This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
of the set P of statements ordered by the logical consequence relation , viz. the join ¬φ ∨ –φ of the statements. For φ and Ψ, the “weaker” common part entailed both by φ and by Ψ is obviously φ ∨ Ψ, since φ (φ ∨ Ψ), and Ψ (φ ∨ Ψ). The “weakest” statement entailed by φ is a logical truth, e.g. (χ ∨ ¬χ) . Thus the least weak statement whose information is common to both ¬φ and –φ is ¬φ ∨ –φ. Unfortunately this is not an accurate model of the absence of semantical specification. It is extensional rather than intensional; although the English expression ‘Tom is my neighbor’ is neutral in sense between [TOM IS MY MALE NEIGHBOR], [TOM IS MY FEMALE NEIGHBOR], [TOM IS MY HERMAPHRODITIC NEIGHBOR], and so on, it is not synonymous with the logical disjunction in logical English Tom is my male neighbor ∨ Tom is my female neighbor ∨ Tom is my hermaphroditic neighbor ∨ Tom is my non-human neighbor ∨ . . . Tom is 19What
Noam Chomsky (1996c: 38) now calls “symbolic representations” of an “Internalist Semantics.”
38
LOGIC , MEANING , AND CONVERSATION
my ADJ neighbor (Atlas 1978b, 1984b, 1989). It is even more clear that the negative English sentence-frame containing ‘not’—A(. . . not . . .)—will not be synonymous with the logical disjunction in logical English ¬A ∨ –A, which, as noted, is logically equivalent to ¬A. A logical disjunction, though logically weak, is semantically specific. In Atlas (1979) I had argued that an account in which negative sentences The F is not G were scope-ambiguous was redundant. If pragmatic inferences were a necessary part of a theory of utterance-interpretation, and the sentences were ambiguous, the inferences would be essential to disambiguation, to the selection of the appropriate sense in the context of utterance. On the assumption that the addressee analyzes the speaker’s sentence-token and discovers two or more senses, a choice of the appropriate sense must be made in light of collateral information available in the context. Thus some inferential mechanism must produce a decision on the best “fit” between each sense and the context of utterance. By contrast, if the sentence is unambiguous but semantically nonspecific for semantic features F1, F2, . . . Fn, the inferential mechanism must give in the context an appropriate, more specific or precise, interpretation of the semantically nonspecific sentence. I hypothesized that since the same inferential mechanisms were at work in both the cases of ambiguity and of sense nonspecificity—one to select a reading, the other to construct a more specific or precise interpretation—the difference between selection and construction was that the construction of a specific interpretation operated on a pre-“propositional” semantic representation that was too underdeterminate, not merely too underdetermined, to carry a truth-value (Atlas 1979; Bach 1982). Contrary to Grice’s view that pragmatic mechanisms operated only post-“propositionally” to give what the speaker implied by asserting a sentence, what statement the speaker asserted was determined by pragmatic inference in addition to the semantic interpretation of context-dependent indexicals, demonstratives, reference-fixing of singular terms, tense, and others. I (Atlas 1979) also observed that classical Gricean theory could not give an adequate explanation of the inferential mechanism: sentence-negations Not (The F is G) were transformed into predicate negations The F is non-G, and the references of singular terms were fixed, but no coherent account was given of the case in which Grice’s sentence-negation meaning (the exclusion negation)—sometimes expressed in English by It is not the case that P, though, as I pointed out in Atlas (1974), It is not the case that P can also express the predicate/choice-negation interpretation; see Horn (1989)— passed through the pragmatic mechanism unchanged, the case in which informative enrichment was absent. According to the classical Gricean theorist, this case should have been the semantically “unmarked” case, but no Gricean principle in the theory could noncircularly explain it to be the unmarked case, the case of saying what one means. In fact, it is linguistically the “marked” case, as Sir Peter Strawson (1950) noted. By contrast, my claim (Atlas 1974, 1975a,b, 1977b, 1978b, 1979), that the negative sentence-meaning [THE F IS NOT G] is semantically nonspecific between the classical scope-interpretations meant that the inferential mechanism operated equally in both cases: it produced the more specific choice-negation interpretation [– (THE F IS G)], and it produced the more specific exclusion-negation interpretation [¬ (THE F IS G)]. In neither case has the speaker “said” something “propositional,” if by ‘saying’ one means merely the act of uttering a meaningful sentence of a semantically nonspecific sort. Or,
SEMANTICAL UNDERDETERMINACY
39
in either case the speaker has “said” something interpretable, if by ‘saying’ one means performing a locutionary act of producing a token of an utterance-type that is interpreted by one’s addressee to have a determinate sense and reference.20 Or, in one case or the other the speaker has “said” what he meant, if by ‘saying’ one means performing an illocutionary act the content of which is correctly interpreted by the speaker’s addressee to have the speaker’s intended sense and reference. Having previously attempted in Atlas (1989) to put in canonical form my explanation of the nature of semantically nonspecific sentence-meaning by developing and applying to sentences the lexical notion of the semantical nonspecificity of subsentential expressions, in this book I attempt to formulate, but not “reduce” or “metaphysically explain” (whatever that means), principles underlying an addressee’s interpretation of a speaker’s utterance. As in Atlas (1979), I radically revise and extend Grice’s (1961, 1967, 1975a,b, 1989a) account of generalized conversational implicata to an account of generalized conversational inferenda. My subject is not just generalized conversational implicature; it is generalized interpretative inference. My two semantic and pragmatic inquiries together provide philosophers and linguists with the framework they need to reconceptualize traditional philosophical problems in the spirit of Grice’s Linguistic Turn (see chapter 2, section 1) and in the spririt of Chomsky’s (1996c, 2000) Internalist Semantics.21 My book Philosophy without Ambiguity (1989) examined negation, presupposition, definite descriptions, negative existence statements, and natural kind terms, the construction of the correct theory for which requires the distinction between ambiguity and semantical nonspecificity. Atlas (1977a,b, 1978a,b, 1979) and chapter 4 of Atlas (1989) showed why the concept of semantical nonspecificity was also essential for a correct account of speaker’s-utterance interpretation. My earlier book having delimited the semantics-pragmatics boundary from the semantics side of Chomsky’s (1996c, 2000) Internalist Semantics, this book attempts to construct a post-Gricean theory of generalized interpretative inference of the sort first adumbrated in Atlas (1978a,b, 1979) and Atlas and Levinson (1981), in order to delimit the semantics-pragmatics boundary from the pragmatics side of Chomsky’s (1996c, 2000) Performance Systems. Together the two books constitute a systematic study of the semantics-pragmatics interface.
20See Ziff (1972a). Similar notions of interpreting utterances surfaced in others’ theories as well, notably the “explicatures” of the Relevance Theory of Sperber and Wilson (1986b); the “implicitures” of Bach (1994a,b); the Pragmatic Intrusion of Katz (1972), Walker (1975), and Levinson (1988, 2000) for the interpretation of singular terms; and the two stages of pragmatic inference in Récanati (1989), who in Récanati (1993: 236) acknowledges the earlier work of Katz (1972) and Walker (1975) on reference-determination of definite descriptions and on disambiguation and of Atlas (1979) on interpretative specification of nonspecific literal meaning (see Horn 1992a: 265). 21 Not, it may have disappointed Grice to realize, in exactly the letter of Grice. In June 1980 I asked him whether, in light of my own and others’ suggested revisions of his Maxims of Conversation—e.g., Atlas and Levinson (1981) and Sperber and Wilson (1986b)—he thought his maxims as originally formulated in 1967 were both accurate and complete. He replied succinctly, “Yes.” He was being slightly disingenuous, as he himself, in Grice (1981, originally a lecture of 1970), proposed a new Maxim of Manner in order to generate the existential “presuppositions” of definite descriptions from negative definite description sentences. See chapter 4 in this volume.
LOGIC , MEANING , AND CONVERSATION
40
As I have suggested elsewhere (Atlas 1975, 1977b, 1978a,b, 1979, 1981, 1989), for theoretical purposes I distinguish among sentence-types or -tokens, utterancetypes or -tokens, literal meanings of sentence-types or -tokens, speaker’s interpretations of utterance-types or -tokens, and addressee’s interpretations of a speaker’s utterance-types or -tokens. My Chomksyan Internalist Semantics claims that a sentence-type or -token, which is a syntactical entity—a well-formed syntactical combination of lexical items—has a literal meaning. I represent that literal meaning by a semantic representation. An assertion (statement) or utterance of a sentence-token will have its literal sentence-type meaning but also a semantic interpretation in the context of utterance. A semantic interpretation of the utterance in a context will be determined when it fixes the references of singular terms, indexicals, demonstratives, and tense and will be determinate when it makes “precise” or makes “specific” the interpretation of semantically nonspecific general terms in the utterance. The semantic representation of a sentence will be semantically underdeterminate, by virtue of its semantical nonspecificity, so that it might not “express a proposition” or carry a truth-value (depending on the relevance of the specific information to the context of evaluation), as well as semantically underdetermined, by virtue of its lacking values for its referential variables, so that it would not “express a proposition” or carry a truth-value (depending on the relevance of determining the values of the referential variables to the context of evaluation). The semantic interpretation of a statement or assertion will be sufficiently determined and, in addition, determinate so that it does identify the proposition expressed by the utterance and can be assigned a truthvalue. Truth-conditional semantics is for the semantic interpretations of utterance-types or -tokens, not for the semantic representations of sentence-types or -tokens. Classical Gricean pragmatics takes a semantic interpretation of an utterance and its context and generates a pragmatic interpretation of the utterance in the context, something intuitively described as “the addressee’s understanding of the speaker’s intended meaning,” by combining the semantic interpretation of the assertion with the asserter’s conversational implicata: something the addressee understands the speaker to convey, suggest, or imply by, in, or when asserting the sentence-token in the context. The Atlas (1979) version of post-Gricean pragmatics allows pragmatic inference to map (a) semantic representations of sentences into semantic interpretations of utterances of the sentences and (b) semantic interpretations of assertions (statements) of the sentences into further semantic interpretations, which are further interpretations by the addressee—what the addressee understands the speaker to have communicated in making the asertion in that context. The latter interpretations are the classical Gricean conversational implicata. Thus, the architecture of the theory is this: 1.
2.
Internalist sentence semantics. Sentence strings, well-formed sequences of lexical items, are meaningful; they have semantic representations. Externalist statement semantics. Utterances of sentence-string-tokens will, when nonspecific items have specific interpretations, when vagueness is sharpened, and when variables have values, have welldefined truth conditions in a context. An utterance with a well-
SEMANTICAL UNDERDETERMINACY
3.
41
defined truth condition in a context will be said to “express a proposition.” The truth condition of an utterance in the context is described by the semantic interpretation. Some utterances are assertions (statements). Externalist speaker/addressee pragmatics. A speaker, in, by, or when asserting a sentence-string-token, will convey, suggest, or imply to an addressee further propositions [implicata]. The contents of the assertion and of the implicata constitute the “total signification” of the speaker’s utterance for the addressee.
Psycholinguistically the picture is this: the mental realization of language L, L = <Syntax, Vocabulary>, generates φ. Speaker S asserts utterance U, a phonetic realization of φ. Addressee A interprets Uφ in the context K in order to know what assertion S has made and what else S conveys by making the assertion. This twoleveled semantics—internalist and truth-conditional—is an essential feature of the Atlas (1979) version of post-Gricean pragmatics. (For discussion of these three levels of representation and interpretation, see Atlas 1989, Levinson 1995, and Lyons 1995b). Chomsky (1996c, 2000) limits his semantics to Internalist Semantics of the syntactically well-formed strings and relegates both truth-conditional semantics and Gricean implicatures of utterances to the pragmatics of Performance Systems. This difference between me and Chomsky is merely terminological, but a parallel difference between me and Jerry Fodor is not merely terminological (see section 2). In section 1, “Metaphor, Nonspecific Meaning, and Utterance Interpretation,” I took a classical problem, that of metaphor, and approached it from the point of view of my Internalist Semantics notion of semantical nonspecificity (with respect to a semantical property). That section provided a gentle introduction to the notion of semantical nonspecificity in an application to an old and familiar problem. In section 2 “Semantical Nonspecificity, Utterance Interpretation, and Psychological Modularity,” I discussed the philosophical consequences of my semantical and post-Gricean pragmatic theory for Fodor’s account of utterance-interpretation. I posed the problem of utterance-interpretation in a new and, I suggested, conceptually more satisfactory way. Unlike me, Fodor thinks that language is an exception in the realm of intentional action—that what “is usually required in the intentional analysis of behavior is a kind of hermeneutic sophistication that’s as far as can be from the execution of a rote procedure. The notable exception is inferring intentional content from utterance form” (1990b: 213). On my view, though hermeneutic sophistication is, as Fodor notes, normally a symptom of an analysis of intentional acts whose results are never “all likelihood,” “enormously reliable” ones, the challenge, rather, is to explain how utterance-interpretation can be both hermeneutic and enormously reliable. That is the challenge that has directed the construction of the accounts discussed in Atlas (1979, 1989) and in the present work. In my view, a satisfactory explanation of utterance interpretation requires the recognition of two linguistic phenomena: semantical nonspecificity of sentence and predicate expressions and generalized interpretative inferences, either from semantic representations or from assertions.
LOGIC , MEANING , AND CONVERSATION
42
Without semantical nonspecificity of the sentences, generalized conversational interpretative inferences are “blind”; and without the contents of generalized interpretative inferenda, the contents of semantically nonspecific sentences are “empty.” Together they yield by hermeneutic means a reliable interpretation of a speaker’s “meaning” in asserting a sentence. That is the post-Gricean thesis stated in Atlas (1979), Atlas (1989: chap. 4) and developed further in this volume. In chapter 2, “Grice’s Theory of Conversational Inference,” I expound on Grice’s views, selectively and critically, from his first publication on the subject in 1961 to his last remarks from his collected essays Studies in the Way of Words of 1989. My treatment is not encyclopaedic or exhaustive. My focus is on a critical evaluation of Grice’s views. It has been clear to me at least since Atlas (1975a,b, 1977a,b, 1978a,b, 1979) that assumptions of the classical, 1967 version of Grice’s account in the William James lectures in Harvard University would require revision in a radical way, for two types of reasons: 1.
2.
Since the literal meanings, and so the semantic representations, of many sentences being asserted are semantically nonspecific (for some semantical feature) and thus in the philosophers’ sense nonpropositions (not articulations of truth conditions and not bearers of a determinate truth or falsity), independently even of deixis, reference-fixing, tense, and so on, “what is said” is not a simple function of context and literal meaning, as I demonstrated in Atlas (1978a,b, 1979). Grice’s Maxims of Conversation and his “idealized model” or “theory” of conversational implicature entail inadequate taxonomies of inference and are insufficiently constrained to avoid “overgenerating” implicata that are logically contrary or contradictory. The post-Gricean account expounded here is intended to remedy the descriptive inadequacy in Grice’s and the neo-Griceans’ account, without despairing of Grice’s program entirely (cf. the enthusiastic attempt at “debunking” Grice in Davis (1998), who throws out the baby with the tub and keeps the dirty bathwater) and without begging further questions of the account’s explanatory adequacy.
On the first problem of semantical nonspecificity, negative, definite description sentences are a notorious case in point (see Horn 1989). As I argued in Atlas (1974, 1975b, 1977b, 1978a,b, 1979, 1989), pragmatic inference is antecedently required even to construct a truth-value bearing interpretation (a “proposition,” Grice’s “what is said”) from the literal, semantically nonspecific meaning of a negative sentence uttered. Noam Chomsky recently suggested that “we cannot assume that statements (let alone sentences) have truthconditions. At most they can have something more complex: ‘truth indications’ in some sense” (1996c: 52). He added: There is no question of how human languages represent the world, or the world as it is thought to be. They don’t. . . . There is no reference-based semantics. . . . There is a rich and intriguing internalist semantics, really part of syntax, on a par in this
SEMANTICAL UNDERDETERMINACY
43
respect with phonology. Both systems provide ‘instructions’ for performance systems, which use them . . . for articulation, interpretation, inquiry, expression of thought, and various forms of human interaction. (Chomsky 1996c: 53)
I had written in Atlas (1979: 278 n.2), “The sentence The A is not B in one context may be understood as L– [an external, exclusion negation] and in another as L+ [an internal, choice negation]. These understandings [not senses] are related to the [literal] meaning . . . of the sentence as allophones are to the phoneme to which they belong.” I had been engaged in a study in Chomskyan Internalist Semantics. In chapter 2 I also discuss Grice’s (1989a) and A. P. Martinich’s (1980) views on the role of Relevance in Grice’s account, arguing that Relevance cannot subsume Grice’s Second Maxim of Quantity or Atlas and Levinson’s (1981) Maxims of Relativity and Informativeness. I show that the phenomena of hyperbole present obstacles to Horn’s (1984b) and Sperber and Wilson’s (1986b) attempts to reduce Informativeness to Relevance. I conclude by discussing the central philosophical problem of Grice’s pragmatics: his reduction of the semantics of utterances to the epistemology of mental states. In chapter 3 “The Rise of Neo-Gricean Pragmatics,” I discuss the project of Atlas (1978b, 1979), Atlas and Levinson (1981), Horn (1984b, 1989, 1993, 1996b), Levinson (1987a,b, 1988a,b; 1991, 2000), and Huang (1994, 2000). We revise Grice’s original account of the Maxims of Conversation so that both speaker-centered maxims of production and addressee-centered maxims of comprehension (interpretation) are formulated as part of a complete account (see Blutner 2000). Horn (1984b, 1989, 1993) has suggested a simplifying dualistic schema of Grice’s maxims—two antithetical categories of informational adequacy, Q (Quantity), and economy, R (Relevance)—the speaker saying enough for successful uptake by his addressee (Q) but saying no more than is needed (R).22 Levinson (1987b, 1988a,b; 1991, 2000) and Huang (1994, 2000) have developed and applied the theory to anaphora in English and Chinese. In chapter 4, “The Post-Gricean Theory of Presupposition,” I review the reduction of Strawsonian and Fregean presupposition to entailment and Gricean implicata, comment on the inadequacies of Stalnaker’s (1974, 1999) notion of pragmatic presupposition, and discuss the flaws in Grice’s (1981) attempted reduction. In chapter 5, “Assertibility Conditions, Implicature, and the Question of Semantic Holism,” I examine several theoretical problems in specifying the logical forms, truthconditions, and implicata of English sentences containing comparative adjectives, equatives, adverbials of degree, and adverbial “approximatives” (e.g., almost). The range of linguistic data explained and the depth of semantic analysis achieved is, I believe, not to be found in competing theories.23 I also derive some surprising consequences for the philosophical dispute concerning “meaning holism.” 22Horn (1984b: 17) notes: “The most detailed and careful discussion in the literature of Q vs. R clashes in English is due to Atlas and Levinson (1981) (cf. Levinson 1983: Section 3.2 for related discussion).” 23Manfred Bierwisch (1989: 129) remarks of the analysis of comparatives and equatives of Atlas (1984a) and in this book, “The most careful account of the relevant facts and relations concerning the comparative and the equative is given in Atlas (1984[a]).”
44
LOGIC , MEANING , AND CONVERSATION
In the final chapter 6, “The Third Linguistic Turn and the Inscrutability of Literal Sense,” I examine a controversial question in the study of the semanticspragmatics interface: What does the English numerical adjective ‘three’ actually mean? I reach the controversial conclusion that the meaning of numerical adjectives is highly underdeterminate, and I introduce a Context Principle for Chomskyan Internalist Semantics: only in the context of a noun phrase does a numerical adjective have a determinate meaning. I attempt to put post-Gricean pragmatics in historical perspective and consider its consequences for a philosophical theory of meaning and for Chomskyan internalist semantics.
2
Grice’s Theory of Conversational Inference A Critical Exposition
1 The third linguistic turn Thirty-five years ago Richard Rorty (1992 [1967]) published an anthology of essays on the philosophical methods of the logical empiricists in the interwar years and of the “ordinary language” philosophers in prewar and postwar Oxford, an anthology entitled The Linguistic Turn. The phrase has caught on. But there has been a third linguistic turn in philosophy—less dramatic than that of the early Wittgenstein’s Tractatus Logico-Philosophicus and more systematic than the nuanced analysis of language characteristic of J. L. Austin’s essay “A Plea for Excuses” (1956–57). It has not been as ideological as either of the first two; it has been more tentative in its claims, more sophisticated in its methodology, more sensitive to the demands of theory construction. Its tutelary deities have been Noam Chomsky and W. V. O. Quine; its progenitors Donald Davidson, Zeno Vendler, H. Paul Grice, and Robert Fogelin; and its charter the texts of Fogelin (1967), Evidence and Meaning; Grice (1967), “Logic and Conversation: The 1967 William James Lectures”; Vendler (1967), Linguistics in Philosophy, Davidson (1967), “The Logical Form of Action Sentences”; Davidson and Harman, eds. (1972), Semantics of Natural Language; and Davidson and Harman, eds. (1975), The Logic of Grammar. The Gricean strand of this development began with P. H. Nowell-Smith’s concept of “contextual implication” in his 1954 book Ethics; it emerged with Paul Grice’s notion of “conversational implication” from his 1961 essay on the causal theory of perception, and it matured in Grice’s Maxims of Conversation of his 1967 William James lectures “Logic and Conversation” and in Robert Fogelin’s “Rule of Strength” 45
46
LOGIC , MEANING , AND CONVERSATION
from his 1967 Evidence and Meaning. Patrick Nowell-Smith and Paul Grice were part of J. L. Austin’s Saturday morning meetings in Oxford in the 1950s. Fogelin expresses his intellectual indebtedness to both Austin and Nowell-Smith. What distinguishes Grice’s approach is his emphasis on “the radical importance of distinguishing (to speak loosely) what our words say or imply from what we in uttering them imply; a distinction seemingly denied by Wittgenstein, and all too frequently ignored by Austin” (Grice 1986: 59). By his own testimony, Grice (1986: 59) was influenced by Quine’s work, which “helped to throw light on the problem of deciding what kind of thing a suitable theory [of linguistic phenomena of the kind with which in Oxford we had long been concerned] would be, and also by his example exhibited the virtues of a strong methodology”; he added that “Quine’s influence on me was that of a model: I was never drawn towards the acceptance either of his actual methodology or of his specific philosophical positions” (1986: 60). Geoffrey Warnock (1973) reports that Austin’s Saturday Morning Meeting group read Chomsky’s (1957) Syntactic Structures in the Michaelmas term of 1959. Grice (1986: 60) describes Quine and Chomsky as his “chief theoretical mentors.” He then describes his work of the 1960s as “principally . . . an attempt to show, in a constructive way, that grammar (the grammar of ordinary discourse) could be regarded as, in Russell’s words, a pretty good guide to logical form, or to a suitable representation of logical form” (1986: 60). For Grice this was principally “the suggestion of notational devices together with sketchy indications of the laws or principles to be looked for in a system incorporating these devices; the object of the exercise being to seek out hitherto unrecognized analogies and to attain new levels of generality” (1986: 60). Grice’s notational devices—for example, his square brackets—of the 1967 William James lectures and of the 1970 University of Illinois lecture “Presupposition and Conversational Implicature” (Grice 1981), have seemed to me unsuccessful in reaching his objective, as I discussed in my essay “On Presupposing” (Atlas 1978b). Ironically, given his goal, his informal discussion of the Maxims of Conversation have been more adapatable to systematic theoretical linguistic explanations than were his syntactical innovations. The syntactic way in which grammatical form could be a guide to logical form was not successfully demonstrated by Grice, but the GriceStrawson Condition (Strawson 1954, 1964) that statements carry presuppositions of the existence of a noun phrase designation only if the NPs are topic NPs has proven fruitful in the linguistic and philosophical analysis of negative existence statements (Atlas 1988, 1989) and justified Grice’s confidence in Russell’s Theory of Definite Descriptions. A development of this idea in my Focal Noun Phrase Limitation Principle (Atlas 1991a) determined the choice of logical subject NPs in logical forms for cleft statements (Atlas and Levinson 1981; appendix 3 here) and for statements of the surface form Only Tom VP, where ‘VP’ is a metavariable taking as values verb phrases (Atlas 1991a, 1993, 1996b, 1997a, 2001). Grice’s interest in and emphasis on the logical form of ordinary language sentences parallel contemporaneous interests of Quine’s in chapter 4 (“Vagaries of Reference”) and chapter 5 (“Regimentation”) of Quine’s 1960 Word and Object; of Anthony Kenny’s in his 1963 Action, Emotion, and Will; and of Davidson’s and
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
47
Vendler’s in their individual essays of 1967. Vendler self-consciously notes in his preface to Linguistics in Philosophy that his essays represent “the gradual introduction of a new technique into analytic philosophy,” the point of which is “to show that the more or less sophisticated data provided by structural linguistics can be used in philosophical arguments” (1967: vii–viii). Grice’s project—like other projects of the Third Linguistic Turn, and unlike that of the Tractatus Logico-Philosophicus and logical empiricism or that of the later Wittgenstein, Ryle, and Austin—is characterized by the conscious selection from, or creation of, concepts in Chomskyan theoretical linguistics whose appropriation, development, and application to problems in philosophy proper can reconceptualize, and so overcome, the stalemates of competing, traditional philosophical “solutions.” Cases in point, in addition to the Russell-Strawson debate on definite descriptions, negation, and presupposition (Atlas 1977b, 1978a,b, 1989) and the question whether ‘ought’ logically implies ‘can’ (Sinnott-Armstrong 1984)1, are (a) the meaning of causal statements (Vendler 1967), (b) the logic and linguistics of negative existence statements (Atlas 1988), (c) the adequacy of propositional attitude sentences as analyses of intentional locutions (Searle 1983; Marti 1993), (d) the semantics of natural kind terms (Atlas 1980b, 1989; Chomsky 1995b, 2000: 148–53; Donnellan 1983, 1993), and (e) the reference of singular terms (Bach 1987). In Grice’s view, his investigation of “logic and conversation” was a prolegomenon to any future metaphysic (Grice 1986: 53, 59). That is the larger philosophical significance of Grice’s account of “logic and conversation.” But it has been Grice’s concept of conversational implicature, and for philosophical purposes his notion of a generalized conversational implicature, that has been a focus of both philosophical and linguistic discussion.2 As Grice emphasized, the importance of the notion was its role in distinguishing what our words say or imply from what we in uttering them imply. This distinction and its ramifications took Grice beyond Wittgenstein and Austin, as he notes (Grice 1986: 59).3 It also significantly influenced debates in linguistic theory on the semantical reductions of pragmatic facts, 1See Atlas (1974, 1975a,b, 1977b, 1978b), Bach (1987), Boër and Lycan (1976), Grice (1981), Russell (1905, 1919, 1959), Sinnott-Armstrong (1984), Strawson (1950, 1952, 1964), and White (1993). 2Generalized conversational implicata are contents conveyed, suggested, or implied by a speaker in, when, or by asserting a sentence, as a default, standard, or normal feature of the asserting of the sentence in any and every speech context. Particularized conversational implicata, by contrast, are nonce implications by a speaker that are implied by virtue of the sentence’s being asserted in a particular speech context. 3Grice himself thought that his theory of implicature would cast light on problems of perception, knowledge, and modality so as to correct what he saw as a mistaken and unreflective Wittgensteinian formulation of these problems popular in the 1950s and 1960s among British philosophers (Grice 1961, 1986: 66). As P. M. S. Hacker remarks:
Grice’s distinction between the meaning of an expression and its conversational implicatures can be brought to bear upon Wittgenstein’s accounts of the meaning of various expressions in terms of use. For if Grice’s argument is correct, then Wittgenstein attributed features of the use of expressions to their meaning, which are correctly ascribable not to their meaning but to pragmatic principles of discourse. (Hacker 1996: 245)
48
LOGIC , MEANING , AND CONVERSATION
illustrated by Gordon and Lakoff’s highly influential 1971 paper “Conversational Postulates” and by the seminal Horn (1972).4 To quote R. A. Harris: Ordinary language philosophy grew more important for generative semantics in the seventies. Beginning with [J. R.] Ross’s fairly direct importation of Austin’s insights about performatives [Ross 1970], which [Jerrold] Sadock and [Alice] Davison took up at Chicago under [James] McCawley, it gained considerable momentum under the influence that philosopher H. P. Grice’s [1967] conversational implicature work had on both Lakoffs [Robin and George], on their students, and on [James] McCawley. It is from this period that linguists began to develop a sense of something they called pragmatics, as distinct from what [is] called semantics. (Harris 1993: 185)
A decade later, in 1983, Geoffrey Leech published his Principles of Pragmatics and Stephen Levinson published his survey Pragmatics. The latter book summarized the extraordinary amount of creative work in philosophy of language and linguistics that began in 1969 with Searle’s Speech Acts and in 1970 with Ross’s “On Declarative Sentences” and that continued through the 1970s. Levinson chose to discuss deixis, conversational implicature, presupposition, speech acts, and conversational structure, establishing a curriculum that continues to influence textbooks years later— for example, Green (1989), Grundy (1995), and Thomas (1995).
2 The language of perception In 1961, in an essay in the Proceedings of the Aristotelian Society, H. Paul Grice, then a tutorial fellow of St. John’s College, Oxford, addressed a problem in the theory of knowledge that had exercised distinguished Oxford philosophers, in particular H. H. Price in his 1932 book Perception. According to Price’s account, if one is perceiving a material object M, the statement so describing one’s experience, namely ‘M is present to my senses’, is equivalent to ‘M causes a sense-datum with which I am acquainted’, and “perceptual awareness” amounts to an inference from a psychological effect to its material cause. Grice observed that since the phrase ‘present to my senses’ was used by Price for one sense of the verb ‘perceive’, Price’s account of the causal theory of perception gives it that (1a) and (1b) are equivalent: (1)
a. I am perceiving M. b. I am having (or sensing) a sense-datum which is caused by M.
This talk of ‘sense-datum’ and ‘acquaintance’ is resonant with reminders of the 1910– 11 lectures of G. E. Moore (Some Main Problems of Philosophy, 1953) and Bertrand Russell’s “shilling shocker,” Problems of Philosophy (1913). By the 1950s and 1960s,
4Gordon and Lakoff’s position was criticized in J. L. Morgan’s 1977 “Conversational Postulates Revisited” and in Gazdar’s 1979a Pragmatics: Implicature, Presupposition, and Logical Form (see Newmeyer 1996: 139–40).
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
49
Oxford philosophers were concerned to show how the philosophical term of art ‘sensedatum’ should be understood, and thus to show what sense, if any, could be made of the claims of a causal theory of perception. The philosophical statement in (2a) would be explained by the ordinary statement in (2b) or by (2c): (2)
a. I am sensing a red sense-datum. b. It looks red to me. c. I seem to see something red.
In the Oxford of the late 1950s (at the time his essay appeared Grice was forty-eight years old) a sense-datum theory of perception of this kind was under attack.5 What interested Grice was the type of this attack, which he described as follows: When someone makes such a remark as “It looks red to me” a certain implication is carried. . . . It is implied either that the object referred to is known or believed by the speaker not to be red, or that it has been denied by someone else to be red, or that the speaker is doubtful whether it is red, or that someone else has expressed doubt whether it is red, or that the situation is such that though no doubt has actually been expressed and no denial has actually been made, some person or other might feel inclined towards denial or doubt if he were to address himself to the question whether the object is actually red. . . . Let us refer to the condition which is fulfilled when one or other of the limbs of this disjunction is true as the D-or-D condition (‘doubt or denial’ condition). (Grice 1965: 441)
Grice then admits that “there would be something at least prima facie odd about my saying ‘That looks red to me’ (not as a joke) when I am confronted by a British pillar box in normal daylight at a range of a few feet.” But the critic’s thesis goes beyond this, as Grice observes: (a) that it is a feature of the use, perhaps of the meaning, of such locutions as “looks to me” that they should carry the implication that the D-or-D condition is fulfilled, and that if they were uttered by a speaker who did not suppose this condition was fulfilled he would be guilty of a misuse of the locutions in question . . . , (b) that in cases where the D-or-D condition is unfulfilled the utterance employing the “looks to me” locution, so far from being uninterestingly true, is neither true nor false. (Grice 1965: 442)
But then, as Grice observes, the critic can close his attack upon the sense-datum theorist: The sense-datum theorist wants his sense-datum statements to be such that some one or more of them is true whenever a perceptual statement is true; for he wants to go on to give a general analysis of perceptual statements in terms of the notion of
5Grice (1986: 62) remarked, “I have never been very happy about Austin’s Sense and Sensibilia, partly because the philosophy which it contains does not seem to me to be, for the most part, of the highest quality, but more because its tone is frequently rather unpleasant.”
50
LOGIC , MEANING , AND CONVERSATION
sense-data. But this goal must be unattainable if “looks to me” statements (and so sense-datum statements) can be truly made only in the less straightforward perceptual situations [since in the straightforward situations, ones without doubt or controversy, the “looks to me” idiom would be out of order and neither true nor false]; and if the goal is unattainable the [Causal Theory of Perception] collapses. (Grice 1965: 442)
Since Grice, in his essay, wishes to defend his own version of the Causal Theory of Perception, he was forced to consider the strategy of the critic that he has described. He mentions a famous, parallel case, one that is both intellectually important for Grice and of historical interest as well. Suppose that one asserted (anomalously): (3)
??It is raining, but I don’t believe that it is raining.
What is it, G. E. Moore (1968: 535–44) had asked, that is peculiar about this statement? Clearly there is something odd about the statement. The oddity might be explained if there were an “implication,” ‘I believe that it is raining’, that was “part of the meaning” of ‘It is raining’, for then statement (3) would be logically contradictory. But it seems clear that such an explanation is mistaken. ‘I believe that it is raining’ cannot be, literally, part of the meaning of ‘It is raining’; the oddity seems to arise because of what “asserting” a sentence consists in. Sentence (3) is, literally, logically consistent: it is possible for it to be raining and for the speaker not to believe that it is. Grice’s diagnosis of his philosophical opponent’s thesis follows a similar pattern: If I were to say ‘it looks red to me’ in a situation in which the D-or-D condition is not fulfilled, what I say is . . . true, not “neuter”; while admitting that though true it might be very misleading and that its truth might be very boring and its misleadingness very important, one might still hold that its suggestio falsi is perfectly compatible with its literal truth. Furthermore one might argue that though perhaps someone who, without intent to deceive, employed the ‘it looks to me’ locution when he did not suppose the D-or-D condition to be fulfilled would be guilty in some sense of a misuse of language, he could be said not to be guilty of a misuse of the particular locution in question; for, one might say, the implication of the fulfillment of the D-or-D condition attaches to such locutions not as a special feature of the meaning or use of these expressions, but in virtue of a general feature or principle of the use of language. (Grice 1965: 442)
To show that the critic of the Causal Theory of Perception was mistaken, Grice undertakes an examination of the concept of implication that figures centrally in the dispute. Grice (1965: 444–45, 448–49) distinguishes four cases: (4)
a. Smith has left off beating his wife. b. She was poor but she was honest.
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
51
c. Jones has beautiful handwriting and his English is grammatical. d. My wife is either in the kitchen or in the bedroom.
In (a) “what is implied is that Smith has been beating his wife.” In (b) “what is implied is (very roughly) that there is some contrast between poverty and honesty, or between her poverty and her honesty.” In (c) “I am reporting on a pupil at Collections. All I say is ‘Jones has beautiful handwriting and his English is grammatical’. We might perhaps agree that there would here be a strong, even overwhelming, implication that Jones is no good at philosophy.” In (d) “it would normally be implied that he did not know in which of the two rooms she was.” Of these examples Grice notes that (a) is a traditional case of “presupposition” in Grice’s pupil’s P. F. Strawson’s (1950, 1952, 1964) sense: the truth of what is “implied” is a necessary condition of the statement’s being either true or false, while (b) could be false, rather than neither true nor false, whether or not there was a contrast between honesty and poverty, if she were poor and dishonest. Grice notes that “what is said (or asserted)” in (a) is the source of the implication, but “what is said (or asserted)” in (b) is not. He also observes that in both (a) and (b), the speaker implied whatever is implied; that in (b) but not (a) the speaker’s words, in a precise sense, imply whatever is implied; and that in neither (a) nor (b) “would it be evidently appropriate to speak of his saying that, or of his saying that in that way, as implying what is implied” (1965: 446). Also in discussing examples (a) and (b), Grice (1965: 446) introduces the notions of the detachability and cancelability of the “implication.” Since these notions become important in the development of Grice’s views, it is of interest to consider the first account that Grice gives. Of detachability in example (4a) Grice writes: One cannot find a form of words which could be used to state or assert just what the sentence ‘Smith has left off beating his wife’ might be used to assert such that when it is used the implication that Smith has been beating his wife is just absent. Any way of asserting what is asserted . . . involves the implication in question. I shall express this fact by saying that in the case of (a) the implication is not detachable from what is asserted (or simpliciter, is not detachable). (1965: 446–47)
And then of noncancelability Grice writes: One cannot take a form of words for which both what is asserted and what is implied is the same as for (a), and then add a further clause withholding commitment from what would otherwise be implied, with the idea of annulling the implication without annulling the assertion. One cannot intelligibly say ‘Smith has left off beating his wife but I do not mean to imply that he has been beating her’. I shall express this fact by saying that in the case of (a) the implication is not cancellable (without cancelling the assertion). (Grice 1965: 446–47)
By contrast, the implication of (b) is, in Grice’s view, detachable because “what is asserted” is the same as “what is asserted” in She is poor and she is honest, but the implication is not as resistant to cancelation as that of (a) is. Grice does not go so far as to say that a denial of the implication of (b) is “unintelligible,” but he admits that
52
LOGIC , MEANING , AND CONVERSATION
saying ‘She is poor but she is honest, though of course I do not mean to imply that there is any contrast between poverty and honest’ is a “peculiar” way of conveying that she was poor and honest. As I have pointed out in Atlas and Levinson (1981: 20), Grice (1961), unlike Karttunen and Peters (1979), distinguished between “presupposition” and what he later called “conventional implicature” by saying, as in the case of (a) and (b), that while the conventional implicatum was detachable and, to a high degree, not cancelable, the presupposition was not detachable and not cancelable. Finally, Grice recognizes that the implication in (b) is attributable to the meaning of a particular word: ‘but’.6 When Grice considers examples (d) and (c), he observes, interestingly, that in asserting P or Q the implication that the speaker does not know whether P or whether Q is a default implication: the implication is standardly, or normally, attributable to the speaker, and to the speaker’s asserting the sentence rather than some other sentence, but there are contexts in which the implication would not be carried. The default implication is ceteris paribus nondetachable. It is also cancelable: it is intelligible and logically consistent to say ‘My wife is either in the kitchen or in the bedroom; mind you, I’m not saying that I don’t know which’. Likewise the implication of (c) is cancelable: ‘Jones has beautiful handwriting and his English is grammatical; I do not of course mean to imply that he is no good at philosophy’. But the implication is context-dependent: it is not standardly involved in the assertion of the sentence, and it is not a default implication—a special context is required to attach the implication to the utterance—for example, that it be uttered at Collections (student evaluations) in an Oxford college. It was important to Grice to distinguish the case of ‘or’ from the case of ‘but’. The implication of ‘but’ is attributable solely to its lexical meaning. Grice (1965: 450) wished to argue that the implication of ‘or’ is not attributable solely to its lexical meaning but to a “general principle governing the use of language.” His “first shot,” as he put it, at formulating the principle was: One should not make a weaker statement rather than a stronger one unless there is a good reason for so doing. On the explicit assumptions that P is “stronger” than P or Q because P entails P or Q and not conversely, and that an “obvious” reason for not making a statement is lack of justification, warrant, or adequate evidence, Grice suggests that his proposed principle of language use will explain the implication of the assertion P or Q. The form the Gricean argument takes must be something like this: 1. S asserted Ψ or χ. 2. S did not assert the logically stronger Ψ or assert the logically stronger χ. 6It is notable that in Grice’s (1965: 446–47) argument for the non-cancelability of the presupposition, he only considers the cancelability of the presupposed sentence for the affirmative statement Smith has left off beating his wife. Strawson (1950) had claimed that presuppositions were preserved under negation, so Grice should have considered the implication of the negative Smith has not left off beating his wife as well. He does finally do so in Grice (1981), with the result that he concludes that Strawsonian presuppositions are not detachable but are, in fact, cancelable. See chapter 4 in this volume for more on this subject.
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
53
3. A good reason for not asserting a statement φ is not having adequate grounds for asserting φ. 4. So the speaker, it is presumed, does not have adequate grounds for asserting Ψ and does not have adequate grounds for asserting χ. 5. If so, then the speaker does not know Ψ, and the speaker does not know χ. 6. So the speaker means to imply in asserting Ψ or χ that the speaker does not know that Ψ and does not know that χ. Not only is it assumed in this form of argument that a speaker is capable of having and of recognizing good reasons for his or her actions, it is assumed by Grice that what counts as a reason for the action of making a statement—actually, in this case, an inaction: not making the statement Ψ—is the same as what the speaker means to imply in performing the alternative action by asserting the disjunctive sentence Ψ or χ. But this assumption on Grice’s part is gratuitous, as I shall discuss further later in this chapter. It follows from Grice’s claim about (4d), that the speaker does not know in which of the two rooms his wife was; that in asserting P or Q, the speaker does not know that P and the speaker does not know that Q [¬KsP & ¬KsQ]. So ¬P is compatible with all that speaker S knows, and ¬Q is compatible with all that speaker S knows. But when one considers assertions of logical disjunctions in mathematical reasoning, a paradigmatic use of a disjunction is to prove a theorem θ by deducing it from a disjunction, from both a statement φ and its contradictory ¬φ, since whatever is a logical consequence of the necessary truth φ ∨ ¬φ is itself a necessary truth. The advantage of such a proof is that one need not know whether φ is a mathematical truth or whether ¬φ is a mathematical truth. One may, of course, use this proof technique even when one does know which is true. When I assert (R) My car keys are either in my pants pocket or on my desk, it is true that an explanation for my act of asserting (R) may appeal to my not knowing which, but I do not think that I should be interpreted to imply, convey, suggest, or mean by, in, or when asserting the disjunctive sentence (R) that I do not know which, anymore than in asserting It’s raining Moore actually intended to convey, suggest, or imply in, when, or by asserting that sentence that he believed that it was raining, so that the total signification of the utterance would be [IT’S RAINING & I BELIEVE IT], even if the best explanation for Moore’s act of asserting ‘It’s raining’ was that he believed that it was (see section 5 in this chapter). Thus I disagree fundamentally with Grice (1965, 1967, 1975a) and especially with linguists Gazdar (1979a) and Levinson (1983, 2000), who construct a theory of “clausal” implicature to explain this misguided intuition about or sentences (see chapter 3, section 2). Grice’s view here seems just a lapse, as he later appeals to Moore’s Paradox sentences—for example, It’s raining and I do not believe it, to distinguish between what a speaker “implicates” and what a speaker “expresses.” In asserting U a speaker S expresses his belief that U is true; he does not implicate that he believes U. A speaker can assert You can/may have chocolate or vanilla and not implicate that the speaker does not know which you can have. In fact, the speaker implicates
54
LOGIC , MEANING , AND CONVERSATION
what is not entailed: you can/may have chocolate, and you can/may have vanilla. You just cannot have them both. Possible (P or Q) Possible P & Possible Q “Possible (P or Q)” » Possible P & Possible Q “Possible (P or Q)” » ¬Possible (P & Q)
Similarly for You can go to Grandma’s or you can go to the movies. In S’s offer of a choice of alternatives, S is normally in a position to assert both limbs of the disjunct, which is why S is in a position to assert the disjunction. (Shades of Mathematical Intuitionism!) When I say pointedly to Jason Linder of a “choice” of dessert at my dinner table: (T) You may have tartufo or you may have tartufo, I am not in doubt as to which he may have, nor do I convey in or by asserting the disjunction that I do not know which he may have. Such alleged implicata do not arise as generalized conversational implicata from sentential ‘or’ as features of the meaning of P or Q, at least in the case P or P. The classical Gricean theorist may riposte that my example merely illustrates the intervening role of the Maxims of Manner: P or P is a circumlocution for P, but an arch or joke one, so one should not expect the usual conversational implications to be produced. But the mistake is already made: P or P is not a circumlocution for P. It is certainly the case that one allegedly typical conversational implication but not both of asserting a disjunction P or Q, such as You may not have both an eclair and pie in asserting You may have an eclair or pie, would not be produced in asserting P or P, namely, You may not have both the tartufo and the tartufo—that is, You may not have the tartufo. The modest point of the or statement was to offer the tartufo or nothing, not to imply conversationally that Jason could have no tartufo at all. Since 1973 I have been pointing out the problem of conflating two claims: (a) “P or Q” » ¬(P and Q) and (b) a speaker S’s asserting P or Q is explained by ¬KsP and ¬KsQ. The first claim is about the implicata of asserting ‘or’. The second claim is purportedly explanatory of the behavior of asserters of ‘or’. Claim (b) does not entail (a), and (a) is false as a general claim about ‘or’. Rather, English ‘or’ is semantically nonspecific (rather than ambiguous) between inclusive disjunction and exclusive disjunction. By Atlas and Levinson’s (1981) inference to the best interpretation (discussed in chapter 3) the form P or Q is interpreted by default as exclusive disjunction, unless blocked—for example, by the use of ‘and/ or’, by the different form P or P, or by collateral contextual information (Fowler 1965: 422). This interpretation entails ¬(P & Q). Thus the alleged scalar implicature from or to not and is actually another Atlas (1984a) hybrid: implicaturecomposed-with-entailment inference. Thus the correct claim is (a') “P or Q” ° » ¬(P & Q). Both the ambiguity theorists and the classical Griceans were mistaken about ‘or’, but the ink continues to run (see chapter 5, section 1). I do not, unlike Grice in 1961, think that (b) entails that asserting the or statement implicates that the speaker does not know which alternative holds. But I shall pretend otherwise in the rest of this section, for the sake of Grice’s argument. It seems to me that to this day the phenomena of clausal implicatures have been misdescribed
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
55
and misunderstood. The theoretical situation is nearly as bad as the one for English conditional if . . . then sentences (see Sanford 2003). The philosophical interest of the semantic distinction between ‘but’ and ‘or’, let us recall, is an issue in the philosophical theory of perception. The critic of the causal theorist’s explanation of sense-datum statements as statements using ‘looks to me’ or ‘seems to me’ claimed that a ‘looks to me’ statement implied that there was, or naturally could be, doubt or denial that an object is the way it looks. Since ordinary statements describing the perception of observable properties of material objects are not cases in which there is normally a doubt or denial, the critic rejects the account of ‘looks to me’ statements, and so the account of sense-datum statements, that the causal theorist requires. Furthermore, the objections are (a) that the doubt-or-denial condition is a semantic “presupposition” of a ‘looks to me’ perceptual report and (b) that the occurrence of the doubt-or-denial implication is attributable, according to the critic, to the presence of the words ‘looks to me’ or ‘seems to me’. These two features are philosophically important in the debate because they allow the critic to claim that the causal theorist is committed to statements that, on the theorist’s own view, are, in general, neither true nor false; since such statements are obviously true or false, in adopting ‘looks to me’ language, the causal theorist has shown himself or herself not to be using ordinary English and, presumably, not understanding what he or she was saying. It is also important that the implication be nondetachable and (to a high degree) noncancelable, because it would not be enough for the critic of the causal theory to show that just one way of characterizing sense-datum statements failed. It is essential that all ways of (equivalently; therefore nondetachably) characterizing sensedatum statements should fail. Grice then proceeds to argue that That pillar box looks red to me does not semantically presuppose doubt-or-denial. His argument was this: Suppose that I am confronted in normal daylight, by a perfectly normal pillar-box; suppose further that I am in the presence of a normal, unsceptical companion; both he and I know perfectly well that the pillar-box is red. However, unknown to him, I suffer chronically from Smith’s Disease, attacks of which are not obvious to another party; these attacks involve, among other things perhaps, the peculiarity that at the time red things look some quite different colour to me. I know that I have this disease, and I am having (and know that I am having) an attack at the moment. In these circumstances I say, ‘That pillar-box looks red to me’. I would suggest that here the doubt-or-denial condition is not fulfilled; my companion would receive my remark with just that mixture of puzzlement and scorn which would please my objector; and yet when he learnt about my attack of Smith’s Disease, he would certainly think that what I had said had been false [rather than neither true nor false]. (Grice 1965: 452–53)
Although this is an intuitively plausible counterexample to the critic’s claim that the doubt-or-denial condition is semantically presupposed by statements of That pillarbox looks red to me, the critic also wants to hold that the implications of these statements are nondetachable and noncancelable.
56
LOGIC , MEANING , AND CONVERSATION
In the course of discussing these claims, Grice (1961/1965:456) notes that if the doubt-or-denial condition is nondetachable (from what is asserted), it is not nondetachable by virtue of the implication’s being a logical implication. It is simply not a logical implication that if this pillar-box looks red to me, then I, or someone else is, or might be, inclined to deny that it is red or to doubt whether it is red. And even if the implication is nonlogical but nondetachable, there is no reason to think that it is, in consequence, not cancelable.7 What Grice now argues is that That pillar-box looks red to me nonlogically “implies” the doubt-or-denial condition, but the implication is nondetachable and cancelable. Like the implication of ‘or’, the implication is standardly implied by the speaker, or by the speaker’s asserting what he did. Since such standard implications are a matter of the use of language, not solely a matter of the meaning of words, the implication is not solely a part of the meaning of the expression ‘looks to me’. So the objection to explicating sense-datum statements in a causal theory of perception by paraphrase into ‘looks to me’ statements fails. It is particularly useful to note that the principle of language use in question is “giving preference to the making of a stronger rather than a weaker statement in the absence of a reason for not so doing” (Grice 1965: 459). One instance of such a case, the stronger φ and the weaker φ or Ψ, was explained by the semantic
7Indeed,
Grice’s example of ‘or’ would show why this is true. Interestingly Grice (1961, 1965: 456– 57) feels the need to dispose of a potential counterexample to his last thesis. He assumes that the implication of ‘but’ is nonlogical, and he has argued that it is (to a high degree) noncancelable. Now he argues that if the implication of ‘but’ is noncancelable, the reason for its being noncancelable is that it is detachable (by the use of ‘and’). Then, surprisingly, he wishes to conclude that if an implication is nonlogical and not detachable, then it is cancelable: (D → ¬C) ∴ (¬D → C). Grice (1961, 1965: 457) writes that “if you say that the implication of the fulfillment of the [doubt-or-denial] condition is (a) not logical in character and (b) not detachable, then you must allow that it is cancellable.” If this were correct, the objection to the Causal Theory would fail, as the objection claims that the D-or-D implicatum is neither detachable nor cancelable. The implicit argument here seems to be that since ‘but’ and ‘and’ are truth-functionally equivalent yet distinct lexical items with distinct uses in the language, and since the occurrence of two nearly synonymous items in the language would be superfluous unless some difference in use attached to them, so the difference in form makes a difference in use by way of a difference in (noncancelable) “conventional implicata” that the word-forms carry. Plausible as this argument is, even if detachability of an implication φ from “what is asserted” were to entail φ’s noncancelability, surely one should not conclude, as Grice seems to, that its nondetachability entails its cancelability; that is just the Fallacy of Affirming the Consequent, as well as an example of a conversational inference from ‘if’ to ‘if and only if’ (see chapter 3 in this volume). What Grice has to show to defeat the critic is that it is possible for an “implication” to be cancelable while not detachable; on Grice’s own showing the default implication of ‘or’ is nondetachable yet cancelable. He has shown what he needs to show. He does not need yet a further, bad argument. It is noteworthy that Grice was surely assuming here, plausibly enough, that semantic presupposition is not a classical logical implication, since Strawson (1950, 1952) had explicitly tried to distinguish presupposition from entailment, but since Grice had granted that presupposition is nondetachable and (erroneously) that it is noncancelable, if it is also nonlogical, its properties would be a counterinstance to the very thesis he has just tried to defend (by fallacious reasoning!).
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
57
entailment: φ φ ∨ Ψ, and the failure of the converse: φ ∨ Ψ φ (assuming it is false that φ Ψ). Grice notes that It looks red to me and It is red are not instances of an entailment ordering; they are logically independent.8 Nonetheless, Grice suggests that, intuitively, one regards It looks red to me as the weaker. If that is correct, his principle of language use will justify the “standard implication”: if a speaker asserts It looks red to me, he is in some doubt, or someone could have denied, that it is red. Yet the implication is not part of the meaning of the sentence ‘It looks red to me’. Thus Grice concludes that he has disposed “of what [he has] found to be a frequently propounded objection to the idea of explaining the notion of a sensedatum in terms of some member or members of the suggested family of locutions,” as in ‘It looks red to me’ (1965: 460). And he suggests that a budget of philosophical objections of a like kind will be undercut by a like analysis: 1. You cannot see a knife as a knife, though you may see what is not a knife as a knife. 2. When [G. E.] Moore said he knew that the objects before him were human hands, he was guilty of misusing the word ‘know’. 3. For an occurrence to be properly said to have a cause, it must be something abnormal or unusual. 4. For an action to be properly described as one for which the agent is responsible, it must be the sort of action for which people are condemned. 5. What is actual is not also possible. 6. What is known by me to be the case is not also believed by me to be the case. (Grice 1965: 459–60)9
It is, I hope, clear that a distinction between semantic features of statements, like semantic entailment, semantic presupposition, and what Grice later called “conventional implicature,” and pragmatic features, like what he later called “particularized conversational implicature” and “generalized conversational implicature,” is crucial to the success of Grice’s undercutting the criticism of the anti-sense-datum theorist. In particular, the distinction between ‘but’ and ‘or’—that is, between what Grice
8I have referred to such orderings of lexical items as “Levinson scales,” but, as we see, Grice discusses the sentential version in 1961. To my astonishment a referee for a prominent linguistics and philosophy journal questioned my introduction of such an ordering on the grounds that no such notion had been previously discussed in the published linguistics literature! Aside from the attitude to conceptual innovation that such a remark betrayed, Paul Grice himself had published the sentential version of my notion in the philosophical literature thirty years before I appropriated the lexical version of it for my use in discussing implicature. For students of Grice’s work the matter was, I thought, common knowledge, but apparently not. I withdrew the essay and published it elsewhere as Atlas (1991a); it has since been the subject of discussion by de Mey (1991), Blok (1993), Horn (1992b, 1996b), and others. It is now part of the linguistics literature. 9On
(1), see “seeing as” in Wittgenstein’s (1967) Philosophical Investigations; on (2), see Wittgenstein’s (1969) On Certainty; on (3), see J. Searle (1969b), “Assertions and Aberrations”; on (4), see SinnottArmstrong (1984) and M. White (1993: 48–49); on (6), see Vendler (1975a), “On What We Know”; Aune (1975), “Vendler on Knowledge and Belief,” and Vendler (1975b), “Reply to Professor Aune.”
LOGIC , MEANING , AND CONVERSATION
58
(1967) later called a case of conventional implicature and a case of generalized conversational implicature—is the crux for Grice’s refutation of the criticism of the causal, sense-datum theory.10 Having made good philosophical use of his division between the semantic and pragmatic against the casual Wittgensteinianism of some 1950’s Oxford philosophizing, six years later Grice presents a more systematic account of his principles of the use of language. It is clear from his 1961 discussion that Grice wishes to characterize, at least in the paradigm cases, the differences among semantic entailment [E], semantic presupposition [SP] (4a), conventional (let us call it “lexical” for convenience) implicature [LI] (4b), and particularized (4c) and generalized (4d) conversational implicatures [PCI and GCI].11 If we look at the logical properties of the classical semantic entailment relation—its reflexivity, transitivity, and monotonicity—and compare semantic presupposition, conventional (lexical) implicature, and conversational implicature with it, we have the mostly obvious results shown in tables 2.1 and 2.2. The pragmatic properties of entailment, conventional implicature, presupposition, and particularized and generalized conversational implicature—namely, detachability, cancelability, and calculability (to be discussed in section 3)—are as shown in table 2.2 (see Walker 1975 and Sadock 1978): Grice was certainly not wrong to see differences here. In particular, his emphasis on the importance of the distinction between conventional implicata and generalized conversational implicata (e.g., between the implicata of ‘but’ and ‘or’), however subtle their manifestations, seems well founded. Although Sperber and Wilson (1986b, 1995) reject (incorrectly, I believe) the distinction, without it the philosophical defense of Grice’s version of the Causal Theory of Perception would have foundered. 10It
is inexplicable or worse that Stephen Neale (1992: 524, n.18) should write, “The distinction between ‘generalized’ and ‘particularized’ conversational implicature is not represented in this diagram . . . because it is theoretically inert (for Grice).” Grice himself writes: I also distinguished . . . particular conversational implicatures that depended on particular contextual features . . . and ones that I thought of as relatively general which I called GENERALIZED IMPLICATURES. These are the ones that seem to me to be more controversial and at the same time more valuable for philosophical purposes, because they will be implicatures that would be carried (other things being equal) by any utterance of a certain form, though, as with all implicatures, they are not to be represented as part of the conventional meaning of the words or forms in question. (It is important that what is conversationally implicated is not to be thought of as part of the meaning of the expressions that are used to get over the implication.) And I thought that this notion of a GENERALIZED conversational implicature might be used to deal with a variety of problems, particularly in philosophical logic, but also in other areas. In these areas there seemed to me to be quite good grounds for suspecting that some people have made the mistake of taking as part of the conventional meaning of some form of expression what was really not part of its conventional meaning, but was rather a non-conventional implication which would normally be carried, except in special circumstances, by the use of that form. (Grice 1981: 185)
11Grice
notes that particularized conversational implicatures are “cases in which an implicature is carried by saying that p on a particular occasion in virtue of special features of the context, cases in which there is no room for the idea that an implicature of this sort is normally carried by saying that p” [a case of a generalized conversational implicature]. (1989b: 37)
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
59
2.1. Logical Properties of Semantic and Pragmatic Relations (1 = “yes”; 0 = “no”)a TABLE
Semantic Entailment
Lexical Implicature
Semantic Presupposition
Conversational Implicatures
Reflexive
1
0
0b
0
Monotonic
1
1
0c
0
Transitive
1
1
1
0
of a relation R among sentences: for every φ, φRφ. Transitivity: for every φ, Ψ, χ, (φRΨ & ΨRχ) → φRχ; Monotonicity: for any set A, ARφ → A∪{φ} R Ψ. See D. S. Scott (1971, 1973) and van Fraassen (1971). aReflexivity
bThe
reason that The king of France exists does not presuppose that the king of France exists is discussed in Atlas (1988, 1989: 91–118).
cShowing
the failure of monotonicity of semantic presupposition is left as an exercise for the reader.
3 Rationality, cooperation, and the imperatives of conversation In the second of his William James lectures “Logic and Conversation” (Grice 1975a,b, 1989b, 1989c), at Harvard University, 1967, Grice briefly presents a description of conversation as rational, cooperative, goal-directed behavior and formulates the Cooperative Principle: (5)
Make your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged.
Then Grice describes the means by which this overarching goal of rational, cooperative, conversational exchange is to be achieved: by following, or at least conforming to, the conversational maxims, or what Grice (1989f: 370) later called “conversational imperatives,” the observance of which promotes conversational rationality. Following the example of one of his favorite philosophers, Grice divides linguistic rationality into four categories: Quantity, Quality, Relation, and Manner (Grice 1986: 66, 103). (6) Maxims of quantity 1. 2.
Make your contribution as informative as is required (for the current purposes of the exchange). Do not make your contribution more informative than is required.
(7) Maxims of quality Try to make your contribution one that is true. a. Do not say what you believe to be false. b. Do not say that for which you lack adequate evidence.
LOGIC , MEANING , AND CONVERSATION
60 TABLE
2.2. Pragmatic Properties of Semantic and Pragmatic Relations (1 = “yes”; 0 = “no”)
Cancelable
Semantic Entailment
Lexical Implicature
Semantic Presupposition
Generalized Conversational Implicature
Particularized Conversational Implicature
0
0
0a
1
1b 0d 1
Detachable
0
1
0
0c
Calculable
0
0
1e
1f
aCompare
to treating presupposition as a GCI; see Grice (1978).
He didn’t {a. {b.
STOP} stop}
beating his wife —{he never beat her} — {?he never beat her}
The focal stress negation in (a) interrupts the default GCI and permits the acceptable continuation (Atlas 1991a) (see chapter 3, section 1). The (b) implicature is a GCI He used to beat her, which makes the continuation odd. See chapter 4, section 6, and chapter 6, section 5. bSadock (1978: 293) points out that cancelability will not be sufficient to show that a sentence is implicated, as the failure of the utterance to be inconsistent with the denial of the sentence may be due to the ambiguity of the utterance: the denial of the sentence forces an alternative reading on the ambiguous utterance. More plausibly, cancelability is necessary; Grice (1989c: 44) certainly thought so. cDefault
value.
dNondetachability,
as Grice (1989c: 43) himself points out, is neither necessary nor sufficient for the presence of a conversational implicatum: it is not necessary since the implicatum may depend on manner of utterance, not on its sense, so another utterance with the same sense but uttered in a different manner may not carry the same implicatum; it is not sufficient since entailments and semantic presuppositions are also nondetachable. Its importance depends on its distinguishing between conversational and conventional implicata. Sadock observes: Whether there are absolutely equivalent paraphrases or not, the fact that it is difficult to tell if two expressions have the same meaning makes the nondetachability test less useful in practice. Suppose the claim is made that a sentence such as Can you open the door? does not conversationally implicate a request to open the door but rather conventionally implicates it and that this claim is backed up by the observation that the implicature fails to go through if the synonymous periphrastic modal be able is substituted for can. Since the paraphrase detaches the implicature, the argument goes, it cannot be conversational. This claim can all too easily be countered with the claim that can and be able are not synonymous and that in fact this example PROVES that they are not. (Sadock 1978: 289) For what it is worth, I can get the implicatum from Are you able to open the door?; in any case, Sadock’s point is well taken. Happily for Grice, Sadock concludes: Nevertheless, [detachability] is a pretty good [test], at least in extreme cases. Detachability in the absence of obvious meaning differences of the right kind is a suspicious fact, and the more apparent paraphrases there are that succeed in detaching an implicature, the more it looks as if the implicature must be conventional. On the other hand, the more apparently synonymous expressions there are that fail to detach an implicature, the less the situation looks accidental and the more it looks as if some principle, such as the Cooperative Principle, is in force. (Sadock 1978: 290) eSee
Grice (1978, 1989a: 269–282); nonmonotonicity suggests calculability.
fThough
calculable, generalized conversational implicata are not typically calculated; see Morgan (1978).
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
61
(8) Maxim of relation Be relevant. (9) Maxims of manner12 Be perspicuous. a. b. c. d.
Avoid obscurity of expression. Avoid ambiguity. Be brief (avoid unnecessary prolixity). Be orderly.
Of course Grice (1989b: 28) recognizes that conversation is pursued in conformity with other sorts of maxims—for example, Be polite—and that one may convey messages by conforming or failing to conform to such norms of conversation (see Brown and Levinson 1978, 1987). What Grice explicitly has in mind is a description of the means by which the conversational goal of “a maximally effective exchange of information” can be achieved. These maxims and the Cooperative Principle are offered in explanation of what a speaker “implies,” “suggests,” or “means,” as contrasted with “what he said (asserted),” in uttering a sentence. Grice writes: Suppose that A and B are talking about a mutual friend, C, who is now working in a bank. A asks B how C is getting on in his job, and B replies, Oh quite well, I think; he likes his colleagues, and he hasn’t been to prison yet. At this point, A might well inquire what B was implying, what he was suggesting, or even what he meant by saying that C had not yet been to prison. The answer might be any one of such things as that C is the sort of person likely to yield to the temptation provided by his occupation, that C’s colleagues are really very unpleasant and treacherous people, and so forth.13 . . . It is clear that whatever B implied, suggested, meant in this example, is distinct from what B said, which was simply that C had not been to prison yet. (Grice 1989b: 24)
The explanation of this example that Grice gives is this: In a suitable setting A might reason as follows: “(1) B has apparently violated the maxim ‘Be relevant’ and so14 may be regarded as having FLOUTED one of the
12See
also Grice (1981) and chapter 4, section 10 in this volume.
13 This
‘so forth’ is important. The speaker’s utterance-interpretation of the sort illustrated in Grice’s example is obviously “open-ended.” We shall encounter other cases in which the interpretation of what the speaker meant by the statement is much more constrained. 14Grice evidently thought that a violation of relevance (Maxim of Relation) implied a failure to be perspicuous (Maxim of Manner), or he thought (more likely) that one way for the speaker to violate relevance would be to fail to be perspicuous, or he thought (even more likely) that if an addressee thought that a speaker had failed to be perspicuous, the addressee would surely also be unable to see the relevance (if any) of the speaker’s remark. I take it that Grice was not being perspicuous.
62
LOGIC , MEANING , AND CONVERSATION
maxims conjoining [sic] perspicuity, yet I have no reason to suppose that he is opting out from the operation of the Cooperative Principle; (2) given the circumstances, I can regard his irrelevance as only apparent if, and only if, I suppose him to think that C is potentially dishonest; (3) B knows that I am capable of working out step (2). So B implicates that C is potentially dishonest.” (Grice 1989b: 31)15
By “flouting a maxim” Grice means ‘blatantly failing to satisfy it’ and remarks: On the assumption that the speaker is able to fulfill the maxim and to do so without violating another maxim (because of a clash), is not opting out, and is not, in view of the blatancy of his performance, trying to mislead. . . . How can his saying what he did say be reconciled with the supposition that he is observing the overall Cooperative Principle? This situation is one that characteristically gives rise to a CONVERSATIONAL IMPLICATURE; and when a conversational implicature is generated in this way, I shall say that a maxim is being exploited. (Grice (1989b: 30)
In his later reflections, reported in “Retrospective Epilogue” to his Studies in the Way of Words, Grice describes the arising of an implicatum from an assertion in this way: An implicatum . . . is the content [my emphasis] of that psychological state or attitude which needs to be attributed to a speaker in order to secure one or another of the following results; (a) that a violation on his part of a conversational maxim is in the circumstances justifiable, at least in his eyes, or (b) that what appears to be a violation by him of a conversational maxim is only a seeming, not a real, violation; the spirit, though perhaps not the letter, of the maxim is respected. (Grice 1989f: 370)
It is essential to conversational implicature that the content of the psychological state of the speaker be one that the addressee is capable of reasoning out, the general pattern of which Grice outlines as follows: Calculability: Implicature argument schema He has said that P; there is no reason to suppose that he is not observing the maxims, or at least the Cooperative Principle; he could not be doing this unless he thought that Q; he knows (and knows that I know that he knows) that I can see that the supposition that he thinks that Q is required; he has done nothing to stop me thinking that Q; he intends me to think, or is at least willing to allow me to think, that Q; and so he has implicated that Q. (Grice 1989b: 31)
Of course, if a speaker violates or flouts one of the conversational maxims at the level of “what is said (asserted),” it is important to have some, even rough, charac15Obviously
Grice meant “enjoin” instead of “conjoin” in this passage. It is also worth remarking on the form of Grice’s argument. In step (2) Grice supposes that the speaker believes that C is potentially dishonest. In step (3) the speaker implicates that C is potentially dishonest; he does not implicate that he believes that C is potentially dishonest.
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
63
terization of what “what is said (asserted)” consists in. (After all, the reasoning just described above begins with [The speaker] has said that P’.) So let us consider more carefully “what is said” or “asserted.” Grice (1989b: 25) notes that in the case of an ambiguous sentence, such as (10), one’s knowledge of the English language would permit one to know that there are two possible senses. (10) He is in the grip of a vice.
This sentence, independently of knowledge of context, could be expressed by either (a) a male person or animal is unable to rid himself of a bad character trait or (b) a male person is caught in a type of clamping tool. Context permits semantic values to be given for parameters like tense and the reference of ‘he’ and a choice of sense(s) to be made. Then one possesses an interpretation of the utterance in the context, not merely the sense(s) of the sentence, that Grice calls “what is said (asserted)” (Grice is indifferent whether or not what is asserted in uttering ‘John is handsome’ and ‘The first student whose last name begins with ‘w’ in the Pomona College class of 1980 is handsome’ is the same, when ‘John’ and ‘the first student whose last name begins with ‘w’ in the Pomona College class of 1980’ denote to the same individual.) In sum, fixing of reference, of tense, of deixis, and disambiguation is assumed to yield, with literal sense(s), “what is said (asserted)” in asserting a sentence-token.16
4 Relevance: Quantity [informativeness] subsumed by relation [relevance]? In “Retrospective Epilogue” Grice (1989f: 371–72) reflects on certain dependencies among the maxims. The first concern, present even in the William James lectures of 1967, is whether the Second Maxim of Quantity (“Do not make your contribution more informative than is required”) is superfluous, being supplanted by the Maxim of Relation (“Be relevant”). The second concern is whether violations of the Second Maxim of Quantity amount to violations of cooperativeness at all. On this second concern Grice is inclined to think that overinformativeness can be misleading to an addressee, and thus uncooperative, in that the addressee will think that there is some point in the speaker’s discursiveness. (Grice [1989b: 27] writes, “there may also be an indirect effect, in that the hearers may be misled as a result of thinking that there is some particular point in the provision of the excess of information.”) On the first concern, Grice (1989f: 372) reiterates his view that relevance and information are dependent notions, but neither early nor late does he explain just how the Second Maxim of Quantity could be explained, or entailed, by the Maxim of Relation. A. P. Martinich (1980: 218) provides a provocative discussion of this difficulty. The reductive claim would argue: (a) if what a speaker says is relevant, it will not 16The notion of “what is said” has since come in for considerable discussion and notable disagreement among philosophers and linguists: see Bach (1994a), Carston (1988), Levinson (1988b, 2000), Récanati (1989), and Sperber and Wilson (1986b).
64
LOGIC , MEANING , AND CONVERSATION
provide more information than is required, and (b) if what a speaker says is not relevant, it is more informative than is required. If (a) were true, the Maxim of Relation would subsume the Second Maxim of Quantity; if (a) and (b) were both true, the Maxim of Relation and the Second Maxim of Quantity would be equivalent. Martinich’s counterexample to (a) is a line from François Mauriac’s Vipers’ Tangle. Recalling his deceased grandson Luc, the miser says, “Everyone loved him, even I.” In Gricean terms, the statement is, Martinich alleges, overinformative: ‘everyone loved him’ entails ‘the miser loved him’; thus, alleges Martinich, the miser’s phrase ‘even I’ is overinformative. Yet, observes Martinich, the statement is relevant: it indicates that Luc was highly lovable and that, although the miser does not love widely, he was at least able to love Luc. Martinich takes this example to show that some statements are both more informative than is required and relevant, contrary to the reductionist claim (a). If φ entails Ψ, the statement φ & Ψ is logically equivalent to φ and is truthconditionally neither more nor less informative than φ. If ‘even’ carried a conventional implicatum, so that Even I loved him were logically equivalent to I loved him (as Karttunen and Peters 1979 once suggested), the statement Everyone loved him, even I would be, like φ & Ψ where φ entails Ψ, logically equivalent to Everyone loved him. Since what is relevant in the context is the even I, what is overly informative is not the phrase even I, contrary to Martinich’s claim, but rather its conjunction with Everyone loved him, since that conjunction is, on the assumptions made here, logically equivalent to Everyone loved him. And the proposition Everyone loved him is, in the context, overly informative if taken literally (and not, as would be appropriate in this kind of case, hyperbole for Luc was lovable). With this redescription of Martinich’s case, I think it is plausible to say that the speaker has literally “said” (asserted) more than is required but was not thereby failing in relevance. Thus the reductionist claim (a) is, indeed, false. The observation that I have just made concerning hyperbole, which was not considered by Martinich (1980), itself raises an interesting issue for Grice’s views. Grice considers hyperbole under a class of conversational implicatures in which the First Maxim of Quality (“Do not say what you believe to be false”) is flouted—that is, blatantly violated.17 Grice’s (1989b: 34) example was Every nice girl loves a sailor. (Apparently love and universal quantification just don’t mix.) Curiously, Grice gives 17The classic example of a flout (C-type) implicature is a flout of the First Maxim of Quantity, namely of “exploitation, . . . a procedure by which a maxim is flouted for the purpose of getting in a conversational implicature by means of something of the nature of a figure of speech” (Grice 1989b: 33). This is a variant of an example originally introduced in Grice (1965: 448–49). This is Grice’s account:
A is writing a testimonial about a pupil who is a candidate for a philosophy job, and his letter reads as follows: “Dear Sir, Mr. X’s command of English is excellent, and his attendance at tutorials has been regular. Yours, etc.” (Gloss: A cannot be opting out, since if he wished to be uncooperative, why write at all? He cannot be unable, through ignorance, to say more, since the man is his pupil; moreover, he knows that more information than this is wanted. He must, therefore, be wishing to impart information that he is reluctant to write down. This supposition is tenable only if he thinks Mr. X is no good at philosophy. This, then, is what he is implicating.) (Grice 1965: 448–49) The other cases Grice discusses, briefly, are the assertions of logical truths, such as War is war.
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
65
no gloss on his example at all. Hyperbole should have been of greater theoretical interest to Grice, to Horn (1984b), and to Sperber and Wilson (1986b), for it demonstrates the incorrectness of the subsuming of the Second Maxim of Quantity by the Maxim of Relation [relevance], pace Horn (1984b) and Sperber and Wilson (1986b). The statement John was loved by everyone could be relevant in a context, as hyperbolic statements are, but, of course, it is “stronger” than would be required for the simple exchange of information and, if taken literally of John in the example, surely incorrect. At the level of what is implicated, by contrast, it would communicate precisely what is required by the context: John was lovable. Thus it is the Second Maxim of Quantity (“Don’t make your contribution more informative than is required [for the current purposes of the exchange]”) that is more efficacious than the Maxim of Relation (relevance), just as the First Maxim of Quality (“Don’t say what you believe to be false”) is more efficacious than the Maxim of Relation: they both guide utterance-interpretation at the level of what is implicated. The Maxim of Relation is incapable of guiding the interpretation of hyperbole, and violations of the First Maxim of Quality alone will not distinguish hyperbole from meiosis (understatement), from metaphor, or from irony. At the level of “what is said (asserted),” the reduction thesis (a) is clearly false, as my reanalysis of Martinich’s example shows. At the level of “what is implicated,” the Second Maxim of Quantity plays an explanatory role that the Maxim of Relation cannot play, and it cannot be subsumed by it. As Levinson (1987a: 76) points out, both Wilson and Sperber (1981) and Horn (1984b) incorrectly hold a reductionist view. Horn (1984b: 12) poses a rhetorical question: “What would make a contribution more informative than required, except the inclusion of material not strictly relevant to and needed for the matter at hand?” The answer to this question commits Horn to reductionist theses: (a) (if what a speaker says is relevant, it will not provide more information than is required) and (b) (if what a speaker says is not relevant, it is more informative than is required). I have argued against (a). Thesis (b) is strictly a nonstarter: its falsity is obvious. Consider the context about John’s lovableness, and suppose that War is war is, gratuitously, and irrelevantly, uttered. Thesis (b)—if what a speaker says is not relevant, it is more informative than is required—would predict that War is war is more informative than is required, which is obviously absurd. Martinich then offers a different argument against the subsumption of the Second Maxim of Quantity (Quantity-2) by the Maxim of Relation (relevance). It is an important instance of a familiar philosophical tactic: If [Relation] and [Quantity-2] are jointly de trop, one could just as easily argue that [Relation] is superfluous. For, if a speaker contributes more information than is necessary, then the excess information of his contribution is irrelevant; and, conversely . . . what is wrong with the objection against [Quantity-2] is that it assumes that . . . faced with a choice . . . one should sacrifice [Quantity-2]. (Martinich 1980: 218)
If, as in Horn’s (1984b) rhetorical question, one were committed to a reductive equivalence, so that on the grounds that φ Ψ, one reduced Ψ to φ, on the same grounds
66
LOGIC , MEANING , AND CONVERSATION
one could just as well reduce φ to Ψ. Suppose further that one had a reductive definition of “more information” in terms of “relevance.” That, alone, would not show that the notion of “more information” was irrelevant (pardon the expression); it would merely show that whether or not it was logically irrelevant, it was no longer logically primitive in the vocabulary of the theory, and that if the notions used in the supposed definition were justifiable in the theory, then the notion of “more information” would be justifiable in the theory. The remaining case, one in which some of the other maxims entailed the Maxim of Relation, subsuming it, would be the safest claim to make if one thought that the Maxim of Relation were redundant, and, conversely, if one thought that the Maxim of Relation could be specialized to yield the other maxims, the other maxims would be redundant. What conversational relevance consists in is a large and disputed subject. I shall briefly consider some of Grice’s own, late remarks in the “Retrospective Epilogue,” which give us some clear indication of what Grice had in mind by relevance. In assessing the Maxims of Quantity, Grice remarks: To judge whether I have been undersupplied or oversupplied with information seems to require that I should be aware of the identity of the topic to which the information in question is supposed to relate; only after the identification of a focus of relevance can such an assessment be made; the force of this consideration seems to be blunted by writers like Wilson and Sperber who seem to be disposed to sever the notion of relevance from the specification of some particular direction of relevance. (Grice 1989f: 371–72; see also Wilson and Sperber 1981 and Sperber and Wilson 1986b)
The role of topic in “what is said,” as in the logical form of cleft statements in English, was emphasized in Atlas and Levinson (1981) (see appendix 3 of this volume), and its role in “what is implicated” was emphasized in Atlas (1984a) (see chapter 5), where it is needed to resolve contradictions in the implicata that standard applications of the First Maxim of Quantity can generate from statements like John is not as tall as Brian. Statement-topic reappears again decisively in determining the correct logical form for Only John loves Sonia (see Atlas 1996b). Statement-topic is crucial for a theory of implicature, as Grice (1989a) noted. Any theory that ignores it, like Sperber and Wilson’s (1986b) Relevance Theory, cannot be a descriptively adequate theory. Grice’s notion of “a particular direction” in a conversation was anticipated, and elaborated, by linguistic work in Conversational Analysis (see Levinson 1983: 284– 370) on rule-specified adjacency of two speech-acts (Sacks 1972; Levinson 1987a: 78; Martinich 1980: 220–21), where what is “relevant” is what “should be done next” (as in (11a)), and on sequential expectations, where what is “relevant” is what is “required or wanted to achieve goals of the interaction” (as in (11b); Levinson 1987a: 77): (11)
a. A: Hello. B: Hello. b. A: Do you stock LT 188? B: Big or small?
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
67
Finally, there is a notion of conversational topic, as contrasted with statement topic, that describes the coherence of a discourse (see “formulation” in Sacks (1972)).18 The philosopher A. P. Martinich independently observed the function of adjacency-pairs and of goal-directed sequential expectations in meeting the (vaguely described) needs of “relevance.” Following Strawson (1964) he also emphasized topical connectedness. In light of his observations, he suggests two Submaxims of Relation: a. Make your contribution one that moves the discussion towards its goal. b. Express yourself in terms that will allow your hearer to tie your contribution into the conversational context. (Martinich 1980: 220–21)
Martinich understands ‘context’ to include “the speaker and hearer, their common perceptible environment, their previous utterances, and all of their relevant [my emphasis] beliefs” (1980: 221). This notion of context is about as roomy as one could ask for; unfortunately, it is not helpful to characterize the admittedly polymorphous notion of relevance by introducing submaxims intended to specify the notion and then undermine the analysis by reintroducing the notion of “relevant belief” in characterizing the concept of “conversational context” employed in the submaxims themselves. What is important and clear from Martinich’s (1980) discussion is the nonreducibility of Grice’s Second Maxim of Quantity to the Maxim of Relation (relevance). This is a significant result, and it is one to which I shall return, as a revised and restructured, addressee-centered version of Grice’s speaker-centered Second Quantity Maxim will play a significant role in the post-Gricean theory of inferenda (e.g., Atlas and Levinson’s 1981 Maxims of Relativity and Principle of Informativeness to be discussed in chapter 3 of this volume). Grice is looking at “information” from only one point of view: the speaker’s. Grice’s point of view on quantity of explicit information was this: “Don’t say Ψ & φ rather than φ if only φ is required.” A different post-Gricean point of view on the Second Maxim of Quantity (“Don’t make your contribution more informative than is required”) would be this: “Don’t say χ & φ rather than χ if it is evident to
18In his Principle of Relevance, Strawson (1964/1971a: 92) anticipates, the “topical connectedness” of both conversational topic and statement topic, although he is not concerned to develop a theory of their relationship:
We do not, except in social desperation, direct isolated and unconnected pieces of information at each other, but on the contrary intend in general to give or add information about what is a matter of standing or current interest or concern. There is a great variety of possible types of answer to the question what the topic of a statement is, what a statement is “about”—about baldness, about what great men are bald, about which countries have bald rulers, about France, about the king, etc.—and not every answer excludes every other in a given case. This platitude we might dignify with the title, the Principle of Relevance. (Strawson 1964, 1971a: 92)
LOGIC , MEANING , AND CONVERSATION
68
both speaker and addressee that what is required—namely, χ & φ—is inferrable from χ by the addressee (i.e., the speaker knows or is justified in presuming that the addressee can infer it, and the addressee knows or is justified in presuming that the speaker intends him to infer it).” In this case, too, if the speaker does provide an excess of explicit information—something beyond χ—the addressee may be misled, or, a possibility that Grice ignores, he may be correctly directed to the intended interpretation by thinking that there is some particular point in the provision of the excess of explicit information. Grice had simply ignored the possibility of implicit, “evidently” inferrable, information playing a role in the Maxims of Quantity.19 A post-Gricean theory will not ignore that possibility.
5 Nontrivial applications, violations, and flouts of the maxims I have already indicated that Grice requires that a conversational implicature must “be capable of being worked out” and that “to calculate a conversational implicature is to calculate what has to be supposed in order to preserve the supposition that the Cooperative Principle is being observed” (1989b: 31, 39–40). In the third William James lecture “Further Notes on Logic and Conversation,” Grice puts an interesting restriction on the character of implicata. He writes: When I speak of the assumptions required in order to maintain the supposition that the Cooperative Principle and maxims are being observed on a given occasion, I am thinking of assumptions that are nontrivially required; I do not intend to include, for example, an assumption to the effect that some particular maxim is being observed, or is thought of by the speaker as being observed. (Grice 1989c: 41–42)
Let us refer to this condition as: The nontriviality restriction on implicata No implicata are of the types: a. [Maxim] is fulfilled. b. Speaker believes that [Maxim] is fulfilled.
Grice (1989c: 42) points out that this restriction has a consequence for understanding G. E. Moore’s “paradox,” which, I observed, is a touchstone for Grice. The “paradoxical statement” is an anomalous assertion (12b) instantiating the schema (12a): (12)
19This
a. ?P but I do not believe P. b. ?It’s raining but I do not believe it.
was a role for information that I had imputed to “presupposition” in my (1975a) discussion of Frege and Dummett. See my discussion of the post-Gricean theory of presupposition in chapter 4 of this volume.
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
69
The interpretation of this statement seems to require a speaker to have made a statement the acceptance of which by any addressee (not excluding himself) the speaker himself undermines by denying that he himself believes what he has said. Grice writes: On my account, it will not be true that when I say that P, I conversationally implicate that I believe that P; for to suppose that I believe that P (or rather think of myself as believing that P) is just to suppose that I am observing the first maxim of Quality [“Do not say what you believe to be false”] on this occasion. I think that this consequence is intuitively acceptable; it is not a natural use of language to describe one who has said that P as having, for example, “implied,” “indicated,” or “suggested” that he believes that P; the natural thing to say is that he has expressed (or at least purported to express) the belief that P. He has of course committed himself, in a certain way, to its being the case that he believes that P, and while this commitment is not a case of saying that he believes that P, it is bound up, in a special way, with saying that P. (Grice 1989c: 42)
There are British linguists such as Stephen Levinson (1983: 105), and American philosophers such as Al Martinich (1980: 224–25), who have rejected Grice’s view of Moore’s “paradox.” In fact, the conversational implicature analysis of Moore’s paradox now seems to have become a “received view” (see comments in D. S. Clarke 1994: 145). Levinson’s explanation of the pragmatic anomaly is that the second conjunct of statement (13) contradicts an alleged Quality Maxim (“Don’t say what you believe to be false”) implicature of the first conjunct of the statement, which is that one believes what one asserts. (13)
??It’s raining and I don’t believe that it is.
Although he claims that “when one asserts something one implicates that one believes it,” Levinson (1983: 105 n.7) nicely observes a major flaw in his own claim and a fortiori in the current “received view” among philosophers. If S believes P were implicated by S’s asserting that P, the implicatum would be consistently and intelligibly cancelable—for example, like the cancelability of the implicatum not both of or in The PM is patriotic or nationalistic—in fact she’s both. Note the absence of linguistic anomaly in this utterance-type. By contrast, Moore’s “paradox” statements are anomalous. Conversational implicata are deniable without inconsistency (with “what is asserted”) and without anomaly. This evidence suggests that Moore’s “paradox” statement is not a case of implicature, despite the current popularity of the view among philosophers.20
20Daniel Sperber (in conversation) questioned this argument, but his concern rests at best on an implicature of what I said, not on what I said. My claim was that cancelability was a necessary condition on the existence of an implicatum; I was not claiming that cancelability was a sufficient condition for the existence of an implicatum. Once one shows that the anomaly of Moore’s Paradox cannot be explained by appeal to implicature, it remains open just what the explanation should be. If I understand Sperber’s views (Sperber and Wilson 1986b), he is inclined to explain the anomaly by some kind of “entailment,” and it is a thesis of this section that “entailment” is not a correct explanation.
70
LOGIC , MEANING , AND CONVERSATION
2.1 Inadequacies of Grice’s implicature argument schema Is there any reason for one to believe that an implicatum Bs[P] of S’s assertion P exists? In fact, there is, and Grice himself unfortunately provides a rationale for one.21 Recall Grice’s sketch of implicatural reasoning: Implicature argument schema He has said that P; there is no reason to suppose that he is not observing the maxims, or at least the Cooperative Principle; he could not be doing this unless he thought that Q; he knows (and knows that I know that he knows) that I can see that the supposition that he thinks that Q is required; he has done nothing to stop me thinking that Q; he intends me to think, or is at least willing to allow me to think, that Q; and so he has implicated that Q. (Grice 1989b: 31)
Martinich plausibly suggests the following instance of the Argument Schema: “He has said that it is raining; there is no reason to suppose that he is not observing the maxims, or at least the Cooperative Principle; he could not be doing this unless he thought that he believed that it is raining; etc.” (Martinich 1980: 224–25). On Grice’s grounds and Levinson’s grounds there is no implicatum Speaker believes that P from the speaker’s assertion P, so the oddity of a lack of cancelation of the (nonexistent) implicatum Speaker believes that P by the speaker’s denial I don’t believe that P, and hence an inconsistency between what is implicated and what is asserted, cannot be the correct explanation of the linguistic anomaly in Moore’s paradox statements. There must be something wrong with the formulation of Grice’s Argument Schema. The flaw in Grice’s formulation of his Argument Schema is in the clause he could not be [observing the maxims or at least the Cooperative Principle] unless he thought that Q. On Grice’s (1989f: 370) own view, an implicatum arises as the content of that psychological state that a hearer needs to attribute to a speaker in order to explain, from the speaker’s point of view, why a maxim has been violated—for example, because its fulfillment is less imperative than the fulfillment of another maxim, a case of maxim clash (Grice’s class B implicata)—or arises as that part of the content of the total signification of the utterance that constitutes what the speaker “meant,” in order to explain, from the speaker’s point of view, why, contrary to appearances, a maxim or the Cooperative Principle has not really been violated or flouted—that is, blatantly violated (Grice’s class C implicata).
21Interestingly, from the historical perspective, in 1942 Moore (1968:542), unlike Grice (1989b: 31), apparently thinks that speaker S’s assertion of P does “imply” BS[P], in some sense of ‘imply’. In the same year, in response to Moore, the American logician C. H. Langford (1968: 333) gives an explanation of the Moore statement of the variant (Grice 1989b: 31) form B1st pers.[P] & ¬P, a form later favored by Wittgenstein, that uses the Sincerity Condition on assertion and a Consistency Condition of rational belief to derive the contradiction B[P] & ¬B[P]. Nevertheless, as I have discussed elsewhere (Atlas 1995), the linguistic anomaly of Moore’s “paradox” statements is not explained by the rational speaker’s insincerity or by the sincere speaker’s irrationality. Likewise, logical inconsistency is not an explanation of linguistic anomaly. See appendix 1 for a discussion of Moore’s sense of ‘imply’.
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
71
I shall argue that Grice’s own formulation confuses the class B problem of explanation of cooperative linguistic behavior with the class C problem of speaker’s utterance-interpretation. The letter of Grice’s formulation in the Implicature Argument Schema is too imprecise to discriminate between these problems, and I will argue that the conflation of these problems is the result of certain philosophical assumptions about the explanatory roles of the semantic contents of assertions and the mental contents of belief-states. In particular, Grice argued that semantic contents of assertions could be subsumed, by analysis, under mental contents of psychological states—of communicative intentions (Grice 1989a: 213–23, 117–37), for instance— and that whatever was to be explained by semantic contents ultimately would be explained by those mental contents. I shall show how Grice’s reductionist program in the theory of meaning undermines his account of conversational implicature. (Ironically there is an immediate parallel to Russell’s 1903 error in his theory of denoting concepts. Russell’s assumption that meanings were intentional objects undermined his theory of reference.) Grice suggests that conversational implicata can arise from violations of the maxims or from exploitations (via blatant violations) of the maxims. (I discussed an example of exploitation in section 3.) His example in which a maxim is violated but its violation is to be “explained by the supposition of a clash with another Maxim” is this: A is planning with B an itinerary for a holiday in France. Both know that A wants to see his friend C, if to do so would not involve too great a prolongation of his journey: (14)
A: Where does C live? B: Somewhere in the South of France. (Gloss: There is no reason to suppose that B is opting out; his answer is, as he well knows [my emphasis], less informative than is required to meet A’s needs. This infringement of the first maxim of Quantity [”Make your contribution as informative as is required . . .”] can be explained only [my emphasis] by the supposition that B is aware that to be more informative would be to say something that infringed the second maxim of Quality, “Don’t say what you lack adequate evidence for,” so B implicates that he does not know in which town C lives.) (Grice 1989b: 32–33)
The gloss that Grice provides is interesting because it indicates the distinct roles that Grice, explicitly, thinks that implicature plays. He has written in the recent “Retrospective Epilogue” to Studies in the Way of Words that “an implicatum . . . is the content of that psychological state or attitude which needs to be attributed to a speaker in order to secure one or another of the following results; [e.g.] . . . that a violation on his part of a conversational maxim is in the circumstances justifiable, at least in his eyes” (Grice 1989f: 370). But this content is not content that, intuitively and pretheoretically, I would have thought that a speaker “implied,” “suggested,” or “meant” by, in, or when saying what he said. It is, rather, as Grice notes in his gloss, a hypothetical explanation of the speaker’s infringement of the First Maxim of Quantity, the hypothesis being that the speaker does not know in which town C lives, and in
72
LOGIC , MEANING , AND CONVERSATION
conforming to the Second Maxim of Quality, the speaker says no more than Somewhere in the South of France.22 A’s hypothesis that B does not know where C lives is inferred as A’s best explanation why, assuming that B is conforming to the Cooperative Principle, B has failed to conform to the First Maxim of Quantity and was less informative than the context required. Nevertheless, it is not clear in the example that when B “said” (asserted) Somewhere in the South of France, B meant or suggested or implied in “saying” Somewhere in the South of France that he did not know in which town C lived. It is not clear that B intended by his choice of words to convey that, and, hence, it is not clear that B meant it in saying what he said. (Despite animadversions by some linguists, I claim, in agreement with Grice, that there are default implicata standardly conveyed by the uttering of sentences, whether or not the speaker specifically intended them to be conveyed—for example, generalized conversational implicata— but in such cases, the speaker may have misled his addressee by his choice of words.) What we have is a plausible explanation of B’s uninformative response. What we do not have is a case of B’s intentionally conveying to his addressee a content that he has not explicitly asserted. It certainly need not be the case that B utters ‘Somewhere in the South of France’ with the intention that by virtue of B’s utterance A will come to believe that B does not know in which town C lived. For B to have this communicative intention (M-intention in Grice’s 1969: 155; 1989a: 105 sense) it must be true that B utters the sentence ‘Somewhere in the South of France’ intending A (a) to believe that B does not know in which town C lives, intending A (b) to think B intends A to have the belief in (a), and intending A (c) to think B intends the fulfillment of the first intention, that of (a), to be effected by the fulfillment of the second intention, that of (b). The example, as Grice describes it, gives us no reason to think that these conditions for a communicative intention are fulfilled. There is a further peculiarity in Grice’s account. He has stated that B’s infringement of the First Maxim of Quantity can be explained only by the hypothesis that B is conforming to the Second Maxim of Quality. Why is it the only such explanation? Couldn’t one also say that if B were aware that to be more informative would be to say something (e.g., C lives in Marseilles) that infringed the First Maxim of Quality, “Do not say what you believe to be false” (or even, “Do not say what you don’t believe to be true”), B should not say it? So, according to Grice’s reasoning, B could implicate, again, that he does not know in which town C lives. There is more than
22Although
Grice leaves the matter undiscussed, there is reason to think that he would accept the following generalization: if the Maxims of Quantity and Quality compete, the imperative to conform to quality supersedes the imperative to conform to quantity. As he remarks in the “Retrospective Epilogue”: The [super] maxim of Quality, enjoining the provision of contributions which are genuine rather than spurious (truthful rather than mendacious) does not seem to be just one among a number of recipes for producing contributions; it seems rather to spell out the difference between something’s being and (strictly speaking) failing to be, any kind of contribution at all. (Grice 1989f: 371)
And he (Grice 1989b: 27) had said already that “other maxims come into operation only on the assumption that this [super] maxim of Quality is satisfied.”
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
73
one plausible explanation of B’s infringement of the First Maxim of Quantity, more than one reason to attribute the same psychological content to B’s mental states. There are two subjects of concern to Grice. One is the rational explanation of behavior, especially verbal behavior: the hypothesizing of reasons for which an agent does what he or she does. The other is the nature of linguistic meaning: utterance(type)-meaning and sentence(type)-meaning. The two concerns intersect at speaker’s-meaning, what a speaker means by, in, or when saying what he or she has said. So Grice wishes implicata to do double duty: first, as part of explaining why agents, including speakers, do what they do; second, as part of interpreting the sentence-tokens produced by speakers. In philosophy this is a familiar demand to put on propositional contents. In the case of belief, the attribution to an agent of an attitude toward the content of a beliefreport, the content of the that-clause, is part of commonsense explanations of the agent’s actions, while the attribution of a particular content to an agent’s utterance is part of commonsense interpretations of the agent’s speech. The propositional content is both a psychological item, the content of a belief, and a semantic item, the meaning of a speaker’s utterance. The two roles converge when the agent intentionally states his beliefs in words: when he makes sincere assertions. The difficulty is that the contents of implicata are not suited to play the double role, in psychological explanation and in semantic interpretation, that Grice expects them to play. It is plausible that whatever (in the way of reasons) explains why an agent did (in a context) what he did is, at least in part, what the agent believed. By contrast, whatever (in the way of reasons) explains why a speaker said (in the context) what he said is not necessarily, even in part, what the speaker meant. B did not mean, in saying Somewhere in the South of France, that he did not know in which town C lived, even if his ignorance was his reason for saying it. Grice (1989f: 370) was mistaken to view implicata as coherently playing both roles: disant-en and disantpour.23 The role for implicata is the one of conveyed speaker’s meaning. Grice himself, not surprisingly, understood the difficulty. As he summarized his analysis of ‘A meantNN something by x’ (1989a: 89), he claimed that the analysandum of nonnatural (NN) meaning was equivalent to ‘A intended the utterance of x to produce some effect in an audience by means of the recognition of this intention’; and we may add to that to ask what A meant is to ask for a specification of the intended effect (though, of course, it may not always be possible to get a straight answer involving a ‘that’ clause, for example, ‘a belief that . . .’).
And then he proceeded to raise the pertinent question: Will any kind of intended effect do, or may there be cases where an effect is intended (with the required qualifications) and yet we should not want to talk of meaningNN? Suppose I discovered some person so constituted that, when I told him
23 For another unsatisfactory reductionist monism of meaning to action, see Donald Davidson (1980: 238–39, 1984a: 148–49).
74
LOGIC , MEANING , AND CONVERSATION
that whenever I grunted in a special way I wanted him to blush or to incur some physical malady, thereafter whenever he recognized the grunt (and with it my intention), he did blush or incur the malady. Should we then want to say that the grunt meantNN something? I do not think so. (Grice 1989e: 220–21)
And Grice makes the significant observation: This points to the fact that for x to have meaningNN, the intended effect must be something which in some sense is within the control of the audience, or that in some sense of “reason” the recognition of the intention behind x is for the audience a reason and not merely a cause. (Grice 1989e: 221)
But then Grice notices the problem: It might look as if there is a sort of pun here (“reason for believing” and “reason for doing”), but I do not think this is serious. For though no doubt from one point of view questions about reasons for believing are questions about evidence and so quite different from questions about reasons for doing, nevertheless to recognize an utterer’s intention in uttering x (descriptive utterance), to have a reason for believing that so-and-so, is at least quite like “having a motive for” accepting so-and-so. Decisions “that” seem to involve decisions “to” (and this is why we can “refuse to believe” and also be “compelled to believe”). . . . It looks then as if the intended effect must be something within the control of the audience, or at least the sort of thing which is within its control. (Grice 1989e: 221)
Actually, Grice’s worry was appropriate. There was a pun here, and unlike Grice I do think that it was serious. When Grice makes a remark of the form “a is at least quite like b,” one wonders just what he had in mind. For on the face of it, having a motive for accepting the divorce court judge’s statement that you owe your former wife $2 million does not seem at all the same as having a reason to believe that you owe your former wife $2 million; in such a case having a reason for believing soand-so does not at all seem quite like having a motive for accepting so-and-so. How would a decision “that” involve a decision “to”? Brevet’s decision that Throckmorton was offside does not logically or causally require that Brevet should decide to do anything, not even if Brevet is the referee. (If Brevet is the referee, he has an obligation, under the rules of the game, to make his decision known, but is that what Grice meant by ‘involves’?) Decisions are a misleading case for Grice’s purposes. And Grice’s observation that decisions “that” involve decisions “to” because one can “refuse to believe” and be “compelled to believe” is even more peculiar. I believe “that” there is a pair of spectacles perched on the bridge of my nose, and, in fact, there is a pair of spectacles perched on the bridge of my nose, but am I compelled to believe that there is a pair of spectacles perched on the bridge of my nose? Some “evidence” is said to be “compelling”—in fact, the feelings of pressure upon my nose and the visible rim of my spectacles might constitute such—but to translate the “compellingness” of the evidence into my “being compelled” to acquire a belief-state does seem just a bad pun.
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
75
Can I refuse to believe that there is a pair of spectacles perched on the bridge of my nose? Suppose I am wearing my spectacles, and in the face of the supposed compulsion to believe that I am wearing my spectacles, can I refuse to believe that I am wearing my spectacles? I really don’t know what to say here. Suppose there is no pair of spectacles perched on the bridge of my nose; now, can I refuse to believe that there is a pair of spectacles perched on the bridge of my nose? My face is a spectacle-free-zone, but Rogers Albritton says to me, “There’s a pair of spectacles perched on the bridge of your nose!” Under what circumstances would I say, “I refuse to believe it”? In another case, I know what I would say. Rogers says, “Atlas, take out the garbage!” I say, “I refuse.” Now, in this case of a response to a command, I know what a refusal is. But when Rogers asserted, “There’s a pair of spectacles perched on the bridge of your nose,” in what sense was his assertion a command that I should believe it? Rogers may have, in a Gricean way, an intention that I should believe it, and I might come to accept it by virtue of recognizing Rogers’s intention that I should accept or believe it, but how does having that motive for accepting his statement come even close to showing that my recognition of Rogers’s intention that I should believe it itself constitute a good reason for my believing that the proposition is true? It would seem to be necessary, at least, that I not believe that Rogers is a trickster, joker, or all-around con artist, believe that he is a reliable observer of my face, that his reliable observations are reliably connected with his verbal assertions, and so on. Since an assertion is not a command that one believe what is asserted, the application of notions like refusal and compulsion to assertion, and so to belief, seems an unjustified extension of those notions if one is to take the notions literally. Can I refuse to believe that there is no pair of spectacles perched on the bridge of my nose? I certainly do not believe that there is no pair of spectacles perched on the bridge of my nose, but am I by not believing it refusing to believe it? Not taking out the garbage tout court is not refusing to take out the garbage (though one can refuse to take out the garbage without saying that one is refusing to take out the garbage, of course.) If all that ‘I refuse to believe it’ means is ‘I do not believe it’, I have no difficulty with Grice’s claims. But in that case, the claims make no philosophical point about a connection between reasons for believing and reasons/motives for doing. In sum, the observations from Grice’s essay “Meaning” do not succeed in securing his claim that in the audience’s recognizing the intention of the speaker for the audience to accept the utterance x the recognition is a reason for believing x and not merely a cause of accepting it. He was acutely aware of the problem, wondered whether it was serious, and found an interesting strategem to deny its seriousness. The strategem, unfortunately, fails in its purpose. The audience’s recognition of a reason or motive for or cause of a speaker’s saying x—namely, the intention to produce a belief in the audience by means of the recognition of the speaker’s intention— is not so far a reason for rather than a cause of believing the content of what is said, at least not for the reasons that Grice offers. In his original 1957 “Meaning” essay (reprinted in Grice 1989e), Grice put his finger on a central difficulty in his own account of “meaningNN.”
76
LOGIC , MEANING , AND CONVERSATION
We can see Grice’s difficulty more explicitly if we consider an example of a flouting of a Maxim, not merely a violation of one. If, in the same context Grice described above, the discourse had been: (15)
A: Where does C live? B: Somewhere on the planet Earth.
B’s remark would have been a flouting of the First Maxim of Quantity, not a violation of it. Like the tautologies discussed by Grice (1989b: 33)—War is war—Somewhere on the planet Earth is a blatant infringement of the injunction to be as informative as required to meet A’s needs for information. If B is not opting out of the conversation, A can explain B’s response by hypothesizing that B does not know more precisely where C lives. Explanation, though, is an inquiry-specific and question-specific activity: what needs explanation, and of what type, depends on how problematic the “facts” are taken to be (Bromberger 1992b). For us Earthlings, B’s response is highly uninformative. For Luke Skywalker, possessing an intergalactic spaceship and having this discussion with Han Solo in a far-off galaxy, B’s (Solo’s) response is not quite so uninformative. If B’s (Solo’s) reply is “informative enough”—for the purposes of the questioner (Luke), merely specifying the galaxy might have been sufficient to answer ‘Where does C live?’—that is because the point of the question could have been the identification of C’s place of origin, broadly understood, or his biological identification as a carbon-based organism of an Earth species, or . . . et cetera. Luke’s explaining Solo’s choice of utterance Somewhere on the planet Earth requires Luke to judge whether Solo has understood the point of Luke’s question and to hypothesize what point Solo has in fact taken Luke’s question to have if it is not the one Luke intended. But explaining Solo’s choice to utter Somewhere on the planet Earth is not eo ipso to interpret the content of Solo’s utterance. The interest of this point about explanation is that in the case of War is war, the blatantly uninformative logical truth of the statement makes Grice say that the statement is “informative at the level of what is implicated, and the hearer’s identification of [its] informative content at this level is dependent on his ability to explain the speaker’s selection of this particular patent tautology” (1989b: 33). That is, Grice asserts the dependency of interpretation on explanation that I have just denied. Grice explicitly makes the semantic content of the utterance-object’s “conveyed meaning” depend on the psychological content explaining the speaker’s intentional utteranceact. Yet, though it is inferrable from Solo’s behavior that Solo does not know more precisely where C lives, I claim that there remains an “open question”: What did Solo “imply,” “suggest,” or “mean” in “saying” the hypothetical sentence-token Somewhere on the planet Earth? (Let me dub this point contrasting meaning with explaining “the Open Question Argument.”) Now, to return to our original example (14), why should A take a sentence to be an implicatum of B’s utterance when the sentence is not a relevant response to A’s WH-question, ‘Where does C live?’? Even if Grice’s statement “I don’t know” is taken partially to explain why B answered as he did, why is it taken to be what B
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
77
meant when he answered as he did (the Open Question)? At the level of what is implicated, the Maxim of Relation (“Be relevant”) is not being satisfied by that putative implicatum, nor is “I don’t know” satisfying the requirement of the First Maxim of Quantity, insofar as what is needed is an informative reply to a WH-question. A reason for saying P, or “I couldn’t justifiably say anything else,” is not necessarily the same as what the saying of P is good for accomplishing in the way of communicative goals: conveying an “implication.” If one didn’t know in which town C lived, and one thought the name of the town was “relevant”—that is, what A needed (and in his gloss Grice does describe B as knowing that he has said less than A needs), one could have said I don’t know where C lives. That utterance, or the shortened I don’t know, would have been as brief a reply as Somewhere in the South of France. One was not obeying a Maxim of Manner (“Be brief”) in uttering Somewhere in the South of France that one would have failed to obey by uttering I don’t know. Why would one bother to implicate I don’t know which town C lives in when one could have asserted it? The answer cannot be merely ‘to be truthful’—to obey the Maxim of Quality at the level of implicature rather than at the level of assertion, since one was being truthful, in the example, in asserting ‘Somewhere in the South of France’. Is one being informative at the level of what is implicated? No, not if what is desired is information about where C lives and not about what one doesn’t know. That one does not know something is only in special circumstances useful information about one: for example, what one’s inability to answer a question or solve a problem tells an examiner, an interviewer, or a court of law. It tells A something informative, in that sense, only if one should have known where C lives, but it still doesn’t answer A’s WHquestion. If, at the level of the implicatum, one is not being notably truthful, more informative about where C lives, more perspicuous, or more relevant than at the level of “what is said (asserted),” in what sense is one being more “cooperative” in implicating than in asserting? Surely the answer for this example is “in no sense.” So why does Grice think that I don’t know where C lives is implicated? Of course, one could object that I am using an example from Group B in Grice’s lecture, an example in which the point is only to illustrate the violation of a maxim and to illustrate the explanation of the violation by the hypothesis of a clash with another maxim. It is the Group C examples, not violations but flouts—blatant violations—where the Cooperative Principle is satisfied at the level of what is implicated. The objector thinks that I should not judge example (14) by the standards set for examples like (15), and so I should not conclude that there is something odd in Grice’s claiming that in (14) B “implies,” “suggests,” or “means” that he does not know in which town C lives by, in, or when asserting Somewhere in the South of France. But it seems to me that every question that I can raise about the flout in (15) can also be properly raised about the violation in (14) as well, for not only are the grammatical features of the examples the same, but the “flouting” character of (15) also applies to (14) in the right context. (14)
A: Where does C live? B: Somewhere in the South of France.
78 (15)
LOGIC , MEANING , AND CONVERSATION
A: Where does C live? B: Somewhere on the planet Earth.
Just imagine a conversation (14) taking place when A and B are both in the South of France and think that the South of France is the world. In that context, example (14) would be both a Group B and a Group C example: there is a violation of Quantity to be explained by a clash with Quality, and the violation is so blatant, as blatant as a tautology (see Quine 1960: 66), as to count as a flout of the First Maxim of Quantity. Example (14) in that context would be like (16): (16)
A: Where does C live? B: Somewhere.
In cases (14), (15), and (16), one would infer from B’s reply that he does not know more precisely where C lives, but it does not follow from that inference-to-anexplanation of his “saying” what he did (and not something more precise) that he is “implying,” “suggesting,” or “meaning” (as contrasted with “asserting”) in asserting what he did that he does not know where C lives. A good explanation of, or reason for, his saying what he did is not necessarily a good interpretation of what he meant in saying what he did. Of course I am not denying that sometimes the two might coincide, nor am I denying a quite different claim: the converse claim that a good interpretation of what he meant in saying what he did is at least part of, or necessary to, a good explanation of, or reason for, his saying what he did.24 The tendency by philosophers, especially Grice, to substitute one of these distinct claims for its converse is just a logical mistake. It is also a mistake, as Richard Rorty (1998b: 296) emphasizes, of a representationalist theory of knowledge (or perception) in which causality (causal explanation) and justification are conflated, a conflating common to both British empiricism and British idealism. What discourses (14), (15), and (16) also show is that this putative distinction between Class B violation-implicata and Class C flout-implicata is not a real distinction. Grice’s B-type “implicata” are not implicata at all. Grice uses B-type “implicata” for purposes of psychological explanation, but they will not simultaneously serve the purposes of C-type implicata as semantic interpretation. To distinguish B-type explanations of maxim-clash from C-type flout implicata, Grice would have to draw a principled distinction between discourses (14), (15), and (16), but there is none to be drawn. Since Grice conflates speaker’s-utterance interpretation with psychological explanation, Grice’s Implicature Argument Schema is inadequate. That the speaker could not be observing the maxims and the Cooperative Principle unless he believed that Q does not show that Q is what the speaker “meant,” or even part of what the speaker “meant,” in his utterance. (It does not even show, in Grice’s terms, that the
24This
is a point occasionally emphasized by Hilary Putnam—for example, in “Reference and Understanding” (Putnam 1978b: 110–11). See also Atlas (1979: 276–77).
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE
79
speaker S intends his addressee A to believe that the speaker S believes that P by virtue of A’s recognizing that speaker S has the intention that A believe that speaker S believes that P.25) The First Maxim of Quality enjoins the speaker not to say what he believes to be false, and the Supermaxim of Quality enjoins him to try to make his contribution one that is true. If the speaker aims to speak truly, we would expect the speaker to say what he believes he has good reason to believe, but that he believes that P is not part of what he “means” in uttering P (as if, for any P, S believes that P could be logically entailed by the speaker- and context-dependent utterancemeaning of P!). Thus, satisfying the implicature argument schema is not sufficient for a sentence to be an implicatum of an utterance, because the schema conflates the psychological explanans of an utterance-act with the linguistic implicatum of an utterance-content. By Grice’s Implicature Argument Schema, the First Maxim of Quality fails to justify treating the Speech Act Sincerity Condition for assertions (Searle 1969a) as an implicature. Grice’s Nontriviality Restriction on Implicatures (that the satisfaction of a maxim is not implicated) is defensible. If a sentence is “calculable” from the Implicature Argument Schema, it is not necessarily a conversational implicatum. Some, such as Martinich (1980), treat the satisfaction of the Argument Schema as a sufficient condition for a sentence being a conversational implicatum. Grice himself merely claims that it is necessary: if a sentence is to be a conversational implicatum, it must be possible to “work it out” according to the inferences in the Argument Schema. Grice’s Argument Schema will not distinguish between sentences needed for the psychological explanation of a speaker’s mental or verbal actions and sentences needed for the interpretation of a speaker’s utterances. Semantic interpretation is not equivalent to psychological explanation, and the interpreted semantic contents of statements are not to be identified with the explanatory mental contents of belief-states, whatever Grice’s hopes in this matter for his reductionist program of meanings to speaker’s intentions (Avramides 1989; Schiffer 1987). Grice’s intuitions that conversational implicature did not provide an explanation of Moore’s “paradox” were correct.
25See T. Baldwin (1992: 228) on Moore. M. Burnyeat (1967–68) makes Grice’s (1989d: 105) Mintending essential to the concept of assertion. But even though speaker S asserts P BECAUSE S believes that P, it does not follow that a