You are on page 1of 12

Journal of Pragmatics 38 (2006) 22232234 www.elsevier.

com/locate/pragma

Book reviews
After Grice: Neo- and post-perspectives Review article of Stephen Levinson, Presumptive Meanings: The Theory of Generalized Conversational Implicature. The MIT Press, Cambridge, MA (2000). xxiii + 480 pp., Paperback, ISBN: 0-262-62130-4 Review article of Robyn Carston, Thoughts and Utterances. The Pragmatics of Explicit Communication. Blackwell, Oxford (2002). x + 418 pp., Paperback, ISBN: 0-631-21488-7 1. Premise As is known, Grices philosophical conception of conversational implicature has been a source of inspiration for much work in pragmatics at the borders between philosophy of language, theoretical linguistics and cognitive science. Two main trends can be distinguished: the post-Gricean, dominated by Relevance Theory, and the neo-Gricean, represented (with differences) by Larry Horn, Stephen Levinson and Jay David Atlas. Relevance Theory and neo-Gricean approaches share a view of linguistic communication as a matter of getting the hearer to grasp the speakers communicative intention. They share also the conviction that the rules of language alone arefor many reasonsnot enough to guarantee that this task is successfully performed and that, therefore, some role has to be assigned to inferential activity. Both post-Gricean and neo-Gricean approaches are concerned with how much inferential activity is needed to explain the understanding of linguistic expressions of all kinds, how this activity is triggered and guided, and what kinds of premises the inferences rely upon. As is known, they give different replies to these questions. Relevance Theory rejects Grices cooperative principle and replaces the multiplicity of the Conversational Maxims with the sole Relevance Principle, whose most complete, two-fold formulation I recall here: (1) Human cognition tends to be geared to the maximisation of relevance. (2) Every act of ostensive communication communicates a presumption of its own optimal relevance (Sperber and Wilson, 1995:260). Relevance itself is dened as a balance of positive cognitive effects (assumptions activated in the receiver) and processing effort. Thanks to the Relevance Principle, the hearers inferential activity, triggered by the speakers utterance, aided by previously stored background assumptions and possibly guided by linguistically encoded procedural constraints, works towards a reconstruction of the speakers communicative intentions, which aims at maximising their relevance.
0378-2166/$ see front matter # 2006 Published by Elsevier B.V. doi:10.1016/j.pragma.2005.09.007

2224

Book reviews / Journal of Pragmatics 38 (2006) 22232234

Relevance Theory offers solutions to several issues left open by Grice. It gives an alternative and more consistent explanation to those implicatures that, according to Grice, depend on the outing of conversational maxims and contributes to the elimination of the puzzling category of conventional implicature by means of the notion of procedural meaning. It also extends inferential contributions to meaning to certain aspects of the communicated proposition, thus introducing the notion of explicature. The role of linguistic meaning in communication is reduced to a minimum and the role of cognition, as well as that of contextual assumptions, is boosted. Neo-Gricean approaches reject the reduction of the Gricean maxims to one Relevance Principle and maintain that multiple rules or principles are at play in linguistic communication. There is no complete agreement about such principles (cf. Atlas and Levinson, 1981; Horn, 1984; Levinson, 1987; Atlas, 2005), but all proposed rules are inspired by Grices Cooperative Principle and conversational maxims. Conversational cooperation remains the main horizon within which the analysis of linguistic communication is conducted and the norm to which communicative activity should conform. Language as a system is still considered as an interesting object of research, but its conventional side is, whenever possible, shown to rely on motivations grounded in conversational practice. Particular attention is paid to how Grice-inspired rules contribute to trans-contextual routines of meaning production and understanding and, beyond those, to grammaticalization and lexicalization processes. Levinson, in the book under review, has levelled a substantial challenge at Relevance Theory, particularly as regards its reduction of all conversational maxims to the Relevance Principle and its massive reliance on contextual information in the derivation of pragmatic meaning. Carston, in the other book under review, has restated the relevance-theoretic framework, offering a broad overview of it and several solutions to problems that have been debated for years, and accentuating its contextualist features. Here I will be concerned with these two books as recent, inuential outcomes of the neo-Gricean and post-Gricean trends of research, respectively. However, the characteristics of the two volumes do not allow for point-to-point comparison. Levinson focuses on one specic phenomenon in linguistic communication (Generalized Conversational Implicature), with applications to the resolution of anaphora. General implications of the proposed views are hinted at but not developed in detail. Carston addresses a broad theoretical problem, that is, the relationship between thoughts and utterances, and in so doing reveals the complexity of so-called explicit communication, that is, the communication of propositions, and discusses in detail a great number of specic problems, among which are conjunction, negation, and metaphor. Both books are very rich as to the number of issues they tackle: I will extract from each only a limited number of themes and then put forward some tentative, partly comparative comments. 2. From conversational implicature to preferred interpretations Levinson presents his own research on Generalized Conversational Implicature as an attempt to transpose Grices efforts to mediate between formalist and functionalist approaches to language, from philosophy to a more thoroughly linguistic ground. His target is, rather than the philosophical formalism that Grice had in mind, Chomskys linguistic formalism. Grice-inspired analyses are employed by Levinson to discover function beneath linguistic form. Specically, Levinson draws on Grice in claiming that, in order to develop an adequate theory of linguistic communication, we need to admit of an intermediate level between coded

Book reviews / Journal of Pragmatics 38 (2006) 22232234

2225

meaning and occasional speaker meaning. It is Grices philosophy of language that provides us with such a level by distinguishing utterance-type meaning from occasional speaker meaning on the one side and sentence meaning on the other (Grice, 1989:8893). Utterance-type meaning depends on the type to which an utterance belongs: it may remain the same across varied occasions of utterance, insofar as utterance type remains constant. Thus, it provides the linguist with regularities which, albeit not yet linguistic conventions, may suggest reliable generalizations. Among the various kinds of implicature described by Grice, Levinson focuses attention on Generalized Conversational Implicature, which he takes to be one kind of utterance-type meaning. A conversational implicature is generalized (as opposed to particularized) when it can be worked out on the sole basis of the assumption that the Cooperative Principle holds (Grice, 1989:3738). This means that in order to understand a Generalized Conversational Implicature (or in order to tell that such an implicature is to be expected), one need only be presented with an utterance of a certain type and assume that it conforms to the Cooperative Principle (or possibly, more specically, to one of its maxims). Generalized Conversational Implicature is also one kind of preferred or presumptive meaning (hence, the title of Levinsons book). It is inherent to Levinsons idea of utterance-type meaning that receivers take utterance tokens to mean what their utterance type means unless there is some indication that they should not take them so. Utterancetype meaning is, therefore, default meaning and like default (or non-monotonic) inferences, is cancellable. This feature further ts Grices characterisation of conversational implicature, as opposed to conventional implicature. Gricean cancellability is reinterpreted as defeasibility, and defeasibility (or non-monotonicity) becomes, for Levinson, the hallmark of the pragmatic. In his exploration of Generalized Conversational Implicature as one kind of utterance-type or presumptive meaning, Levinson raises two main criticisms at RelevanceTheory, concerning, respectively, the context dependency of utterance processing and the kind of reasoning of which utterance processing consists. Relevance-driven inferences, he remarks, always require the framework of a specic cognitive context (consisting of assumptions represented in the minds of speaker and hearer or accessible to them), so that all inference in utterance understanding is reduced to nonce inference (context-bound and ad hoc). Relevance Theory, being all about occasional speaker meaning, cannot give us access to any intermediate level between that and coded meaning. Moreover, since it draws on deduction (which is inherently monotonic), it cannot account for those specically pragmatic features of utterance processing which consist of preferred, defeasible interpretations. Levisons project, albeit neo-Gricean, departs considerably also from Grices philosophy. In Grice, the ultimate source of non-natural meaning remains the intention of the speaker: utterance-type meaning appears as a step away from utterance-occasion meaning in the direction of abstraction and is not given any special role. Also the emphasis put by Levinson on Generalized Conversational Implicature goes beyond Grice and his philosophical motivations. When an implicature is regularly associated with the use of a certain linguistic expression, Grice shows a preference for giving it (insofar as possible) a conversational account, rather than a conventional one, because the conversational account is explanatory and generalizable while the conventional account would merely be ad hoc. Generalized conversational implicature is, therefore, to be preferred, for Grice, only as an explanatory alternative to conventional implicature. But Grice does not seem equally interested in the contrast between generalized conversational implicatures and particularized ones: after all, as also Carston (110) notes, most of

2226

Book reviews / Journal of Pragmatics 38 (2006) 22232234

the examples of implicature that he discusses are particularized conversational implicatures, not generalized ones. In fact, it is as a linguist that Levinson is particularly interested in utterancetype meaning. Consideration of meaning at this level provides linguists with opportunities of generalization about how meaning is affected by those features of utterances, which may be constant across occasions: from intonation patterns to word order, from the lexicon used to syntactic structure and morphology. In concentrating on Generalized Conversational Implicature, Levinson is attempting to test how far utterance processing can go without invoking shared assumptions, encyclopedic knowledge of the domain being talked about, or mind-reading abilities. Levinson distinguishes three main kinds of Generalized Conversational Implicature, depending on three different inference rules or heuristics inspired by Gricean conversational maxims: the Q-Heuristic, What isnt said, isnt or What you do not say is not the case, related to Grices rst maxim of Quantity (Make your contribution as informative as required); the I-Heuristic, What is expressed simply is stereotypically exemplied, related to Grices second maxim of Quantity Do not make your contribution more informative than is required; the M-Heuristic, What is said in an abnormal way isnt normal, related to Grices maxim of Manner Be perspicuous and its sub-maxims Avoid obscurity of expression and Avoid prolixity. Here are some examples (where the symbol +> stands for implicates): (1) Q-implicatures: (1a) Some of the boys came +> Not all of the boys came (1b) John tried to reach the summit +> He didnt succeed (1c) Her dress was red +> not red and blue (2) I-implicatures: (2a) John turned the switch and the motor started +> John turned the switch and then the motor started, Johns turning the switch caused the motor to start (2b) a secretary +> a female one (2c) The picnic was awful. The beer was warm +> the beer was part of the picnic (3) M-implicatures: (3a) Bill caused the car to stop +> Bill stopped the car indirectly, not in the normal way, e.g., by the use of the emergency brake (3b) The corners of Sues lips turned slightly upward +> Sue didnt exactly smile (cf. Levinson:3639) As is easily seen, Q-implicatures and M-implicatures are derived through a basically Gricean inferential path: the inference to the implicature is required for the utterance to be considered as conforming to Grice-inspired maxims or principles. The implicature in (1a), for example, is activated because unless it is communicated, the utterance of Some of the boys came would give insufcient information and, thus, violate the rst maxim of Quantity. The implicature in (3a) is activated because unless it is communicated, the utterance of Bill caused the car to stop would be uselessly prolix and thus violate the maxim of Manner. The pattern proposed for I-implicatures invokes the observance of the second maxim of Quantity, according to which the speaker should not be more informative than required. This maxim is observed by omitting stereotypical information to be easily retrieved by the hearer, which ensues in communicating stereotypical information by means of an alleged Generalized Conversational Implicature. But attention should be paid to asymmetries between Q- and M-implicatures on the one side, and

Book reviews / Journal of Pragmatics 38 (2006) 22232234

2227

I-implicatures on the other. While the Gricean inspiration of the Q- and M-implicature pattern is pretty clear, I-implicatures, in spite of supercial similarities, are signicantly different. Utterances triggering I-implicatures, such as those in (2) above, are easily seen to trivially satisfy the second maxim of Quantity: being general or even elliptical, how could they provide too much information? The I-implicature does not seem to play any role in assuring the conformity of the utterance to the maxim. This makes I-implicatures quite un-Gricean. To a certain extent, Levinson is himself aware of this asymmetry. Although he does not put it in such terms, he does admit that I-inferences are different from Q- and M-inferences. They are, he says, much less metalinguistic in character and much more dependent on common presumptions about the world. The fact that, this notwithstanding, he admits of I-inferences as Generalized Conversational Implicatures may be taken to show that he is less interested in the inferential path through which implicatures are derived than in their function within grammaticalization and lexicalization processes. The last part of Levinsons book is devoted to anaphora. Here, what is at stake in his defence of utterance-type meaning becomes most clear. He attempts to bridge the gap between intra-sentential anaphora (anaphors in the terminology of generative grammar) and discourse anaphora by showing how anaphoric interpretation is constrained by pragmatic principles (the same principles on which Generalized Conversational Implicature depends). He examines two of Chomskys Binding Conditions (1981, 1995): (a) If Alpha is an Anaphor, interpret it as coreferential with a c-commanding phrase in the relevant local domain D; (b) If Alpha is a pronominal, interpret it as disjoint from every c-commanding phrase in D; and he argues that at least one of them can be pragmatically derived according to principles of preferred interpretation. According to Levinson, there are languages in which condition (a) is grammatically specied, while condition (b) is pragmatically derived from it and languages in which condition (b) is established rst, on the basis of a stereotypical presumption of clausemate co-argument disjointness (itself depending on the I-principle), while condition (a) is derived from it in its own turn (344). It is suggested that the languages in which condition (a) is grammaticalized represent a later stage of diachronic development with respect to those in which condition (a) is pragmatically derived from condition (b) (346348). Levinsons discussion is rich in examples and outlines of pragmatic derivations (sometimes controversial, but always worth attentive consideration). Presumptive meaning is here at its best, while, perhaps, the farthest away from Gricean inferential paths. 3. From implicature to the inferential approach to explicit communication Carstons book is highly representative of the relevance-theoretic approach, which has moved from the revision of the Gricean notion of implicature towards a general theory of the role of inference in linguistic communication. Carstons research focuses on the role of inferences in determining what an utterance says, and therefore, what is explicitly (as opposed to implicitly) communicated. While it is obvious that linguistic form underdetermines what an utterance implicates, which is then derived inferentially, it is not obvious whether and to what extent the understanding of what is said by an utterance requires pragmatic inferential processing. In consideration of the several sources of underdeterminacy that affect what is saidambiguity, indexicals, propositional forms requiring completion, context sensitivity of predicates, scope of operators,

2228

Book reviews / Journal of Pragmatics 38 (2006) 22232234

gurative uses of languageCarston claims that any project dening what is said in terms of linguistic meaning fails to account for representation levels that are active in utterance processing and should therefore be rejected. She also argues that underdeterminacy is essential to natural language: no sentence ever fully encodes the thought or proposition it is used to express, nor is it ever possible to replace a sentence with another fully encoding the same meaning. What is said, i.e., what is explicitly communicated, is always at least in part inferentially derived. There is a distinction between what is said and what is implicated by an utterance, nevertheless. Carston claims that there is a derivational difference: explicatures (belonging to what is said) are derived by pragmatically lling in and adjusting the semantic scaffolding provided by the linguistic expression used (either by saturation of variables or indexicals in the logical form or by free enrichment), while implicatures are derived wholly pragmatically, at most under linguistically encoded procedural constraints. In order to draw her distinction between what is said and what is implicated, she does not rely directly on the derivational difference, which cannot be exhibited directly, but on two tests: the embedding test (a communicated assumption belongs to what is said by an utterance as opposed to what is implicated if it falls within the scope of logical operators when these are applied to the uttered sentence) and the functional independence principle (an assumption inferentially communicated by an utterance is an implicature of it as opposed to an explicature only if, in the mental life of the hearer, that assumption and the proposition expressed by the utterance play independent roles) (189 ff.). These tests enable her to class as explicatures several cases of inferentially communicated assumptions, which neo-Griceans typically consider as Generalized Conversational Implicatures. Consider such inferences as causal readings of conjunctions (e.g., when The vase fell and broke is taken to mean The vase fell and therefore broke): on the one hand, the and-conjunction preserves its causal reading when embedded in a conditional or a disjunction; on the other hand, its truth-functional reading (according to which The vase fell and broke is true if and only if both The vase fell and The vase broke are true, independently of the order in which the two reported events occurred) clearly plays no independent role in the mental life of a speaker or hearer. Thus, according to Carston, causal readings of andconjunctions are explicatures. Carston examines, at length, the cases of and-conjunction and of negation. Both andconjunctions and negative utterances often convey meaning other than their bare truth conditions: and-conjunctions convey various kinds of relations between the conjuncts and negative utterances often (albeit not always) convey the presuppositions of the corresponding afrmative sentences. In both cases, there have been attempts at semantic analyses proposing some kind of ambiguity or polysemy associated with the operator involved or with the structure of the sentences to which it applies. Also pragmatic, inferential analyses have been proposed, with the intention of keeping the semantics of the operator involved as lean as possible. Carston sides with the proponents of pragmatic analyses in criticizing semantically oriented approaches, but replaces their positive proposals with her own. For and-conjunction, she is inclined to assign to and at most a procedural meaning (but suggests also that its encoded meaning may be limited to its function as a syntactic operator). Following a suggestion by Blakemore (1987), she maintains that the and-operator makes a new unit out of the two utterances it conjoins, so that what carries the presumption of optimal relevance is the conjoined proposition. From this, there arises the need of enriching the encoded meaning of the and-conjunction in a direction yielding an optimally relevant reading not of the individual conjuncts, but of the conjunction itself. Chronological or causal relations between the conjuncts are good candidates for such enrichments, while explanatory relations or elaboration relations are not suitable, because they

Book reviews / Journal of Pragmatics 38 (2006) 22232234

2229

require that the two conjuncts are considered as distinct processing units. For negative utterances, after detailed confrontation with previous pragmatic analyses of negation (such as Grice, 1981; Atlas, 1989; Horn, 1989) and with the analysis of presupposition projection due to Burton Roberts (1989), Carston proposes a two-step enrichment process starting from truth functional, wide-scope and presupposition-cancelling negation, which yields rst a presuppositionpreserving, narrowed meaning, but may proceed beyond it towards a presupposition-cancelling, echoic (or metarepresentational) reading. Like the relations between conjuncts communicated by and-conjunctions, also the narrowed (presupposition-preserving) and the metarepresentational (presupposition-cancelling) meaning of negative utterances are, according to Carston, not implicatures of any kind, but explicatures: pragmatic, inferentially derived aspects of explicit communication. What is the rationale of classifying certain communicated assumptions as explicatures rather than implicatures? They are implicit from the point of view of what the linguistic surface actually displays as well as from the point of view of the amount of reformulation work needed in case a hearer wants to focus attention on them and spell them out in words. Kent Bach (1994) stressed this by proposing, for at least some of the inferences that Relevance Theory calls explicatures, the label implicitures, which distinguishes them from implicatures but does not conceal they are communicated implicitly. But, in Carstons theory, a hidden premise seems to be at work according to which, since what is communicated explicitly is to be identied with what is said, and what is said by an utterance is to be identied with its truth conditions, explicitness belongs to all that is assessed when we come to a truth/falsity assessment. It seems to me that the problem here is not so much whether implicatures may contribute to truth-evaluable content (as Levinson would have it), but what exactly is focused upon as subject to evaluation in terms of truth and falsity and why. There would be nothing odd in considering certain implicatures of an utterance when discussing its truth or falsity, if the truth/falsity assessment were (as in Austin, 1962:142 145) a matter of global correctness of the speech act with respect to the relevant portion of the world. And it might not be impossible to explain why certain alleged implicatures contribute to determining the truth/falsity of complex sentences, in terms of the speech acts to be assessed. But if truth and falsity are the semantic values of thoughts, i.e., that which it is the whole point of a thought to determine, then of course implicatures cannot have any bearing on them. Carston seems to make the Fregean choice here, a choice for the eternity of thought and its denition as an essentially true/false kind of object. It is a respectable philosophical choice, but it might have problematic consequences for cognitive science in case also thoughts (as mentalese sentences) turn out to underdetermine their truth conditions (cf. 74 ff.). Propositional thoughts would become again abstract objects, quite distinct from any psychological entity or process and cognitive science would no longer be about them. However, coming back to the pragmatic processing of utterances as Carston conceives it, there is more to explicatures than the inferences associated with and-conjunctions and the pragmatic narrowing of the negation operator scope. Carston classes as explicatures and, in particular, as free enrichments also many inferences connected with encoded lexical meaning. Here again, she disagrees with Levinson in describing the narrowing of lexical meaning as explicature as opposed to Generalized Conversational Implicature of the I-kind. But she also puts forward an innovation with respect to standard views in Relevance Theory (Sperber and Wilson, 1995), proposing to consider also the broadening of encoded lexical meaning as a process of the same kind as free enrichment and, therefore, as explicature. Free enrichment is usually conceived as a narrowing: a selection among possibilities provided by the linguistically encoded semantics or a specication of linguistic meaning by addition of

2230

Book reviews / Journal of Pragmatics 38 (2006) 22232234

details. The pragmatic broadening of the encoded linguistic meaning (also called loose use) is exactly the opposite: it does not add details but subtracts or suspends them (or possibly replaces some of them with others). Loose use is not considered by standard Relevance Theory as contributing to explicature, but as generating implicatures: when in an utterance words are used loosely (for example, metaphorically), the proposition expressed, although developed as usual from the encoded linguistic meaning, fails to be communicated and serves only as a vehicle for the implicatures. Against this treatment of loose use and particularly of metaphor, Carston convincingly argues that narrowing and broadening are two symmetrical processes of pragmatic development of the linguistically encoded logical forms, both yielding communicated explicatures. Thus, metaphorical meaning, traditionally seen as the prototype of non-literal meaning, belongs for Carston to what is said and is only one of the many, varied ways in which what is said fails to be directly derivable from the decoding of a linguistic form. This claim of Carston might appear as an adjustment and improvement of Relevance Theory, but has consequences that go deep into the nature of those alleged building blocks of meaning and thought that are called concepts. Concepts (in the standard Fodorian, atomistic view; cf. Fodor, 1998) are nodes in memory, associated with various kinds of information; on the one side, they act as building blocks of mentalese sentences and therefore of thought, while on the other side, they provide linguistic expressions with their encoded meaning (insofar as the latter is dened translationally, that is, as the translation of linguistic expressions into mentalese). Now, when a concept is narrowed by free enrichment, the denotation of the concept occurring in thought falls within that of the linguistically encoded one; but if also broadening contributes to explicature, the denotation of the concept occurring in thought turns out to include instances not comprised in the denotation of the linguistically encoded concept. It may even happen, as Carston observes in her discussion of metaphor, that the pragmatically retrieved ad hoc concept featuring in the thought conveyed by a metaphorical utterance has a denotation totally disjoint from the denotation of the linguistically encoded concept. This is highly problematic: if metaphorical meaning is generated by free enrichment, Carston argues, encoded lexical meaning can no longer be conceived as consisting of full-edged concepts (360 ff.). Consistently, her preferred picture of word meaning is cast in terms of concept schemas or pointers to a conceptual space: on the basis of these schemas or pointers, actual concepts (components of thoughts) are pragmatically inferred on actual occasions of use. But if so, concepts as symbols of mentalese and building blocks of actual thoughts no longer seem to provide the basis for the semantics of natural languages. Carston turns at this point towards an account of word meaning quite surprisingly recalling the old empiricist story of the construction of ideas from impressions: she seems to admit of some process of abstraction or extraction from the ad hoc concepts associated to a phonological form on its different occasions of use, to the more general meaning, which then functions as a gateway both to the existing concepts and to the materials needed to make new ones on further occasions of use of the same phonological form (364). I am not convinced that this is a good solution; and moreover, I doubt its compatibility with the Fodorian semantics for mentalese. Thus, an adjustment apparently aimed to improve the symmetry of a theory ends up threatening its presuppositions and inviting further restructuration. 4. Meaning, thought and rationality Research on utterance understanding (or, in the cognitive science jargon, utterance processing) has important connections with conceptions of meaning, thought, and human rationality. The connection with meaning is quite intuitive: when we understand an utterance, we

Book reviews / Journal of Pragmatics 38 (2006) 22232234

2231

grasp its meaning, so that a theory of utterance understanding should be expected to shed light on the nature of meaning. The connection with thought is one step forward: do our linguistic utterances express thoughts? Can their linguistic meaning be called a thought? Or is there a gap between language and thought, and if so, what kind of a gap? Finally, the connection with human rationality comes in insofar as the accounts of utterance understanding we are concerned with involve inferences, that is, at least on the face of it, some form of reasoning. So what do Levinson and Carston, in the books under review, both (with differences) concerned with utterance understanding, suggest us about meaning, thought, and human rationality? As to meaning, we may gather hints from either book. Consider Carstons ad hoc concepts: this is what speakers really mean, what they have in their minds. So, meaning appears to be quite a concrete matter. But that is not linguistic meaning; what the uttered words mean is never what the speaker really means by uttering them. Linguistic meaning is poor, sketchy, abstract and if language has the great expressive powers it does have, most of the merit is not of language itself, but of the inferential abilities that make effable (in the sense of communicability) what is not effable (in the sense of encodability). Levinsons picture of meaning, insofar as I understand it, is remarkably less concrete and more linguistic. The very existence of presumptive or preferred meanings suggests that the view of meaning as something concretely existing in the mind of a speaker does not hold, or at least, is not the whole story. As all entities or events subject to defeasibility or cancellability, presumptive meanings cannot be straightforwardly material; there must be a stipulative or normative element to them, intersubjective rather than merely individual, which lets them vanish when non-default conditions obtain. Besides, there seems to be a viable transition between pragmatic meaning and linguistically encoded meaning, since Generalized Conversational Implicatures allow for, and contribute to explaining, grammaticalization and lexicalization strategies. Even the need for pragmatic inferences is traced back to facts about language: it is the nature of the linguistic signier which allows for the issuing of one morpheme at a time, generating what Levinson calls the encoding bottleneck (67), and makes additional pragmatic meaning indispensable for fast and rich communication. While Carston is inspired by the idea (which she attributes to Grice, see 205) that communication should be possible even without the use of a code, in Levinsons view language as a code is both presupposed to exist and argued to be diachronically constructed. As to thought and the language/thought relationship, it is not clear to me if there are any morals to be drawn from Levinsons book. His formulation of the bottleneck problem seems to presuppose that a speaker may want to communicate something before, and independently of, his or her choice to utter certain words drawn from the linguistic system at his or her disposal. Is this the locus of thought in the framework of his view? That is difcult to say. The comparison of utterance understanding with visual communication (the understanding of a drawing of Rembrandts, see 24) also suggests that Levinson may want to recognize non-verbal and even non-conceptual kinds or levels of thought besides verbal, conceptual ones. It seems, however, that for him thoughts take a denite shape when cast in the form of sentences, whether uttered or implicated.1 Carston is by far more diffuse and explicit about the language/thought relationship, which falls within the main topic of her book. For her, as we have already seen, thoughts are in general not encodable in a natural language form. The resulting gap between thought and language has two main aspects. On the one side, thoughts are extremely ne-grained if not

It should not be forgotten that Levinson has been also concerned with problems of linguistic relativity (cf. Gumperz and Levinson, eds., 1996; Levinson, 2003).

2232

Book reviews / Journal of Pragmatics 38 (2006) 22232234

idiosyncratic (they contain, for example, ad hoc concepts), while linguistic meanings appear abstract and poor. On the other side, logical forms (which result from the decoding of linguistic forms) are incomplete and fail to determine truth values, while thoughts are propositional and possess truth conditions. The gap between thoughts and linguistic forms is bridged by means of pragmatic inferences, but this does not cancel all of its consequences. In particular, as Carston admits, the semantics of the apparent counterparts in natural language of logical operators need no longer be identical to the semantics of those operators. In fact, if natural language sentences do not encode propositions, and fail, therefore, to be straightforwardly truth conditional, natural language connectives need not encode truth functional properties. So natural language and logical or ideal language come apart again, in spite of Grices project to defend a unied semantics for both, like in the old times of ordinary language philosophy (let alone logical empiricism and behind it, Frege, Russell, and Wittgenstein). Carstons view of the relationship between language and thought has, in fact, much of a Wittgensteinian avour, and also (if I may) some Wittgenstein-like ambiguity. Carstons trust in the richness of the inferential resources of human minds resembles Wittgensteins trust, expressed in his Tractatus, in our languages being in order as is (1922, see section 5.5563). In both cases, though, language wraps up thought so that no correspondence can any longer be seen between elements of a linguistic utterance and elements of the underlying thought, thanks to which alone the linguistic utterance can mean something. The relevance-theoretic apparatus, as elaborated upon by Carston, makes a technically convincing attempt to bridge the gap between linguistic form and communicated thought, but the process should not be mistaken for a construction of thoughts (with their ad hoc concepts) from utterances. After all, it is thoughts that are foundational, and their ingredients, innumerable ad hoc concepts, guarantee (like Tractarian elementary objects) that the limits of our thought coincide with the limits of our world. But (perhaps misleadingly, and most likely unintentionally) Carstons theory may also remind the reader of the later Wittgensteins critique of the concept Wittgenstein (1953, see sections 71 and 99, and passim). She admits that natural languages are not directly truth conditional, or even, perhaps, not truth conditional at all. If linguistic meanings are not full-edged concepts but pointers to a conceptual space, they could be conceived not as referential labels for things or kinds, but as instruments for establishing temporary, contextually changing resemblance relations. Carston herself does not follow the later-Wittgensteinian path, insofar as she sticks to the propositional format for communicated thought. Last but not least, Levinson and Carston are both concerned with human rationality. Both their accounts of utterance processing involve inferences; both stress the rationality of utterance production and understanding. Levinson sees implicature as an economic and efcient (hence, rational) solution to the bottleneck problem; Carston, in using the Relevance Principle, relies on rational criteria of efciency, economy of effort and so on for the explanation of pragmatic, i.e., non-encoded, meaning. Legitimate doubts can be raised about whether the inferences either author describes as the source of the kinds of pragmatic meaning at issue should indeed be considered as actual reasoning of speakers and hearers. If so, they should not only constitute the actual processing path of the examined utterances, but also be accessible to the consciousness of the individuals involved. It is not clear to me whether Levinson holds that utterance-type meaning is actually always inferentially derived. I nd his initial, provocative comparison between utterance understanding and the understanding of Rembrandts sketch (24) rather puzzling under this point of view. In what does the analogy consist? The understanding of the drawing relies, as a rst step at least, on good-form laws, which (according to the Gestalt psychologists at least) are not inferential. Does

Book reviews / Journal of Pragmatics 38 (2006) 22232234

2233

Levinson, this notwithstanding, consider visual perception as inferential or is he willing to see utterance-type meaning as driven by some kind of good-form law? I have no direct reply to this question. Perhaps, beyond the language used in formulating his rules, Levinson does not intend to describe the competence of actual language users but merely, so to say, translate it into formulas suitable for being handled by the linguist. Such an attitude would also be consistent with his assimilation of rules for Generalized Conversational Implicature to heuristics (30 ff.). By this term it is usual to refer to routines that our cognitive system adopts because they spare time and effort but, nevertheless, yield correct results most of the time. Although incorporating some kind of reasoning and satisfying criteria of meansends rationality, a heuristic is not possessed by the individual as knowledge, let alone conscious knowledge, but rather as a habit. So, its activation does not constitute an instance of reasoning. In contrast, deductive, monotonic inferences play a major role in Carstons picture of utterance processing (as in general in Relevance Theory). About such inferences too, it makes sense to wonder whether there is a claim that they are actually implemented by speakers and hearers in the course of utterance production and understanding. On the one side, Carstons relevance-theoretic framework is concerned with the actual processing of utterances and with principles, such as the Relevance Principle, which are claimed to be actually active in human minds. But on the other side, it is clear that people do not reason consciously from the linguistic input to explicatures and implicatures. So the claim to be attributed to Carston is, more plausibly, a weaker one: that unconscious utterance processing follows steps parallel to the series of explicit, linguistic inferential steps outlined by the theory. If thought is conceived as mentalese language, it is even possible to contend that actual inferential processes involving mentalese sentences take place in the minds of speakers and hearers hardly ever emerging to consciousness. These processes may deserve to be called inferences, but I am doubtful about whether they can count as instances of reasoning.2 So, about human rationality, the two books we have considered give two diverging suggestions, while sharing one deeper feature. According to Carston, the internal workings of the human mind which are employed in utterance processing follow a deductive and monotonic model, while according to Levinson, they follow at least in part a non-monotonic, heuristicsguided one. But in neither case are people represented as properly reasoning; the rationality of the inferential route they follow is, therefore, more a matter of efciency of causal connections and economy of processing effort than a matter of acting or assuming attitudes upon reason. This similarity marks a difference between the two books we have examined as representative of neoor post-Gricean trends of research, and the original Gricean project: Grice, as a philosopher, was interested in the rationality of utterance processing in the sense of its being grounded in reasons. References
Atlas, Jay D., Levinson, Stephen C., 1981. It-clefts, informativeness, and logical form: Radical pragmatics (revised standard version). In: Cole, P. (Ed.), Radical Pragmatics. Academic Press, New York, pp. 161. Atlas, Jay D., 1989. Philosophy Without Ambiguity. Oxford University Press, Oxford. Atlas, Jay D., 2005. Logic, Meaning, and Conversation. Oxford University Press, Oxford. Austin, John L., 1962. How to Do Things with Words, second revised ed., 1975. Oxford University Press, Oxford. Bach, Kent, 1994. Conversational impliciture. Mind and Language 9, 124162.

Carston herself has discussed this problem in her lecture Pragmatic inferences: reective or reexive? at the 9th International Pragmatics Conference, Riva del Garda, 1015 July 2005.

2234

Book reviews / Journal of Pragmatics 38 (2006) 22232234

Blakemore, Diane, 1987. Semantic Constraints on Relevance. Blackwell, Oxford. Burton Roberts, Noel, 1989. The Limits to Debate. Cambridge University Press, Cambridge. Chomsky, Noam, 1981. Lectures on Government and Binding. The Pisa Lectures. Foris, Dordrecht. Chomsky, Noam, 1995. The Minimalist Program. MIT Press, Cambridge (MA). Fodor, Jerry, 1998. Concepts. Where Cognitive Science Went Wrong. Oxford University Press, Oxford. Grice, Paul, 1981. Presupposition and conversational implicature. In: Cole, P. (Ed.), Syntax and Semantics 9: Pragmatics. Academic Press, New York, pp. 183198 (reproduced in Grice 1989). Grice, Paul, 1989. Studies in the Way of Words. Harvard University Press, Cambridge (MA). Gumperz, John J., Levinson, Stephen C. (Eds.), 1996. Rethinking Linguistic Relativity. Cambridge University Press, Cambridge. Horn, Larry, 1984. Toward a new taxonomy for pragmatic inference: Q- and R-based implicature. In: Schiffrin, D. (Ed.), Meaning, Form, and Use in Context. Georgetown University Press, Washington, pp. 1142. Horn, Larry R., 1989. A Natural History of Negation. University of Chicago Press, Chicago. Levinson, Stephen C., 1987. Minimization and conversational inference. In: Verschueren, J., Bertuccelli Papi, M. (Eds.), The Pragmatic Perspective. John Benjamins, Amsterdam, pp. 61129. Levinson, Stephen C., 2003. Space in Language and Cognition: Explorations in Cognitive Diversity. Cambridge University Press, Cambridge. Sperber, Dan, Wilson, Deirdre, 1995. Relevance, second ed. Blackwell, Oxford. Wittgenstein, Ludwig, 1922. Tractatus Logico-Philosophicus. Routledge & Kegan Paul, London. Wittgenstein, Ludwig, 1953. Philosophische Untersuchungen. Blackwell, Oxford. ` has studied philosophy of language and semiotics, with particular attention to ordinary language Marina Sbisa philosophy and pragmatic issues such as speech acts, presupposition and implicature. Her research interests include the use of pragmatics in discourse analysis, and contextualism. Among her recent publications in English there are: Intentions from the other side (in: G. Cosenza, ed., Paul Grices Heritage, Turnhout: Brepols, 2001, 185206) and Speech acts in context (Language and Communication, 22, 2002, 421436). She teaches Philosophy of Language at the University of Trieste.

` Marina Sbisa Department of Philosophy, University of Trieste, Andr. Campo Marzio 10, 34123 Trieste, Italy E-mail address: sbisama@units.it 12 September 2005