Marina Sbisà

University of Trieste

 

The rationality of conversational implicature*

(draft, 2003)

 

 

1. From calculability to rationality

 

There is general agreement on the fact that conversational implicature is something “rational”. It is less clear what exactly this means. One explanation, all too obvious, is that conversational implicature is rational because is it inherently “calculable” and calculations require reasoning.

 

But if it is calculability that provides the connection between conversational implicature and rationality, this connection is not one devoid of philosophical consequences both about the kind of phenomenon conversational implicature is and about the way we should conceive of rationality. As I hope to show, Grice’s account of conversational implicature, albeit contentious, stands out among competing accounts of implicit meaning thanks to its author’s sensitivity to such themes. 

 

In this paper, I first focus on the calculability of conversational implicatures, comparing three different accounts of how a conversational implicature arises: Grice’s own (1989), the relevance-theoretic account (Sperber and Wilson 1995) and  Levinson’s heuristics-based account (2000). The relevance-theoretic account is not concerned with Gricean conversational implicature alone, but with any kind of inferential contributions to comprehension, which include those dealt with by Grice as conversational implicatures. Levinson’s account is limited to generalized conversational implicature, a sub-class (albeit a theoretically salient one) of Gricean conversational implicatures. It is therefore on the terrain of generalized conversational implicature that the three competing analyses can be compared. I propose some examples in which generalized conversational implicatures of the “scalar” kind arise (or fail to arise) and compare the accounts that the three competing approaches provide for each.

 

The results of this comparison open the way to a scrutiny of the background assumptions of the rival approaches as to both calculability and rationality. In conclusion, I sketch  the role that the conception of rationality I attribute to Grice should play in a theory of conversational implicature and more generally of text comprehension.

 

 

2. Three competing approaches to scalar implicature

 

Scalar implicatures are typically associated with the use of lexical items connected to one another by a scalar ordering. “Some” and “all” are two such lexical items. It has often been observed that the utterance of a sentence such as “Some guests have gone home” suggests more or less strongly that not all of the guests have gone home. Scalar implicatures may come “naked” and therefore as information about the world as in this example, but may also come as attributions of an epistemic position to the speaker; like all implicatures, in certain situations or connections they can be cancelled or just fail to arise. To illustrate the limits and main varieties of scalar implicature, I propose the following examples: 

 

(1)       World-oriented scalar implicature

[Is the party still going on?]

Some guests have gone home.

---> Not all of the guests have gone home.

 

(2)       Speaker-oriented scalar implicature

[Politicians are often corrupt]

Some politicians are honest.

---> The speaker does not know whether all politicians are honest.

 

(3)       No scalar implicature

[If any of your neighbours has pets, you should not use that pesticide in your garden.]

Some of our neighbours have pets.

---> 0

 

Actual situation of utterance need not come directly into play in the arising or not arising of implicatures associated with the use of “some” (henceforth, “some”-implicatures). The sentences in square brackets in the examples above are meant as possible previous conversational turns and characterize the kind of situation in which the sentence containing “some” (or “some”-sentence) is used, in fact in each case a typical situation for the use of that “some”-sentence. As to (1), if we want to hear whether a certain party planned for that very day and at which we are not present is still going on, we are interested in hearing whether all of the guests or most or only few, or none, are there and at the same time we may assume that whoever knows about the ongoing party knows about that. As to (2), in discussion about the honesty of politicians the participants’ knowledge is likely to be an issue: does the speaker who defends politicians know about all the relevant politicians?1 As to (3), if we tell somebody about our neighbours having pets, it is difficult to imagine why the audience should be interested in whether all of them have pets or only some (consider, moreover, that neighbours typically form a fuzzy set); rather, the contrast at issue is likely to be one between there being or not being any pets in the area. So, in all our three cases the very linguistic material in the “some”-sentence orients the attention of the receiver towards situations of certain kinds. I maintain that the contribution of other words in the “some”-sentence to the arising of the “some”-implicature is not enough to make it a “particularized” implicature, i.e. an implicature relying on contextual information. So, we may consider the implicatures in (1) and (2) as generalized conversational implicatures and the example (3) as an example in which no such implicature arises.

 

I now briefly review the three approaches to generalized conversational implicature I am going to compare.

 

According to Grice (1989), a generalized conversational implicature can be derived if we assume that the speaker is cooperative and that he or she, perhaps contrary to appearances, follows the Maxims of Quantity, Quality, Relation and Manner. From the observation of the linguistic form of the utterance and this assumption, the audience can reason to the implicature without invoking contextual features specific of the current conversational situation. Grice proposes a “general pattern for the working out of a conversational implicature” that I will apply to our examples (1)-(3) below.

 

According to Relevance Theory, henceforth RT (Sperber and Wilson 1995), an implicature is derived as the result of automatic processing governed by the Relevance Principle “Every act of ostensive communication communicates a presumption of its own optimal relevance”. In turn, “optimal relevance” means that the ostensive stimulus is relevant enough to make it worth the addressee’s effort to process it and the most relevant one compatible with the communicator’s abilities and preferences. There is no distinction between generalized and particularized implicature, since the steps through which the implicature is arrived at always include premisses from background knowledge.

 

According to Levinsons’s theory of Generalized Conversational Implicature (GCIT) (2000), generalized conversational implicatures are default inferences we arrive at thanks to some heuristics to which speaker and hearer are mutually oriented. Such heuristics can be spelled out in terms of a “Speaker’s maxim” guiding speech production and a “Recipient’s corollary” guiding speech comprehension. Resort to the heuristics is motivated by the need to convey more information in a given stretch of time than the linear production/reception of speech allows for. One source of implicature cancellation is competition among different heuristics.

 

 

2.1. Discussion of three examples

Let us now consider how the three approaches deal with our examples.

 

2.1.1. Grice

 

In a Gricean perspective, scalar implicatures are Quantity implicatures, that is, they depend on the assumption that the speaker (contrary to appearances) does follow the first maxim of Quantity “Give as much information as is required”. In our first example:

 

(1) [Is the party still going on?]

Some guests have gone home.

---> Not all of the guests have gone home.

 

a Grice-inspired reasoning that permits the audience to derive the implicature (cp. 1989:31) should run as follows:

 

 

In our second case

 

(2)       [Politicians are often corrupt]

Some politicians are honest.

---> The speaker does not know whether all politicians are honest.

 

the Gricean pattern suggests the following path:

 

As to our third case

 

(3)       [If any of your neighbours has pets, you should not use that pesticide in your garden.]

Some of our neighbours have pets.

 

the reasoning does not go through, because the third step:

 

 

is not to be taken. In fact, the pertinent contrast that the some-sentence in (3) is about is just that between “no pets” and “some pets”, therefore its utterance perfectly fits the Quantity requirements.

 

 

2.1.2. Relevance Theory

According to RT, the steps through which the implicature in our first example

 

(1) [Is the party still going on?]

Some guests have gone home.

---> Not all of the guests have gone home.

 

is worked out are the following:

 

In our second case

 

(2)       [Politicians are often corrupt]

Some politicians are honest.

---> The speaker does not know whether all politicians are honest.

 

the processing might run as follows:

 

 

As to our third case

 

(3)       [If any of your neighbours has pets, you should not use that pesticide in your garden.]

Some of our neighbours have pets.

---> 0

 

Sperber and Wilson (1995: 277) deal with it by saying that the fact that some of the speaker’s neighbours have pets is relevant enough and therefore there is no reason to assume he or she meant that not all of them do. But being “relevant enough” is not enough to establish optimal relevance, if the latter also requires the stimulus to be “the most relevant stimulus compatible with the speaker’s abilities and preferences”. If we want to apply the same inferential route as above, we should say:

 

 

 

2.1.3. GCI Theory

 

GCIT proposes to view generalized conversational implicatures as default inferences we perform in following some heuristics based on general human rationality. One of these heuristics, the Q-principle, applies when we are faced with the utterance of a sentence which is weaker than its linguistic alternates. The Q-Principle comprises a speaker’s maxim, “Do not provide a statement that is informationally weaker than your knowledge of the world allows” or “Select the informationally strongest paradigmatic alternate that is consistent with the facts” (Levinson 2000: 76) and an addressee’s corollary. More neutrally phrased as a heuristic to which both speaker and hearer are mutually oriented, it might sound “What isn’t said, isn’t” (Levinson 2000: 31) or “What you do not say is not the case”(Levinson 2000: 36).  In our first case:

 

(1)       [Is the party still going on?]

Some guests have gone home.

---> Not all of the guests have gone home.

 

the steps that lead to the default conclusion (and thus to the implicature) are:

 

 

As to our second case:

 

(2)       [Politicians are often corrupt]

Some politicians are honest.

---> The speaker does not know whether all politicians are honest.

 

the steps (apart from the first one, which I do not repeat) should be the following:

 

 

But this runs contrary to our intuition that the speaker, although willing to defend politicians insofar as possible, is not in an epistemic position to assert “All politicians are honest”. In consideration of the discussion of similar issues by Levinson himself,  it is fair to complicate the Recipient’s corollary to the Q principle as follows:

 

 

Thus we might choose to get:

 

 

In our third case:

 

(3)       [If any of your neighbours has pets, you should not use that pesticide in your garden.]

Some of our neighbours have pets.

---> 0

 

I am afraid that GCIT has no way to hinder the implicature from arising, possibly with the vaguest epistemic modification conceivable, such as “The speaker is not in an epistemic position to assert that all of their neighbours have pets”.

 

 

2.2. How each approach fares

Do all the three approaches consistently cope with all of our three examples? Do the inferential paths they propose smoothly and convincingly lead to the intuitively grasped implicatures (or the absence thereof)?

 

 

2.2.1. Grice

 

The Gricean reasoning about (1) may raise various perplexities, two of which are relevant here:

 

 

As to the former perplexity, it should be noted that to conclude that the speaker has implicated that q is not, in the Gricean framework, very different from saying that the hearer, in addition to grasping the overt message that p, grasps also the implicature that q. But this would amount to saying that, if q itself says something about the world, the hearer gathers information about the world, rather than about the speaker. 

The second perplexity is more difficult to dispel. Grice’s general pattern for the working out of a conversational implicature does not in fact explain why the receiver should think that what the speaker should be supposed to think is exactly that q (rather than q1, q2, q3, qn). The connection between p and q is at best mediated by the maxim invoked: in the case of a scalar implicature, is it something that would complete the content of the overt message that p, so as to make the utterance give as much information as required and thus cope with the first maxim of Quantity.

Apart from the problem of how the implicated assumption is selected, the Gricean pattern seems to work well, is easily transferred from example (1) to (2) and provides an explanation of the absence of the implicature in example (3). Case (2) is easily treated by modifying the second step of the pattern: the default assumption that the speaker is following both the Cooperative Principle and its maxims (including the first maxim of Quantity) is replaced by the report of an anomaly, that is, the speaker’s not following the maxim of Quantity. The implicature suggested rescues the speaker’s cooperativity because the first maxim of Quantity is violated only in order to follow another maxim (the second maxim of Quality). Once the implicated assumption is identified, the steps that enable us to impute the implicature to the speaker are the same as in case (1). Case (3) is immediately rejected as a possible case of conversational implicature because no hidden condition is needed in order to make the speaker’s utterance compatible with his or her following the first maxim of Quantity.

 

2.2.2. Relevance Theory

 

In RT (according to the revised Presumption of Optimal Relevance, Sperber and Wilson 1995: 270), the audience’s beliefs about the speaker’s preferences and knowledge are a necessary ingredient of the processing of the utterance. All implicatures are therefore “particularized”, that is, derived from premisses coming from the current particular context. This is not per se a defect of the inferential path through which the implicature is derived. But in the case of certain scalar implicatures at least, resort to beliefs about the speaker’s preferences and knowledge may make the inferential path useless.

 

Another problem for the RT approach comes from the claim that the ostensive stimulus is understood as “the most relevant one” compatible with the speaker’s preferences and abilities. For a stimulus to be relevant, it must offer a good balance between amount of cognitive contextual effects and processing effort. For it to be “the most relevant one” in a certain set, it must offer the best balance, that is, either more cognitive contextual effects than any other at the same cost, or the same amount of contextual effects as some other stimulus at a lower cost. This suggests that alternates to a certain ostensive stimulus should be compared with it as to amount of contextual effects and processing effort. But exactly which alternates are claimed to be taken into consideration? Making consistent choices on this across examples (1) and (2) generates defects either in account of the former or in that of the latter.

 

In case (1), one might take it that not only “All the guests have gone home” but also “Not all guests have gone home” would have a greater amount of contextual effects than “Some guests have gone home” (and processing costs would not be substantively higher). If we consider this comparison as one that the efficient processor would make, the only admissible implicature for “Some guests have gone home” becomes “The speaker does not know whether all the guests have gone home”. But this does not correspond to intuitions. We get the intuitive result if we do not take it to be necessary to compare “Some guests have gone home” with “Not all guests have gone home”. But if having fixed this, we make a consistent choice in case (2) and take it that the efficient processor would compare “Some politicians are honest” only with “All politicians are honest”, it is the account of (2) that gets into trouble. There is no way to decide between the two disjuncts arrived at in the last-but-one step of the inferential path (“If the speaker lacks the ability to state “All politicians are honest”, this may be due to two different causes: either the speaker knows that this is not the case, or the speaker does not know whether this is the case”), apart from resort to background knowledge about the speaker’s state of knowledge. But then the assumption that the speaker does not know whether all politicians are honest is not arrived at as the final step of some inferential processing, rather, is retrieved from background knowledge directly.

 

In case (3), I find it really difficult to explain why the processing of the “some”-sentence does not yield any “some”-implicature. In fact, the first premiss (invoking optimal relevance) exactly corresponds to those used in cases (1) and (2), so where is the difference? The second premiss (“In the current context, “All of our neighbours have pets” would not yield more contextual effects than “Some of our neighbours have pets” does”) runs opposite to those in cases (1) and (2) (and thus makes the difference), but why should that be the case? Note that in order to explain the utterance’s optimal relevance not on the basis of abilities and preferences (thus generating an implicature) but on the basis of contextual effects, we should perhaps go so far as to admit that “All of our neighbours have pets” would yield less contextual effects than “Some of our neighbours have pets” in the current context, or that it would require greater processing effort. But it is doubtful that either of these conditions obtains.

 

 

2.2.3. GCI Theory

 

The main merit of GCIT is that it explicitly states a language-grounded reason for the implicatures arising in cases (1) and (2):

It is the connection between “some” and “all” in the lexical structure of the language that creates the effect of non-compliance with the rules or principles of conversational cooperation and explains why the implicature will have a certain content. However, the GCIT account of scalar implicatures suffers also from grave flaws.

 

In the account of example (1), the “naked” implicature “Not all of the guests have gone home” is arrived at through the attribution to the speaker of the knowledge that “All the guests have gone home” would be false. This is puzzling. The step “The speaker knows that not all of the guests have gone” is directly licensed by the Recipient’s corollary of the Q-principle, so that its soundness rests on whether that corollary is well grounded in its own turn. But the corresponding step in case (2) does not work at all: there, in order to reach the intuitively correct conclusion (the implicature “The speaker does not know whether all politicians are honest”), we simply cannot include “The speaker knows that not all politicians are honest” among our inferential steps. So perhaps the Recipient’s corollary of the Q-principle should be revised. In order to cope with this problem we have proposed a modification of the Recipient’s corollary partly inspired by Levinson himself, but this would only work in conjunction with criteria for deciding which one of the disjuncts that the new formulation of the corollary licenses the audience to derive is to be actually inferred in each particular occasion. But no such criteria are specified by Levinson. He does not explain what factors make an implicature vary from the “naked” form q to  “The speaker does not know whether q” or “For all the speaker knows, q”, etc. Rather, he avoids making a decision as regards what he calls the “epistemic modification” on implicatures. This is odd, because he seems to maintain that the basic form of implicatures is the “naked” one, while his Q-principle under no formulation yields it without going through one of the “modified” forms.

 

Levinson (2000: 78n) cites a slightly different explanation of scalar implicatures by Atlas, according to which the series of inferences to be drawn goes as follows

 

where the additional premiss “Not A (all) is consistent with what the speaker knows” covers all the disjuncts in our reformulation of the Recipient’s corollary to the Q-principle. But even here the conclusion “Not A (all)” fails short to be guaranteed, since it is not clear why the mere consistency of a sentence with the speaker’s state of ignorance should be taken as licensing an inference to that very sentence. Maybe we should assume, in addition, that the speaker intends that not A(all) should be recognized to be consistent with what he or she knows; but even so, the conclusion remains a mere suggestion.

 

As to case (3), we have already seen that GCIT seems to be at a loss with it. A way out might be to consider the implicature as pre-empted by some background assumption, such as that there is no interest in whether all neighbours have pets or in the epistemic position of the speaker about this.

But such an assumption does not fit any one of the standard forms for canceling or suspending phrases (Levinson 2000: 81). Instead, we might say that according to this approach, the scalar implicature is still available (perhaps in its vaguest form “The speaker is not in an epistemic position to assert A (all)”), while no participant is interested in focusing attention on it.

 

 

2.3. Why does Grice fare best?

 

From this discussion, the following provisional conclusions can be drawn:

 

 

 

Perhaps these conclusions should not come as a surprise. For, let us stop for one moment and consider by what standards or with what an ideal in mind the three competing analyses of conversational implicature have been examined. We have focused on how implicature is calculated according to the three approaches.

 

Maybe the fact that Grice fares best depends on the particular way, not shared by the other two approaches examined, in which he conceives of the role of these calculations.

 

Differences between Grice’s way of working out conversational implicatures and RT’s or GCIT’s use of inferential steps in the modeling of utterance processing have perhaps their root in the contrast between the psychological and normative dimensions of language use. But to get clearer about this, we have to further examine how these three competing approaches  conceive of calculability and furthermore, what background assumptions about rationality they rely upon.

 

 

3. What is calculability?

 

It may seem that the calculability of conversational implicature is of some use for a theory of linguistic communication or of text comprehension only if the inferential path it makes available also constitutes or models the way in which the implicature is understood (and intended to be understood). It seems therefore sensible to assume that the claim that conversational implicatures are calculable entails certain assumptions about the psychology of speaker and hearer, namely that, in order to get from an utterance to its conversational implicature, they actually follow an inferential path of the kind suggested. Thus the calculability requirement generates hypotheses about actual calculations.

 

However, this way of dealing with calculability is not fully satisfactory. Calculability does not entail actual calculation by the hearer. Also the claim that calculations must actually be implemented in the mind of the speaker, if he or she really implicates something, raises doubts. I argue that it makes sense to consider calculability as a requirement for conversational implicature without assuming it that the psychological reality of meaning production and understanding needs to be characterized by the corresponding calculations.

 

 

3.1. Does calculability require actual calculations in the hearer’s mind?

 

The view of calculability as entailing the occurring of actual calculations in the hearer’s mind suffers from a basic ambiguity. We may mean by “hearer” just any hearer, or a hearer who understands the implicature correctly and completely. In both cases the reduction of calculability to calculation yields counter-intuitive or undesirable results.

 

If the focus is on actual utterance processing, calculation by any actual hearer should be equally relevant. But do all actual hearers, faced with an utterance that suggests a conversational implicature, regularly perform implicature calculation? Of course not. Sometimes they do not even understand the implicature, in other cases they grasp it but there is no evidence of how they came to do so. Jennifer Saul has recently argued that a certain assumption can be conversationally implicated even if it is not actually recognized as such by the hearer (Saul 2002a: 243). This fits well empirical evidence from text comprehension. In my research on the comprehension of textbooks, during which various comprehension tests were administered to adults and kids2, questions aimed at eliciting the working out of a conversational implicature were answered correctly and completely only by a minority of the subjects. But is this good reason to conclude that the text whose comprehension is tested does not carry any conversational implicature? Clearly, it is not. Rather, replies that fail to work out the implicatures that the text makes available to its readers should be counted as (to varying degrees) wrong. Of course, particularly in the case of written communication (where the author is typically out of reach), there might be discussion as to whether a text does or does not carry a certain conversational implicature. But the very fact that we may be right or wrong about this shows that an utterance’s carrying a certain conversational implicature does not depend on the implicature’s being actually calculated. (Failures to understand implicatures remain data to be explained, but I will not tackle this problem here).

 

If the focus is on the successful processing of an utterance, that is, on the processing that leads the competent hearer to correct and complete understanding, the claim that the inferential path leading to the implicature is actually implemented only concerns those hearers who understand the implicature correctly and completely. But in this case too we run into problems. If conversational implicatures are such, that the hearers who understand them do so by actually going through a certain inferential path, then before characterizing a certain implicature as conversational, we should have evidence about the way in which hearers work it out. But how can such evidence be acquired?  Once we see that a hearer has grasped the implicature, should we ask him or her how he or she got it and rely on his or her reply? People are often able to point at those features of an utterance that suggest a certain implicature to them, but (to wit) the Gricean inferential path is only explicitly laid down by Grice. Introspection does not provide evidence in support of calculability (so intended), or perhaps is not the right source for evidence for or against it. In fact, the approaches that assume that conversational implicatures are actually calculated through a certain inferential path collocate the relevant inferential processing at the sub-conscious or even sub-personal level. We need not be aware of applying Levinson’s heuristics. In Relevance Theory, utterance processing is carried out by the automatical functioning of our mind/brain, which we do not necessarily introspect. So perhaps support for the claim that conversational implicatures, when grasped, are arrived at through a certain inferential path can only be provided by experimental research,  such as research on processing times. But the findings of such research, even when they show that utterances requiring conversational implicature have a longer processing, do not by themselves decide the issue. They are quantitative and cannot tell anything about the details of the supposed processing. Moreover, the fact that the processing of certain utterances takes a longer time might be explained in different ways and it is still an hypothesis that its explanation lies in the implementation of an inferential path.

 

It may also be doubted that the claim that those hearers who understand a conversational implicature always arrive at it through a certain inferential path, can count as an empirical hypothesis. In order to make empirical hypotheses about such hearers, we should first be able to single them out, which is far from trivial. If we rely on our own judgement about whether a certain utterance carries a certain implicature, the hearers who understands the implicature correctly and completely are just those who agree with our judgement. If we invoke some norm or rule, aiming at the objective correctness of understanding, our way of identifying the hearer who understands cannot fail to include a normative element. Finally, if we say that the relevant hearers are those who approximate to the ideal of the rational hearer, we fall into a circle. If we claim that the rational hearer would derive a a certain implicature by following a certain inferential path, this amounts to assuming that the suggested inferential path is rational, and anything not conforming to such an inferential path would not count as the understanding of a conversational implicature by a rational hearer.

 

So, if we consider actual hearers, we have to face the gap between conversational implicature and its actual recognition and calculation: in many cases it is simply false that actual hearers go through the inferential steps designed for the implicature at issue. But if we limit consideration to those hearers who succeed in grasping the implicature, the best we can do is assimilating them to ideal, “rational” hearers and this move destroys the empirical character of the claim that conversational implicature is actually understood through a certain inferential path.

 

Among the three approaches to conversational implicature we have been considering, only Grice’s avoids getting entangled in these problems. As Saul has correctly noticed (2002a), the actual working out of the implicature by the hearer is not included in the conditions for conversational implicature laid down by Grice (1989: 30-31). Moreover, Grice (1989: 31) grants that conversational implicature may well be grasped intuitively. For an implicature to be calculable and therefore conversational, he only requires that the intuition be "replaceable by an argument".

 

 

3.2. Does calculability require actual calculations in the speaker’s mind?

 

Insofar as RT and GCIT theory are concerned, the role of the speaker mirrors that of the hearer. The same inferential steps through which the hearer will go in order to understand the implicatures of an utterance, also play a role in speech production: they are part of the mental equipment and activities both of hearer and speaker. However, there are circumstances in which it is not sensible to take the actual speaker to contemplate the content of his or her implicature in advance and consider, or even plan, the inferential path leading to it.

 

So, in RT, at least in the case of "weak implicatures" it is admitted that the speaker need not always intend all implicatures of his or her utterance individually.  In weak implicature, the speaker communicates any one among a certain set of assumptions, leaving the hearer free as to which one he or she chooses to pick up (Sperber and Wilson 1995:  195-202). By granting this, however, RT does not give up its main assumptions about communication, whose main point remains to make the hearer reflect the speaker’s thoughts: only, weak implicature is deemed to be an imperfect form of communication, since the weaker the implicature, the less confidence the hearer can have that in working it out he or she reflects the speaker’s thoughts. Moreover, doubts may arise about how consistent this picture of weak implicature is with respect to the whole RT framework. It seems to me that if the speaker intends to communicate the presumption of his or her utterance’s optimal relevance (as required by the RT approach) and is sincere, he or she must believe his or her utterance to be optimally relevant. But where can such a belief come from? Can it arise at all, without the speaker having any idea of the cognitive effects his or her utterance is likely to produce (which would involve at least some actual calculations)? If, instead, the speaker is not credited with any detailed awareness of the cognitive effects of his or her utterance, his or her belief that his or her utterance is optimally relevant (or will appear as such to the hearer) can only arise from the automatical functioning of the Relevance Principle in his or her mind: but in this case, calculations must be supposed  to be performed at the sub-conscious or sub-personal level.

 

In GCIT, the role of the speaker in planning the hearer’s default inferences is ambiguous between the conscious expectation that the hearer will go through certain inferential steps and the unreflective acceptance of default procedures. Basically, the speaker is supposed to reason like this: “If the speech situation is normal, and I say that p, the hearer, who (like myself) relies on certain heuristics, will infer that q”. But it is counterintuitive to assume that speakers explicitly take into consideration the fact that the speech situation is normal; in fact, one of the points of introducing default procedures is to make such explicit consideration unnecessary. So, it is counterintuitive to  attribute to the speaker the full-fledged calculation of the possible implicatures of his or her utterance. Such full-fledged planning of conversational implicature may either be sub-conscious or pertain to the ideal, rational speaker. In the latter case, however, it is a normative idealization as opposed to a psychological requirement.

 

Also Grice, while not concerned with the psychological processes on behalf of the hearer, is perhaps committed to require of the speaker that he or she actually calculate the implicature of his or her own utterance. Indeed, Grice does not neglect the psychological dimension of conversational implicature. He mentions at least one psychological requirement on behalf of the speaker among the basic conditions an implicature should satisfy in order to be a conversational one: the speaker must think (and expect the hearer to think that the speaker thinks) that it is within the competence of the hearer to understand the implicature (1989: 31). Saul has argued that this condition on conversational implicature keeps Gricean calculability from being reducible to calculation by the hearer: Grice plays the speaker’s psychological dimension against the hearer’s and limits the authority of the audience on conversational implicature (Saul 2002a). But the same condition also suggests that Gricean calculability, even if it does not impose psychological requirements on the hearer, imposes them on the speaker. If the speaker must think that the hearer is capable to understand that the supposition that the speaker thinks that q is required, it is reasonable to conclude that the speaker must also be aware that the supposition that he or she thinks that q is required, and therefore both of the inferential path and its conclusion. This makes Grice’s account of calculability vulnerable to much the same problems illustrated above with reference to other approaches.

 

This notwithstanding, I believe that a Grice-inspired view of conversational implicature is not necessarily committed to the idea that conversational implicature requires actual calculations to be implemented in the speaker’s mind. In fact, it is still open what the speaker has to do exactly and in what condition he or she has to be, in order to count as "thinking" what he or she is required to think. If we take it that the speaker has to entertain an occurrent, activated thought, then we are committed to the view that the speaker has to entertain the implicated assumption in his or her mind and be aware of the corresponding inferential path. However, we might take it that the speaker, in order to "think" what he or she is required to think of the addressee's ability to understand the implicature, must merely merit to be attributed that thought: and this is regularly the case if his or her general attitude towards the addressee is of the kind one has towards a fellow rational being. This reading of Grice's third condition for an implicature to be conversational avoids commitment to claims about the psychological implementation of calculations in the speaker's mind.3 

 

It could be objected that this reading is illegitimate, because grounded on a disputable view of attitude attribution and in any case, one not shared by Grice.  I cannot go into the details of the issue of attitude attribution, which would lead us far away.4 Be this as it may, I believe that in conversational implicature, the unavoidability of going through reference to the speaker’s subjectivity depends on something subtler than the psychological reading of calculability as applied to the speaker. Conversational implicatures do not arise unless the hearer considers the speaker as a thinking being, capable of and keen to cooperation, who considers his or her hearer as a thinking being capable of and keen to cooperation. There must be an interpersonal relationship in the background and the recognition or acceptance of each other as subjects. In fact, whenever it is clear enought that a certain implicature would be required in order for the speaker to count as cooperative, but the audience lacks clear indications that the speaker is actively aware of the implicated assumption and intends to convey it, the decision of dealing with the speaker as conversationally implicating that q  rather than merely displaying or betraying his or her belief or willingness to believe that q is a delicate one that involves the overall relationship between audience and speaker. We might not be able to impute a conversational implicature to a speaker without assimilating him or her to an ideal, rational speaker (addressing to ideal, rational partners). In this connection too we are introducing a normative element into the picture.

 

 

3.3. The normative role of calculability

 

If the calculability of conversational implicature is not a matter of actual psychological processes, what is its role?

 

As Saul (2002a: 245) has felicitously remarked, a speaker who conversationally implicates something makes the implicated assumption available to his or her hearers and is responsible for making it available. As we have seen, this does not entail that the hearer always gets the implicature right or that the speaker is always fully aware of his or her own implicature; a fortiori it does not entail that the corresponding calculations are always actually implemented in their minds. But if an utterance’s conversational implicature does not depend on what and how the hearer understands nor on what and how the speaker is aware of, this means that there is a normative factor coming into play, something that permits us to say that the hearer or speaker should grant or admit that a certain utterance carries a certain implicature and that this implicature should be retrieved by going through a certain inferential path. This amounts to saying that Grice's theory can or perhaps should be read as concerning the normative dimension of language use rather than its psychological dimension. In this perspective, the calculability of conversational implicature has no longer the role of outlining the way in which implicatures are actually understood, but that of guaranteeing for the availability of an argument in support of the attribution of a certain implicature to the speaker. That an implicature is calculable just means that the attribution of that implicature to the speaker who has issued a certain utterance can be supported by argument. This has direct consequences on the issue of the correctness of implicature attribution.

 

In observing conversational practice as well as the practice of reading, it is often difficult to distinguish conversational implicatures from mere psychological associations, fanciful assumptions of the hearer’s that the hearer projects on an utterance. The calculability requirement, non-psychologically understood, offers us a criterion: conversational implicature attribution must be supportable by argument, and by an argument following an inferential path of a certain kind. So the calculability requirement, non-psychologically understood, provides us with a normative means for monitoring text comprehension (insofar as conversational implicature is concerned) and in general for improving our ability in grasping and making explicit that which a text does not explicitly say but makes available to its audience.

 

 

4. Two ways of being rational

 

We have just seen that calculability, independently of any claim about the actual implementation of the corresponding calculations in the minds of speaker or hearer, has a point thanks to its role in guaranteeing that the attribution of a conversational implicature to a speaker can be supported by argument. It is tempting, and would fit very well some ideas of Grice’s, to say that what calculability thus guarantees is the rationality of conversational implicature.

 

Grice deals with rationality in connection with argumentation in his Conception of Value (1991) and much he says in the first two chapters of Aspects of Reason (2001) further confirms the central role he assigns to that connection. A conception of rationality can be extracted from Grice’s work that I propose to call “argumentative rationality”. In my opinion, argumentative rationality provides the background against which the non-psychological calculability of conversational implicature makes sense.

 

RT has turned the calculability requirement into a claim, or set of claims, about actual psychological processes in the minds of speaker and hearer. Also GCIT endorses claims of the same kind. We have seen that such claims are not without problems. But the reason why, difficulties notwithstanding, these approaches are bound to the psychological reading of calculability and tend to understate the problems I have reviewed in § 3.1-3.2, lies in their background assumptions about rationality. I call the conception of rationality underlying these approaches “instrumental rationality”, because it emphasizes the connection of rationality with the efficacy and optimization of means-ends relations.

 

I turn now to an outline and brief comparison of these two conceptions of rationality.

 

 

4.1. Argumentative rationality

 

The conception of rationality that emerges from Grice (1991; 2001) can be summarized as follows: rationality is a concern that one’s moves are justified and a capacity (to some degree) to give effect to that concern. Since justification consists of one or more assertions that support a certain conclusion or decision, and citing an assertion in support of a conclusion is the basic move underlying arguments, I believe that this conception of rationality merits to be called “argumentative”.  

 

Argumentative rationality plays an important role in Grice’s defense of absolute value (1991: 80-91, 114-120; cf. Sbisà 2001). Value can be absolute, according to Grice, insofar as there are essentially rational beings, whom he calls “persons”, who have absolute value and can attach absolute value to what they evaluate. Humans turn themselves into such essentially rational beings when they consider rationality, which they possess contingently and accidentally, as their essential property. But what is, according to Grice, an essentially rational being? He or she is just a being whose essential role is giving justifications. Thus the Gricean notion of an essentially rational being presupposes a high rating of the concern and capacity to give justifications. Not only it is something worth becoming the essential feature of a new kind of beings, persons, but also, persons have absolute value just because they possess argumentative rationality essentially. It should also be noted that according to Grice, the demand for absolute value is itself “rational”. In fact, it is raised by the concern of objectively validating one’s acceptances and attitudes, which is in turn (together with the corresponding capacity) integral to argumentative rationality. So, argumentative rationality plays the role of the ultimate source of value.

 

If we consider argumentative rationality as a contingent property of humans (or perhaps other beings too), we should say that somebody who refuses to provide a justification for one move of his or her, or fails to provide appropriate reasons or evidence in support of one of his or her claims, is not, on that occasion, rational. Argumentative rationality, however, admits of borderline cases, in which for example there is the concern for giving justifications but the justification provided is under some respect flawed (cf. Grice 2001).  But if we accept Grice’s definition of personhood, it becomes possible to consider all humans as essentially rational, even in those circumstances or stages of development at which argumentative rationality is not contingently present: the justification of one’s moves remains a task for every human being who wants to take him or herself seriously as a person.

 

If we take the rationality of conversational implicatures to be “argumentative”, we can in a new way make sense of Grice’s declared preference for implicature assignments based on conversational as opposed to conventional implicature. Only conversational implicature is calculable and therefore inherently rational in the argumentative sense. Explaining a certain part of the meaning of an utterance as a  conversational implicature, rather than as part of what is said or conventionally implicated, requires that there be an inferential path leading to that implicature and thus an argument in support of it.

 

So, on the one hand, Grice’s Modified Occam’s razor (1989: 47) applies to conventional meanings not as a mere matter of achieving theoretical economy, but as a matter of preferring argumentatively justifiable meaning assignments whenever possible. On the other hand, philosophers who find it comfortable for their semantic theories to consider part of the meaning conveyed by an utterance as consisting of conversational implicatures and therefore a matter of pragmatics, are implicitly admonished thay should do so only if there is a clear inferential path (possibly one including the cooperative principle or a maxim of conversation among its premisses) that leads to the desired conclusion.

 

 

4.2. Instrumental rationality

 

Also those approaches to implicature that make the whole point of calculability reside in modeling the actual inferential path of speaker or hearer, take it that conversational implicature is a “rational” matter. Under such approaches, conversational implicature has been credited with “rationality” for reasons that do not coincide with its mere calculability, and particularly, with calculability as the mere availability of an argument in support of implicature assignment. The conception of rationality which  these approaches to implicature rely upon can be traced back to what I call "instrumental rationality".

 

Instrumental rationality is very likely the most widespread conception (or perhaps family of conceptions) of rationality. According to it, a course of behavior is rational if it is characterized by the agent’s non-accidental use of effective means, or of means believed to be effective, for achieving his or her goals. Such rationality is thus typically concerned with means-ends relations.

 

Grice too has relied upon the instrumental conception of rationality in his attempt to qualify implicature as rational by estabilishing the rationality of the Cooperative Principle.  He says that he would like to show that the Cooperative Principle is something that it is rational for speakers to follow, in consideration of the fact that “anyone who cares about the goals that are central to conversation/communication... must be expected to have an interest... in participation in talk exchanges that will be profitable only on the assumption that they are conducted in general accordance with the CP...” (1989: 30). Relative to the goal of participating in profitable talk exchanges, it is rational for a speaker to assume that the talk exchange he or she is participating in is conducted in general accordance with the Cooperative Principle, because this very assumption is a necessary conditions for the talk exchange to be profitable. Grice, however, adds that he doubts this conclusion and does not say why. Indeed, he seems to have given up supporting the rationality of conversational implicature along these lines.

 

RT very clearly relies upon instrumental rationality. According to it, human cognition tends not merely to sufficient results, but to optimal ones and these have to be achieved by optimizing relevance, which is itself a balance of costs and benefits and therefore a matter of instrumental rationality. The addressee’s aim in interpreting an utterance is to identify the speaker's intention and this is done, as with any attribution of an intention to an agent, by observing the means the speaker chooses and assuming that these are appropriate to his or her goals, given his or her beliefs (Sperber and Wilson 1995: 271). So, the processing of an utterance by the addressee is both an exercise of instrumental rationality and an attribution of instrumental rationality to the speaker as an individual (or organism). The speaker in his or her own turn speaks in order to get the hearer identify his or her intention and intends to use means effective to this aim, in particular, chooses stimuli he or she deems to be capable to appear relevant enough to the addressee to be appropriately processed. Such rationality does not require self-awareness: Sperber and Wilson (1995: 261-262) remark that the principle of cost-benefit optimisation generally holds in the sphere of biology. Nor is it a matter of choice and justification: the Relevance Principle is supposed to function automatically. The gap between the instrumental rationality at work here and argumentative rationality becomes perhaps more apparent if we consider that the Relevance Principle as a psychological mechanism extends also where “argumentation” cannot reach, that is, before the proposition expressed by an utterance (its "explicature") is identified (Sperber and Wilson 1995: 182-193; Carston 1988). It is no chance that in this connection the language used is rather that of "pragmatic processes" than that of reasoning and argumentation.

 

Levinson gives a slightly different characterization of the rationality of conversational implicature. He claims that language use suffers from a "bottleneck" problem (2000: 6): we most often need to convey more information in a given stretch of time than the linear production/reception of speech allows for. So for him too, conversational implicature responds to a problem of optimization: how to convey all that one needs or wants in a reasonably small amount of time. This problem generates resort to “heuristics”, which are, in turn, shortcuts to achieving rich communication by relatively poor linguistic means. In order to appreciate the gap between Grice’s argumentative rationality and the instrumental rationality Levinson appeals to, we should consider that the application of the heuristics-based rules has nothing particularly “rational” in itself, let alone “argumentative”: it is the default rules that are applied which are supposed to be “rational”. But in what sense are they “rational”? Levinsons seems to take it for granted that they are grounded in general principles of rationality, but does not specify which ones. I suspect that, in fact, heuristics and default rules are made “rational” by the fact that they serve well enough the optimization of linguistic communication.

 

 

4.3.  What a difference does it make?

 

At first sight it might not be clear whether instrumental rationality and argumentative rationality single out  mutually exclusive or overlapping forms of rationality and, therefore, whether the two conceptions are incompatible or may be reconciled. Indeed, Grice’s philosophical psychology makes use of the received folk-psychological connections between belief and desire (1989: 284 ff.) and so does his analysis of practical reasoning in Aspects of Reason (2001). So, Grice uses both conceptions of rationality without opposing them to each other. It is likely that he intended to subsume instrumental rationality under argumentative rationality, in consideration of the fact that an agent may justify his or her behavior as a means for a certain goal.

 

But there are also reasons for considering instrumental rationality and argumentative rationality as opposed to one another: two conceptions between which we, as philosophers, might have to choose. In fact, there are various sources of possible dissatisfaction with instrumental rationality, with which the argumentative conception might be able to cope. These are:

 

·        instrumental rationality does not account for the normative character of rationality, since effectiveness is not a matter of norms but of fact

·        it does not satisfactorily deal with the relationship between rationality and value; effectiveness turns out to be the only value connected with rationality (the only “rational” value?), so that indeed it is effectiveness that makes rationality into something valuable

·        it subtracts goals from undergoing rational screening (except as means for other goals): if rationality is concerned only with the means, how can we use it to discuss whether a certain goal should be adopted? It may be objected that restrictions on acceptable goals can be obtained from other sources, ultimately grounded in our biology (compatibility with the survival of the species); but such a reply is clearly insufficient, because we may want to rationally discuss that too.

 

If we check argumentative rationality for these problems, we find that:

 

·        argumentative rationality can be normative insofar as there are normative (regulative) standards for argumentation

·        as to relationship to values, argumentative rationality is itself the basic value, insofar as it is the essential property of persons (the beings to whom absolute value is attached), and it is also productive of other values (because persons can attach value to what they value)

·        it allows for the rational scrutiny of goals, since goals are in the same need of justification as are means (or anything else we choose, say or do).

 

So, no surprise that it makes a difference whether we take conversational implicature to be “rational” according to either conception. When conversational implicature is viewed as instrumentally rational, it is viewed as a means for the optimization of communication (in various possible senses) and to this aim it is quite reasonable, or perhaps even mandatory, to assume that it has to do with how communication actually functions, that is, with the processes actually occurring in the minds of speaker and hearer. So instrumental rationality is fully compatible with, or even requires, a view of the calculability of conversational implicatures as psychologically implemented in actual calculations. It is not a problem for instrumental rationality that there is often no introspective evidence of the relevant calculations, because these may well be sub-conscious.

Sub-conscious mental processes can be rational mental processes nevertheless (under the instrumental conception), if they succeed in providing efficient and economic means to the given ends. This even more drastically separates relevance-theoretic views from argumentative rationality, which requires of agents personal-level involvement.

 

Under the argumentative conception of rationality, it fully makes sense to conceive of calculability as availability of an argument in support of the assignment of the conversational implicature to the speaker’s utterance. What the rationality of conversational implicature requires to speaker and hearer is only that they be willing to justify their understanding of the implicature and capable (to a certain degree) to provide such justifications, namely replace intuitive graspings by some more or less complete version of the relevant inferential path.

 

All this can better explain why, when in § 1. we focused on the inferential paths proposed for conversational implicature calculation by the three approaches of Grice, RT and GCIT, we found the Gricean approach most flexible as well as most accurate in the formulation of inferential steps and design of the inferential path. We were looking at the route from the utterance to the implicature as a series of steps from one or more premisses to a conclusion, that is, as constituting an argument. In so doing, we were assessing Grice iuxta propria principia: his goal in proposing his general pattern for the working out of conversational implicatures is that of offering argumentative support to the understanding of implicature. By the same token, we were imposing to RT and GCIT a concern that is largely foreign to both.  Relevance theorists and Levinson are concerned with causal chains,  occasionally taking the form of inferential paths, and have no need to be interested in the justificatory value of the latter.

 

 

5. The rationality of text comprehension

 

If the considerations made in 3.1-3.2 above against the psychological reading of conversational implicature calculability are correct and the calculability requirement is not to be read as a psychological requirement, but plays a normative role, there is good reason for adopting argumentative rationality, rather than instrumental rationality, as the background for conversational implicature. Suppose then we endorse this view and take it that the rationality of conversational implicature resides in the willingness and (to some extent) capability of speakers and hearers to spell out an inferential path, argumentatively supporting implicature attribution. What consequences will this have for further research on conversational implicature, and more generally linguistic communication or text comprehension?

 

It seems to me that Saul is right when she claims that Grice's project and the project of such theories of utterance processing as Relevance Theory "can be pursued alongside each other rather than in competition” (2002b: 371). But while agreeing on this, I would like to go further. RT and in part also GCI Theory have originated from the assumption that it can safely be taken for granted that the processes actually leading to comprehension mirror the rational justifications we may give for our understanding of an utterance or text. This assumption should be questioned. It is not undisputably safe to take rational justifications or arguments as a guide for psychological hypotheses. Modeling cognitive processes on possible rational justifications or arguments might generate something that looks very much like a psychological theory without really being such: a science-fiction account of how an up to now non-existent and if existent, humanly unbearable perfectly efficient communicator would function. So, perhaps, in tackling linguistic communication cognitive science should put greater and greater effort in inventing hypotheses and methods of its own.

 

As to what Grice's theory of conversational implicature, non-psychologically read, teaches us about text comprehension, I would like to draw the following morals.  In dealing with text comprehension, we clearly need to foreground the point of view of the audience, that is, of all those who engage themselves in understanding the text (possibly including the speaker, insofar as he or she is engaged in self-monitoring activity). But this should not be taken to mean that we have to be concerned with the audience’s psychological processes in the first place. It is not true that the only way for dealing with text comprehension is to make claims about such processes. We can also provide explanations or elucidations of how an audience should understand a text or why they should understand it so, or perhaps, how they should justify their understanding, namely, we can be concerned with with ways or strategies for justifying conversational implicature attributions and perhaps, beyond these, other kinds of meaning attributions. Now, there have been many discussions in philosophy and neighbouring fields on whether any distinction can be drawn between correct and incorrect readings of a text. If no such distinction can be drawn, then interpretation is endless (as Peirce, Derrida or Rorty have suggested in different ways) (cf. for an overview Collini ed. 1992) and it may hard or even in principle nonsensical to tell mistaken or misleading interpretations from good ones. A Grice-inspired solution, sufficient to exclude irrational readings without resorting to extra-textual factors such as tradition or experts, would be to rely on the availability of argumentative paths in support of the readings at issue. The lack of available argumentative paths or the discovery of faults in the available ones would permit to the audience to exclude irrational readings. What would be left standing might still be an open disjunction (as Grice himself remarks as regards conversational implicature: 1989: 40), but while this is hardly compatible with approaches to comprehension driven by instrumental rationality, it does not raise problems for an approach inspired by argumentative rationality, not bound to make assumptions about actual calculations in the participants’ minds.

 

 

Notes

 

* This paper was presented at the Symposium on Rationality and Implicature (Conference “Mind and Language. 7 Symposia”, Bologna, 15-18 October 2003, organized by Paolo Leonardi within the national research project “Mind and Language”). I am grateful to Claudia Bianchi, Paolo Leonardi, Jennifer Saul, Alessandro Tavano for various suggestions and objections, and to the audience in Bologna for the discussion. Preparatory work for this paper profited from my stay as a Visiting Scholar at Rutgers University in Fall 2002. This research was made possible by joint research funding by the Ministero dell’Università e della Ricerca and the University of Trieste; the local project, on “Belief, Knowledge and Context” is coordinated by myself and the national project, on “Mind and Language”, is coordinated by Paolo Leonardi. 

 

1 It should be noted that if “some” in (2) is stressed, the “naked” implicature “Not all politicians are honest” becomes more plausible. Here I assume that the sentence is uttered without any particular stress.

 

2 I have worked at this research line since 1988, albeit not without interruptions. My researches have profited from the collaboration of Paola Rodari, Simona Regina, Gabriella Nuciforo Paoletti.

 

3 In a slightly different context (his “general pattern for the working out of a conversational implicature”; 1989: 31), Grice rephrases the above discussed condition as follows: "he (the speaker) knows (and knows that I (the hearer) know that he knows) that I can see that the supposition that he thinks that q is required". Here "think" is replaced with "know", which, at least under a non-philosophical reading, is much more of a dispositional verb than "think", therefore more suitable for reference to a dispositional state than an occurrent mental event.

 

4 For a view of belief attribution that defends the claim that belief reports have the nature of true/false assertions, but denies that this depends on beliefs being particulars in the mind/brain, see Kemmerling (2003).

 

References

 

Carston, Robyn (1988), “Implicature, explicature and truth-theoretic semantics”, in R. Kempson (ed.), Mental Representations. The Interface between Language and Reality, Cambridge: Cambridge University Press, 155-81. Repr. in S. Davis (ed.), Pragmatics. A Reader, Oxford: Oxford University Press, 1991, 33-51.

Collini, Stefan (ed.) (1992), Interpretation and Overinterpretation, Cambridge: Cambridge University Press.

Grice, Paul (1989), Logic and Conversation, Cambridge (Mass.): Harvard University Press.

Grice, Paul (1991), The Conception of Value, Oxford: Oxford University Press.

Grice, Paul (2001), Aspects of Reason, Oxford : Oxford University Press.

Kemmerling, Andreas (2003), “Belief ascription: objective sentences and soft facts”, Facta Philosophica 5:2.

Levinson, Stephen (2000), Presumptive Meanings, Cambridge: Cambridge University Press.

Saul, Jennifer (2002a), “Speaker meaning, what is said, and what is implicated”, Nous 36, 228-248.

Saul, Jennifer (2002b), “What is said and psychological reality: Grice’s project and relevance theorists’ criticisms”, Linguistics and Philosophy 25: 347-72.

Sbisà, Marina (2001) “Intentions from the other side”, in Giovanna Cosenza (ed.) Paul Grice’s Heritage, Turnhout: Brepols, 185-206.   

Sperber, Dan and Wilson, Deirdre (1995) Relevance, Oxford: Blackwell. First ed. 1986.