Download Grice's Maxim of Quality: Role in Conversational Implicature and Speaker Belief and more Study notes Logic in PDF only on Docsity!
DE-MAXIM-IZING QUALITY
CHRISTINE GUNLOGSON
University of Rochester
1 Introduction
Quality has long been recognized as having a unique status among Grice’s conversational maxims, shown in 1–4 (Grice, 1989b):
(1) 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. (2) Quantity a. Make your contribution as informative as is required (for the current purposes of the exchange). b. Do not make your contribution more informative than is required. (3) Relation: Be relevant. (4) Manner: Be perspicuous.
Grice himself was the first to remark on the exceptional nature of Quality (Grice, 1989b:27):
... it might be felt that the importance of at least the first maxim of Quality is such that it should not be included in a scheme of the kind I am constructing; other maxims come into operation only on the assumption that this maxim is satisfied.
Grice concludes, partially on grounds of convenience, that Quality behaves enough like the other maxims with respect to implicature generation to justify its inclusion in the list. But in returning to the subject two decades later, he expressed dissatisfaction with that solution (Grice, 1989c:371):
The maxim of Quality... 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. False information is not an inferior kind of information; it is just not information.
2 Gunlogson
In modern Gricean approaches, Quality is invoked in several ways. It is standardly assumed to underlie inferences about the speaker’s propositional attitude toward the content of her contribution. If the speaker is meeting the demands of Quality—as interlocutors normally presume—then addressees can infer that the speaker believes what she is saying and (believes she) has sufficient evidence for it.^1 It does not follow that the addressee necessarily believes what the speaker says is true, but acceptance is often assumed as a simplification. The presumption that the speaker intends to be conforming to Quality is also crucial to the logic of scalar implicatures, where the necessity of satisfying Quality serves an upper bound on the amount of information a speaker can cooperatively provide. Quality has received relatively little attention in comparison to Relation and Quantity. Formal dialogue models often stipulate truth-telling as a mandatory rule of the discourse, or limit attention to dialogues where it is observed — a useful idealization for many purposes, but one that renders the operation of Quality nearly invisible. In this paper I review and add to the ways in which Quality is known to differ from the other maxims. These differences are not superficial but represent important properties which must eventually be accounted for under anyone’s theory. The claim is that to regard Quality as a routinely-observed maxim is to obscure some of the real puzzles that surround its operation and the inferences it gives rise to. I lead off in Section 2 with several non-standard features of Quality-based inferences: their noncancelability, their origin (in the usual case) in satisfaction of the maxim rather than violation, and their capacity for being suspended. Section 3 goes on to consider how the domain of Quality compares to that of the other maxims and raises prospects of generalization to implicit (unuttered) content and to non-assertive and/or non-declarative utterances. The cumulation of the differences discussed makes what I believe to be a strong case for ‘de-maxim-izing’ Quality while retaining it as a fundamental principle of a different sort. Precisely what this ‘different sort’ is, however, remains unclear. In Section 4 I outline some of the issues that arise in attempting to formulate the principle of Quality as a condition on context. In service of the overall goal of this paper, which is to lay out the territory that any full-fledged implementation of Quality should be responsible for, Section 5 summarizes the properties of Quality identified throughout the preceding discussion.
2 Qualities of Quality
The maxim of Quality, like the other maxims, and the overarching Cooperative Principle that they explicate, is stated as a set of guidelines for speakers. Given that language users are, as a rule, both speakers and interpreters, guidelines that speakers reliably follow will constitute reliable guidelines for interpreters as well, triggering inferences based on conversational and linguistic choices made by speakers presumed to be cooperative.
(^1) There has been debate about whether to characterize the speaker as believing or knowing that her claim is true (see
Hintikka, 1962, Gazdar, 1979); more recently, Thijsse (2000) argues convincingly in favor of the speaker believing she knows. The points to be made in this paper do not depend on which of these attitudes is posited. I will generally speak in terms of speaker belief rather than knowledge, simply because the former is more prevalent.
4 Gunlogson
Grice himself did not advocate treating speaker belief as an implicature — quite the opposite, as can be seen from his comments on Moore’s paradox below (Grice, 1989a:42):
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 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.
Rejecting the implicature approach does not, of course, solve the problem of the “special way” in which saying, expressing, believing, and committing are bound up together in the utterance of a declarative sentence. Grice’s thoughts about addressing that problem involve his analysis of sentence mood and its effects, a topic that is beyond the scope of this paper, though it comes up briefly in 3.2. Next, I consider what happens when Quality’s requirements clash with those of other maxims.
2.2 Quality takes precedence
Quality normally operates by being satisfied. In particular, the usual inferences about speaker belief discussed in the previous section, whether or not they are classified as implicatures, accompany the assumption that the speaker is conforming to Quality. It is true that speakers can “quietly and unostentatiously” fail to observe Quality, perhaps in order to deceive; and there are arguably effects arising from the flouting of the first maxim of Quality (Grice classifies irony and metaphor as such cases, e.g.)^2 But there are no cases of cooperative behavior where Quality is traded off against the demands of other maxims, despite the impression given by Grice’s notion of “maxim clash.” Maxim clash is described as violation of one maxim in order to observe another, giving rise to an implicature based on the supposition that the speaker was unable to meet the demands of both. A modified version of Grice’s original example is given in 8, Grice (Grice’s (3), 1989b:32):
(8) [A and B are planning their itinerary for a trip to France. Both are aware that A would like to visit C if it’s convenient given their planned stops] A: Where does C live? B: Somewhere in the South of France
According to Grice’s commentary, B implicates that he does not have more precise information about where C lives. He does so by providing less information than he knows A requires. His underinformativeness violates the first maxim of Quantity (Make your contribution as informative as is required). In the absence of any indication that B is opting out of the conversation, his violation of Quantity must be explained; and it can be by supposing that a more informative answer
(^2) With regard to flouting of the second submaxim, he remarks that such cases are “perhaps not easy to find”, and
tentatively offers a single example.
De-Maxim-izing Quality 5
would violate the second maxim of Quality (Do not say that for which you lack sufficient evidence). The clash between Quantity and Quality is thus resolved in favor of Quality. Maxim clash might sound like a battle any maxim could win, given the right circumstances. But in the event, Quality is always going to defeat Quantity. What would it look like for Quality to lose? Consider the following variation on Grice’s example, where Bob in 9 knows as little about Corinne’s exact location as B did about C’s in 8:
(9) Alice: Where does Corinne live? Maybe we can visit her on our trip. Bob: [knowing only that it is somewhere in the south of France] a. Somewhere in the south of France [Quality wins] b. In Cannes [Quality loses]
Response 9b is perfectly comprehensible, relevant, and as informative as is required. Quantity is observed at Quality’s expense, the mirror image of the previous example. Nevertheless it is not cooperative. And this is generally the case: when an informative answer is desired, we do not consider making up an answer or giving a speculative one (unlabeled as such) to be cooperative behavior on the part of the speaker, no matter how informative the answer would be if true. This point echoes Grice’s observation that “[f]alse information... is just not information.” Grice singles out false contributions, but unsupported “information” (contributions not known to be false but that fail to satisfy the second Quality submaxim) seems to fall into this category as well. We are supposing that Bob really does not know where Corinne lives, so his response is not known to be false. But 9b is uncooperative even if by some stroke of fortune Bob has given a true answer and Corinne does live in Cannes. His response fails on two related counts, corresponding to the submaxims of Quality: he does not believe it to be true (he’s agnostic, by assumption) and he does not have sufficient evidence to support it (if he did, presumably he would believe it). In a clash between Quality and Quantity, then, the (cooperative) outcome is a foregone conclusion: Quality will always win, remaining inviolable itself and forcing violation of Quantity. Quality thus provides an upper bound on the informativeness demanded by the first Quantity submaxim. There are further complications concerning the nature of the implicatures generated and the speaker’s knowledge state, but they need not concern us here. Note, however, that this pitting of Quantity against Quality is crucial to the explanation of generalized scalar implicatures, where the lexical item chosen is assumed to reflect the strongest position on the scale available to the speaker without violation of Quality. What of other maxims and other clashes? Although Grice couches his discussion of maxim clash in general terms, as though describing a productive process that all the maxims are subject to, it is not at all clear that other types are to be found. 8 is the only example Grice provides, and it is not the most illuminating choice, given that the priority of Quality over Quantity could have been anticipated on other grounds, viz. the remarks about Quality already cited. Other maxims do not seem to interact in a similar clashing way, either with Quality or with each other. The maxim of Relation, for instance, is subject to apparent violation, which motivates inferences to reconcile that appearance with the assumption of cooperativity. Actual violation of Relation is “perhaps rare”,
De-Maxim-izing Quality 7
of Relation-based implicatures. As mentioned earlier, the presumption that a contribution will conform to the maxim of Relation is so strong that when something apparently irrelevant is said, the machinery of inference sets to work to discover the way(s) in which it is relevant after all, hence not a true violation. By contrast, effects Grice attributes to flouting of (the first maxim of) Quality, such as sarcasm, do not involve reconciling the mismatch between what the speaker says and what the speaker can be presumed to believe. Instead they require recognizing the disparity and drawing inferences from the speaker’s choice to present it, overriding the inferences of speaker belief that normally obtain.
2.3 Suspension
As we saw in 2.1, Quality-based inferences about speaker belief are not individually cancelable. But there is another way in which inferences about speaker belief can be blocked. When the speaker is understood to be recounting a joke or story, or offering any kind of fictional content, addressees do not take the speaker to believe everything he says. Let us call this phenomenon suspension of the usual speaker-belief inference. Suspension is not equivalent to flouting; although the speaker flagrantly fails to adhere to Quality, the violation itself is not intended to give rise to specific implicatures about the speaker’s intent. The basic phenomenon can be illustrated by a very simple example such as 2.3, to be understood as the beginning of a joke:
(10) There were these three guys stranded on a desert island — a philosopher, a computer scientist, and a psycholinguist....
Naturally we do not assume that the speaker believes in the existence of these three men or has any real-world evidence of three men in such a plight. It’s a story, and while it continues, the audience understands that attributing speaker belief is not appropriate. Fictional contexts involve many sorts of complexities that I will not attempt to enumerate, much less address. The main point of this section is a very simple one: the potential for suspension is unique to Quality. It is simply not possible with the other maxims. There is no established mode comparable to story-telling that allows considerations of relevance, e.g., to be suspended, or to stipulate that certain categories of inference (Quantity-based, say) are to be systematically blocked. It is hard to imagine what the result would be like if there were. Adapting Grice’s observation about false information, we might say that a discourse failing to adhere to Relation and Quantity is not a discourse at all, but just a sequence of utterances. (I leave open the possibility that Manner is not as essential as the other two.) A discourse in which one or more parties does not observe Quality, however, can be perfectly well formed as a discourse — a deceitful one, perhaps, if the non-observance is covert and the intent is to mislead; or an openly non-epistemic one, if the departure from ordinary Quality standards is undertaken publicly and deliberately, as in the case of story-telling. Fictional narratives share another property with factual discourse, one not mentioned specifically by Quality: consistency. Quality includes no clause enjoining speakers to strive for consistency, avoiding logical contradiction. This omission is natural, given the emphasis on truth-telling. Presumably a speaker who is giving an accurate portrayal of reality will achieve consistency, or at least maximize it, as a byproduct of aiming for truth. But fictional narratives, and fictional characters within them, cannot be inheriting their internal consistency from reality.
8 Gunlogson
Furthermore, within non-epistemic contexts, ordinary sorts of implicatures still go through, despite the suspension of Quality-based inferences of speaker belief. If, for example, the narrative features a genie who tells the hero he gets three wishes, we take that to mean exactly three; the use of some by a narrator or character will still implicate not all; and in 11, Green’s implicated no to Hornsby’s proposal is clear even though they are invented characters (The Night Gardener, (Pelecanos, 2007):60):
(11) “Anybody up for a beer?” said Hornsby. “I’ll let y’all buy me one.” “I got practice,” said Green...
It seems that recognizing an implicature, even a particularized implicature as in 11, not only doesn’t require being a participant in the discourse that produced it, it doesn’t require that the implicating speaker even exist, let alone have communicative intentions of the appropriate sort. Much remains to be said on this topic, but leaving fictional implicature for further research, I turn next to consideration of Quality inferences from unuttered contributions.
3 The domain of Quality
The maxim of Quality, repeated below for reference, makes use of the term ‘contribution’ and prohibits ‘saying’ what is false or insufficiently supported.
(1) 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 sufficient evidence.
What does the contribution include, and what counts as saying? Interpreting ‘say’ in Grice’s ‘favored sense’ (Grice, 1989d:25), the answer is quite clear: Quality is stated in a way that narrows its applicability to conventional linguistic content, something like the literal meaning of the sentence. Attempts to formalize the maxims, such as Gazdar (1979), Groenendijk (1999) have tended to implement Quality as a condition on assertive utterances of declarative sentences, compatible with this narrow interpretation. Gazdar’s formulation provides a usefully explicit instance. Noting the exceptionality of Quality as a maxim, he implements it instead as a felicity condition on assertoric utterances. The informal version of the condition is given below (corresponding to Gazdar’s (32)); Kφ is short for the speaker knows that φ :
(12) For any declarative sentence φ , assertion of φ commits the speaker to Kφ.
Gazdar’s condition invokes both sentence type (declarative) and speech act (assertion), restricting its application to a subset of utterances of declaratives. While few analysts are as explicit about assumptions, the implicit restriction of assertions/statements to declarative utterances, and conversely, of declarative utterances to assertion, is widely adopted in practice. In this section I will raise some questions about this narrow conception of what Quality ‘applies to’. The first concern, addressed in Section 3.1, relates to the status of implicit (unuttered) contributions to the discourse with respect to Quality considerations. The second is the issue of whether, and how, to extend Quality-like principles to other kinds of sentence types and functional categories, assuming that they also lead to inferences about the speaker’s state. I will turn to that problem in Section 3.2.
10 Gunlogson
would not be required. Depending on one’s view of presupposition, there are other possibilities. If presuppositions are taken to be part of the literal content of the sentence, associated with their lexical trigger, belief attribution could presumably be covered under a suitable implementation of Quality as it stands. A similar result (for present purposes) would follow from taking a purely pragmatic view of presupposition like that of Stalnaker (1978), where presuppositions have the status of background beliefs the speaker must be assumed to hold if she believes what she asserts. Attribution of speaker belief for presuppositions follows from the attribution for asserted content. Alternatively, a presupposition may be considered an entailment, true whenever the presupposing assertion is true. Assuming that belief attribution for some content is understood to have as a consequence attribution for its entailments, observance of Quality for assertive utterances would have as a side effect its observance for their presuppositions. (Note, however, that the entailment approach becomes less attractive when we include non-assertive utterances such as questions and commands, which also carry presuppositions.) In light of these possibilities, I will assume that presuppositions like those in 13 fall into the domain of Quality with respect to the speaker and that nothing special needs to be done to generate speaker belief inferences when appropriate.
3.1.2 Implicatures
Turning to implicatures, we can consider the same issue. Is Quality as stated applicable to implicatures? Should it be? The answer to the first question is straightforward, to the second less so. The main point of Grice’s introduction of a favored sense of ‘say’, and indeed the notion of implicature itself, was to distinguish on principled grounds the conventional content — what is ‘said’ — from what is inferred from the saying of it. Straightforwardly, if a maxim is limited in application to what is ‘said’ in the literal sense, it does not apply to implicatures. For the maxims of Quantity, Relation, and Manner, application to what is ‘literally said’, however that contentious notion is implemented, is necessary to produce implicatures in the first place. The tension between what is said and what could have been said gives rise to an inference. There is no parallel process for the resulting inference. That is, there is no coherent notion of ‘what was (literally) implicated’ vs. ‘what could have been implicated’, and it is hard to see how there could be. The most obvious illustration of this fact is Manner, since it (potentially) concerns not just content but the way the content is realized in surface form and utterance. Since implicatures are unuttered and have no surface form, Manner cannot (and need not) apply to them. Exempting implicatures from evaluation against the standard of the maxims appears to be the right choice for Quantity, Relation, and Manner. For a concrete instance, consider an utterance of Some of the students showed up at the colloquium. The utterer of such a sentence ordinarily implicates that not all of the students were present. The scalar implicature arises, under standard accounts, from the speaker’s decision not to use the equally relevant and stronger term all. If the generated implicatures were in turn evaluated for conformity to the maxims, what content should they be compared against? The parallel evaluation would seem to involve what is actually implicated against what a cooperative speaker would be expected to implicate. But there is no notion of ‘what a cooperative speaker would implicate’ to be compared against, and it is hard to see how there could be. The Gricean framework for inferring what is meant from what is said is not designed to be recursive, at least not in this way.
De-Maxim-izing Quality 11
All three non-Quality maxims are sensitive to the content of what is said, often at the level of lexical choice, for Quantity at least. The implicature-generating mechanism relies on evaluation of what was said against some notion of what a cooperative speaker, acting in accordance with the maxims, ought to offer. The actual content that the speaker offers must be assessed in this comparison. Quality-based inferences about speaker belief spring from a different source. The inference that the speaker believes the content she is asserting does not depend on evaluating that content against the standard of what a cooperative speaker would assert. In fact, the content of the utterance matters for these Quality inferences only to a limited extent: flouting must be recognized as such, and there must be some sensitivity to sentence type/speech act^4 to restrict applicability to content that can be evaluated in terms of truth and evidence. Given a minimally appropriate utterance, i.e. one whose content can plausibly be taken as a belief of the speaker’s, the attributive inference is the default, regardless of content. If this characterization of Quality’s operation is accurate, there is nothing to prevent broadening its scope to apply to implicatures as well as assertions. The result of extending the domain of Quality would be to activate the default attribution of speaker belief for the implicated as well as the asserted content of a speaker’s ‘contribution’. I believe this is a desirable result — that we do, by default, take others to believe what they communicate via implicature, and understand a speaker to be misleading when she implicates what she does not believe and have evidence for. Possible objections to this claim will be considered shortly. One way to accomplish the extension of Quality’s domain would be to make it a default contextual condition, applying to all (propositional) additions to the context, explicit or implicit. I will take up that idea in Section 4. For now, I will make a simple adjustment to the wording of Quality, at the Gricean level of informality, substituting communicate for say:
(15) Quality [revised]: Try to make your communicative contribution one that is true.
a. Do not communicate what you believe to be false. b. Do not communicate that for which you lack sufficient evidence.
Setting implicatures aside for a moment, there seems to be independent justification for extending Quality to communicated content. Consider communicative gestures such as nodding and pointing, which do not ordinarily count as ‘saying’ or ‘uttering’ something. Surely it is just as deceitful to give a false answer by nodding as it is to utter yes — it is the conflict between the facts and the affirmative response that is the problem, regardless of how the affirmation is conveyed. (For that matter, it is not entirely clear how yes, no, and other response particles come by the appropriate ‘literal content’ to generate Quality inferences.) Similarly, an unsupported accusation delivered via pointing (to someone in a line-up, say) is as deserving of censure as uttering, “That’s the guy,” without sufficient evidence. The broadened version of Quality could encompass such gestures. Implicatures are a different matter, however. Nodding and pointing as just described count as explicit moves, with relatively explicit content.^5 They play the same kind of role in the discourse that utterances do; the difference is in modality. By contrast, implicatures are not analogous to
(^4) More on this equivocation in Section 3.2. (^5) I am not suggesting this is true of all gestures and all uses, of course.
De-Maxim-izing Quality 13
3.2 Beyond statements
Quality is the only maxim to mention truth and falsity (emphasis added below):
(1) 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.
The other maxims make reference more generally to the speaker’s ‘contribution’ without directly invoking truth values and attitudes. The stated goal of making a true contribution means that Quality in its original form does not extend straightforwardly beyond the category of statements/declaratives — that subset of utterances that can be evaluated in truth-conditional terms. I suspect that the problems discussed here arise in some form for the other principles as well, particularly for Quantity, where the notion of relative informedness lends itself to modeling in terms of entailment relations. Nevertheless, I will continue to concentrate on Quality, which in Grice’s rendering is exceptionally explicit about the nature of the contribution required. The restriction of Quality to truth-conditional content raises a number of complex issues. I single out for discussion here the matter of inferences about speaker attitudes arising from utterances falling outside that domain. Do we need multiple versions of Quality, each applying to a distinct kind of utterance? And if so, what are the appropriate utterance categories? One point to be kept in mind is that once we accept that the domain of (the original) Quality is restricted, the necessity of classifying utterances by type is unavoidable. This is true even if we do not adopt multiple versions of Quality. At a minimum there needs to be a binary distinction between utterances to which the original version is applicable and the ‘other’ category of utterances to which it is not. Thus, we must arrive at some method of categorization even to implement the basic version of Quality. To begin the investigation, compare the inferences that would naturally follow from utterances of 16 and 17 in ordinary circumstances:
(16) The server’s down.
(17) Is the server down?
We can take 16 to be the standard case of an assertive declarative, with the usual attribution of speaker belief arising from Quality. Inferences drawn from an utterance of 17 would be quite different. Normally the addressee would be justified in concluding that the speaker does not hold the belief that the server is down, in fact does not know whether that is the case, and does believe that the addressee may be able to supply a true answer. What is responsible for the routine inference of speaker ignorance attendant on an utterance of 17? Similarly, the speaker’s preference for a closed door will ordinarily follow as an inference from an utterance of 18. Inferences about speaker desires, however, are not readily explained by appeal to the truth-oriented standard of the original Quality maxim.
(18) I prefer the door to be closed.
(19) Close the door, please.
Note with regard to 18 – 19 that the inference supported in each case about the speaker’s state (belief, ignorance, preference) is not presuppositional — that is, it does not have to be established in the context before the utterance. Rather, the inference follows from the utterance.
14 Gunlogson
Taking note of the inapplicability of the Quality maxim to non-assertive utterances, Levinson (1983:105) suggests in passing that the truthfulness applies only to assertion and can be considered a subcase of a more general sincerity requirement for speech acts. Sincerity for questions, e.g., does not require speaking the truth – instead, the sincere questioner must truly desire an answer. The questioner cannot truly desire knowledge she already possesses. Therefore, presuming sincerity on the questioner’s part goes hand in hand with presuming ignorance of the answer as well. Along similar lines, Martinich (1980) criticizes the narrowness of Grice’s version of Quality and proposes replacing it with a more general Authenticity requirement, shown in 20 (Martinich’s B′):
(20) Authenticity (Martinich, 1980) Be authentic. That is, do not knowingly participate in a speech act for which the conditions for its successful and non-defective performance are not satisfied.
Martinich’s idea is that Quality in its original form is a specialized instance of the general principle in 20, and thus applies to only to the assertive family of speech acts. For concreteness, I offer an informal generalization of Quality from a functional point of view in 21, with hypothetical speech act categories italicized:
(21) Be sincere. (Quality)
a. Assert only what you believe to be true and have evidence for. b. Ask only for information that you lack, want, and believe the addressee can supply in accordance with clause (a). c. Issue a command only to present an outcome that you wish the addressee to bring about.
I have shown only three clauses above, involving speech acts at a high level of generality. The appropriate number and granularity would depend upon the taxonomy of speech acts in whatever theory is adopted. In principle there could be a very large number of clauses, with sincerity conditions for all speech acts residing under the general Quality principle. Levinson’s and Martinich’s proposals both suggest, without going into detail, that conditions on speech acts can account for the different kinds of inferences that arise from the canonical uses represented by 16, 17, and 19. Presumably the default assumption that speakers are sincere would lead to specific inferences for each speech act, as outlined above for questions. If a sincere questioner is ignorant of the answer, then the presumption of sincerity supports the inference that a speaker posing a question is ignorant. This suggestion is problematic, however, as can be seen most readily by taking an addressee’s perspective. It is not much of a stretch to assume that a speaker embarking on a speech act is aware of the appropriate sincerity conditions for the act he has in mind. The same cannot be said for his addressee(s), who is not privy to the speaker’s state and who therefore must identify the nature of the act from what is publicly available — the utterance itself. Context can be of assistance, to be sure. But as was pointed out in connection with the examples above, the context need not provide any insight into the speaker’s state before the utterance. It is certainly possible for the utterance itself to be the sole or primary basis for inferences about the speaker’s attitude and intentions. The above remarks highlight what I take to be the most serious problem of a speech-act approach to generalizing Quality: the risk of circularity. The potential for circularity lies in
16 Gunlogson
use; they are not simply variants of interrogative forms. Let us assume the question mark indicates a final rise, a common feature of such uses. The force of the objection depends on the validity of several assumptions about sentence type and intonation, including: (i) that all kinds of questioning are prima facie incompatible with all degrees of speaker belief; (ii) that rising intonation on a declarative functions simply to confer questionhood; (iii) that declarative ‘questions’ have the same status as interrogative ones. These assumptions are open to challenge; see Gunlogson (2003, 2008), Bartels (1997) for data and arguments. Without insisting on the adoption of a particular analysis, the point can at least be made that the mere existence of declarative questions does not constitute an objection to the sentence-type approach. There is much more that could be said about the differences between the two kinds of approaches sketched above, and the merits and demerits of each. In view of the circularity problem, I am inclined toward the sentence-type approach. It is not my goal in this section, however, to argue for the superiority of a sentence-type analysis over a speech-act approach, or vice versa. Instead I want to close by pointing out a larger problem, namely a certain disparity between the requirements for Quality arrived at in this section and the considerations that arose in connection with implicit content in Section 3.1. In broadening the scope of Quality beyond declaratives used as statements, it becomes necessary to distinguish between types of utterances, as we have seen. Quality becomes more general in the sense that it applies across more utterance types, however those types are defined. It is more specific, though, in the sense that there are multiple clauses of Quality that correspond to the different types. This subdividing on the basis of utterance type leads to some tension between the extension of Quality proposed here with the different sort of generalization discussed in Section 3.1, where it was argued that the domain of Quality should extend to implicitly contributed content. Such implicit content is most naturally conceived of as propositional. The appropriate clause of Quality to apply to implicit content would be the one applying to truth-conditional content, the clause for declaratives or statements (21a or 22a, respectively). But none of the revised and subdivided principles of Quality considered in this section extend to implicit contributions, which do not have a sentence type and do not constitute (not necessarily, at least) a type of speech act. Broadening 21a or 22a to refer to contributed propositions and not just statements or declaratives is an obvious remedy to try. A simple adjustment could be made to the first clause of 22, e.g., to address this, making reference to semantic representations rather than utterances:
(23) Be sincere. (Quality)
a. Contribute a proposition only if you believe it to be true and have evidence for it.
Uttering a statement or declarative would be the explicit way to contribute a proposition, just not the only way. The above proposal is not entirely satisfactory, however. The reason is that utterances of all types can carry presuppositions and give rise to implicatures. A non-propositional utterance would need to be subject to the appropriate clause for its type AND to have the propositional clause apply to any implicatures or other associated implicit content. This can no doubt be engineered, but it has an unsatisfying patchwork quality about it. More concretely, there is nothing here that appears to connect to other phenomena discussed, such as suspendability.
De-Maxim-izing Quality 17
4 Quality as a contextual condition
In Section 3.1.2, one rather simple possibility presented itself. Suppose that, instead of viewing Quality as governing only assertive utterances, we implement Quality as a more general rule applying to all (propositional) additions to the context, explicit or implicit. Assuming that implicatures ordinarily constitute additions, the rule will extend to them, as desired. In order to flesh out the implications of such a move, I will start by outlining what I take to be a fairly standard set of assumptions about conditions on assertive contributions in contemporary semantic theory and examine whether and how the proposed extension fits in. In formal semantics and pragmatics, assertion is routinely modeled in terms of addition to the Common Ground (CG), adapting ideas of Stalnaker (1978). The CG is a set of propositions representing what is mutually taken for granted by the discourse participants. In factual discourses the contents of the CG will correspond to what the participants take as the agreed-upon facts, and will also represent mutual beliefs. An assertion by a speaker is aimed at adding new information to the stock of mutual knowledge. Canonically, utterances of declaratives are taken to function as assertions, and all assertions are assumed to be utterances of declarative sentences, so that, in effect, a speaker uttering a declarative is understood as intending to add new information to the CG. Typical conditions on assertion are shown in 24. The speaker is expected to be:
(24) a. Informative/non-redundant: offering a proposition that (the speaker believes is) not already in the Common Ground b. Consistent: offering a proposition that (the speaker believes) does not contradict what’s already in the Common Ground c. Truthful: offering a proposition that the speaker believes is true and that the speaker has adequate evidence for
24a can be seen as a minimal requirement of Quantity. Quality makes its appearance in 24c, stipulating that the speaker must speak in accordance with his beliefs and evidence. The three conditions in 24 form a cohesive set from the speaker’s perspective, as in Grice’s statement of the maxims. All three can be understood as guidelines for the cooperative speaker. There is a difference between 24a–24b and 24c, however, that turns out to be important. 24a–24b make reference to the Common Ground, i.e., to mutual assumptions/beliefs, while 24c crucially refers to the speaker’s state only. A standard formulation of requirements 24a–24b in a possible-worlds framework is given in 25:
(25) Let c represent the context set, the set of worlds in which all propositions in the Common Ground are true, and let p be the propositional content of the assertive sentence uttere d, construed as a set of worlds. a. Informativeness requirement: c ∩ p , c b. Consistency requirement: c ∩ p , 0 /
The requirements in 25 are stated in terms of the Common Ground, with the simplifying assumption (often left implicit) that all participants have the same, and accurate, beliefs about the state of the discourse. Given this assumption, the parenthesized references to the speaker’s
De-Maxim-izing Quality 19
beliefs of the interlocutors. Quality should have the result of attributing belief to the contributor, i.e., the speaker, not indiscriminantly to all participants. For one thing, it is possible that not all participants will subscribe to the contributed content. For another, there is a fundamental distinction between being the contributor, the person in the discourse with whom the content originates, and simply accepting the content on someone else’s authority. While there may be some sort of Quality principle governing the latter role of accepting, it is not Grice’s Quality principle, which pertains to the speaker’s knowledge before the utterance, not as a result of it. One necessary step, then, is to disassemble the CG structure into component parts, each representing what one discourse participant is known to assume or believe. If we further stipulate that assumptions correspond to beliefs, then an operation adding a proposition to a participant’s set is effectively an addition to that participant’s publicly recognized beliefs. Presumably the proposition that the participant in question has the specified belief could uncontroversially pass into the CG, without the content of the belief itself following. Rather than trying to retrofit the Common Ground with the detail needed to distinguish speaker contributions from addressee acceptances and so on, a revision which takes it some distance from the original conception, we could contemplate starting with a more flexible set of assumptions, returning to Gazdar’s proposal for a model. His implementation borrows from Hamblin the notion of a ‘discourse commitment’ that is not necessarily a belief. A context is a consistent set of propositions; when associated with an individual participant, it is construed as that individual’s ‘commitment slate’, in Hamblin’s terminology. Commitments are not explicitly defined, but the important point at the moment is that they are not equivalent to beliefs. The additional step Gazdar takes to spell out the operation of Quality for assertion is to specify that the contributor of an assertion that p is also committed to knowing that p holds, or at least acting that way. Gazdar’s informal version of Quality is repeated below:
(27) For any declarative sentence φ , assertion of φ commits the speaker to Kφ.
Adopting this general strategy, the first task would be to generalize beyond the asserted content for attitude attribution. A goal in doing so is to link it to the operation of adding a proposition to a commitment slate, rather than to a specific type of contribution. This amendment has the potential to address the difficulties raised at the end of Section 3.2 involving implicit content added via non-assertive contributions. At the same time, making the attribution explicit allows for its implementation as a default operation, which in turn leaves open the possibility of accounting for the suspension phenomenon by the rather direct method of suspending the default. I believe the latter approach has much to recommend it, though there are significant issues that arise as well. I leave these to be addressed in future work.
5 Summing up
Summarizing, we have seen that the operation of Quality is distinguished in the following ways:
- Quality inferences are not cancelable; attempts to deny belief produce a Moore’s paradox effect
- Quality takes precedence; it cannot be traded off against the requirements of other maxims
- Quality inferences can be globally suspended in non-epistemic contexts, contrasting with the behavior of implicatures in general
20 Gunlogson
- Quality-like inferences arise for unuttered as well as uttered content, so that attribution of belief to a speaker making a statement will (in the default case) extend to implicatures and presuppositions of the statement
- Quality can be violated stealthily, without generating violation-related implicatures, unlike other maxim violations (real or apparent), which are inherently public
- Adherence to Quality must be evaluated with respect to the personal beliefs of individual discourse agents, unlike the other maxims, which can more plausibly be represented as contextual constraints
- Quality calls for truth-telling and is thus inapplicable on its face to non-assertive utterances, though inferences about speaker attitude arise with such utterances as well
These differences are presented in condensed form in Table 1. I do not wish to claim that the operation of the remaining maxims provides some clear and uniform standard that the behavior of Quality is measured against. However, with respect to the properties discussed, it is fair to group the three together as ‘other maxims’.
Property Quality Other maxims Inference cancelable No Yes Inference can lose in maxim clash No Yes Inference suspendable Yes No Inferences about implicit content Yes No Unobtrusive violation possible Yes No Evaluable against CG/context No Yes Crucially refers to truth Yes No
Table 1: Summary of differences
This catalogue of distinctive properties provides ample evidence that Quality is not just another, slightly aberrant, conversational maxim but a principle of another sort, deserving of attention in its own right. Together with the points raised in discussion, the list provides a set of desiderata for implementing a more general, ‘de-maximized’, version of the principle governing matters of quality.
References
Bartels, C. 1997. Towards a compositional interpretation of english statement and question intonation. Doctoral Dissertation, University of Massachusetts, Amherst. Gamut, L. T. F. 1991. Logic, language, and meaning. Chicago: University of Chicago Press. Gazdar, Gerald. 1979. Pragmatics : implicature, presupposition and logical form. New York: Academic Press. Gordon, David, and George Lakoff. 1975. Conversational postulates. In Syntax and semantics, MLA-IB, 83–106. Los Angeles, CA. Grice, H. P. 1989a. Further notes on logic and conversation. In Studies in the way of words, 41–57. Cambridge, Mass.: Harvard University Press. Grice, H. P. 1989b. Logic and conversation. In Studies in the way of words, 22–40. Cambridge, Mass.: Harvard University Press.
