attributions of representation from personal ones, it may still be thought that my advertised endeavour to render representation free of representa is simply incoherent, even if the account is isolated to theoretical linguistics. This is because the very idea of a representation, however finagled, involves a represented, even if no object at all manages to be represented. So much is enshrined in answerability. As Burge (2010: 45) puts it, ‘[r]epresentation is rather like shooting. Some shots do not hit anything, but they remain shootings’. Perhaps so, but, without further ado, representations entailing things represented (to be shot at) is merely a reflex of the standard transitive reading of to represent. It is the project of the following to attempt to figure out how the idea of representation attribution can be fruitfully employed in theoretical explanation in a way that doesn’t trade on the transitivity of the notion. If one wants to insist that a theoretical inquiry is beholden to the quotidian interpretation of its theoretical terms, then I am more than happy to change topic to representation’ that I stipulate to be intransitive, which leaves both the nature of representation’ and the relation between representation and representation’ live issues. A far more serious concern is how provisionally to spell out the idea of a representation in a neutral way that at least allows for the conceptual possibility of a non-relational or non-transitive notion of representation; taking on such a burden is as much as is owed to the above objection. I shall propose a sufficient condition for a cognitive, non-personal system to be representational, which I hope is acceptable even for those who would insist that such a condition requires supplementation with an

36

John Collins

answerability constraint in order to render a fully relational notion of content that may be properly called representational. The condition I have in mind is structural: (REP) A system is minimally representational if the competence it enables requires the use of variables for its description. Here I assume that we posit an underlying cognitive system subserving some competence, and take the system to trade in representations to the extent to which our account of the competence makes essential appeal to variables.9 This claim bears some elaboration. Firstly, REP concerns the kind of theory required to capture a certain phenomena. The theory posits primitive and complex structures, where for the moment, to be such a structure is to satisfy whatever conditions the theory at hand specifies, or, in other words, to be an instance of the variables the theory involves. A complex representation has parts that are representations, each complex being some function of a set of primitives. Secondly, I do not here understand ‘competence’ to presuppose a transitive conception of representation. REP is certainly not inconsistent with such a conception, but it does not entail it. Competence in the present context can be understood as an idealisation of how the system performs once we abstract away from any factors not parameterised by the internal design of the system. For example, competence is not specified in terms of memory limitation or the embedding system’s capacity for attention on the relevant task. So, we want a theory to capture the possible range of states of the system such that it behaves the way it does. Thirdly, a type/token distinction of especial interest arises with representations. On my understanding, a theory of the relevant kind describes 9

This claim has a strong affinity to the thesis of Marcus (2001), who takes the notion of a variable to be central to the assessment of different models of cognitive architecture (connectionist vs. symbolic). My intent, however, is to use the notion of a variable as featured in a competence description to constrain our attribution of representations to a system. My claim, therefore, is not directly about cognitive architecture, but about how best to describe the competence the architectures are supposed to enable.

Representations without Representa: Content and Illusion in Linguistic Theory

37

the system as being able to token representations of the same type. Hence it is that variables are essential in order to cover the range of types the system can token. A type here is, in effect, a variable. The condition captures the thought that the types under which the tokens fall are not mere artefacts of our abstracting from differences over the states of the system; the types, rather, mark the design of the system, what states the system can fall into. Fourthly, in one sense, REP denies or at least eschews the distinction between vehicle and content, at least in certain applications. There are, of course, many occasions where it is crucial to distinguish between vehicles and their contents. Numerals are not numbers. The latter, for example, stand in algebraic relations to one another, numerals do not; the idea of dividing ‘4’ by ‘2’ is gibberish. Similarly, the inscription ‘dog’ or the sound wave /dg/ is not a dog, the kind of thing that barks and wags its tail. In the present case, however, where we are talking of states of the mind/brain, it is simply unclear how the distinction should be made, if at all. To be sure, we should not confuse a verb phrase or a Euclidean figure with patterns of synaptic firing, but it just doesn’t follow that such patterns represent (are the vehicles for) verb phrases or Euclidean figures. We might simply think of the abstract properties as the means of describing or individuating kinds of brain state that are subsumable under certain generalisations couched in terms of the type-individuating abstracta. Under such a construal, the brain states do not represent the abstracta, or stand to them in an external relation; rather, the abstracta simply individuate the would-be vehicles of content for the explanatory endeavour at hand. This idea will be more fully spelt out as we go along; presently, something must be said about variables. By a variable I mean a symbol occurring in a formula or principle that is substitutable for a range of other symbols, including variables, where the symbols count as belonging to the class precisely because they substitute for the given variable under the principles and operations that constitute the design of the system. This characterisation departs from the standard semantic notion in that I do not presume a domain of items that is fixed independently of the variables that range over it, although given the variables, we can, to be sure, think of the items that are the variable’s values as being of the relevant type. My point here is not to

38

John Collins

demand that the domain of the variables is not fixed, but simply to leave it open that a system’s employing a variable allows for items to become its values and so acquire a type-hood simply through being such a value. The variable allows for two characteristic properties of representational systems. Firstly, a system that operates by way of variables is responsive to new information in a robust way because it is able to treat novelty as ‘being the same’ if the new information is substitutable for information that is already possessed or variable-typed by the system. The system thus operates in abstraction from the continuous novelty that pertains for any physical process. Secondly, the variables potentially allow for indefinitely complex structures by typing a given complex by way of a variable that makes the structure simple relative to other operations that can embed with the symbols that may substitute for the variable. The above was abstract, so let us be more concrete. Consider the difference between a finite-state machine and a phrase structure grammar (Chomsky 1957). The point of the comparison here is not that it bears upon language, but merely that it offers a clear way in which variables can be essential to a competence description along the lines REP indicates. The essential difference between the two systems is that a phrase structure grammar necessarily includes a variable. The output of a finite state machine, on the other hand, is restricted to concatenated strings, where the nth concatenate determines the next potential concatenate. Being characterisable in terms of a variable, a phrase structure grammar defines a set of strings in terms of the member strings being values of variable types (potentially complex). The output of a phrase structure grammar, therefore, is not in fact a set of strings at all, but a set of variable structures that have strings as tokens or values. Thinking of a system as a finite state machine is to think of it essentially in terms of its output, how it behaves, as it were; this is because there is no symbol of the machine that is not itself part of the string output. A phrase structure grammar, on the other hand, depicts a system essentially in terms of how it structures or produces whatever it outputs; better put, its output is in fact a set of structures that oblige one to understand the strings produced as organised in a definite way, where such an organisation is wholly abstract.

Representations without Representa: Content and Illusion in Linguistic Theory

39

Here is an example of what I mean. Consider the string in (1): (1) aabb Viewed just as a string, it has no definite structure at all, being generable in various different ways, right-left or left-right, for example. Viewed as the output of a finite-state machine, it is a concatenation from left to right, a letter per step. Here is a problem. If (1) is just a concatenated string, then it has no determinate relation to other strings, such as those in (2). (2) a. aaabbb b. aabbb To be sure, a finite-state machine that can produce (1) can produce these strings too, but what it cannot do is produce (1) and (2a) without also producing (2b). This is because the machine cannot embody the rule Produce the same number of bs as as (for that the machine would require a variable ranging over all previous strings (see below)). Instead, the machine embodies a rule of the form Produce another b, which does not require a variable at all. Thus, (2b) is necessarily producible by a finite-state machine that produces (1) and (2a). Evidently, however, a pattern is not being captured here, viz., strings of the set {anbn}, which does not include (2b). A variable solves the problem. Let ‘X → Y’ mean ‘Y is of type X’: (3) a. Z → ab b. Z → aZb From (3), we can generate the structure (4) [Za[Zab]b] corresponding to (1). It will be noted that (3) does not generate (2b), as desired. We can here say that the string in (1) only counts as a representation if it is structured in a definite way that makes it part of a system of

40

John Collins

representations, i.e., a value of a variable; after all, if it was a one-off occurrence, then its structure could be wholly adventitious. Only a system at least as rich as a phrase structure grammar that essentially employs a variable can appropriately structure the string, in the way indicated. Hereon, I shall assume that we can talk about representational systems in the wholly structural way just sketched. It will be noted that representations in this sense have internal structure and are generalisable in the sense of admitting substitution. What does not follow from this minimal notion, though, is any sense that such representations are available to the embedding system in terms of answerability, accessibility, or promiscuity. All we have is a bare idea of a set of structures that we posit in order to understand a competence that is lawful in treating otherwise disparate states or conditions as being the same, which is what the employment of variables captures. We might want to go further and suggest that exactly such a system will take such structures to be answerable and therefore have promiscuous access to them, but that is an additional claim. In the next section, I want to look at the notion of an empty category in linguistic theory in order to bring into relief the question of answerability for linguistic representations. 3. Empty Categories in Linguistics Empty categories are part of the modern linguist’s toolkit. I shall assume some familiarity with the notion and the variety of such categories posited in contemporary linguistics; in fact, in principle, any category can have an empty instance. Simply for orientation, therefore, the following remarks will suffice. (5) a. Mary wants to leave by herself b. *Mary wants to leave by himself c. Sarah wonders who Mary wants to leave by herself d. Sarah/Jim wonders who Mary wants to leave by himself

Representations without Representa: Content and Illusion in Linguistic Theory

41

(5a) is OK when non-embedded and when embedded in a further sentence whose subject is feminine, as in (1c). On the other hand, (5b) is not OK when non-embedded, but becomes OK when embedded in certain clauses, regardless of the gender of the matrix subject, as in (5d). We immediately recognise the difference between the cases, even though, for most readers, I imagine such a pattern had never occurred to them. The pattern can be explained by the positing of empty categories, where we take a competent speaker/hearer’s understanding to be characterised in terms of representations whose particular structure explains the pattern exhibited in (5). Such structures do not reflect how sentences appear or sound. In particular, imagine that the representation that explains a speaker/hearer’s understanding of the ‘subject-less’ English infinitive features an item in subject position relative to the verb. If the subject is not questioned, then it is represented as a generic pronoun (PRO) that is bound as if an anaphoric pronoun: (6) Maryi wants [PROi to leave by herself/*himself]10 Hence it is that (5b) is unacceptable, for only herself can agree in gender with its clause-mate empty category subject PRO, which is bound by Mary (marked by the indexes). The mystery now is why both (5c-d) should be acceptable; after all, (1b) is part of (5d), or is it? On reflection, we may note that (5b) is not semantically part of (5d), for in the latter construction, the subject of the infinitive is being questioned, and so is, in a sense, unknown, not necessarily Mary as it is in (1b). Hence it is that the reflexive may be masculine or feminine: (7) Sarah wonders [whoi [Mary wants to leave by herself/himself]] Such explanation of otherwise puzzling phenomena is how empty categories earn their keep. The above sketch, suffice it to say, is snow blown 10

There are current disputes about the status of PRO, but these may be sidelined for present purposes. See Landau (2013) for a survey of positions on ‘control’ phenomena that are often taken to necessitate a PRO-like item.

42

John Collins

from the tip of the iceberg that is modern syntactic theory. The question before us is whether the species of representation apparently integral to the kind of explanation just sketched involves a move beyond the purely structural notion presented above. On that model, PRO serves simply as an item to satisfy the structural demands of the verb of its host phrase, which, in the present constructions, is bound by some antecedent value of a nominal structurally related to the instance of PRO. This minimal conception does not suggest that PRO has no systematic effect on the interpretation of English infinitival constructions – such effects are well known and part and parcel of the analysis of control. The question that concerns me presently is how to understand the attribution of representations of PRO to the linguistically competent subject, not how PRO relates to semantic interpretation. To move beyond the minimal conception is to think of the subject who has a representation of PRO as being specifically answerable in a PRO-like way, as it were. Any kind of causal story looks sunk from the off. Being a noun, as it might be, cannot cause anything. Nouns are abstract, not concrete. So, a sound wave token of /dg/ might cause me to represent it as a noun, but the noun-hood of /dg/ is a consequence of this representation, it is not its cause. After all, there just is no unity to being a member of any linguistic category such that causal relations can be supported by tokenings of the category qua linguistic. This becomes all too obvious when we consider empty categories that have no external signature at all. PRO-hood, by definition, cannot cause representations of PRO, for PRO is just a feature of representations; it is not something external to which we are related. 4. Linguistic Illusion Without further ado, it seems that there is no way of understanding the intentionality or transitivity of linguistic representations: there isn’t anything for them to be about! Faced with such a problem and wishing to preserve the relationaity of representations, Rey (2003, 2005) has suggested that the representations linguistic theories posit have intentional inexistents as their contents. This thought is intended to place PRO on the same footing as Father Christmas and Zeus, or the representation of a

Representations without Representa: Content and Illusion in Linguistic Theory

43

scene being so and so, where the agent is suffering from a visual illusion. In such cases, there is nothing external to the agent that is Zeus, say. Even so, we hardly want to say that Zeus is in anyone’s head (Quine 1951). In some sense, then, we may represent Zeus, even though there is nothing independent of our representation that answers to [Zeus] that may make our representation correct or incorrect. We make as if to represent Zeus. In a phrase, representations can be answerable to intentional inexistents. According to Rey, noun phrases, words, sentences, PRO, et al. simply do not exist; rather, they are the contents we represent as part of our competence with language. I certainly think that this ‘illusion model’ is a vast improvement on an understanding of representation modelled on causation, but the proposal has problems due to its essentially conflating theoretical representational attribution with the personal notion. Firstly, I take it to be granted on all sides that intentionality or answerability, should the two notions be distinct, remains an obscure and somewhat imponderable notion, at least if approached from a theoretical perspective. It might still be, of course, that the notion is indispensable to linguistics and, more generally, any explanatory appeal to representations. That said, if one is concerned about the opacity of intentionality showing up in linguistics precisely because it threatens the naturalistic credentials of the science, one’s qualms will not be allayed by the reassurance that the relevant representations pertain to inexistents. Rey’s denial of externally individuated linguistic types and tokens is more than reasonable and is broadly accepted within linguistics, but such a denial can be motivated entirely independently of any worries about representation. One may simply note that modern linguistics neither presupposes nor entails external linguistic types. If this is right, then any philosophical quandaries associated with external linguistic types are simply irrelevant to the linguist. So, denying the external existence of noun phrases, say, does not eo ipso suggest an intentional inexistent view of noun phrases. Perhaps noun phrases do exist, just not externally; they are theoretical abstracta, whose ultimate ontological status is not in any way decided by the host theory itself. Secondly, the notion that we suffer an illusion in our consumption or production of language strikes me to be misleading, at best, as not a very

44

John Collins

useful way of describing the fact that linguistic properties are not answerable to external properties. The crucial characteristic of illusions is that we suffer them, in the sense that we can be debriefed or get to be in a position to see the visual set-up that wrought the effect. In Austin’s (1961) terms, illusions differ from delusions in that the former presuppose that our sensory capacities are working fine, while the latter presuppose that the capacities are functioning in some abnormal way. Illusions persist or are unavoidable, but their provenance can be understood and their effects cognitively isolated by simply being on one’s guard or being familiar with the effect. On the other hand, if one has delusions, a visit to the doctor is recommended. Now, we scarcely want to say that being linguistically competent is being delusional, for being so competent is precisely to be functioning normally – nothing whatsoever has gone wrong. Yet if the relevant content (e.g., the representation of syntactic structure) is illusory, then it seems that we have positioned the content at the wrong level, as it were; we have made it seem as if the contents at issue are conscious, whereas they are supposed to be largely sub-personal. My point here is not to trot out the familiar homunculi fallacy or make appeal to what we ‘ordinarily’ say, whatever that might be. The point, rather, is that the claim that linguistic representation is illusory presupposes intentionality rather than sheds light on linguistic representation in the first place, for something must be represented as thus and so for a system such that the system can represent its occurrence as an illusion. An illusion is brought about by the normal and proper operations of the mind/brain producing a state that is at variance with some other source of knowledge about the matter at hand, such as the length of a pair of lines or the relative position of a sound within a sequence of sounds. Thus, illusion requires a meta- representational system and so presupposes a representational state. The mind/brain just does its thing, indifferent to whether an illusion is produced or not. At any rate, if there is no variance between the normal functioning of the mind/brain and some source of knowledge about the same matter, I can’t see how an illusion is supposed to arise. It is, for example, unhelpful to view the table before you as an illusion on the basis of your failure to perceive it as mostly empty space, as physics supposedly informs us it is. Quantum physics cannot debrief us about tables and their shapes and colours, be-

Representations without Representa: Content and Illusion in Linguistic Theory

45

cause the physics recognises no such properties for us to be right or wrong about. Physics, ultimately, is supposed to explain the phenomena of colour, solidity, etc., not preserve the content of our experiences. Thus it is in the linguistic case. I do think we suffer from various kinds of ‘linguistic illusion’, but these are ‘special circumstance’ illusions and so do not support Rey’s thesis of universal illusion (I take it as given that no-one should be tempted by an ‘argument from linguistic illusion’ that takes us from one illusion to everything being an illusion). Consider the familiar ‘garden path’ sentence: (8) The horse raced past the barn fell The naïve subject is here led to misconstrue the first verb she comes across (race) as the main verb. So construing (8) leads to apparent gibberish with the final verb (fall) dangling without a subject. The ‘illusion’ arises from an agent following a rule of the form: Interpret the first verb of a sentence as the main verb until further notice or in the absence of any evidence to the contrary. The rule works fine in the vast majority of cases, but can lead a speaker astray, where, say, a verb of a reduced passive relative clause modifying the subject is the first verb. Thus, (8) has a fine interpretation as The horse that was raced past the barn fell, but the interpretation is mostly inaccessible. Kim Sterelny has contended that there are no linguistic illusions: There are, it is true, ‘garden path’ sentences and other difficult to process constructions. But these do not prompt false beliefs about what is said; rather, they result in comprehension failures. (Sterelny 2010: 285)

The argument here is that illusions must consist of a content that is false regarding what is said, much as a visual illusion consists of a content that misrepresents how things are. A failure of comprehension, on the other hand, does not necessarily consist of any belief at all about what is said. This reasoning is mistaken. Firstly, even a comprehension failure instead of a false belief about what’s said might still plausibly be construed as an illusion, if the cause of the failure is a result of the systematic misrepresentation of the struc-

46

John Collins

ture; after all, a ‘garden path’ does have a potential interpretation. So, the false belief generated here is that the sentence expresses no content. Sterelny does not explain why a false belief about what is said is necessary for a linguistic illusion. Secondly, there are other cases, anyway, where a false belief about what is said is produced, such as the familiar case of No eye injury is too trivial to ignore, which is naturally construed as an injunction against ignoring eye injuries, but it in fact encodes the contrary advice, as evidenced by the structurally equivalent No atom bomb is too small to ban. This counts as an illusion by Sterelny’s criterion in terms of generation of false belief. Similarly, tokens of Many more people have been to Paris than I have are liable to generate false beliefs. The structure does not support any coherent interpretation, as becomes evident when the ellipsis is filled: Many more people have been to Paris than I have been to Paris. These are very familiar examples; any textbook of psycholinguistics abounds with more. There are, therefore, linguistic illusions much as there are visual illusions. In both cases, it is a fair hypothesis that the illusions reflect design features of the respective cognitive systems. Rey, however, does not have the kind of illusions just discussed in mind for his general purpose of spelling out an intentional inexistence model of linguistic representation. His idea is more that we suffer the illusion of seeing non-linguistic material items (sound waves, etc.) as if linguistic, which is quite distinct from the above illusion, where one linguistic structure is interpreted as another. It is not clear, however, whether there is any illusion according to Rey’s understanding. If the relevant linguistic technology is all unconscious, then it is unclear in what sense a speaker/ hearer may be mistaken about it. That is, to be under an illusion, the speaker/hearer must systematically represent things to be thus and so, whereas things are really such and such. The problem here is to make sense of speaker/hearers initially representing things to be thus and so in a way that is wholly divorced from their ability to assess or meta-represent such representations so as to be answerable to them in a way that might be systematically, unavoidably mistaken. That is still a notion up in the air, which is not made any clearer by the attribution of illusory content. It is difficult to see how a speaker/hearer can be under an illusion that sound waves feature PRO or copies of who. The speaker/ hearer lacks all such

Representations without Representa: Content and Illusion in Linguistic Theory

47

notions, so she can’t judge any experience in terms of them. The notions are the theorist’s, which he employs to explain the speaker/hearer’s competence with the language; no-one imagines the concepts to be used by the speaker/hearer herself to describe her own linguistic experiences or offer opinions about her language. In sum, it is difficult to see how a speaker/hearer could be under the illusion that x is F, if she lacks the concept of F. The intentional inexistent model, therefore, strikes me as fundamentally obscure, not because the relevant linguistic properties are obviously externally instantiated (I think it is obvious that they aren’t), but because we have no understanding of the kind of phenomenon intentional inexistence as a property of unconscious representation is supposed to be. It can’t be an illusion, as we have seen; nor can it be a systematic error of some kind, for linguistic agents are not in error in relation to the relevant concepts, i.e., those of theoretical linguistics. One might still want to say that speaker/hearers are under the illusion that words and sentences are out there, but such notions are not what linguistic theory posits; besides, there is sufficient opacity about what is involved in our common-sense notion of word or sentence that I really haven’t a clue what it would be to represent something as a word or sentence such that it is illusory to do so. After all, who decided that our linguistic common sense is naïve realism? It bears emphasis that I think Rey is certainly correct to claim that the only reality there is to language is as represented, as a feature of human cognition. Where we should part company is over his further claim that this position is best spelt out by way of preserving a relational conception of representation with the object being an intentional inexistent. 5. Representations without Representa Our problem is that linguistic representations – representations of x, where x is a linguistic notion – should have linguistic kinds as relata, but such kinds are not independently instantiated so that our representations might be answerable to them. The world (the environment of the organism), we might say, just isn’t the kind of place where one finds PRO. Rendering the representations to be of inexistents does not help us much,

48

John Collins

for we still need to explain the representation of x in the first place, whether x is inexistent or not; for, it is not obvious how a sub-personal system can be under illusion (as opposed to producing one). The way out of this problem, I think, is not to focus on the relata, but on the relation, i.e., deny the relational character of representations. It bears emphasis that the decision of whether to take representation to be relational or not is orthogonal to whether the representational state itself is individuated narrowly or widely, i.e., with reference to the environment of the system or not. Let us agree that representational states, linguistic or otherwise, can obtain within the physical confines of an individual organism. It remains an open question whether the individuation of such states factors in that which the state purports to represent as an independent entity, or, indeed, other entities to which the organism might stand in causal and cognitive relations. Our issue, in other words, is not about individualism (whether or not externalia enter into the individuative conditions of representations), but more narrowly about whether representation can be construed non-relationally, even if representational states do depend upon externalia for their individuation, somehow or other.11 Our minimal notion of representation is at least consistent with such a non-relational approach, for it is purely structural. The minimal notion, however, does not allow us to make sense of a theory attributing representations to a system; it only allows us to understand that which is attributed as non-relational. For example, if we have a theory that trades in PRO, then we can understand PRO as not being answerable to anything; it is understood in structural terms relative to a theory that as a whole seeks to capture linguistic competence. But how are we to understand the theorist talk, as it were, of the linguistic agent representing PRO? Even if PRO is non-relational and not answerable to the way the world is, the linguist appears to credit the speaker/hearer with a representation of PRO. The question, therefore, is how to understand the representa11

For example, one could deny that linguistic representations of PRO are genuinely relational, but still hold that such representational states depend upon a communicative capacity to encode semantic information that essentially involves interlocutors. One may remain perfectly non-committal on this question while still holding that linguistic representations are non-relational.

Representations without Representa: Content and Illusion in Linguistic Theory

49

tional talk of the linguist in a non-relational manner, for the posited structure of the linguist’s theory, presumably, must be represented, even if the structures have no external bearing, i.e., do not take external values. I shall answer this query by way of an appeal to the ‘method of hyphenation’. Chomsky (2000: 159-161) takes his intended notion of representation to be non-relational, but does not spell out what he has in mind. McGilvray (1999: 127-162) elaborating on Chomsky’s position takes representation of x to be a categorical rather than a relational notion, which renders ‘representation of x’ as ‘x-representation’. This method of hyphenation, as we may call it, goes back to Goodman (1949, 1976) and Dennett (1969). They had their particular concerns (McGilvrary mentions Goodman, but not Dennett). Goodman (1949) appeals to hyphenation, not as a positive proposal, but to show that no two words could properly be said to share the same meaning: for any two orthographically distinct words, each could enter into a hyphenated compound that differed in extension from the other. So, a morning-star-picture is extensionally distinct from an evening-star-picture, even though morning star and evening star both refer to Venus. Later, Goodman (1976) did offer hyphenation as a positive view, but only as regards depiction. So, we might think of a painting of a dog in terms of a dog-painting. This is ingenious, but also quite intuitive. A painting of a dog seems not to be relational, for there need be no dog at all, not even a notional one, which the painting is of; the painting is merely of the dog type, as it were. For his part, Dennett (1969) employs the idea, without reference to Goodman, to avoid unwanted ontological commitments. Suppose one is leery of voices as entities. How would one accommodate Sue has a lovely voice, which appears to relate Sue to voices. Hyphenation comes to the rescue by rendering the problematic sentences as involving monadic predicates, which perforce do not relate the subject to the problematic entities: Sue has-a-lovely-voice. Dennett does not commend this account as any kind of logical analysis, but merely as a way of foreclosing ontological worries for those cases where one’s ontological scruples are in tension with the vernacular. In this sense, the method of hyphenation is simply a provisional way of being parsimonious in one’s ontological commitments; the hyphens may be removed where a subject’s commit-

50

John Collins

ment to the relevant entities is recognised to play an essential role in the explanation of the mental life of the subject.12 If we abstract away from the particular concerns that animated Goodman and Dennett, we can think of the bare method of hyphenation as a way of categorising particular entities or states in terms of other entities (real or notional). The crucial feature is that, where the vernacular would express this categorisation dyadically, in terms of a painting of a dog, say, the hyphenation renders the dyadic attribution monadic: the painting is a dog type, much as it might be green or expensive or French. As I noted above, elucidating the dyadic attribution into a monadic one is neutral as regards individualism. The crucial feature of the method is that monadic states are the truth makers, if you will, of the dyadically specified states, which leaves it open whether and how the monadic states are dependent on features of the environment. On this model, then, the contents of the representations (the ‘x’ of ‘representation of x’) are now understood as ways of categorically typing states as kinds of representations or structures, where such categories – the xs – are the posits of our theories. So, when we attribute a representation of PRO to a speaker/hearer, we are saying that a state of the subject is to be typed as PRO, where such a state stands in an explanatory relation to whatever phenomena a theory that posits PRO explains (see section 3). In effect, this provides us with the notion of representations without representa. It bears emphasis that I am commending this model only for the representation of theoretical posits, ones of the kind witnessed in linguistics. As we shall soon see, there is good reason for such narrowness, as the model looks to be infirm when applied to cases where there isn’t a distinction between the perspective of the theorist and the perspective of the object of the theory.

12

I favour the general tenor of Dennett’s approach, but since I think that ‘logical form’ is no guide to what exists or even to what we think exists, the hyphenation is not required to foreclose ontological worries about voices and the rest. After all, who among us believes in voices as disembodied entities even though we are all happy using voice as a nominal occurring as an argument of all kinds of verb.

Representations without Representa: Content and Illusion in Linguistic Theory

51

6. Personal Problems with the Hyphen The method of the hyphen has not found wide appeal over the last forty years or so; the reasons are not hard to fathom, if the method is understood to apply to personal representations as characterised above. Although the insertion of hyphens in complement clauses of representational verbs certainly does avoid the problem of ‘aboutness’ of representations, because the relevant predicate is rendered monadically, not dyadically, the victory comes at the cost of utter failure to respect the properties of personal representations. Thus, if the method of the hyphen is construed as an analysis or an account of logical form, then it is, almost, the worst theory possible.13 Recall that personal representations include as a paradigm case structured entities to which agents are related in terms of a variety of attitudes. So it is that we can draw inferences and generalisations over the contents represented, which depend upon the contents being both structured and independent of any particular attitude. The method of the hyphen obliterates both conditions, and introduces unique oddities. On the first count, the relevant points are trivial. If complex contents are fused/hyphenated along with the attitudinal predicate as ‘unbreakable’ monadic predicates, then no generalisation or inference can hold over the contents, which are now mere aspects of the single fused category. In stark terms, Bill in Bob thinks Bill is a cheat is as open to generalisation or inference as cat is as it occurs in cattle. For example, if we make the attribution in (9)a, then (9)b-c should follow: (9) a. S represents P & Q b. S represents P c. S represents Q At the very least, to represent a complex is to represent the parts of the complex; that looks like a simple syntactic truism, regardless of the in13

I am reluctant to agree that it is the worst theory possible, for the account does offer some insight, I think, on so-called intensional transitives, such as depiction. Such cases are not, if at all, straightforwardly analysable as propositional attitudes (cf. Forbes 2006).

52

John Collins

terpretation of the complex. This banality, however, is flouted by our proposed category notation: (10) a. S P-&-Q-represents b. S P-represents c. S Q-represents Clearly, (10b)-(10c) do not follow from (10a), for having one kind of representation does not entail having another, unless the content of one is included in the content of the other or, in syntactic terms, if a mechanical rule permits the transformation of one symbol into another. Either way, the model in (10) precisely banishes the componential structure of the representations that allow for such relations between representations. In (10), there are simply three distinct representations. What we call them looks to be irrelevant. We could have called them pain, misery, and grief. So, the relations in (9) are simply not preserved in (10). The additional oddity is that the account entails that the same attitude cannot be tokened with different contents; the attempt to do so results in something like a pun, for since each token is fused with some different complement, then it is simply a distinct monadic predicate. One cannot bear the same relation to different contents. Indeed, there aren’t really different attitudes at all, for P-belief has as much in common with Q-belief as it does with P-desire. It seems, therefore, that we have an indefinite number of primitive monadic predicates to describe our attitudes, i.e., they have no common parts between them. The basic problem with the account is that it makes accidental the way an attitude is specified in terms of its content, but there is nothing accidental about it, precisely because representations are systematically related to one another, which, perforce, the method of the hyphen has to elide. If Bill believes that Bob is a cheat, presumably it matters that it is Bob, as opposed to Sam, or whoever, that is the target of Bill’s attitude, but the hyphen account can’t even provide that much save by stipulation (let’s call the first

Representations without Representa: Content and Illusion in Linguistic Theory

53

state ‘belief-that-Bill-is-a-cheat’, which could easily have been called ‘belief-that-Sam-is-a-cheat’.). All of this is quite absurd.14 7. Reviving the Hyphen Notwithstanding the absurdity of the hyphenated account as an analysis or account of logical form, I do think it offers genuine insight into subpersonal representation of linguistic properties. What is right about the account is that it depicts the representations posited by linguistic theory as monadic (non-relational) states of a system that are typed according to the categories of the theory. So, on this account, to talk of a system representing x is not to relate the system to x in anyway at all, but to type a state of the system as an x-type (x-representation), where ‘x’ is a bit of our theoretical technology. So far so good, but how can the theoretical attributions be fused as hyphenated predicates without loosing all explanation? I think this just question has an answer. The representations the linguist employs to explain linguistic competence are simply linguistic categories, not propositions speaker/hearers know or believe. Thus, Chomsky (1981: 223) writes: [S]aying that a person knows his grammar and knows the rules and principles of his grammar... does not imply that he has propositional knowledge that these are the rules and principles of his grammar. The linguist may develop such propositional knowledge, but that is quite a different matter.

14

The above kind of criticisms were forcefully made by Fodor (1978). The points are perfectly obvious, though, and do not militate for any particular account of personal-level representation, save for a non-hyphenated one, of course. In particular, I do not intend here to take a stand on the correct analysis of propositional attitude constructions, whether, say, they are (dyadic-) relational or not (see Russell 1912, for a classic non-propositional analysis of judgement, and Moltmann 2003 for a somewhat different analysis in the same spirit for the attitudes in general). My present intent is only to be able to conclude that a ‘fused’ analysis is forlorn, independently of whether or not some other non-dyadic analysis is feasible.

54

John Collins

So, in our theories, we want to preserve the distinction between x and its representation insofar as we want our theories to encode generalisations and capture inferences that turn on componential structure. From the theorist’s perspective, therefore, there are just the contents free of the notion of representation. The x-representation notion only becomes relevant when we wonder how the speaker/hearer stands to our theoretical statements. The statements would be true just in case the speaker/hearer falls into a pattern of states as recorded by the theory, where the states are typed in terms of the theoretical technology required to explain the phenomena of the speaker/hearer’s competence. One problem with the story so far is that the speaker/hearer must apparently make inferences and arrive at generalisations, but we have effectively said that this cannot happen unless the speaker/hearer possesses representations of x, not mere x-representations. The simple answer to this problem is that speaker/hearers do not make inferences or draw generalisations about the kind of states that our theories attribute to them, no more than a boat on a river does vector algebra. The theory may well, say, model a particular process or state of understanding as a derivation of structures, but it does not follow that the theory is true only if the speaker/hearer makes the derivation (whatever precisely that means). It suffices that the state-transitions the theory models are reflected in the system itself without the system necessarily doing inferences. For example, there are numerous models of sentence comprehension and various ways in which such models embody the syntactic representations that are our concern (see, e.g., Townsend and Bever 2001). To be sure, presumably only one such model is true, but no particular model appears to be entailed by syntactic theory as currently conceived. The syntactic theory places abstract constraints upon components of the mind that can, as far as we know, be variably satisfied.15 In simple terms, then, the problem 15

The ‘paradox’ of parsing is that it appears to be a top-down, left to right process, whereas syntax is a bottom-up, right to left process of merging lexical items. It might be that a syntactic representation is still operative in parsing as a check on the parse. Alternatively, it could be that the full set of syntactic relations (as recorded in a tree) are never represented explicitly, but only constrain the parse step by step and reveal themselves in parse failures. Unfortunately, this topic is too broad to be further discussed here.

Representations without Representa: Content and Illusion in Linguistic Theory

55

here is severe only if we imagine that the kind of inferences and generalisations recorded by the theory that posits the relevant states must be recapitulated by the object of the theory, if the theory is to be true. I can see no reason to think that such a strict recapitulation is required.16 Here is another way of understanding the general claim. The theoretical statements we take to be true of a system are not hyphenated, precisely because such a method would remove all content from them. The inferences and generalisation they support, however, are not attributed to the system itself; rather, we take the system to realise a set of monadic states, which the theory describes as a set of interrelated principles in terms of its posited technology, not as a theory which is represented. So, no explanation of the theory involves the method of hyphenation. Only when one asks how the system stands in relation to the theory may one appeal to hyphenation, and even here only as a means of saying that a monadic state is specified dyadically in terms of the system realising states systemically constrained by the theory’s principles. Hence, the ladder of hyphenation is kicked away, but may still be offered to those below. I shall now look at a problem for the account. The account looks like a species of anti-realism or some kind of instrumentalism. Thus there arises a familiar problem: how can the representational explanation be successful, if there are no representations with the structure the theory posits? Put generally, the thought here is that the best explanation of why a theory is successful is that it is true, and for a representational theory to be true is for there to be representations as representa of dyadic relations. In the present setting, however, this thought begs the question against my approach, which precisely denies that representational attributions are made true by dyadic states. Let us not, however, descend into a dialogue of the deaf. I am sympathetic to the realism that animates the concern, so let me speak to that issue. Clearly, the account on offer is not a behaviourist one, according to which a system merely needs to behave or be disposed to behave in a 16

Of course, the child has often been depicted as a little theorist, which is just to make the mistake I am warning against. I think, however, that ‘little theorist’ talk need not and should not be taken too seriously. See Collins (2007) for extensive discussion of this issue.

56

John Collins

certain pattern for the contents to be rightly attributed. The theory is of the internal states of the system, not its behaviour. In this sense, the distinction between a system following rules or laws, and merely falling under them is completely wrong-headed, if one imagines that the relevant laws cannot be stated for the internal states of the systems as opposed to its mere behaviour. It does not follow, however, that we therefore depict the theory as being in the system. The theory is intended to capture the possibilities of the system without making any assumptions about how the system is so much as capable of falling into such states and what determines whether it is in this or that state at a given moment (consider, say the range of ambiguities available given a single morphophonemic form). So, in this sense, the laws are of the internal states of the system. The theory is going to be true, as best as we can tell, if it proves to be uniquely explanatory of the relevant linguistic phenolmena. The theory does not make any additional claims about how the system realises the kind of possibilities depicted, so the theory is not liable to the ultimate truth about such matters. In the remainder of this section, I shall defend this account by showing that it makes best sense of the attribution of the representation of both principles and structures over which they are defined. The account on offer, therefore, is not an eccentric reinterpretation of linguistic theory, but perhaps the only coherent interpretation available. First, then, consider syntactic principles, which are, if you will, laws of the system. The general picture I am commending is inescapable here (cf. Collins 2004, where a number of principles are discussed). Consider the Projection Principle (PP) of GB syntax.17 PP basically claims that lexical items that enter a derivation at a level Li, should impose the same constraints (or contribute the same interpretive information) at Li+n as they do at Li, where ‘n’ ranges over all further levels. In effect, PP is a ‘no-tampering’ condition that derivations retain information in the sense that a derivation draws the information out of a lexical selection in line with general combinatorial principles rather than alters or deletes the information that is in the system. I am not presently concerned with the 17

Similar generalisations remain in more recent theories, but here I am just interested in the principle as an example.

Representations without Representa: Content and Illusion in Linguistic Theory

57

truth of the principle, but its virtue is easy to see in a simple example. In a wh-question such as Who does Bill love?, the transitive verb love cannot be interpreted intransitively or contextually, but must be interpreted relative to the wh-item. So, although the wh-item moves and the verb looks as if it were intransitive (it does lack an item in object position), the system in some sense has not lost any information, precisely because who is serving two roles: providing an object for love and also questioning that argument position. Just from so much, I think a dilemma can be mounted for someone who insists that PP must be represented by the linguistic system. (i) Assume PP is represented by the system. (ii) Lexical items either (a) contain the information PP covers or (b) they don’t. (iii) If (a), PP merely replicates the information the lexical items already possess (e.g., a transitive verb is interpreted as having an objectual argument). PP is thus redundant. (iv) If (b), then a lexical item will be interpreted relative to its merged position, which the item will not select given that it lacks the relevant information. Again, PP is redundant, for there is no fixed lexical content it covers. The conclusion we should draw from this is that PP is true (if true) because of how the system as a whole operates, not because of how the system generates its own generalisations about how it operates in terms of PP. The system does not need to do that; this is signalled by PP being transparently redundant when we depict the system in such terms. It might be thought that although such reasoning works for principles or laws, it does not work for syntactic representations (tree structures). I think, however, that much the same reasoning holds. It is a mistake to think of a syntactic representation as a description of a sentence or a representation of anything. A tree structure is best viewed as a particular hypothesis derived from the relevant laws about a particular state of the system, how it will consume, produce, and otherwise understand a particular recognisable form. It does not follow from this that the tree structure is explicitly represented. The system, rather, operates according to

58

John Collins

the constraints the particular hypothesis encodes. Consider an ambiguous sentence, such as Coward’s familiar line: Mad dogs and Englishmen go out in the mid-day sun. The ambiguity is resolvable in two ways depicted by the following bracketing: (11) a. [[[Mad dogs] and Englishmen]...] b. [[Mad [dogs and Englishmen]]...] Semantically speaking, the ambiguity arises from whether the sentence refers to dogs that are mad and Englishmen or to dogs and Englishmen, both of whom are mad. Clearly, any adequate grammar must be able to account for the difference, for any competent speaker-hearer is sensitive to the difference. But now consider the perfectly competent speakerhearer who never notices the ambiguity or even reflects on it; indeed, if asked what he meant when reciting the line, he might not even be able to give one an answer. In such a case, it seems arbitrary to say that the speaker-hearer employs either (11a) or (11b). We are tempted, I think, to say that he employs neither. Still, the fact that the speaker-hearer can recognise the difference and might otherwise employ one or another construal, shows that an adequate grammar must record the difference. The point of the representations, therefore, is to account for the parameters of freedom a speaker-hearer has in his consumption and production of language, at least in a modally robust, idealised way. We do not need to imagine that every speaker represents a sentence to themselves in some particular way prior to or coeval with consumption or production of it. 8. Non-Theoretical Representations The kind of account offered for theoretical attributions of linguistic representations does not translate into an account of general intentional inexistence as it arises in colloquial discourse. As argued above, however, this is not a defect of the account in the absence of a plausible argument that representation is a univocal notion across and within theoretical and colloquial discourse. As it is, the difference between the two cases is

Representations without Representa: Content and Illusion in Linguistic Theory

59

easily explicable. The account presented denies that the relevant linguistic states are answerable from the perspective of the system: the states are those a theory identifies in terms of sui generis abstract categories, not states that represent the world to be a certain way such that the subject is answerable to how the world happens to be. The general problem of representations without representa is how to understand the relationality of representation when the external relatum is missing. The account on offer is directed at a local instance of the problem that does not presuppose or entail fully relational intentionality, insofar as linguistic theory does not attempt to record what, if anything, our mind/brains are answerable to such that their states can be said to be true or false, satisfied or not. Furthermore, as previously noted, any generalisation from our Goodmanesque local solution to intentional inexistence in the broad sense that covers little Johnny’s thoughts about Father Christmas meets with an insuperable problem. The problem is the very one we evaded earlier by making a distinction between the theory and its object. No such manoeuvre is available in the general case. Even if we were to construe ourselves as theorists when we attribute thoughts to little Johnny, the problem remains that Johnny himself has the thoughts, or something similar. When, with our theoretician’s hat on, we attribute PRO to little Johnny, he has no thoughts about PRO at all.18 It might still be thought that an account of representation in linguistic theory should be subsumed under some general account of representation as such that covers human language, the beliefs of little Johnny about father Christmas, navigating insects, singing whales, and much else besides. While the search for such generality is philosophically commonplace, there is to my mind no empirical conceptual reason to constrain the theoretical attribution of representation in our area of concern (theoretical linguistics and related disciplines) to answer to whatever we think of little Johnny and the rest. This is not to deny that we are able to conjure up a suitably abstract notion, perhaps representation-as, 18

None of this is to suggest the impossibility of an account fiction/non-existence consistent with the relational construal of representation (see Everett and Hofweber 2000, for discussion). For present purposes, my claim is only that such an account cannot be the one that works for linguistic representation. See Azzouni (2010), however, for a model congenial to the present perspective.

60

John Collins

which leaves it open whether the system is conscious or not and whether the putative representation has any represented object (cf. Burge 2010). My only complaint is that such an abstract, anaemic notion offers no constraint on the kind of account we are after, and so is redundant, even if not incoherent. That is to say: linguistic explanation appears to have no recourse to a universal notion of representation. Besides, we may not be able to be precise about just what we mean when we speak with the folk of fictions and other entities that enjoy a somewhat recherché ontological status (numbers, Euclidean figures, etc.), at least as compared to the stable physical properties bees and other insects appear to represent as understood in our theories. The worldly conditions under which common representational talk is true is not so clear. The truth conditions of our personal representations might be answerable to all manner of worldly arrangements that are not definitely recorded as part of the linguistic meaning of the words involved. Walt Disney, for instance, might be relevant to the truth of claims about Donald Duck, but one need have no notion of Disney to be competent with Donald Duck. On the other hand, no such issues arise for our talk of representations in linguistics (and other areas of cognitive science, for that matter). What makes the claims true is opaque insofar as the science is hardly complete, but we commit ourselves to the categories attributed to speaker-hearers as abstracta featuring in successful explanations. There is no other hurdle to clear. 9. Concluding Remarks If all of the above is right, we can have representations without representa. The representations are states of a system that are typed in terms of the representa, which are the theoretical notions employed by the relevant theory. The linguistic mind doesn’t represent verb phrases, or PRO, or anything else. We use ‘verb phrase’, ‘PRO’, et al. to categorise states in patterns that our theories track. This doesn’t make the mind/

Representations without Representa: Content and Illusion in Linguistic Theory

61

brain states unreal, no more than using vector algebra to track the trajectory of a cannon ball makes the trajectory unreal.19 References Anderson Stephen and Paul Kiparsky (eds.) 1973. A Festschrift for Morris Halle. New York: Holt, Rinehart & Winston. Antony, Louise M. and Norbert Hornstein (eds.) 2003. Chomsky and His Critics. Oxford: Blackwell Publishing. Austin, John L. 1961. Sense and Sensibilia. Oxford: Oxford University Press. Azzouni, Jody 2010. Talking about Nothing: Numbers, Fictions, and Hallucinations. Oxford: Oxford University Press. Burge, Tyler 2010. Origins of Objectivity. Oxford: Oxford University Press. Byrne, Richard and Andrew Whiten 1988. Machiavellian Intelligence: Social Expertise and the Evolution of Intellect in Monkeys, Apes and Humans. Oxford: Clarendon Press. Carruthers, Peter and Peter Smith 1996. Theories of Theories of Mind. Cambridge: Cambridge University Press. Cherniak, Christopher 1986. Minimal Rationality. Cambridge, MA: MIT Press. Chomsky, Noam 1973. Conditions on transformations. In: S. Anderson and P. Kiparsky (eds.), 232-286. Chomsky, Noam 1975. Reflections on Language. London: Fontana. Chomsky, Noam 1980. Rules and Representations. New York: Columbia University Press. Chomsky, Noam 1981. Knowledge of language: its elements and origins. Philosophical Transactions of the Royal Society B295, 223-234. Chomsky, Noam 2000. New Horizons in the Study of Language and Mind. Cambridge: Cambridge University Press. Chomsky, Noam and Howard Lasnik 1977. Filters and control. Linguistic Inquiry 8, 425-504. Collins, John 2004. Faculty disputes. Mind and Language 19, 503-533. Collins, John 2007. Meta-scientific eliminativism: a reconsideration of Chomsky’s review of Skinner’s Verbal Behavior. British Journal for the Philosophy of Science 58, 625-658. 19

A version of this paper was earlier delivered at The Personal and the SubPersonal conference (Institute of Philosophy, University of London, May 2012). My thanks go to Barry Smith, Tim Crane, and Dan Dennett for comments. I also thank Georges Rey, for many conversations on these issues, and Nicholas Allott for useful comments.

62

John Collins

Davidson, Donald 1980. Essays on Actions and Events. Oxford: Clarendon Press. Dennett, Daniel 1969. Content and Consciousness. London: Routledge. Dennett, Daniel 1978. Brainstorms: Philosophical Essays on Mind and Psychology. Cambridge, MA: MIT Press. Everett, Anthony and Thomas Hofweber 2000. Empty Names, Fiction and the Puzzles of Non-Existence. Chicago: Chicago University Press. Forbes, Graham 2006. Attitude Problems: An Essay on Linguistic Intensionality. Oxford: Oxford University Press. Fodor, Jerry 1978. Propositional attitudes. The Monist 61, 501-23. Fodor, Jerry 1983. Modularity of Mind. Cambridge, MA: MIT Press. Gallistel, Charles R. 1990. The Organization of Learning. Cambridge, MA: MIT Press. Gallistel, Charles R. and Adam P. King 2009. Memory and the Computational Brain: Why Cognitive Science Will Transform Neuroscience. Oxford: WillyBlackwell. Goodman, Nelson 1949. On likeness of meaning. Analysis 10, 1-7. Goodman, Nelson 1976. Languages of Art. Indianapolis: Hackett. Graves, Christina, Jerrold Katz, Scott Soames, Yuji Nishiyama, Yuji, Robert Stecker, and Peter Tovey 1973. Tacit knowledge. The Journal of Philosophy 70, 318330. Knowles, Jonathan 2000. Knowledge of grammar as a propositional attitude. Philosophical Psychology 13, 325-353. Landau, Idan 2013. Control in Generative Grammar: A Research Companion. Cambridge: Cambridge University Press. Marcus, Gary 2001. The Algebraic Mind: Integrating Connectionism with Cognitive Science. Cambridge, MA: MIT Press. Matthews, Robert 2007. The Measure of Mind: Propositional Attitudes and Their Attribution. Oxford: Oxford University Press. McGilvray, James 1999. Chomsky: Language, Mind and Politics. Cambridge: Polity Press. Moltmann, Friederike 2003. Propositional attitudes without propositions. Synthese 135, 77-118. Newell, Alan 1980. Physical symbol systems. Cognition 4, 135-183. Quine, Willard Van Orman 1951. On what there is. In: W. V. O. Quine From a Logical Point of View. Cambridge, MA: Harvard University Press, 1-19. Perner, Joseph 1991. Understanding the Representational Theory of Mind. Cambridge, MA: MIT Press. Rey, Georges 2005. Mind, intentionality and inexistence: an overview of my work. Croatian Journal of Philosophy 5, 389-415. Russell, Bertrand 1912. The Problems of Philosophy. London: Unwin.

Representations without Representa: Content and Illusion in Linguistic Theory

63

Sterelny, Kim 2010. Moral nativism: a sceptical response. Mind and Language 25, 279-297. Stich, Stephen 1978. Beliefs and Sub-Doxastic States. Philosophy of Science 45, 499-518. Stich, Stephen 1981. Dennett on intentional systems. Philosophical Topics 12, 3962. Townsend, David and Thomas Bever 2001. Sentence Comprehension: The Integration of Habits and Rules. Cambridge, MA: MIT Press. Whiten, Andrew 1991. Natural Theories of Mind: Evolution, Development, and Simulation of Everyday Mindreading. Oxford: Blackwell. Whiten, Andrew and Richard Byrne 1997. Machiavellian Intelligence II: Extensions and Evaluations. Cambridge: Cambridge University Press.

Luis Fernández Moreno

Complutense University of Madrid [email protected]

Kripke, Putnam, and the Description Theory Abstract: This paper aims to propose a version of description theory of reference for natural kind terms that could constitute an alternative to Kripke’s causal theory. This sort of description theory is based, on the one hand, on extension to natural kind terms of some proposals by Searle and Strawson about proper names and, on the other hand, on incorporation of a fundamental component of Putnam’s theory of reference for natural kind terms, that is, Putnam’s thesis of the division of linguistic labor. Keywords: reference, description theory, natural kind term, proper name, Kripke, Putnam, Searle, Strawson

0. Introduction In most manuals of Philosophy of Language it is asserted that Kripke and Putnam are the main advocates of the causal theory of reference for natural kind terms, and indeed very often it is spoken of the PutnamKripke or the Kripke-Putnam theory. Although in this paper we shall mainly deal with the debate between Kripke and the description theory of reference, we will also allege that, although there are some similarities between Kripke’s and Putnam’s reference theories for natural kind terms, there are also some important differences between them (see section 4), and that the description theory can incorporate a fundamental component of Putnam’s theory. It is appropriate to begin by pointing out one similarity between Kripke’s and Putnam’s theories. This similarity concerns their view of natural kinds, which can be put forward via the following three theses. Firstly, within the set of entities that constitute a natural kind there is a subset

66

Luis Fernández Moreno

whose elements are regarded as paradigmatic members of the kind. Secondly, this feature is attributed to such entities because they share a series of properties concerning their external appearance, to which these author allude as superficial or normal distinguishing properties. Thirdly, we assume that such entities share underlying or structural properties that are responsible for their superficial properties; in the case of samples of water, the structural property in question is their molecular structure, while in the case of samples of gold – their atomic structure. It is the possession of this last sort of properties that determines the reference or extension of the respective natural kind term. A difference between Putnam’s and Kripke’s theories is that Putnam has mainly dealt with natural kind terms and hardly with proper names, while Kripke’s considerations of natural kind terms in his main writing on semantics of this type of expressions (Kripke 1980) are secondary with respect to his thoughts about proper names. Only after having dealt with proper names in the first two lectures of 1980, does Kripke, in the third and last lecture, pay attention to natural kind terms, one of his aims being to allege the existence of similarities between natural kind terms and proper names. At the bottom of these similarities lies Kripke’s claim that the description theory is inadequate with respect to proper names or natural kind terms. For this reason, Kripke proposes, in the second lecture of 1980, an alternative reference theory of proper names, which he extends, in the third lecture, to natural kind terms. 1. Kripke’s Theory of Reference and the Characterization of Description Theory Kripke’s theory of how the reference of terms (proper names and natural kind terms) is determined consists of a theory of how their reference is fixed, that is, how it is initially determined, and another one about how the reference is transmitted and hence how it is determined for the speakers that have not introduced the term. According to Kripke, a term is introduced in an initial baptism in which its reference is fixed by ostension or “by a description” (Kripke 1980: 96-97), and although he tends to give more weight to the introduc-

Kripke, Putnam, and the Description Theory

67

tion by ostension, he concedes the possibility of subsuming the introduction by ostension under the introduction through description. Thus, he asserts that …the case of a baptism by ostension can perhaps be subsumed under the description concept also. (Kripke 1980: 96, n. 42, my emphasis)

Once a term has been introduced in an initial baptism, in which its reference has been fixed, the term is transmitted by the speakers present in the initial baptism, which will include the introducer or introducers of the term to other speakers, thus establishing causal chains of communication – also called by Kripke “chains of communication” (1980: 91 f.), as well as “causal (historical) chains” (1980: 139) or “historical chains of transmission” (1980: 163). A causal chain of communication has its origin at an initial baptism and connects the uses of a term, and hence its users, with the reference (extension) of the term in virtue of the speakers’ belonging to the same linguistic community through which the term is being transmitted. However, for a speaker to be a member of a chain involving a term it is required that, when he learns the term, he intends to use it with the same reference as it was used by the speakers from whom he learnt the term.1 In his critique of the description theory for natural kind terms, Kripke assumes that the descriptions, and therefore the general terms constituting them, to which an advocate of such a theory will resort, would only express the so-called superficial properties. These are purely general or purely qualitative properties, i.e., properties expressed exclusively by means of general terms that do not involve a reference to an individual entity – for instance, in the case of the term “water”, the properties of 1

Kripke often expresses himself as if the links of the causal chains of communication were speakers themselves (see, e.g., Kripke 1980: 91) rather than uses of terms by speakers, but the causal connections in a chain of communication take place between uses of terms; for this reason it would be more appropriate to regard uses of terms as the links of the chain. However, for the sake of simplicity, and as we have already begun to do, we will make use of the first, briefer, way of exposition. In this case, a speaker becomes a link of a chain when he first uses a term that he has acquired from his use by other speaker(s).

68

Luis Fernández Moreno

being colourless, transparent, tasteless, etc. Kripke attributes to the description theory the thesis that such sorts of properties would constitute the meaning of natural kind terms and would also determine their reference. However, in his critique of description theory, Kripke suggests that this theory can be interpreted as a theory of meaning or as a theory of reference (cf. Kripke 1980: 32-33, 53-54, 59). This distinction is plausible since, although the authors who advocate a description theory of meaning for a given type of terms will naturally sustain a description theory of reference for these terms, maintaining a description theory of reference for one type of terms does not necessarily entail sustaining a description theory of meaning for such terms.2 A description theory of meaning would be only relevant for our purpose in case it is sustained that the meaning of an expression determines its reference. Even then we can leave aside the description theory of meaning, extract the descriptive proposal concerning the meaning of the terms in question, and formulate it as a proposal concerning the reference of the respective terms. In what follows, we shall conform to that procedure and take into account only the description theory of reference. The justification of Kripke’s claim that description theory is inadequate with respect to proper names and natural kind terms presupposes a characterization of this theory. In the first and second lecture of 1980, Kripke characterized the description theory of proper names in a detailed way by means of six theses, some of which are only constitutive of a description theory of meaning, while others form a description theory of reference, although Kripke adds to the latter sort of theory a condition 2

There are versions of description theory of reference that are linked to a description theory of meaning, like those by Frege and Searle concerning proper names (see Frege 1892 and Searle 1969), according to which the meaning (sense) of a name determines its reference, and its meaning is standardly given by a description or by a cluster of descriptions. However, a description theory of reference does not need to involve a description theory of meaning. Thus, e.g., Strawson supported a description theory of the reference of proper names, but he sustained a Millian view of their meaning, according to which the meaning of a proper name, i.e., its contribution to the propositions expressed by the sentences containing that name, is its referent (see Strawson 1997a).

Kripke, Putnam, and the Description Theory

69

for its fulfilment, the non-circularity condition, according to which the properties or descriptions that determine the reference of a term cannot “involve the notion of reference in a circular way” (Kripke 1980: 97, n. 44). In short, the resort to the notion of reference in such properties or descriptions should not result in a vicious circle. Leaving aside for now this circularity condition, Kripke’s characterization of the description theory of reference for proper names, though extended to natural kind terms, is in the main the following: a term (proper name or natural kind term) refers to an entity3 if and only if it satisfies most, or a weighted most, of the properties or descriptions associated with the term by the speakers. Having presented the main thesis of description theory of reference for proper names, in its characterization by Kripke, and having extended it to natural kind terms, it is advisable to compare it with the one supported by some of the advocates of description theory of proper names in its contemporary version, e.g. Searle and Strawson. Firstly, when Kripke characterizes and critically examines description theory of reference, he talks indistinctly – as we have done – of descriptions or properties, like many of the present advocates of description theory do4 and as we shall also do in the following. Secondly, it should be pointed out that Kripke deals with reference as a property of linguistic expressions, while Searle and Strawson claim that reference is not a property of expressions, but of the speakers who use them. However, we will remain neutral as to whether reference is a property of linguistic expressions or of speakers, although our considerations using one of these options could be easily paraphrased in terms of the other. In any case, the debate between the causal and the description theory of reference does not centre on that question. 3

4

We talk of “an entity” for sake of simplicity. In the case of natural kind terms, which are a sort of general terms, we should talk of a set of entities, which constitutes the extension or reference of the term. It is appropriate to indicate that both, Searle and Strawson, talk occasionally not of descriptions or properties but of propositions, statements or facts, but this terminological difference is hardly relevant. The propositions, statements or facts in question would play – as the descriptions or properties – the role of singling out the object that constitutes the referent of the name.

70

Luis Fernández Moreno

Thirdly, Searle and Strawson do not allude to most, or a weighted most, of the properties or descriptions associated with a name, but rather, as we shall do henceforth, to a sufficient number of them.5 Fourthly, the descriptions that, according to Searle or Strawson, determine the reference of proper names are identifying descriptions.6 Yet these authors, especially Searle, understand this notion in a broad sense. For him, identifying descriptions include expressions of any of the three following types. Firstly, an ostensive or a demonstrative presentation of an object, e.g., “that — over there”, where the blank “—” has to be substituted by a general descriptive term; secondly, a description containing only general terms and being true of an only object;7 lastly, an expression that combines descriptive and demonstrative devices – in his words, a “mixed demonstrative and descriptive identification”, for instance, “The man we saw yesterday” (Searle 1969: 86, 92). However, it is noteworthy that, since only the second type of identifying descriptions mentioned by Searle would express general properties or purely qualitative ones, Kripke’s interpretation of description theory seems to be mistaken, as he sustains that according to that theory the reference of terms is exclusively determined by properties of this type. Fifthly, since present advocates of description theory appeal to a very wide notion of (identifying) description, it is hardly surprising that they employ the notion of property in a broad sense. In fact, current supporters of description theory, e.g. Frank Jackson, fall back on a comprehensive notion of property. Thus, Jackson asserts:

5

6

7

Searle (1969: 169) alludes literally to a “sufficient but […] unspecified number”, and Strawson (1959: 192) to “a reasonable, or sufficient” number. By “identifying description” it is understood a description sufficiently specific to singularize only one single entity. Strawson (1959: 182, n. 1) rejects the idea that identifying descriptions have to be put forward exclusively in general terms and claims that, as a rule, they will contain indexical terms: “it is impossible, in general, to free all identification of particulars from all dependence upon demonstratively indicatable features of the situation of reference.” In this regard there seems to be no discrepancies between Strawson and Searle, since the latter claims that descriptions in purely general terms constitute only a limited case of identifying descriptions (Searle 1969: 86).

Kripke, Putnam, and the Description Theory

71

The term “property” […] is simply a short word for a way things might be in the wide sense that includes relational and dispositional ways things might be. (Jackson 1998: 202)

Given the broadness of this notion of property, which is clearly not restricted to general properties or purely qualitative ones, it is understandable that the authors in question do not commit themselves to any specific conception concerning properties nor to a precise criterion of identity for them. However, the absence of these commitments can perhaps be excused given the diversity of properties that can be expressed by means of descriptions or by general terms that appear in these descriptions. According to the preceding considerations, based on the wide notion of property we have just mentioned, the version of the description theory of reference resulting from extending to natural kind terms, Searle’s and Strawson’s description theory of proper names could be characterized as the theory according to which the reference of a natural kind term is determined by a sufficient number of properties or descriptions that the speakers associate with the term. 2. Reference Fixing and further Details of Description Theory It could be alleged, in order to differentiate the causal theory from description theory at least with respect to reference fixing, that according to description theory, reference fixing of a term to denote an entity (see note 3 above) requires that the entity satisfies the identifying description involved in the introduction of the term. Meanwhile, according to the causal theory, it is not necessary that the description by means of which a term is introduced be true of the denoted entity. However, this contrast cannot be as sharp as sometimes intended or, at least, it cannot be so in the case of Kripke’s. According to some of his assertions, although the description by means of which the reference of a term has been fixed could be not true of the denoted entity, it has to be approximately true. The relevant assertions are related to Kripke’s thesis that the statements in which the fixing of the reference of a term is expressed are a priori

72

Luis Fernández Moreno

true. Kripke asserts the following regarding the descriptions through which a term is introduced that are not true of the referred object: In such cases, the description which fixes the reference clearly is in no sense known a priori to hold of the object, though a more cautious substitute may be. If such cautious substitute is available, it is really the substitute which fixes the reference. (Kripke 1980: 80, n. 34)

However, a description that constitutes a “more cautious substitute” of another cannot be very different from it; in particular, in some of the examples put forward by Kripke the difference would lie in the use of “more cautious sortals” (Kripke 1980: 116, n. 58); for example, the use of the term “planet” instead of the term “star” in case of the description employed to fix the reference of the name “Hesperus”. Therefore, a term is introduced in an initial baptism in which its reference is fixed. Such baptism can be considered to have taken place by a description – given the possibility of subsuming the baptism by ostension under the baptism by a description – and that description has to be (approximately) true of the baptized (i.e., referred) entity.8 As already said, in his critique of description theory of reference Kripke presupposes that the properties that according to such theory would determine the reference of natural kind terms are exclusively superficial ones; they are general or purely qualitative properties, and not structural or underlying ones. The question that arises is whether description theory can accept that not only superficial but also structural or underlying properties play a role in the determination of reference of natural kind terms. Our answer will be affirmative (see section 4). In order to gear up our justification of this answer it is convenient to appeal to certain components of Searle’s and Strawson’s reference theories that we have not mentioned so far, especially those concerning speakers’ association of properties or descriptions with proper names, which we will extend to natural kind terms. In consequence, we shall simply talk here, as we have already often done, of “terms”. 8

It is plausible to suppose that if the introducer of a term were completely wrong concerning the properties of the object or objects for which the term is introduced, the term would lack reference.

Kripke, Putnam, and the Description Theory

73

Firstly, the association of properties with terms is generally tacit or implicit rather than explicit (see Strawson 1959: 182, n. 1), but these properties can become explicit in the speakers’ answers to questions about what or whom they refer to by a term. Secondly, the properties associated with a term by various speakers can be different, but these differences are not important insofar as they identify the same entity (Searle 1969: 171; Strawson 1959: 183). Thirdly, the properties or descriptions associated by a speaker with a term may involve the reference of the term as used by other speakers. Thus, Strawson asserts: The identifying description, though it must not include a reference to the speaker’s own reference to the particular in question, may include a reference to another’s reference to that particular. […] So one reference may borrow its credentials, as a genuinely identifying reference, from another; and that from another. But this regress is not infinite. (Strawson 1959: 182, n. 1)

This last claim and, in particular, the admission of parasitic descriptions, as we shall call them, will be important for some of our later considerations. Searle also recognizes this sort of descriptions when he concedes that a description that a speaker can associate with a term can allude to the description that other speakers associate with the term: My reference to an individual may be parasitic on someone else’s but this parasitism cannot be carried out indefinitely if there is to be any reference at all. (Searle 1969: 170)

A description of this sort, that is, a parasitic description, will only be identifying if it remits to another identifying description which does not allude any longer to the reference of the term in their use by other speakers, i.e., parasitic identifying descriptions have to remit lastly to other identifying descriptions that are no longer parasitic ones. In this regard it is noteworthy that neither Strawson nor Searle (cf. Searle 1969) allege in principle any reason to justify that the deferral to the descriptions of other speakers would not involve us in a regress ad infinitum, although Searle in (1983) does mention a reason, namely, that the deferral to the

74

Luis Fernández Moreno

descriptions of other speakers can end at the term introduction in an initial baptism (cf. Searle 1983: 244). When comparing Kripke’s theory and description theory, it is noteworthy that the advocates of the latter can accept Kripke’s notion of an initial baptism in which the reference of names and natural kind terms has been fixed. Let us bear in mind that according to Kripke (1980: 96 f.), an initial baptism can proceed by ostension or “by a description” but, as already indicated, he concedes the possibility of subsuming the introduction by ostension under the introduction through description. In fact, Kripke does not seem to find significant differences between description theory and causal theory with respect to reference fixing; he concedes that there is a set of cases in which the description theory is true, these being “usually initial baptisms” (Kripke 1980: 78), although he points out that the use of terms (proper names and natural kind terms) in initial baptisms is “only […] a rare class of cases” in the use of terms (Kripke 1980:78). This qualification is, of course, obvious, since most uses of terms are not the uses involved in their (initial) introduction. 3. Reference Transmission and a Social Version of Description Theory As already indicated, Kripke concedes that description theory is usually adequate as a theory of reference fixing. Therefore, his arguments against that theory have to be mainly addressed not towards the explanation delivered by such theory concerning the reference of the terms as they are used by the speakers who have introduced them, but as employed by the speakers to whom they have been transmitted, that is, by most speakers. Thus, Kripke asserts: For most speakers, unless they are the ones who initially give an object its name, the referent of the name [or natural kind term] is determined by a “causal” chain of communication rather than a description. (Kripke 1980: 59, n. 22; my emphasis)

As already indicated, the notion of causal chain of communication alluded to in this passage is the core of Kripke’s theory about transmission of

Kripke, Putnam, and the Description Theory

75

reference. The question to be posed is whether description theory has a proposal to explain transmission of reference, and in this regard we have to again take up the sort of descriptions we have denoted as parasitic descriptions. This sort of descriptions was mentioned by Strawson, and by Searle, although they play a rather secondary role in their relevant versions of description theory; Searle (1983) grants them a more significant role than Searle (1969). However, these sorts of descriptions have to be regarded as most important, since when some of the descriptions that the average speaker associates with a term are questioned, he can defer the reference of the term (and as a rule he will be willing to do so) to other speakers that he considers more knowledgeable about that referent. It is noteworthy that Kripke has taken into consideration this descriptivist strategy and put forward some objections against it (cf. Kripke 1980: 90, 92 f., 161). The strength of these objections is unequal, but the most important one grounds on the assumption that the speaker who defers or is willing to defer the reference of a term to other speakers has to be able to identify such speakers. In this regard, it has to be pointed out that if the speakers in question are those from whom he learnt the term, he will probably be able to allude to them only in a generic way, because he will not remember them or will not be sure of who they are. However, going a step further, the speaker can resort to the speakers of his linguistic community socially recognized as authorities on the entity that constitutes the referent of the term, since the average speaker knows that there are such speakers in his linguistic community. He also knows how to identify some of them, although there can be a small margin of fallibility in this regard. These considerations connect in a natural way with one of the main components of Putnam’s reference theory of natural kind terms, the (hypho)thesis of the division of linguistic labor. In this regard it is suitable to indicate that according to Putnam there are two contributions involved in the determination of the reference of natural kind terms, which he denotes as the contribution of the environment and the contribution of

76

Luis Fernández Moreno

the society.9 On the one side, the extension of a natural kind term depends on how our world is, since it is determined by underlying properties of the members of the kind belonging to our world. On the other side, the discovery of the underlying properties is a matter of scientific research; for that reason those who carry out this research or make use of its results will have a better knowledge than an average speaker has on the membership conditions into a natural kind and hence into the extension of the corresponding natural kind term. In this regard, there is, in Putnam’s words, a division of linguistic labor. He formulates the “hypothesis of the universality of the division of linguistic labor” in the following way: Every linguistic community exemplifies the sort of division of linguistic labor just described: that is, possesses at least some terms whose associated “criteria” are known only to a subset of the speakers who acquire the terms, and whose use by the other speakers depends upon a structured cooperation between them and the speakers in the relevant subsets. (Putnam 1975b: 228)

Regarding this passage, it is appropriate to make some remarks. Putnam alludes to the speakers belonging to such subsets as “expert” speakers or simply as experts. Although it does not remain clear enough in that passage, the “cooperation” between the rest of speakers – i.e., average speakers or lay speakers or non-experts, and the experts – points to the claim that the former are willing to defer to the latter the determination of the reference of natural kind terms as they use them. This is the case, for instance, in the use of the term “gold” by the lay speaker. Thus, an entity falls into the extension of a natural kind term used by the average speaker if it a) falls into the extension of the term as used by the experts, and b) the average speaker is linked by relationships of cooperation with the expert or experts in question.10 In this regard we should point out that, although the experts par excellence are members of the relevant scientific community, it is advisable to conceive the group of experts in a 9

10

On these contributions, see Putnam (1975b: 227-234, 245, 265, 271) and Putnam (1988, chapter 2). It seems reasonable to qualify this assertion, since if a lay speaker were completely wrong concerning the sort of entity to which the expert speakers refer by a term, it could be questioned that in the idiolect of such a speaker the term has the same extension as in the use by the experts.

Kripke, Putnam, and the Description Theory

77

broader way, including, for instance, those members of the linguistic community that, while not being properly members of the relevant scientific community, employ techniques and procedures based on results of scientific research. This remark is relevant for the way the term “criteria” that appears in the quoted passage is to be understood, since the criteria in question constitute means to recognize whether an entity belongs to the extension of a natural kind term. However, these criteria do not have to include necessarily the explicit description of the underlying properties of the members of the kind – in case this set of properties are known – which is the determining factor for the extension of the natural kind term. Lastly, it is appropriate to point out that in later writings, like (1988), Putnam does not talk any longer of the “hypothesis of the universality of the division of linguistic labor”, but simply of the division of linguistic labor, as a thesis he subscribes to (see Putnam 1988: 22). We shall henceforth follow this path of reasoning. Description theory can incorporate Putnam’s thesis on the division of linguistic labor and sustain that the resort to the notion of expert can play an important role in a reference theory of natural kind terms. In addition, concerning the notion of expert, it has to be pointed out that, on the one hand, this notion is relative to a natural kind or to a sort of such kinds; on the other, the experts with respect to a natural kind will have a better knowledge than an average speaker as to the identifying properties of the members of a natural kind, which, according to authors like Kripke and Putnam, are their structural or underlying properties. Nevertheless, the version of description theory that we propose says that not only the experts with respect to a natural kind term, but all speakers, associate descriptions with the natural kind in question. Still, some of the descriptions associated by non-experts have the function of deferring the reference of the term to their reference in its use by experts; these descriptions, generally implicit, will tend to become explicit when other descriptions associated with the term are questioned. According to that version of description theory, non-expert speakers with respect to (the kind) gold implicitly associate with it a description equivalent to the following: the substance to which the experts on gold refer when they utter the term “gold”. In other words, the substance about which the members of my community socially recognized as au-

78

Luis Fernández Moreno

thorities on gold speak when they utter the term “gold”. And, as already indicated, the average speaker knows that in his linguistic community there are such sorts of speakers. However, the experts on gold will associate with that natural kind, and hence with the natural kind term “gold”, a set of properties, many of them identifying ones. These properties – or at least, many of them – will not involve the notion of reference; thus, the condition of noncircularity required by Kripke of the description theory is fulfilled. Come to this point, the theory will sustain that the extension of the term “gold”, as it is used by the experts on gold belonging to our linguistic community and hence also by the rest of its members, consists of entities that possess a sufficient number of the properties that such experts associate with that term. Therefore, the thesis of description theory, according to which the reference of a natural kind term is determined by a sufficient number of the properties or descriptions that speakers associate with the term, only applies in strict sense to those speakers who are experts in the kind that the term designates. They will associate with the term in question not only superficial properties and other sorts of macroscopic properties (of members of the kind), like the physical properties of the samples of the kind, as the boiling point, the melting point, etc., but also structural properties. The reference of a natural kind term will be determined by a sufficient number of such properties. Thus, this social version of description theory, which incorporates Putnam’s contribution of the society, can accept that not only superficial but also structural or underlying properties play a role in the reference determination of natural kind terms. 4. Putnam’s and Kripke’s Theories of Reference Obviously, this version of description theory gives priority, as far as reference determination is concerned, to the social links over the historical links that would require that the reference of the use of a natural kind term by a speaker should ultimately depend on what the referent of the term fixed in the corresponding initial baptism was. However, it is noteworthy that in section (d) of the addenda to Kripke (1980) the author

Kripke, Putnam, and the Description Theory

79

lessens the importance of the resort to an (in principle identifiable) initial baptism in his theory of reference fixing. Kripke concedes there that the appeal to an identifiable initial baptism involves an excessive simplification, quite clearly in the case of natural kind terms and less apparently in the case of proper names: The notion of an initial sample appealed to there [in the proposal put forward in the third lecture of (1980) concerning how the reference of the term “gold” would have been fixed in an initial baptism11 – LFM] gives an oversimplified picture of the case. Analogously for proper names, of course I recognize that there need not always be an identifiable initial baptism; so the picture is oversimplified […]. It is probably true, however, that in the case of proper names, examples with no identifiable initial baptism are rarer than in the species cases. (Kripke 1980: 162).12

In this way, Kripke himself comes to recognize that the recourse to an identifiable initial baptism regarding some proper names, and moreover concerning many natural kind terms, constitutes an excessive simplification. In this regard, it is convenient to point out that with respect to the majority of natural kind terms there is no identifiable initial baptism; thus, for example, we do not know and there is no reason to suppose that we could get to know at what time and with respect to which entities natural kind terms like “water” and “gold” were initially introduced. However, that acknowledgment implies a modification of a significant part of his theory, since Kripke just appealed to an (in principle identifi11

12

Kripke tells us that we could imagine that the reference of the term “gold” would have been fixed in the following way: “Gold [i.e., the referent of the term ‘gold’ – LFM] is the substance instantiated by the items over there, or at any rate, by almost all of them.” (1980: 135). The reason for the qualification “almost all” is that some of the entities belonging to the sample involved in the reference fixing of a natural kind term like “gold” may not be instances of the kind. If it is discovered that the number of such entities is not small, the sample being so little homogenous, there may be different reactions to it, among which Kripke mentions those of considering that the sample instantiates two sorts of kind, or of dropping the natural kind term we had supposedly introduced (see 1980: 136). In the following, I will pass over that qualification. As it is the case in this passage, Kripke occasionally uses the terms “species” and “kind” indistinctly (cf. Kripke 1980: 121, 127).

80

Luis Fernández Moreno

able) initial baptism at the introduction of proper names and natural kind terms and thus at their reference fixing. In any case, our proposal is to substitute the resort to causal chains that ultimately bind our uses of natural kind terms with the uses of these terms in an assumed initial baptism by the resort to deference links which bind our uses with the uses by other members of our linguistic community and ultimately by the experts in the kinds that those terms designate. Our claims agree with assertions in a not frequently mentioned paper by Putnam (1975c). Thus, after indicating that “we may no longer care about the original use of the term” (Putnam 1975c: 274), he asserts that the person placed at the beginning of a causal chain – he literally says “chain of transmissions or cooperations” (1975c: 275) – does not need to be the original introducer of the term, but may be the “relevant expert” (1975c: 275). Moreover, concerning a fictitious proper name “Ab-ook”, Putnam makes the following remark: In my view, the “introducer” [of the name “Ab-ook”] need not be the person who first “dubbed” Ab-ook, nor need the causal chain go through the first person the hearer learned the name “Ab-ook” from. The “causal chain” is a chain of cooperations connecting the hearer to the relevant experts, as determined by the society. (Putnam 1975c: 287, n. †)

Although this passage deals only with a proper name, the assertions contained in it would not only be applicable to many proper names – for instance, to the names of historical figures – but also to other sorts of terms, in particular, to natural kind terms. Hence, the “introducer” of a natural kind term need not be the first person who introduced the term; it can be another person, though it has to be an expert.13 13

This claim can be supplemented with the thesis of multiple grounding put forward by M. Devitt and K. Sterelny (see Devitt and Sterelny (1999): 75 f. and 89 f.), according to which the reference of a proper name or a natural kind term is not only fixed in the initial introduction of the term, but other, later, uses of the term are as relevant for the determination of its reference as its use in the initial introduction. Thus, proper names and natural kind terms are multiply grounded on their referents and not all causal chains need to have their origin at their initial introduction.

Kripke, Putnam, and the Description Theory

81

An objection usually put forward against that sort of social theory of reference is that the experts can be mistaken about the reference of natural kind terms, and that according to the causal theory, the factor which ultimately determines the reference of natural kind terms is the underlying properties of the members of the kinds belonging to our environment, instead of the human beliefs or theories on those properties. In this regard, it is advisable to mention a conflict that seems to arise within Putnam’s theory. As already said, he sustains that there are two contributions involved in the determination of the reference of natural kind terms: on the one side, the contribution of our environment or of our world; on the other, the contribution of the society, based on the existence of experts in our linguistic community. Nevertheless, Putnam tends to believe that these two sorts of contributions are at the same level. This tendency is already found in Putnam (1975b), for instance, where he asserts that The extension is, in general, determined socially – there is a division of linguistic labor […] [and] is, in part, determined […] [by] the actual nature of the particular things that serve as paradigms. (Putnam 1975b: 245; the second emphasis is mine)

However, it can be alleged that these two contributions cannot be situated at the same level, since even the experts could be mistaken concerning the underlying properties of the entities of the kind, and these properties are those that in Putnam’s theory ultimately determine the reference of natural kind terms. This is so in the case of Kripke, who does not accept the notion of expert in Putnam’s sense; for Kripke the only experts are those speakers who have introduced the terms in question in an initial baptism – see Kripke (1986). Nonetheless, in Putnam’s theory once a term comes to exhibit the division of linguistic labor there are experts of the community who introduce and reintroduce the natural kind terms. Now, if in Kripke’s causal theory the introducer of a term in an initial baptism cannot be very mistaken concerning the properties of the referent of the introduced term, since in that case it could be alleged that the term has no reference (see note 8 supra). Likewise, in Putnam’s theory, the experts, who introduced and reintroduced the natural kind terms, cannot be very mistaken concerning the properties of the members of a

82

Luis Fernández Moreno

natural kind, which are not only the so-called superficial properties, but also the underlying properties. The version of description theory of the natural kind terms’ reference that we have put forward is an extension of the description theory of reference of proper names proposed by Searle and Strawson. Furthermore, it incorporates Putnam’s thesis of the division of linguistic labor and it admits the contribution of the underlying properties of the members of a natural kind to the determination of the corresponding natural kind term’s reference. Our claim is that it is a sort of description theory of reference that constitutes a noteworthy alternative to Kripke’s causal theory of reference for natural kind terms.14 References Cauchy, Venant (ed.), 1986. Philosophy and Culture. Proceedings of the XVIIth World Congress of Philosophy. Montreal: Éditions de Beffroi. Devitt Michael and Kim Sterelny 1999. Language and Reality. An Introduction to the Philosophy of Language. Cambridge, Mass.: MIT Press, 2nd revised and enlarged edition. Frege, Gottlob 1892. Über Sinn und Bedeutung. Zeitschrift für Philosophie und philosophische Kritik 100, 25-50. (English translation: On Sense and Reference. In: G. Frege, 1952, 56-78). Frege, Gottlob 1952. Translations from the Philosophical Writings of Gottlob Frege. P. Geach and M. Black (eds.), Oxford: Blackwell. Jackson, Frank 1998. Reference and Description Revisited. In: J. E. Tomberlin (ed.), 201-218. Kripke, Saul 1980. Naming and Necessity. Oxford: Blackwell (Revised and enlarged edition, first published in Donald Davidson and Gilbert Harman (eds.), Semantics of Natural Language. Dordrecht: Reidel, 1972). Kripke, Saul 1986. A Problem in the Theory of Reference: the Linguistic Division of Labor and the Social Character of Naming. In: V. Cauchy (ed.), 241-247. Putnam, Hilary 1975a. Mind, Language and Reality (Philosophical Papers, vol. 2). Cambridge: Cambridge University Press. Putnam, Hilary 1975b. The Meaning of “Meaning”. In: H. Putnam (1975a), 215271. 14

This paper has been supported by the Spanish Ministry of Economy and Competitiveness in the framework of the research project FFI 2011-25626.

Kripke, Putnam, and the Description Theory

83

Putnam, Hilary 1975c. Language and Reality. In: H. Putnam (1975a), 272-290. Putnam, Hilary 1988. Representation and Reality. Cambridge, Mass.: MIT Press. Searle, John 1969. Speech Acts. An Essay in the Philosophy of Language. Cambridge: Cambridge University Press. Searle, John 1983. Intentionality. An Essay in the Philosophy of Mind. Cambridge: Cambridge University Press. Strawson, Paul F. 1959. Individuals. An Essay in Descriptive Metaphysics. London: Routledge. Strawson, Paul F. 1997a. Direct Singular Reference: Intended Reference and Actual Reference. In: P. F. Strawson (1997b), 92-99. Strawson, Paul F. 1997b. Entity and Identity. Oxford: Oxford University Press. Tomberlin James E. (ed.) 1998. Philosophical Perspectives. Vol. 12, Language, Mind, and Ontology. Oxford: Wiley-Blackwell.

Chris Fox

University of Essex [email protected]

The Meaning of Formal Semantics* Abstract: This paper considers some foundational issues of formal semantics. These issues include the nature of the data that is to be accounted for, the nature of the formalism used to characterise that data, and the criteria for determining whether a particular formal account is adequate. If the relevant data is merely the arguments and truth conditions that subjects agree with, then the terms in which the theory is expressed may be irrelevant. But a case can be made that it is also legitimate to be concerned with the intuitions that people have about how they reason with language, as well as technical and philosophical issues relating to the chosen formalism. The diversity of issues and objectives is highlighted by considering a selection of different problem domains, and by questioning the sometimes implicit assumptions about what counts as an appropriate formal framework. Ultimately the meaning of formal semantics depends upon assumptions about the nature of the data that is to be accounted for, and in what sense the formal theory itself needs to be sympathetic to that data. Keywords: formal semantics, formalisation, set theory, model-theoretic analysis, proof-theoretic analysis, ontology

0. Introduction Those of us engaged in formal semantics are sometimes asked to justify what it is that we do. In particular we may be questioned as to whether what we are engaged in constitutes some form of testable, rigorous scientific endeavour, or whether it is more about telling plausible sounding stories about aspects of language for which we have stories to tell. If the latter, then semanticists may be vulnerable to accusations of ignoring confounding data and issues, or failing to take seriously the need for some objective criteria by which the proposed account can be evaluated, * This paper develops some of the arguments presented by Fox and Turner (2012).

86

Chris Fox

and determined to be better or worse than some alternative account. In this context it is worth considering, at an abstract level, the nature of theories that are proposed by formal semanticists, and the kinds of problem domains in which they might be applied. A typical semantic theory will seek to characterise some certain aspects of meaning for a fragment of the language. Perhaps the most common contemporary approach is to find a way of translating linguistic examples into a set-theoretic representation, for which the rules of set theory itself then mimic the behaviour of the linguistic examples in some sense.1 We might describe this as the model-theoretic tradition, where any semantic feature of the language is to be characterised as a set. In this setting, even a relation between sets is just another set. The translation might be mediated by a logical representation, but usually the settheoretic model of this logic is taken to be the “real” semantics. We may question to what extent such characterisations constitute a semantic theory: what do they tell us that is new or informative? Of course, if we take an extensional view – a view which is consistent with the set-theoretic approach – we could say that such theories themselves characterise what it is to be a semantic theory. But rather than accept this impredicative characterisation, we could break the circle and ask some more reflective questions about this approach. For example, is it right to conflate seemingly distinct notions by collapsing everything to sets? Furthermore, if this is an empirical account, the question remains as to the nature of the data that is being captured. In particular, the question arises as to whether essential aspects of the data, or our intuitive understanding of language, are being lost or overlooked by such characterisations. Such questions can also be asked of other approaches and paradigms besides the model-theoretic tradition, where for example various notions are captured in terms of expressions in -calculus or some form of type theory. In such cases, is it appropriate for distinct notions to be conflated when formulated as -terms or as types?

1

A similar methodology can be adopted with frameworks other than set theory, such as constructive type theory (see Ranta 1994, for example).

The Meaning of Formal Semantics

87

1. Problem Domains While it may be appropriate to consider formal semantic interpretation independent of any particular problem domain, it can be useful to consider what role a formal semantic analysis might play in various contexts: different issues can come to the fore in different domains. Here we considered text-based computer games, statue law, as well as plurals and mass terms.2 The first of these exemplifies a narrow, controlled domain, the second involves the use of rigorous language, and the last exemplifies a general linguistic phenomena that is not constrained to a particular field of use. One early area for computer-based natural language “understanding” was text-based “adventure games”.3 Such games provide a very clearly defined domain, with a finite number of objects, places and actions, but with a potentially unlimited number of commands and questions that can be issued by the player. The problem is then how such commands and questions can be interpreted mechanically. In contrast to other problem domains, this provides a clear criteria for correctness, and relatively “clean” data.4 In the case of the formal interpretation of statute law, some relevant issues are whether it is possible to determine, mechanically, when the law is being complied with, just by interpreting its language. Intuitively at least, one would expect such laws to have a moderately clear criteria for correctness of interpretation. In contrast to the adventure game problem, the language used is complex, and the data to which the law is applied is quite “messy” and unclear. At the very least, to determine 2

3

4

The areas described here illustrate a range of issues, but they are not intended to be paradigm cases, or in any sense exhaustive. An early example of this genre is the “Colossal Cave Adventure” (Montfort 2003: 10). The view that the domain is limited and straightforward might in part be based on an understanding of how such games actually behave, which may limit expectations about what is appropriate and reasonable given the constraints of the time. Players of contemporary computer games might now expect more sophisticated language comprehension. Certainly, the language processing abilities of early text-based games were rather primitive.

88

Chris Fox

whether a law is being complied with, the relevant circumstances have to be described at an appropriate level of abstraction. More generally, those working in formal semantics are interested in capturing certain universal aspects of language and its meaning. An example of this is the interpretation of sentences with mass terms and plurals. In such cases, the data is often complex and uncertain. It is not always clear what entailments hold. We give three somewhat arbitrary examples to illustrate the range of fundamental conceptual problems that can arise even with what appears to be a relatively elementary linguistic notion. In the case of the sentence “Three men lifted two pianos”, we may wonder how many people, pianos and lifting events were involved. For the sentence “The water is dirty”, if we take “the water” to be defined as referring to some salient collection of water molecules, and nothing else, we may wonder in what sense such a collection of molecules can be considered “dirty”. The statement “The cards are on top of each other” can be used correctly to describe a stack of cards, even though the bottom card is not on top of any other card (and there is no card on the top card). There are many other problematic cases, but these are perhaps sufficient to indicate that there are difficulties in giving highly reductive accounts of meaning.5 The appropriate ontological structures for human reasoning can be vague. For example, while we may know that material objects composed of atoms, the meaning of language does not depend on such an understanding of the nature of the physical world. In that sense at least, meaning and the nature of our reasoning appear to be independent of scientific theories of the world. Accounts of plural and mass entities also highlight the question of whether first-order theories are adequate. One structure proposed for analysing such terms is a complete lattice (Link 1983). Yet the axiom for lattice-theoretic completeness is second-order in nature (requiring quantification over collections of entities). Such examples highlight the problem that the criteria for correctness are uncertain and the data messy, even for apparently simple, single5

These examples are discussed in more detail by Fox (2000).

The Meaning of Formal Semantics

89

sentence examples. For this reason we may be forced to limit the scope of the formal semantic analysis, and leave some aspects of meaning unanalysed. 2. Issues Some salient issues for a formal account of meaning include the nature of meaning, the nature of the data, the kinds of expressions (“speech acts”) involved, and whether we are aiming for an implementation, a formal characterisation, or an explanation. We need to reflect on the kind of meaning we are seeking to formalise, and whether it depends on the specific problem, or whether there is a general notion of linguistic meaning common to such problems. If there is a coherent notion of linguistic meaning, we may also wonder to what extent it can be characterised by a formal theory. The data is more than just the language. For the adventure game, knowledge of the game state is required. For statute law, knowledge of salient parts of a process or procedure are required, together with knowledge of how to formalise descriptive information at an appropriate level of abstraction. In the case of linguistic phenomena such as plurals and mass terms, we seem to require some knowledge of folk ontology, and a way of dealing with vague and incomplete meanings. In all cases, the data to be characterised is some kind of abstraction, of both language and the world. It is clear that there are a range of linguistic expressions other than propositional assertions. For example, the adventure game requires an analysis of imperatives and questions, and statute law, obligations and permissions. It is possible that ultimately we may decide it appropriate to have some propositional interpretation of these apparently distinct notions.6 Even so, ontologically we still need the means to reflect on whether this is an appropriate thing to do. 6

For example, Tichý (1978); Hamblin (1973); Karttunen (1977), and others, argue that questions and propositions should be the same type and that any distinction resides in our relationship to them. Similarly in the case if imperatives, some propose reducing them to properties (Portner 2005, for example).

90

Chris Fox

Semantic theories may have different intended uses. This may effect what counts as an appropriate form of analysis. For the adventure game, it is important to have an effective interpretation that supports an implementation. For statute law, we require an effective description that characterising the key aspects of the law. For linguistic puzzles, we might be more interested in something that could be characterised as an “explanation”. We may question whether these are really distinct (or even appropriate) characterisations, and to what extent they may share some common elements. The general objectives of the formal aspect of formal semantics can be construed as having at least two related aspects. First, there is the evaluation of language, and linguistic expressions, in logic. This could be achieved by using a formal theory to give truth conditions for propositions, answerhood conditions for questions, and compliance conditions for imperatives, for example. Second, formal semantics can capture structural patterns of entailment, seeking to mirror our intuitions about linguistic reasoning, presuppositions, connotations – formalising general rules about when, and in what sense, one proposition, imperative, etc., follows from another. 3. Formal Semantics Contemporary formal semantic theories assume that a rigorous formal analysis of at least some aspects of meaning is possible. Typically such theories present, or at least assume the existence of a compositional translation. Compositionality guarantees completeness of the coverage of the semantic interpretation for sentences that are generated by the formal syntax, or grammar. It is not necessarily a constraint on the nature of the interpretation itself (Zadrozny 1994) other than it not being completely arbitrary.7 Perhaps the most rigorous version of such an analysis is the “method of fragments” (Montague 1970a, 1970b, 1973; Partee 2001) in which the full details of the grammar and semantic analysis are spelt out 7

See Westerståhl (1998) for a critical appraisal of these arguments.

The Meaning of Formal Semantics

91

for a fragment of the language. In more recent work on formal semantics, such an approach is often presupposed but with the details left unstated. There is still an assumption that, in principle at least, the relevant part of language can be given a rigorous grammatical characterisation, and that we can give appropriate compositional rules of interpretation. In order to make progress on a formal analysis, it has to be assumed that there are some stable, core intuitions about meaning shared by a community of language users. The common canonical core might not include all aspects of meaning (cf. Section 1). There is also the issue of performance vs. competence (de Saussure 1916; Chomsky 1965) which pervades the analysis of language; our intuitions about what language “ought” to mean may differ from actual practice. There is then a question about whether our normative ideas about language are a genuinely reflection of the intrinsic nature of language itself – assuming that notion is coherent – and whether, and in what sense, the formalisation of such norms corresponds to the formalisation of linguistic meaning. Concerning the way a theory is expressed, at a minimum the framework in which meaning is to be formalised consists of a system of symbolic notation and deterministic rules whose use and interpretation is understood by an appropriate community. To constitute a logic, as such, the formal system ought to at least support notions of entailment and contradiction. One key question concerns what features a logic must have in order to adequately and faithfully capture appropriate aspects of meaning, and also to express a compositional analysis without resorting to additional, extra-logical machinery.8 Typically a logic will be characterised in terms of its proof theory (the patterns of entailment that it supports), and a model theory (an interpretation of the symbols within some other formal framework). The behaviours of the proof theory and model theory should be consistent with each other (at a minimum, the model theory should not contradict the proof theory). Technically, the role of a model theory is to show that 8

Including the mechanisms for compositionality in the logical framework helps ensure that all relevant aspects of the compositional analysis have a coherent interpretation in the logical framework. This usually requires the logic to support some mechanism for performing substitution, such as the -calculus.

92

Chris Fox

there is a consistent interpretation of the proof theory (where the logic is described as being sound with respect to that interpretation), and perhaps that the logic completely characterises the model (where the logic is described as being complete with respect to the model).9 We may wonder whether there is a “natural” logic. Informally at least, we may debate the criteria for what is appropriate, or adequate for capturing human reasoning. The question of whether there is a natural logic may then be refined into the question of whether there are theoryindependent criteria for judging the correctness of some a logic system (or its interpretation). Evidence from axiomatic geometry suggests that such questions are non-trivial, and fall outside the realm of logic itself.10 Given the difficulty of formulating objective criteria to identify a “natural” logic, it is conventional to fall-back on some standard systems. These include propositional logic (a logic that can be given a straightforward algebraic definition), first-order logic (a logic that is singled out by Lindström’s theorem, and which supports both compactness and the downward Löwenheim-Skolem theorem), second-order logics (the weakest logics that are categorial with respect to their models), and Higher Order Logic (a logic that can be defined in Church-typed calculus). As can be seen, these “standard” logics can be characterised as formal systems that have, in some sense, mathematically “interesting” properties. But this does not necessarily mean they are “natural”, adequate or appropriate for capturing how we reason with natural language.

9

10

There is some potentially confusing use of terminology, where a model is sometimes described as providing the semantics of the logic, a terminology that perhaps presupposes the model-theoretic view of formal semantics (Section 4). The argument can be summarised as follows. We can formalise rules for geometry. The rules can be considered to be either descriptive or normative for the intended system. They satisfy certain notions of consistency, but there is nothing within the formal theory that allows us to say that some formalisation is objectively “right”, and another “wrong”, as we can demonstrate that two different, otherwise incompatible systems of axiomatic geometry are mutually consistent: if one theory is consistent, then so is the other. In the case of axiomatic geometry this follows from the independence of the Euclidean parallel postulate (Beltrami 1868).

The Meaning of Formal Semantics

93

Those who attribute a special foundational status to possible worlds might take them to provide a criteria for characterising a “natural” logic; one that is sound and complete with respect to a set-theoretic possible worlds interpretation. On such a view, the logic may be considered to be an eliminable layer of representation. Theories can then be expressed directly in terms of a model theory. This can be seen to characterise the “model theoretic” tradition. 4. Model Theoretic Semantics: A Conventional Approach In much of the literature relating to formal semantics, there is a presumption that set-theoretic models are the proper target of formal interpretation. The received story in the Montagovian tradition is that the process of giving a formal interpretation involves finding mappings from natural language to logic, and from logic to a model. Furthermore, this should be done in such a way that composing the two mappings gives a direct translation into the model, allowing the logic to be eliminated (see Thomason 1974, for example).11 The models typically involve possible worlds (or some alternative, such as context change potential) so that the translation is then from natural language into possible worlds (or context-change potential). In turn, these theories are usually expressed in terms of set theory, allowing us to characterise model-theoretic semantics as being, broadly, the translation of natural language into set theory. We might call this the standard model-theoretic account. The model-theoretic approach can be exemplified with the analysis of propositions, questions and imperatives. Essentially, worlds can be taken to be sets of propositions, or (equivalently) propositions can be taken to be sets of worlds. Properties are then sets of world-individual pairs, relations are sets of pairs mapped to sets, and quantifiers (at least the simpler 11

There is a point of controversy here about whether this was Montague’s own view. The received view corresponds to the characterisation of Montague semantics presented by Thomason (1974). But this is not uncontested. Montague (1969) hints at the idea that the logical is important. This view is also considered by Partee (2013) and Cocchiarella (1981, 1997) – P.C. Partee, July 2013.

94

Chris Fox

cases) are set inclusion and set intersection relationships. Ultimately everything is then mapped to sets or structures over sets, which can all be reduced to a “non-representational”, pure set theory. Questions can be treated as partitions of worlds (Groenendijk and Stokhof 1997) corresponding to different answers. Answers then correspond to an identification of a subset of the partition, with appropriate adjustments to account for propositional vs. non-propositional answers.12 For imperatives, broadly speaking the model-theoretic possible worlds analysis is given in terms of preference relationships over possible worlds (see Lascarides and Asher 2004; Segerberg 1990, for example). A similar approach is adopted for possible worlds interpretations of deontic expressions (for example, von Wright 1951, 1953). The conventional approach is perhaps motivated in part by the view that set theory has a foundational status, and that any logical representation itself can only have meaning if it too is interpreted in set theory. On such model-theoretic accounts, set theory, and possible worlds, are taken to be independently “given”, perhaps with a more natural or foundational status than a logic. The justification might be that this avoids a reliance on some form of representationalism, for both logic and model theory (as criticised by Wittgenstein and others). Arguments could be made that semantic accounts that rely on preexisting, independently motivated frameworks, such as set-theoretic possible worlds, also have some explanatory power. There is a sense in which the theory can be said to “predict” the relevant semantic behaviour in that the behaviour is already contained within the theory: the role of our compositional analysis is then to show how language can be mapped systematically to sets that already have the relevant behaviour. In some cases, the translation into set-theory may be somewhat disguised by a narrative that interprets the sets in terms of possible worlds or context-change potential. Semanticists will also talk of “questions”, “partial states”, “answers“, “propositions”, “worlds” and “states” rather than just “sets”. But typically there is an implicit assumption of the par12

One alternative is to interpret questions and answers in terms of abstractions and structured propositions. Depending on how this is formalised, related ontological issues can arise, as sketched later in Section 5 (see also Fox 2013).

The Meaning of Formal Semantics

95

adigm that in their formalisation, such notions can and should ultimately be reduced to pure set theory – even in cases where a logic representation is used. One downside of this set-theoretic reduction is that the intended interpretation, and the distinctions between different conceptual notions, are not necessarily evident in the formal theory itself, but instead may be provided by a narrative, a narrative that is not strictly part of the formalisation. 5. Is the Standard Approach Adequate? The arguments that might be made for the standard model-theoretic view can be summarised as follows. 1. 2. 3. 4.

It takes a pre-existing “foundational” framework as a starting point. There is no need to “justify” the basic rules of a logic. It avoids representationalism. There is a potential to explain, not just describe.

We may question how much weight and credence should be given to such arguments, and whether the standard, set-theoretic possible worlds account is fundamentally different to any other theory or framework. In brief, is it right to credit the standard account as being a non-representational theory with explanatory power? The standard approach implicitly presumes that all relevant notions can be reduced to set-theoretic characterisations. We may wonder whether there are problems in such reductions (Fox and Turner 2012). Here it is appropriate to reprise Benacerraf’s dilemma (Benacerraf 1965). We can purport to define numbers in set theory by identifying those sets that capture the relevant properties and patterns of behaviour. But there are different, equally valid ways of doing this. On the von Neumann account, a number is the set of all its predecessors, with zero being the empty set.

96

Chris Fox

“0”  {} “1”  {{}} “2”  {{},{{}} “3”  {{},{{}},{{},{{}}}} That is, “1” is represented by {“0”}, “2” by {“0”,“1”}, and “3” by {“0”,“1”,“2”}, and so on. The notion of “less than” (