HERBERT PAUL GRICE, "LOGIC AND CONVERSATION: THE OXFORD 'IMPLICATURE' LECTURES" (1964), BANC MSS 90/135c

Speranza

In his theory of implicature, Herbert Paul Grice (sometime university lecturer at Oxford) sets out to demonstrate that certain formal devices of logic, such as sentential truth-functional operators like the connectives, and the quantifiers, have the same "meaning" as their natural-language counterparts (His list: "not," "and," "or," "if," "all," "some (at least one)," "the". 

Is he right?

Contrary appearances had spawned efforts in emantics to marginalise what Grice calls "classical" logic, on the one side, and “correct” natural language (or "ordinary language," as Grice would prefer) on the other.  

However, since these efforts assume that there are divergences in meaning, Grice regards them as wrong-headed. ("Do not multiply senses beyond necessity"). 

Using Sidonius's coinage, 'inplicatura' (which Grice spells "implicature," and which he used in  his 1967 Oxford lectures on "Logic and Conversation"), and before then, at Oxford, Grice argues that we can explain these appearances while remaining committed to the interpretations these devices are given in formal or 'classical' logic. 

Grice’s efforts did not go unrewarded.

Grice's implicature-based approach to the semantics and pragmatics of conversation became the standard. 

Even so, philosophers on both sides of this dispute have challenged Grice’s approach on many fronts. 

Prominent among these is a challenge first issued in L. J. Cohen, of Queen's (Vide: "Can the conversationalist hypothesis be defended?" -- now repr. in Cohen's essay on language and knowledge). Cohen had dealt with Grice in a previous publication, "The diversity of meaning," and his initial criticism to Grice was challenged by R. Walker. 

Critics argue that certain data cut against Grice’s approach, viz., complex examples in which sentences with a specific type of semantics are embedded inside the scope of a sentential connective. I.e. what Grice would call a 'molecular' proposition (following the terminology of Benson Mates, his colleague at Oxford, and George Myro). 

Grice endorses a truth-functional interpretation of the sentential connectives, a move that seems justifiable for a wide range of cases.

However, with respect to these complex examples, this commitment ostensibly forces Grice to see semantic problems where intuitively there are none.  

Moreover, certain examples in which connectives are embedded inside the scope of other connectives appear to undermine entirely the Griceian interpretation of these terms. 

Grice sees this in his "Valediction", in WoW (Studies in the Way of Words). He does not quote from L. J. Cohen explicitly, but I assume he knew what we were talking about!

It is urged that any attempt to respond to these examples with implicatures fails, and so Grice’s account is incapable of explaining the range of cases for which it is intended.  

Therefore, it must be set aside in favor of an "alternative" (non-Griceian) account of the semantics and pragmatics of conversation.

I believe that this argument, which I will call the Scope Argument is unsound.

In this essay I craft a Griceian rejoinder.  

In developing the Scope Argument, I focus on a particular example first introduced by L. J. Cohen -- poor Cohen is working with the MS of the "Logic and Conversation" Harvard version -- since he possibly was not able to attend the seminar at Oxford where Grice introduced 'implicature' -- an example that supports what I take to be the most threatening version of the argument.  

After close analysis of this example, I present a detailed statement of the version of The Scope Argument that it supports. 

This argument is intended to demonstrate that Grice cannot retain his semantic commitments and account for the interplay of meanings in communication, but it fails on two fronts.  

On one front, it underestimates the resources available to the Griceian for dealing with complex examples of this sort. 

On a second front, and more importantly, it misrepresents the nature of Grice’s implicature-based account, evaluating it as if it were a contribution to the mere psychology of conversation -- unless we interpret this psychology as what Grice, pretentiously, following the Kantian tradition, calls 'psychologia rationalis'! 

This treatment reflects a failure to appreciate the specific character of Grice’s contribution to the theory of conversation, a failure that vitiates The Scope Argument. 

In his semantics, Grice interprets sentential connectives truth-functionally, shifting non-truth-functional aspects of meaning to pragmatics (to what the Utterer U has IMPLICATED), where they are modeled with conversational implicatures. 

In developing this theory, Grice uses many examples.

But critics have argued that these are too limited, and that when one considers a broader range, the intuitive appeal of his theory dissipates.  

Cohen, for instance, remarks:

"The Griceian theory gains what 
support it seems to have from 
the consideration of relatively simple examples. Its
weakness becomes apparent when more 
complex sentences are examined...” 

As an instance of such a datum, Cohen supplies a sentence quite like this one: 

(i) If the old king dies and a republic is declared, 
Tom will be content, and, 
if a republic is declared and the old king dies, Tom will NOT be content.

This sentence is formed by embedding complex sentences that contain sentential connectives inside the scope of other sentential connectives. 

In logical form:

(p & q) ) r & (q & p) ) ~r

In fact, there are two layers of embedding.

The principal connective, viz., the second ‘and’, joins two complex conditionals that contain conjunctions as their antecedents.  

If the connectives are treated truth-functionally, and no un-articulated semantic structures are posited, an utterance of (i) cannot be true if the sentences conjoined in the antecedents of the embedded conditionals are both true.  

But surely, Cohen argues, we can and do regard this sentence as expressing a proposition that could be true in just such a case. 

Thus, there would appear to be a conflict between the Griceian take and our intuitive take on the semantic content of these utterances.           

Intuition enjoins us to see (i) as non-problematic.

But the Griceian commitment to truth-functional interpretation of the semantics of connectives suggests otherwise. 

We can cast this as a conflict between two premises.  

First, our intuition premise:

(P1)      Intuition Premise: 

An utterance of (i) is intuitively non-problematic.

And second, our minimalist premise:

(P2)      Minimalist Premise: 

The semantics of natural language connectives is given entirely by their standard truth-functional interpretation.

Each of these premises deserves comment.  

We can begin by remarking on what it is for an uttered sentence to be intuitively non-problematic. 

Emphasis on "intuition" here makes plain the importance of how an utterance strikes us. 

As Griceian utterers, our immediate reaction to an utterance of (i) determines whether it is intuitively non-problematic or not.

But in general, what is it for an utterance to be intuitively non-problematic?  

For our purposes, an utterance will be intuitively non-problematic just in case it is taken to be a vehicle of a substantive and possibly true claim. 

Consider (i). 

It can be used to make a true claim if the conjunctions that form the antecedents of the embedded conditionals are false, but this would be a trivial and not a substantive claim. 

Only in those circumstances where these are true could (i) be a vehicle of a substantive claim.  

To say that utterance of (i) is a “vehicle” of such a claim might only be to say that it is used to convey or imply it comfortably, given the conventional meanings of its terms.

Thus, even if it cannot be assigned a substantive and true claim as its conventional interpretation, it could count as intuitively non-problematic in a circumstance if it could be taken to imply one there.  

What really matters here is that one rather effortlessly interprets the utterance as conveying, or implying, or implicating, a substantive and possibly true claim, regardless of the specific formal relationship between the utterance and the claim. 

Taking an utterance as intuitively non-problematic reflects the perspective of one who “takes” it.

This is meant to capture the fact that, while one could intend an utterance to be intuitively non-problematic, whether it is or not will depend on how it strikes those who attend to it.

As for the minimalist premise, let us begin by considering the motivation behind it. 

Among other things, natural language is a medium for stable and robust inferences. 

One can explain this fact about language by taking a cue from formal logic and interpreting the inferential elements (e.g., the sentential connectives) truth-functionally. 

This would supply the systematicity and generality necessary to underwrite the inferential character of natural language as it is used in communication. 

In addition, it would help explain the stable contributions made by terms and sentences to communication across a wide variety of communicative contexts. 

Explanation and prediction across these contexts require a general foundation and this proposal fits the bill, specifying as it does those properties that account for the truth-bearing and truth-conducting structure of linguistic elements.

The general attitude motivating premise (P2) also motivates a minimalist approach to the semantics of natural language.

This is certainly evident in Grice.  

Grice uses the term of art ‘what is said’ (or what is 'explicated') to refer to the semantic core that comprises the conventionally encoded meanings of sentential constituents, arranged in an order determined by the syntax of the sentence, as well as those contextual determinants necessary to disambiguate and fix indexical elements, i.e., those determinants necessary to make the utterance truth-evaluable. 

Any element of meaning associated with a sentence that is NOT on this short list counts as pragmatic content and so is not a part of what is said -- it is part of 'what is implicated'. 

Commitment to premise (P2) implies that sentential connectives contribute only their standard two-valued truth-functional meanings to what is said by an utterance.  

In what follows, I will treat the term ‘semantic content’ as synonymous with ‘what is said’ or 'what is explicated'. (Grice sometimes used 'phrastic', echoing R. M. Hare). 

With premises (P1) and (P2), so understood, the conflict between the premises generated by an utterance of (i) is even more evident. 

If we limit ourselves to the truth-functional meanings of the connectives in (i), we cannot make it express a substantive and truthful claim, in violation of premise (P1).  

However, it can express such a claim, so it would appear that Grice’s commitment to premise (P2) is mistaken.  

At this point, though, Grice would turn to his "conversational implicature" machinery -- first presented at a seminar in Oxford on "Logic and Conversation" where he translates Sidonius's inplicatura as 'implicature' -- and which is designed to do justice to premise (P1) while retaining premise (P2).  

From his perspective, the conflict between these is merely apparent and rests on an altogether too simple view of the matter. 

We need not reject either premise so long as we construe premise (P1) properly. 

The apparent tension is resolved if we allow that our intuitions about the significance of an utterance might not be all that discriminating.  

Intuitions can indicate when a true proposition figures prominently into the overall interpretation of an utterance.

Hwever, intuitions cannot in general determine whether it forms a part of the utterance’s semantic content (what is explicated) or a part of its pragmatic content (what is implicated). 

Thus, when our intuitions tell us that examples like (i) are non-problematic.

Our intuitions pass judgment on the total content of the utterances, i.e., their semantic content together with their pragmatic content, or what Grice calls their “total significance" (what an utterer means or SIGNIFIES -- using Perceian parlance).

Given that our intuitions do NOT discriminate between semantic and pragmatic content, it is open to Grice to locate theoretical reasons for discriminating them.

In particular, Grice can assert premise (P2), so long as the rest of the content necessary to explain the intuitively non-problematic character of these utterances is accounted for in pragmatic terms.  

This is precisely what Grice does.

Grice embraces premise (P2), and then uses the machinery of conversational implicature to account for satisfaction of premise (P1) in these cases. 

They seem non-problematic because we effortlessly resolve the tension between their minimalist semantics and intuition by identifying the conversational implicatures that dominate our interpretations.  

Thus, Grice endorses the following premise:

(P3)      

One can maintain both premises (P1) and (P2) if one introduces pragmatic elements, and specifically conversational implicature, into the interpretation of utterances.

Grice's move is intended to be general, and it would appear to account for a variety of examples.

However, L. J. Cohen -- who had access only to the Harvard, not the Oxford, version of the "Logic and Conversation" lectures on 'implicature -- argues that it does not work for the uttering of (i), which therefore counts as a counterexample to the Gricean approach. 

Armed with the uttering of (i), Cohen mounts a reductio ad absurdum of the Griceian view (that he infamously refers to as the 'conversationalist hypothesis') arguing that any attempt to restore premise (P1) with pragmatic machinery, and conversational implicatures in particular, undermines a commitment to premise (P2).

To account for our intuitive interpretation of the uttering (i), Cohen notes that the Griceian must introduce a conversational implicature that express the temporal sequence associated with the embedded conjunctions. 

(The 'locus classicus' is of course Strawson's "Introduction to Logical Theory" -- acknowleddging "Mr. H. P. Grice" -- "from whom I never ceased to learn about logic" since Grice was Strawson's tutor at St. John's -- and his example, "She married and had a child").

Grice’s conversational maxims (notably "be orderly" -- cfr. Urmson's example, "He went to bed and took off his trousers" in "Philosophical Analysis") support these implicatures as additional meanings that can explain away the apparent conflict between an utterance of (i) and what Grice in the Harvard lectures calls the Cooperative Principle -- but in the Oxford lectures refers to as the desideratum of conversational clarity, the desideratum of conversational candour, the principle of conversational self-interest, and the principle of conversational benevolence -- (He speaks, incidentally, of 'helpfulness', not cooperation!).

In particular, the maxims of quality and manner -- at Harvard, Grice felt like joking with Kant! -- support calculation of a temporal solution to the apparent truth-conditional problem generated by the semantic content of (i), viz., that if interpreted truth-functionally, the conjunctive antecedents would make it impossible to take an utterance of this sentence to convey a substantive and true claim, a violation of the Cooperative Principle in a normal conversational setting. 

Thus, it would seem that identification of semantic content occurs prior to identification of the implicature, since the former causes the problem solved by the latter. 

In identifying the implicature, we treat what is said by a sentence as a fully formed object that can be evaluated in light of our conversational expectations. 

Call this way of working out pragmatic content the “Serial Generation Approach," since we identify the semantic content and then the pragmatic content in series.  

Proponents of The Scope Argument regard the Serial Generation Approach as part of the Griceian solution.

Thus, we have our fourth premise:

(P4)      If premise (P3) holds, we generate pragmatic content in conformity with the Serial Generation Approach.

Given the Serial Generation Approach, we take what is said by a sentence to cause the conversational problems solved by conversational implicature.

We solve these problems by introducing the pragmatic content into the total content of the sentence uttered. 

Thus, implicature is associated with a sentence, but there is no requirement that this be the top-level embedding sentence.  

In an example like the uttering of (i), there are a number of embedded sentences with which we might associate the implicature.

It is reasonable to expect some correlation between the sentences that underwrite implicature generation and those with which we associate the implicatures generated.

However, this expectation is insufficient by itself to determine whether it is the whole sentence or just some sentential part of the whole that will be associated with the implicature. 

Thus, we appear to have two options for association.

We can associate the implicature with the top-level embedding sentence, which in this case is the conjunctive sentence formed out of the two conditionals.

Or we can associate it with an embedded sentence, either the simple sentences, the conjunctions embedded in the antecedents of the conditionals, or the embedded conditionals themselves.  

Call the first of these options holistic and the second atomistic.  

Thus, we have two more premises of the Scope Argument:

(P5)      If the serial generation approach holds, we associate the pragmatic content with a sentence.
(P6)      If we associate the pragmatic content with a complex sentence like (i), we must associate it either holistically or atomistically.

Beginning with holistic association, note that the serial generation approach requires us to interpret the entire embedding sentence semantically before pragmatic supplementation.  

After calculating what is said by the utterance, we would be compelled by a clash with Grice’s Cooperative Principle to search for an implicature.  

However, this implies that we would detect a problem with the utterance that would require resolution by implicature.

If there were such a problem, it would be reflected in our intuitive reaction to these examples, and it is not, ex hypothesi. 

Therefore, we must not associate the pragmatic content needed to resolve the problems generated by the uttering of (i) holistically.  

Thus, we can add these further premises to the argument:

(P7)      If holistic, we would generate pragmatic content to solve an intuitive problem with (i).

(P8)      (i) is intuitively non-problematic, per (P1)

Therefore, we do not associate pragmatic content with (i) holistically. 

Thus, we are left with an "atomistic," not a "holistic," association.  

On this approach, we would associate implicatures with constituent sentences. 

Given the reasoning above, this cannot be done to solve intuitive problems, since there are none.  

Still, though, the utterance must give us a reason to generate an implicature.

It is here that the "atomistic" approach might supply precisely what the Gricean needs. 

Perhaps implicature generation involves constituent sentences and occurs so quickly and effortlessly at the level of constituent sentences that the process is practically unnoticeable. 

If this is the case, then the atomistic approach would get you around obvious clashes with premise P1) at the level of the whole sentence by distributing clashes out over the constituent parts, each of which would be resolved quickly and so be too small and ephemeral to receive much notice.

Whether this is sustainable or not, a knotty problem remains.  

To bring this out, it will help to regiment the logical form of (i) as per above:

(i’)       ((p ^ q) ) r) ^ ((q ^ p) ) ~r)

There are three points at which we could attach a conversational implicature, consistent with the "atomistic" approach: 

(a) the simplest sentences (i.e., p, q, and r)

(b) the conjunctive sentences that form the antecedents of the conditionals, or 

(c) the conditional sentences. 

Option (a) seems unlikely, since the complex (or 'molecular', rather than atomic) sentences are what create the problem. 

That leaves options (b) and (c).  

If we associate the conjunctions ‘p ^ q’ and ‘q ^ p’ with a conversational implicature that expresses their temporal ordering, then their implicative contents will serve as inputs into computation of the truth values of the conditionals. 

This implies, however, that the truth values of the conditionals will depend on more than just the truth values of their constituent sentences, a fact that conflicts with premise (P2). 

Similarly, if we attempt to resolve the difficulty by associating a conversational implicature with the conditionals, we would be forced to treat the principal conjunction non-truth-functionally.  

Thus, the atomistic attempt to capture our intuitions, and so satisfy premise (P1), forces us to renounce premise (P2). 

To our argument, then, we add these further premises:

(P10)    If atomistic, we must treat connectives in (i) non-truth-functionally.
(P11)    They must be treated truth-functionally, per (P2).
(P12)      Therefore, we do not associate pragmatic content with (i) atomistically.

Since premise (P6) offers no third alternative, we infer that the serial generation approach must be false.

This, however, implies that we must reject premise (P3), and so we have our next premise:

(P13)      

Thus, the pragmatic analysis involving implicature in premise (P3) will not resolve the problems raised by an example like the uttering of (i).

This leaves us without a viable way of using conversational implicatures to maintain both premises (P1) and (P2) in the face of the uttering of (i). 

Thus deprived of conversational implicature and bereft of an alternative, we are forced to relinquish our grip on either premise (P1) or premise (P2). 

Since premise (P1) is the more compelling of the two, premise (P2) is shown the door, and with it goes the Griceian view of the semantics of connectives. 

Call this premise (P14). 

(P14)      Therefore, we must reject premise (P2) and the Griceian view of the semantics of connectives along with it.

The Griceian believes that one can and should maintain the following theses at once:

Utterance of (i) is intuitively non-problematic.

The semantics of sentential connectives is exhaustively given by their truth tables.

We can solve problems of interpretation at the semantic level by introducing content at the pragmatic level.

Call these his central tenets. 

The first thesis is a starting point for all who participate in this debate. 

The second thesis is what sets the Griceian view apart. 

Commitment to truth-functional interpretation of connectives as the stuff of semantics reveals an even more basic commitment to the existence of an objective level of utterance significance that supports robust communicative exchange.  

As such, the second thesis springs from the belief that semantic minimalism supplies the best framework within which to examine fundamental and very general aspects of utterance significance, viz., those associated with sentences that support variable contextual interpretations without being themselves sensitive to these variations. 

However, the Griceian requires a third thesis to smooth over prima facie tensions between the first and second theses.

The third thesis is needed to capture those aspects of utterance significance that are sensitive to contextual variation.

Whatever proponents of The Scope Argument may think of the Griceian commitment to objective meaning, they are of one mind in maintaining that the Griceian view is untenable.  

Proponents of the scope argument believe examples like the uttering of (i) demonstrate that commitment to the second thesis threatens the first thesis, the most intuitively plausible thesis, and that any attempt to preserve the first thesis with with the third thesis undermines the second thesis. 

Thus, the third thesis will not help us save the first thesis. 

Failing any other Griceian alternative, we are left with the option of rejecting the first or the second thesis, and and second thesis loses.  

Therefore, proponents of the scope argument argue that the Griceian view of the semantics of connectives is mistaken and should be set aside, a move that would undermine the Griceian approach to conversation as a whole.

The Scope Argument takes very seriously the processing implications of pragmatics. 

Pragmatics concerns aspects of linguistic significance that are borne out of the intentional use of language, and Grice’s view is a seminal contribution to this discipline. 

Proponents of The Scope Argument note the emphasis Grice places on calculation and generation in his story, which they believe signals his intention that it be a partial explanation of the processing of pragmatic content in utterance interpretation. 

In light of this, proponents of the scope argument believe they are justified in bringing processing considerations to bear on Grice’s view.

As they see it, if the proponents of the scope argument can establish that there is an irreconcilable conflict between utterance processing and his pragmatic theory, his theory will have to go.

I will argue that the Griceian view should not be taken in this way.

But whatever one thinks about the relationship between Grice’s pragmatic story and utterance processing, it is clear that his account of the logic of conversation must at the very least be compatible with the psychology of conversation  -- or 'psychologia rationalis,' to echo Grice -- if it is to be adequate. 

Indeed, if the psychological evidence established a processing model that was inconsistent with his account, so much the worse for his account. 

The Scope Argument urges precisely this conclusion by pressing the uttering of (i) as a counterexample to Grice’s view of conversation. 

In particular, it serves  the uttering of (i) up as proof that we cannot generally process utterances in the fashion described by Grice while remaining fast to his view about the meanings of sentential connectives. 

However, this only follows if we embrace premise (P4), i.e., if we accept that Grice’s implicature approach implies the Serial Generation Approach.

It is perhaps worth noting that there is no compelling textual evidence for attributing this premise to Grice, even though this is the standard interpretation. 

But whatever Grice thought (and the "Valediction" in WoW -- Way of Words -- is not precisely crystal-clear about this), it is NOT the case that Grice's central tenets require him to endorse it. 

One may lay out an alternative to the serial generation appraoch that involves both on-line and parallel processes.  

This alternative is compatible with Grice’s central tenets, and that this compatibility undermines the scope argument.

The content communicated by an utterance of (i) can be distinguished into semantic and pragmatic aspects.

Focusing on the connectives, a Griceian will take their semantics to consist in their truth-functional character, which is a non-cancellable aspect (Grice introduced the 'cancellability' test in his early "Causal theory of perception," repr. in Warnock, "The philosophy of perception," Oxford readings in philosophy -- The "beautiful handwriting" example). 

Their pragmatic aspects, by contrast, are not truth-functional, and they are cancellable and calculable from the conversational maxims. (In the excursus on "Implication" in "Causal theory of Perception," Grice distinguishes between conversational implicature proper, conventional implicature, and presupposition, applying calculability and cancellability tests). 

These aspects are closely related to specific terms, such as ‘and’ and ‘if’. 

Consider the first ‘and’.  

In addition to the explicit conjunctive element, this logical particle contributes the temporal element ‘then’, which is cancelable and calculable and so qualifies as pragmatic content (Grice refers to the maxim, "be orderly," a helpful directive to follow if you are being conversational benevolent, as he would say in the Oxford lectures).

Similar stories can be told about the pragmatic content associated with the conditionals and the ‘and’ that occurs in the antecedent of the second conditional.

It should be pointed out that the whole of the Harvard lectures originated as a response to Strawson's treatment of 'if', so conditionals surely mattered for Grice and note that 'if' is the only connective that receives treatment in a whole lecture -- the fourth lecture in the Harvard series. 

The scope argument assumes a conceptual connection between use of conversational implicature to account for pragmatic aspects of content and serial generation approach.  

If in fact the maxims are used to calculate the pragmatic content, they would appear to require the whole utterance on which to operate. 

However, there is nothing about the Griceian picture of content just sketched that requires all pragmatic elements to be calculated relative to what he calls, echoing Kant, the 'conversational maxims.'

To be sure, they must be calculable, but they need not actually be calculated.

Thus, how they come to be associated with an utterance in a given case can vary, and while it could be via an explicit maxim-driven calculation, it need not be.  

This is a good thing, since in many cases the pragmatic elements seem to come along with the sentence uttered.  

This close association could be due to a process of standardization, or perhaps through the standardized application of general pragmatic heuristics.

These elements would count as pragmatic in the relevant sense---i.e., they are cancellable and calculable---even though they are NOT calculated. 

Further, there would appear to be no reason why the process of standardization can not associate the content with sub-sentential chunks, such as words or phrases. 

The pragmatic elements required to handle the uttering of (i) are related to specific terms, and this relationship could be cashed out in terms of standardization. 

Thus, pragmatic aspects of content could come to an utterance context as pre-packaged associations forged by their regular conjunction with a term. 

These pragmatic associations underwrite interpretive expectations and tendencies that shape the interpretation assigned to an utterance by conversational participants.  

Since we bring these tendencies with us to conversations, they enable us to give shape to interpretations while utterances unfold. 

Thus, we are conditioned through conversational experience to expect certain pragmatic associations, and so we look to save cognitive resources by assigning them immediately, all the while allowing that this assignment is defeasible and could be cancelled explicitly or contextually. 

Consistently with Grice’s account, then, we can embrace a dynamic, on-line processing approach, according to which the pragmatic content of a sentence is generated “on the fly” while the sentence is uttered. 

Consider the infamous utterance by Cohen, (i), once again. 

As it is uttered, we 'interpret' an ‘if’ and we expect a conditional claim.

Futher, we expect the antecedent and consequent to be connected in ways that are not merely truth-functional, unless our expectations are cancelled in some fashion by the surrounding conversational context. 

(This is Grice's way of dealing with Strawson's alleged counter-example to the truth-functionality of the horseshoe).

The ‘if’ in Cohen's infamous uttering of (i) will make the addressee sensitive to a range of antecedent-consequent relationships, e.g., causal, modal, mereological, etc.   

When the consequent appears, the addressee -- what Grice symbolised as "B" in the Oxford lectures on implicature -- will likely opt for a causal interpretation.  

Similarly, when we process an ‘and’, we expect a second conjunct, and further, we expect the order of the conjuncts to be significant. 

With respect to the first ‘and’, the expectation is that it will be expanded into ‘and then,' because the conjuncts are events whose temporal relationship is often significant. (This was the point made by Strawson: "She married and had a children; though I don't obviously want to imply that events occurred in that order.")

Therefore, introduction of pre-packaged associations into the pragmatic mix makes possible recognition of the dynamic and on-line character of utterance interpretation, a fact that undermines premise (P4).  

As a consequence, the scope argument is rendered unsound.

Before celebrating this result, the Griceian must attend to a further problem that cannot be resolved by on-line processing alone. 

However we process the pragmatic content, holding fast to premise (P2) commits us to the view that the full semantic content of  Cohen's infamous uttering of (i) cannot be true when both conjuncts in the antecedents of the conditionals are true.  

On the assumption that the semantic content influences our intuitive reaction to an utterance, this implies that one would encounter an intuitive problem with (i), a consequence that undermines premise (P1). 

Granted, this problem likely will not arise until on-line processing is complete, but at that point we would be left with semantic content that has the wrong truth-conditional profile, creating a problem for our interpretation. 

The on-line processing approach promises relief by eliminating the serial generatioan approach and its associated atomistic/holistic dichotomy, but it leaves us with a problem that is very much like the one created by the holistic resolution, viz., a conflict between intuitive interpretation and semantic content.  

Therefore, even though reconception of the pragmatic content of Cohen's infamous uttering of (i) in terms of pre-packaged associations helps us around the formal problems created by serial generation argument, it leaves us with the familiar intuitive problems, a result that should embolden the proponent of the scope argument.

In fact, this proponent might well argue that rumors of the death of the scope argument were greatly exaggerated. 

The scope argument remains -- is alive, if not kicking -- albeit in modified form.  

The argument in its new form will be identical with the old through premise (P3), but premise (P4) will be replaced by an assertion of the online processing approach.

The online processing approach and premise (P2) generate a semantic content that cannot be true when the conjuncts in the antecedent of the conditionals are both true, undermining premise (P1).  

Therefore, we are forced to give up premise (P1), premise (P2), or the on-line processing approach.  

Of these, it appears that premise (P2) loses out once again, and with it, the Griceian compromise approach.

The Griceian has a principled response to the argument in this new form.  

Underpinning this form of the scope argument is the assumption that our intuitive reaction to the utterance is keyed to the semantic content, which if true would militate against Grice’s account of this content.  

This assumption, alas, is false.

If we are to do justice to the complexity of utterance interpretation, we must recognize that it proceeds at two levels. 

The initial interpretation keys to the words used and involves identification of their semantic content. 

As this proceeds, the interpreter associates terms with the pre-packaged associations she expects them to carry in conformity with the on-line processing approach -- what Grice calls 'pragmatic intrusion'.

As these pragmatic elements are associated with the utterance on the fly, they are treated as provisional and defeasible parts of the total content, and so are open to modification or cancellation. 

These processes run in parallel, with the output of pragmatic processes figuring into the total content that emerges, and the output of semantic processes figuring as input into pragmatic processes and as parts of the total content.  

As the total content emerges, it constrains how we take the output of semantic processes.

If this output conflicts with the total content, we might treat the semantic output as an input to a pragmatic process or we might allow it to force a modification in the total content.  

Our decision will be made quickly and will be open to subsequent revision.

Semantic and pragmatic processes support the overall interpretation and both are essential to it.

Typically, neither process supplies the whole story about the utterance. 

In general, semantic content guides us as we introduce pragmatic elements.

As we construct our interpretation of the total content of the utterance, the total content becomes the sole focus of our attention. 

The semantic content is to the total content of an utterance as a bare wall is to a decorated surface—the wall can be exposed, accentuated with trim, or covered with paintings or posters, but it must be there if there is to be any sort of decorated surface. 

The wall is essential, but we aim at a decorated surface, and once we have that, we can concentrate on the decoration and ignore the unexposed wall.  

The same is true of the semantic content: we use it in constructing our interpretation of the utterance, and while some aspects of it may be prominent parts of the final product of the interpretive process, other parts may serve only as platforms for pragmatic content; semantic aspects of the latter type support our identification of the relevant pragmatic content and then give way to them. 

Of course, if need be, we could get back to the semantic content, just as we can get back down to the bare wall.

 Especially important for our purposes, semantic and pragmatic processes remain distinct even while they contribute to the formation of a unified and coherent interpretation of the total content of the utterance.  Further, it is the total content and not the semantic content that matters in the final analysis. 

The intuitions that make premise (P1) plausible in these cases are keyed to total content.

Interpretation of this is developed dynamically from semantic content, some of which gives way to pragmatic elements as the interpretation goes forward.

Thus, difficulty at the semantic level will not threaten premise (P1) so long as it is not represented at the level of total content, and in these cases it is not. 

Therefore, recognition of utterance interpretation as a bi-level parallel process enables us to embrace premise (P2) and the on-line processing approach without threatening premise (P1), thereby establishing that the second version of the scope argument is also unsound.

As we have just seen, Cohen's infamous uttering of (i) is not a counterexample to Grice’s view of conversation. 

Grice's view is flexible enough to join with an on-line and parallel model of utterance processing, and together they supply an independently principled account of the intuitively non-problematic character of Cohen's infamous uttering of (i). 

With the threat posed by Cohen's infamous uttering of (i) and its ilk dispatched, the scope argument is left without teeth -- metaphorically (cfr. McGinn, "Grice had only one tooth" -- hyperbolically -- cfr. Quine, "He lacked some teeth").  

But while the argument will do no more damage, the same is not true of the spirit that motivated it, and so to that we now turn.

We have established that Grice’s account of conversation can accommodate complex data like (i), but we have not addressed a deeper question, viz., whether TSA fairly represents the Gricean project. As we noted, TSA depends on processing readings of certain key terms, e.g., ‘calculate’ and ‘generate’. These terms signal the relevance of a psychological critique to some, but I am not among this group. These terms are ambiguous between psychological and non-psychological readings, and Grice employed them in their psychologically innocent senses. His account is a contribution to the logic of conversation and not to its psychology. TSA conceals an equivocation, and in what follows, I argue that for this reason it fails to touch on the main business of Grice’s account of conversation.

            By emphasizing logic in his account, Grice calls attention to the structural character of conversation that supports inference and understanding. This character is viewed as general, underpinning conversational episodes in which the Cooperative Principle is observed. As we have seen, part of the logical story involves the truth functional treatment given to connectives, but this provides a somewhat misleading clue about the general nature of the project. His general concern with inferential structures is not as abstract as this, but is rather more concrete and practical; in particular, it focuses on the practical rationality of human communication. As Grice remarks, one of his “avowed aims is to see talking as a special case or variety of purposive, indeed rational behavior.”[25] Elsewhere, he is even more direct: “it is the rationality or irrationality of conversational conduct which I have been concerned to track down.”[26] The logical structure of language is important because it plays a fundamental role in the overall structure of conversation, and it is this structure that underwrites inferences about meanings against the backdrop of practical rationality. The logic of conversation, then, is a practical logic with rationality and not truth as its dominant metric, although truth remains a crucial element because of the implications it has for the rationality of conversational participants.

More detail about this logic is in order, although it can be no more than a sketchy outline at best in this essay. Let’s begin by taking a logic to be an inference modeling system comprising primitive elements and inference rules, and a conversation to involve a sequence of utterances.  Utterances needn’t involve language for Grice, but let’s simplify here and focus on those that involve sentences in some language L. An utterance of this sort can be understood as the tokening of a sentence of L in a context, or more simply, a sentence token/context pair, where the context specifies properties such as the space-time coordinates of the utterance event and the producer of the token. Together, this pair introduces a specific proposition into the conversation, viz., the semantic content of the utterance. On the assumption that the participants in the conversation understand L and hear the utterance, they will likely have an attitude toward the semantic content introduced by the tokening of the sentence in that context, e.g., they might believe it, deny it, etc. In addition, other propositional attitudes will be central to the conversation. The speaker will have intentions execution of which give rise to this utterance, as well as beliefs about his utterance, the audience, the context, and what this utterance will convey to his audience in this context.  The audience will also have intentions that guide their participation in this episode, as well as beliefs about the utterance, the context, the speaker, and what the speaker might have intended to convey with that utterance in this context.  These articulated assemblies of propositional attitudes are all centered on the utterance, in that the propositions involved must include as a constituent either the utterance or some aspect of the episode that is made relevant by the utterance. We now have what we need to provide a rough model of conversation. Take a conversation to be a sequence of utterances each of which serves as the center for an articulated complex of attitudes that the conversational participants have toward propositions about the constituent sentence tokens, contexts, and other participants.

A conversation is thus a sequence of utterances and an articulated and temporally extended assembly of propositional attitudes centered on those utterances. These are the elements that figure into the conversational inferences modeled by the logical system.  Turning to the inference rules, we find that in this system, the rules divide into two difference but related types.  First, there are those that that distinguish arrangements of propositional content that are truth conducing from those that are not.  These rules identify clusters where the intended meaning of an utterance, i.e., its total content, follows via abduction from the other propositions centered on the utterance. Here the inference rules of non-deductive logic and first order deductive logic will apply. In this system, though, there is the added dimension of the attitudes taken toward the propositions by conversational participants.[27] These attitudes guide the participants’ behavior and figure as constituents in the contents of the other propositional attitudes. The second type of inference rule identifies those attitude assemblies where belief in the total content of the utterance is rational, given belief in the other centered propositions. From the perspective of conversational success, inferences concerning rationality are more significant, since rational belief in the total content is generally what the speaker intends to make possible with her utterance. These two rule types are, of course, closely related---if the relations among the propositional contents conform to the proposition-level inference rules, then belief in the intended meaning will likely be rational, given belief in the other propositions.

 It is as inference rules of the second type that we must see the Cooperative Principle and its attendant maxims.  The Cooperative Principle is an overarching rule that underwrites the rationality of certain beliefs about the participants and the utterances. The maxims also underwrite rationality of belief, given other beliefs about the utterances, contexts, and participants.  In general, these two types of inference rules codify patterns of non-deductive inference that are exemplified in episodes of successful communication, i.e., communicative episodes where the total content to be conveyed by the utterance is actually conveyed.

            Grice’s account, then, is a rational reconstruction of conversation designed to reveal its logic. It is in light of this that one should understand terms like ‘calculate’ and ‘generate’. They do not signal psychological maneuvers; rather, they signal the presence of truth and rationality conducing structural relationships among conversational elements. Their use here is similar to their use in mathematics, where one might speak of a theorem generated by a set of axioms, or a numerical claim calculated from other claims. To the extent that these terms do apply to our behavior as conversational participants, they describe what the propositions make possible and not what we actually do in thinking about them. With this in mind, let’s return to TSA. We note that it depends on folding implicatures into an ongoing interpretation to solve a problem at one point and then assessing the connectives at a subsequent point.  But this way of speaking betrays the fact that it construes the various propositions involved here, viz., the semantic and pragmatic contents associated with the embedded and embedding sentences, as temporally related in an extended processing event. But this is not psychology, it’s logic, and so the standard to apply is rationality and not psychological plausibility. From this perspective, the semantic content of (i) conforms to (P2), warts and all, but an utterance of it centers a cluster of propositional attitudes that, taken together, provide abductive support for rational belief in the total content.  There is no semantic problem posed by (i) that threatens the rationality of belief, given the other propositions centered on it.

            “But surely,” one might say, “it is in support of rational belief that the logical rubber meets the psychological road. TSA establishes that there is no way to realize a conversation involving utterance of (i) psychologically and still respect Grice’s central tenets. His account is incompatible with empirically acceptable processing models and so must be set aside.” There are two problems with this view. First, as I argued in the previous section, Grice can help himself to an independently justified processing model that can account for our intuitive feelings about (i) within the constraints supplied by his logic.  This points to the second problem, viz., the nature of those constraints. The critic would have us see these as rigid, forcing the processing model to be very much like and perhaps even identical with the logical model he describes. But identity is too strong.  The exact relationship between the logic and psychology of conversation will likely be complex, but as we noted earlier, the logic must at least be compatible with the psychology, i.e., the story about the inferential structure of conversation must be compatible with the cognitive story about processing if the logical story is to be sustainable. However complex the relationship turns out to be, though, it seems unlikely that the specific details of the processing model will be forced by the logic. Of course, the processing model could be a close implementation of the Gricean “calculations” that trace logical relations among propositional attitudes, but they need not be, and evidence suggests that this is unlikely.  I believe that the correct processing story will be more akin to the one I’ve told, supplemented perhaps by accessibility considerations of the sort adduced in Recanati (1995). Whatever the cognitive story may turn out to be, the processes that figure into it are likely manifold and heterogeneous.  Grice’s goal was not to describe these processes, but only to plumb the logic that supported them; since there is no reason at this point to believe that his logical search should have also issued in descriptions of these processes, the key to acceptability of his account is compatibility. TSA sets out to prove incompatibility, but as we have seen, it fails.

            Some support for the view that the relation between logic and psychology is loose can be gleaned from an examination of the rhetoric of argument. When one wishes to convey an argument to an audience, there will typically be many ways to present it. One must respect the argument, but one needn’t present it in its most logically revealing form. The logic of the argument underdetermines its expression and interpretation, constraining them without forcing them.  This also seems true in the case of conversation, where its logic underdetermines its psychological realization. There are a number of possible processing models that could be the model that conversational participants instantiate, given the Gricean logic, including the one described in the preceding section.  Further, it seems that this will be true regardless of the logical account that one endorses.  Thus, there is reason to believe that compatibility and not identity should be our focus in relating the logic and psychology of conversation.

            One might go further, however, and reject logic altogether as entirely irrelevant to conversation theory. This, however, would be a grave error. No defensible empirical account of conversation will be without an underlying logic. The elements of the conversational story relate to one another in ways that support inferences about meanings---this much is clear---and these patterns of inference depend on structural arrangements that ground the logic of conversation. Of course, there are many who dispute Grice’s logic, and I have supplied little to assuage their concerns, especially given the limited focus on TSA. However, while the logic that underwrites the mature science of conversation may well be quite different from what Grice has limned, TSA does not supply any reason to believe this.

V.        Conclusion

Critics of Grice regularly conflate the logic of conversation with the psychology of conversation. Since his is an account of the pragmatics of language and so focuses on language use, they have taken psychology to be an appropriate lens for examining his view. However, this is a mistake. Grice seeks to supply a rational reconstruction of the structure of conversational pragmatics and not a model of its psychological realization. This is logic, not psychology.  That said, it is important to note that, while different, they are crucially related since the logic of conversation supplies the framework for the processing model. Within the framework, the conversational maxims serve as constraints on propositional attitude assemblies and so serve not as psychological laws but as logical laws that underwrite ascriptions of rationality. As its foundation, this framework employs a minimalist semantics for sentences, as reflected in (P2).

            To the extent that TSA depends on a psychological reading of Grice’s project, it is mistaken and misleading. However, one can interpret it as an attempt to prove that Grice’s logic is incompatible with any processing model that does justice to our intuitions. In particular, it purports to establish the incoherence of any attempt to assuage intuitive disturbances that utterances of (i) create by layering pragmatic content over a truth-functional core. Even charitably interpreted in this form, the argument fails. By combining Grice’s story with a model of utterance processing that is on-line and parallel, we can demonstrate that TSA is unsound and readily accommodate our intuitions about examples like (i). This move also enables us to join the virtues of Gricean semantics as a platform for understanding the sentence’s contribution to communication with the virtues of on-line accounts of utterance interpretation, a move that derives independent motivation from the empirical successes enjoyed by the on-line model. If we take our intuitions to give us insight into the total content of the utterance and we take seriously the on-line processes that yield our representation of this content, we can gratify our intuitions and maintain a commitment to minimalist semantics at the same time.

I am grateful to many for commentary and criticism. The following people deserve special mention: Kent
Bach, Anne Bezuidenhout, Eros Corazza, Bruce Glymour, Ernie Lepore, John Perry, François Recanati,
Stephen Schiffer, Corey Washington, audience members at a UI/WSU Philosophy Colloquium, and a referee
for The Journal of Philosophy.  I am especially indebted to Joseph Keim Campbell, David Shier, and Harry
Silverstein, as they encouraged me to write this essay and gave me helpful advice along the way.

This name is inspired by Recanati, who addresses these considerations while discussing what he calls the “Scope Principle”. It is important to keep TSA and Recanati’s Scope Principle distinct, however.  The Scope Principle is introduced as a universal test for semantic status, while TSA is designed to undermine the implicature approach of Grice through provision of a counterexample.  See Recanati, F. (1991) ‘The pragmatics of what is said,’ in Davis, S., ed., Pragmatics: A Reader (New York: Oxford University Press), 97-120, at pp. 112-114.

Cohen, L. (1971) ‘Some remarks on Grice’s views about the logical particles of natural language,’ in Bar-Hillel, Y., ed., Pragmatics of Natural Language (Dordrecht: Reidel), 50-68.  See p. 67.

Adapted from Cohen (1971), p. 58.  

This conflict can be generated by other data as well. For examples, see Cohen (1971), p. 63; and p. 203 in Bezuidenhout, A. (1997) ‘The communication of de re thoughts’, Nous 31, 197-225. In general, the data appear to vary along two dimensions. First, there is the dimension corresponding to the sentence embedded, which either contains a connective, such as ‘and’ or ‘if ... then’, or a predicate, such as ‘jumps’, associated with enrichments that do not fill syntactic slots. Second, there is the context within which the sentences are embedded. Two types of context are represented in the data: (a) anti-reflexive operators, such as better than, and (b) binary sentential connectives that have the following property: when the first sentence connected is true, the truth value of the complex sentence formed depends on the truth value of the second sentence. With respect to the second context, the conjunction and conditional exhibit the associated property, and they appear regularly in examples found in the literature. By contrast, the disjunction does not have this property and so does not create a context for the purposes of TSA. It would appear that any combination of embedded sentence type and context type produces Scope Argument data, although this essay is not the place for that argument.

While not as big a fan as some, Grice agrees that intuitions play an important role here.  See Grice, P. (1989) Studies in the Way of Words (Cambridge, MA: Harvard University Press), p. 49.  For a bigger fan, see Recanati, F. (1993) Direct Reference (Oxford: Blackwell).  Cf. Bach, K. (2002) ‘Seemingly semantic intuitions,’ in Keim Campbell, J., O’Rourke, M., & Shier, D., eds., Meaning and Truth (New York: Seven Bridges Press), 21-33.

See the discussion of the “conversationalist hypothesis” in Cohen (1971), pp. 50-52. Cohen is careful to avoid attributing this view to Grice, although there are more than a few indications in Part I of Grice (1989) that he would welcome such an attribution. Cf. Allwood, J. (1986) ‘Logic and spoken interaction,’ in Myers, T., Brown, K., & McGonigle, B., eds., Reasoning and Discourse Processes (San Diego: Academic Press), 67-94.

For related distinctions, see semantic meaning/speaker meaning in Kripke, S. (1979) ‘Speaker’s reference and semantic reference,’ in French, P., Uehling, T., & Wettstein, H., eds., Contemporary Perspectives in the Philosophy of Language (Minneapolis: University of Minnesota Press), 6-27.  See also what is said/what is implicated in Recanati (1991) and (1993), and implicature/explicature in Sperber, D. & Wilson, D. (1995) Relevance: Communication and Cognition, 2^nd ed. (Oxford: Blackwell), p. 182.

Grice (1989), p. 41.

My reconstruction of this part of the argument is grounded in the reasoning found in Cohen (1971), pp. 58-59 and pp. 63-64.  See also pp. 194-195 in Posner, R. (1980) ‘Semantics and pragmatics of sentence connectives in natural language,’ in Searle, J., Kiefer, F., & Bierwisch, M., eds., Speech Act Theory and Pragmatics (Dordrecht: Reidel), 169-203.

This assumption is central to the Serial Model of linguistic interpretation that is criticized throughout Recanati, F. (1995) ‘The alleged priority of literal interpretation,’ Cognitive Science 19, 207-232.  In this essay, he presents the case for regarding this as the standard interpretation of Gricean pragmatics. In addition, see Recanati (1993), pp. 271-272.  In setting up the “Scope Principle”, Recanati endorses an argument that he attributes to Anscombre & Ducrot, and in this he includes as a step the claim that “no implicature can be generated at the sub-locutionary level, i.e. at the level of an unasserted clause such as the antecedent of a conditional” (p. 272). Rather, implicatures can only be generated at the level of the “complete utterance”, which must therefore be given if the implicatures are to be worked out. See Anscombre, J.-C., & Ducrot, O. (1978) ‘Echelles argumentatives, échelles implicatives et lois de discours,’ Semantikos 2, 43-67.

Opinions about what to do here vary.  First, we might reject (P1) and claim that the examples are really problematic after all, despite intuitions to the contrary. However, for us to reject this prima facie case in favor of (P1), there would need to be significant theoretical benefit associated with commitment to (P2), and that has yet to be established.

Second, we could reject (P2).  This option comes in several different varieties that we can divide into two groups: the semantic and the pragmatic. The semantic varieties include positing ambiguity (see Grice (1989), pp. 47-49 for critical discussion), introducing slots to be filled (see Recanati (1991), p. 103 for discussion), and introducing features to be killed (Posner (1980), pp. 182-188).  The pragmatic varieties view much of the content that goes beyond (P2) as “free enrichments”, determined pragmatically in unpredictable ways by the interaction of utterance and context, and added to the semantic content of utterances as directed by our intuitions (e.g., Bezuidenhout (1997), and Recanati (1991). I believe, however, that it is too early to give up on a Gricean approach in favor of one of these alternatives.  In what follows, I will argue that upon closer inspection of the data in question, we can retain both (P1) and (P2).

Variations of this reasoning are found in the literature.  In all cases, the data are regarded as counterexamples to Grice and support variants of TSA. In some cases, e.g., Posner (1980), the supporting examples are similar to those of Cohen, whereas other cases build on different types of recalcitrant examples.  For instance, consider this sentence:
(ii)        It is better to meet the love of your life and get married than to get married and meet the love of your life. 

Examples like this motivate a form of reasoning that differs from the argument regimented in the text by ignoring the atomistic approach and relying exclusively on appeal to intuition. For discussion, see Bezuidenhout (1997), and Recanati (1991). In this version, there is no use made of (P4).  Instead, the argument proceeds from (P3) in the following fashion:
(P4')     Use of implicatures to account for the significance of examples such as (ii) requires recognition of a problem with the example.
(P5')     There is no problem with the example, per (P1).
(6')       Therefore, we must reject (P3).

This form of the argument is arguably weaker than the Cohen formulation, as it depends solely on an appeal to intuition, and intuitions about these matters can differ in ways that are difficult to adjudicate. The form of the argument reconstructed in the text demonstrates that intuitive dissonance, assumed for the sake of argument, prompts a response that results in truth conditional distress.  Thus, it works as an internal criticism of Grice, whereas the form described in this note is primarily external.

This is apparent in Bezuidenhout (1997), Recanati (1991 & 1993), and even in Cohen (1971). It is also a widespread attitude in cognitive psychology.  For a list of psychological works that rest their criticism on this assumption, see pp. 51-52 in Wisniewski, E. (1998) ‘The psychology of intuition,’ in DePaul, M., & Ramsey, W., eds., Rethinking Intuition (Lanham, MD: Rowman & Littlefield Publishers), 45-58.

Grice (1989), pp. 39-40 and 43-44.  Calculability is emphasized on p. 31.

We could, for instance, explicitly reject the temporal implication as an aside while uttering (i), thereby establishing it as cancelable.  As for its calculability, see Section II.

For the purpose of countering TSA, we really need only cancelable and calculable content.  Grice would consider these to be generalized conversational implicatures.  For a defense of this classification, see Levinson, S. (2000) Presumptive Meanings (Cambridge, MA: The MIT Press), esp. Ch. 2. Others regard this classification as mistaken, as Bach argues in Bach, K. (1994) ‘Conversational impliciture,’ Mind & Language 9, 124-162. As Bach sees it, these elements qualify as implicitures, or implicit content, as distinct from implicatures. For our purposes, it makes no difference whether these are generalized conversational implicatures or implicitures since both are cancelable and calculable and so fall into the domain carved out by the notion of pragmatic content we are using.

See Grice (1989), p. 21, where he emphasizes that implicatures must be “capable of being worked out.”

For discussion of standardization, see Bach, K. (1995) ‘Standardization and conventionalization,’ Linguistics and Philosophy 18, 677-686. For discussion of this process as well as general heuristics, see Levinson (2000).

Words and phrases function as anchors for this pragmatic content, which operates as a rapidly accessed pragmatic presupposition that shapes on-going interpretive efforts and creates expectations about the trajectory of the interpretation as a whole.  See Martin, C. (1989) ‘Pragmatic interpretation and ambiguity,’ in Program of the Eleventh Annual Conference of the Cognitive Science Society (Hillsdale, N.J.: Lawrence Erlbaum Associates), 74-481; Riesbeck, C. & Schank, R. (1978) ‘Comprehension by computer: Expectation-based analysis of sentences in context,’ in Levelt, W., & Flores d'Arcais, G., eds., Studies in the Perception of Language (New York: John Wiley & Sons), 247-294; Sedivy, J., Tanenhaus, M., Chambers, C. & Carlson, G. (1999) ‘Achieving incremental semantic interpretation through contextual representation,’ Cognition 71, 109-147; and Zukerman, I. (1989) ‘Expectation verification: A mechanism for the generation of meta comments,’ in Program of the Eleventh Annual Conference of the Cognitive Science Society (Hillsdale, N.J.: Lawrence Erlbaum Associates), 498-505.  It should be noted that arguments such as those advanced by Clifton and Ferreira are relevant to determining the precise character of the pragmatic elements, pushing us perhaps in the direction of implicitures and away from generalized conversational implicatures.  See Clifton, C. & Ferreira, F. (1989) ‘Ambiguity in context,’ Language and Cognitive Processes 4, 77-104.

This expectation of connectivity will also inform our initial interpretations of the second and third occurrences of ‘and’ in (i). These expectations will remain in force and help give shape to the interpretation of the third ‘and’, but they will be set aside in favor of a straight truth-functional interpretation of the second.

These processes include what Bach calls “completion” and “expansion”.  See Bach (1994).  Cf. Recanati’s processes “saturation” and “strengthening”.  See Recanati (1991), p. 102.  See also Recanati (1993), pp. 233-266.

Parallel processing is another respect in which the model advocated in this essay agrees with on-line processing models. Constraint-based, on-line models generally assume parallelism.  See Tanenhaus, M., & Trueswell, J. (1995) ‘Sentence comprehension,’ in Miller, J., & Eimas, P., eds., Speech, Language, and Communication (San Diego: Academic Press), 217-262. It should be noted, however, that allowance for parallel treatment of semantics and pragmatics in utterance interpretation does not necessarily imply computation of a full representation at both levels; rather, it might be that the processing system accesses and uses both semantic and pragmatic aspects of the input in building a representation of the total content.  For discussion of this idea in connection with syntax and semantics, see Marslen-Wilson, W., Tyler, L., & Seidenberg, M. (1978) ‘Sentence processing and the clause boundary,’ in Levelt, W., & Flores d'Arcais, G., eds., Studies in the Perception of Language (New York: John Wiley & Sons),  219-246.  For additional discussion, see Clark, R. & Gibson, E. (1988) ‘A parallel model for adult sentence processing,’ in  Program of the Tenth Annual Conference of the Cognitive Science Society (Hillsdale, N.J.: Lawrence Erlbaum Associates), 270-278; Dinsmore, J. (1989) ‘A model for contextualizing natural language discourse,’ in Program of the Eleventh Annual Conference of the Cognitive Science Society (Hillsdale, N.J.: Lawrence Erlbaum Associates), 597-604; Gorrell, P. (1989) ‘Establishing the loci of serial and parallel effects in syntactic processing,’ Journal of Psycholinguistic Research 18, 61-73; Marslen-Wilson, W. (1987) ‘Functional parallelism in spoken word recognition,’ Cognition 25, 71-102; St. Johns, M. & McClelland, J. (1988) ‘Applying contextual constraints in sentence comprehension,’ in  Program of the Tenth Annual Conference of the Cognitive Science Society (Hillsdale, N.J.: Lawrence Erlbaum Associates), 26-32; and Tanenhaus, M., Dell, G., & Carlson, G. (1987) ‘Context effects and lexical processing: A connectionist approach to modularity,’ in Garfield, J., ed., Modularity in Knowledge Representation and Natural Language Understanding (Cambridge, MA: MIT Press), 83-108.  Compare the discussion of interpretive levels in Recanati (1993), pp. 312-318 and propositional form and style.

As such, this proposal closely resembles one described near the end of Recanati, F. (2001) ‘What is said,’ Synthese 127, 75-91; however, I would argue that while this notion of semantic content need not be “entertained or represented” at any point in utterance interpretation, it is nevertheless available to the interpreter and so has psychological reality in this sense. 

One might complain at this point that I have not indicated how Grice would have this process of replacement go; however, for reasons to be given in the next section, I don’t believe that Grice must supply this. This would be a psychological story, and that is not what his account is intended to provide.

Grice (1989), p. 28.

Ibid., p. 369.

For a logical system that is similar in this respect, see Chisholm, R. (1989) Theory of Knowledge, 3^rd ed. (Englewood Cliffs, NJ: Prentice Hall).

These beliefs would not suffice for rationality if, for example, the agent in question also had other beliefs that were inconsistent with them.  This reveals the essential non-deductive character of the attitude level inference rules.

REFERENCES

Grice, H. P. Logic and Conversation: The Oxford Lectures -- now deposited at the Bancroft Library.