SOURCE: "Moore's Paradox, Sincerity Conditions, and Epistemic Qualification," in On Being and Saying: Essays for Richard Cartwright, edited by Judith Jarvis Thomson, The MIT Press, 1987, pp. 133-50.

I

This is not a scholarly paper on Moore's paradox. Many of the points I make have been made by others (long ago in some cases), and I hope they will acquiesce in my putting them in the present context without further acknowledgment. I want to suggest in this paper that a Moore paradox of the statemental type has to do with epistemic force rather than merely with sincerity conditions of illocutionary acts (if with them at all). Although something like a sufficient condition for an utterance to be odd in the Moore-paradoxical way will emerge, I do not try to say what conditions it is necessary to have to have a paradox of this type. The main reason for this is that the original paradigms primarily dealt with in connection with Moore's paradox have involved two forms of sentence that are (supposedly) suited for making statements, viz.

(A) p, but I don't believe that p

and

(B) Not-p, but I believe that p

although it is a feature of the account of Moore's paradox to be given here that the utterance of many other forms of sentence can involve what is apparently the same sort of oddity, that of sentences whose use is somehow impossible or ruled out although they do not involve self-contradiction.

In fact, I have no evidence, other than that of linguistic intuition and the plausibility of my descriptions of them, that the original paradigms of Moore's paradox and the other examples of odd utterances I deal with exhibit one and the same sort of oddity. But because it appears that a unified account of all of them can be given, I conduct my discussion under the assumption that just one kind of oddity is in question. (For further remarks on the larger picture, see section V.)

When statemental utterances are in question, the present account falls under the type of account of Moore's paradox sometimes called the "saying and disbelieving" line. Although popular, it is not particularly well named, because even paradoxes using sentences of form (A) have to do not with disbelieving things but with saying that one does; nor does this name mention implying (or indicating, etc.) what one believes, reference to which is essential. According to this account, the statemental conjunctive utterances that are Moore paradoxical are so because what is said in uttering one of the conjuncts (its propositional content) is logically incompatible with what is merely implied by uttering the other conjunct. Actually, it is a matter here of what would be the case if these utterances were made in the normal assertive way, because in fact they are in a sense impossible utterances (or things that one cannot say) and probably never actually get uttered in this way. Interchanging assertion and denial, as in cases of form (B), gives rise to an analogous type of utterance that we might call the "denying and believing" form of Moore's paradox and make similar comments about.

Thus, as examples of the original paradigms of Moore's paradox, we have such utterances as

(1) It is raining, but I don't believe that it is

and

(2) It is not raining, but I believe that it is.

Actually, completeness requires dealing with a third form, as a result of an ambiguity in (A), as a result of a phenomenon known to linguists as "raising" and resulting in a difference in the scope of the negation with respect to "believe." Let "(Al)" stand for the sentence form (A) interpreted as involving narrow scope of the negation (with respect to "believe"), and let "(A2)" stand for (A) with the wider scope. (The "not" in (A), understood this way, was said to have been "raised" (transformationally) from the complement clause; the one in sentence (A2) was not raised.) (A) on interpretation (Al) thus means

(A1) p but 1 believe that it is not the case that p

whereas on interpretation (A2) it means

(A2) p, but it is not the case that I believe that p.

The "saying and disbelieving" line on Moore paradoxes of this form can still be construed as holding that what uttering the first conjunct implies (in the sense in question) is explicitly denied by the second conjunct, because the second conjunct (like that of an utterance of the same form disambiguated the other way) is still logically contrary to the implicatum of uttering the first conjunct.

A "denying and believing" line would be the analogue of those just mentioned, holding instead that uttering the first conjunct of a Moore-paradoxical utterance of form (B) implied (in the sense in question) that the speaker did not believe what the second conjunct explicitly says the speaker believes.

The account of Moore's paradox that I try to develop in what follows belongs, broadly, to the saying-and-disbelieving, denying-and-believing type of account, but I try to pursue systematically the suggestion (implicit in some of the earlier literature) that the conflict involved is one involving epistemic force or what I have elsewhere (Caton 1966, 1981) called epistemic qualifiers.

II

In this section I explain what I take to be the broad outlines of the system of epistemic qualification, starting with some discussion of the kind of pragmatic relations that I believe define these broad outlines, and then try to locate belief (important in connection with Moore's paradox) within it.

Discourse Relations

As I use the term, a remark is any illocutionary act in the most complete sense, including content, of Austin (1962), Alston (1964), or Searle (1969). Where convenient, I talk about the content of a remark, its illocutionary force ( = what type of illocutionary act was performed in issuing it), the sentence used or that could be used to make it, a clause that could be used to express it, etc.

By a discourse (or d-) relation, I mean a relation between remarks in a discourse. One example of a discourse relation is d-commitment: if making a certain remark in a discourse commits one to making another remark in that discourse, there is a certain relation between the two remarks involved; it is sometimes even ordinarily called commitment (although of course other things are also) and may be called d-commitment. A special case of dcommitment is the relation deriving from the fact that the propositional content of a certain remark one makes logically entails that of a certain other remark. Another example of a d-relation is the relation holding between two remarks as a result of the fact that making one of them in a certain discourse excludes making the other, on pain of incoherence (uninterpretability) of that discourse; this relation between remarks may be called d-exclusion or dincompatibility.

These two d-relations may be related definitionally, because if one is d-committed to making a certain remark R2 by what one has already said R1, surely R1 is d-incompatible with agreeing with the negation of R2. This must be the case at least when statemental remarks are in question.

Epistemic Qualification

Three further concepts are useful in expounding the nature of our system of epistemic qualification. The fundamental one is that of an epistemic state, and deriving from it are those of an epistemic qualifier and of an epistemic qualifying expression.

By an epistemic state here I refer to knowledge (in the propositional sense of knowing that p, for some proposition p), its various species or types, such as remembering and realizing, and the various states that are contrary to it, in the sense that it cannot be at one and the same time that one is in one of those states and also knows that p. For example, if one just thinks that p (and does not know that p), then obviously one cannot also know that p. (I think here of responding to the question of whether one knows that p or (just) thinks that p.) But also, if one suspects that p, then equally, I think, one cannot also then know that p.

Just referring to knowledge, its types, and its competitors might not, I think, delimit a complete natural class of epistemic states, for it might be that there are states functioning in the same sort of linguistic and conceptual way that are compatible with knowledge without being types of it. In particular, belief may be of this sort; yet the epistemic qualifier ' believe'¹ correlated with it is plainly in the same business as those related to the epistemic states mentioned: By itself, it is a qualifier of middling strength, competing on the one hand, for example, with those related to knowing and being certain and on the other with, for example, those related to inclination to believe, suspicion, and thinking it possible that p. So I must confess that up to this point I do not have a satisfactory definition of epistemic states nor therefore of epistemic qualifiers or epistemic qualifying expressions.

By an epistemic qualifier I mean a qualification or modification of a proposition occurring in one's discourse so as to indicate what one takes one's epistemic state to be with regard to that proposition at the time it occurs, so qualified, in the discourse.

By an epistemic qualifying expression I mean a linguistic expression (or other linguistic device) that is used to express an epistemic qualifier, that is, to indicate what epistemic state the speaker takes her/himself to be in with respect to an indicated proposition. Examples of epistemic qualifying expressions are "I know," "I think," "I believe," "I'm certain," "I feel certain," "It seems to me," "I incline to think." Those in this group may be called personal epistemic qualifiers because they explicitly mention the speaker. Ones that may be called impersonal, because they do not explicitly mention the speaker, include "It is certain," "It is virtually certain," "It is definite," "It must be" (or the auxiliary ".. . must..."), "It is overwhelmingly probable," "It is likely," "There is a good chance," "It is possible," and "It may be" (or the auxiliary " . . . may . . ."). I also call the epistemic qualifier expressed by a certain epistemic qualifying expression a personal or impersonal epistemic qualifier, according as the epistemic qualifying expression is personal or impersonal.

Because epistemic qualifying expressions usually have other uses besides their use to express a certain epistemic qualification of an indicated proposition, where necessary I speak of an epistemic qualifying expression "functioning to express an epistemic qualifier." For example, in the conditional statement

(3) When it is certain that she has left, the desk clerk will use his pass key to let the officers in

the epistemic qualifying expression "it is certain" is not being used to indicate what epistemic state the speaker takes her/himself to be in at the time of utterance with respect to some proposition (notably not with respect to the proposition that the person referred to has left); that is, the epistemic qualifying expression is not there functioning to express an epistemic qualifier, as it would be (for example) in the statement that

(4) It is certain that she has left

given, say, in answer to the question of whether it is certain that the person referred to has left.

Personal epistemic qualifying expressions containing one of the verbs that denote epistemic states seem to tend to function to express epistemic qualifiers when in what one might call the Austinian position in their conjugation, viz. the first-person singular present indicative active. This is, of course, the position in which performative (or illocutionary act) verbs have their striking performative force. If there is some reason why there should be this coincidence, it is not known to me. (It may be noted that Austin included as (possible) performative verbs some that might, perhaps, have been better regarded as epistemic qualifying expressions; these were "doubt," "believe," and "know" (Austin 1962, p. 161). Note, though, that not even all personal epistemic qualifying expressions function to express epistemic qualifiers only in the Austinian position, at least not if "It seems to me that p" counts as a personal one.)

One epistemic qualifying device important in connection with Moore paradoxes (and illocutionary acts) is not in English and other familiar languages a word or phrase but rather a device linguists might call a zero-form, that is, the absence of any other indication of the speaker's epistemic state. Because a person who uses this device can often be said to have made a "flat statement," I appropriate this term and call the epistemic qualifier thus expressed the flat-statement epistemic qualifier, denoting it by gfs. (I see no reason to suppose that there is more than one.) Epistemic qualifying expressions do not appear to be ambiguous as to what epistemic state they denote, although what they convey in a particular application may well be what Toulmin (1958, ch. 1) calls "field-dependent." It thus seems legitimate to refer to a mapping from epistemic qualifying devices into epistemic states and of a one-to-one correlation between sets of equivalent epistemic qualifying devices and epistemic states.

Epistemic states, epistemic qualifiers, and epistemic qualifying expressions exhibit what may be termed epistemic strength: epistemic states, in that some of them are preferable to others if one wants to be in as evidentially good a state with respect to some matter as one can; epistemic qualifiers, in that they are associated with this epistemic preferability through being correlated with epistemic states; and epistemic qualifying devices, in that they express these so correlated epistemic qualifiers. A number of conversational moves reflect this ordering of epistemic qualifiers, for example, those involving remarks of the form "It's not just that Q1(p); (but also/rather) Q2(p)," as in "I don't just believe that p; I know that p."

Three broad strength groups of epistemic qualifiers can be distinguished, using the relation of d-incompatibility, as follows: a strong epistemic qualifier Qs may be defined as one such that saying that Qs(p) is d-incompatible with saying that it may be that not-p; for example, "I know," "I'm certain," "It is certain." The flat-statement qualifier is among these. A moderate epistemic qualifier Qm may be defined as one such that saying that Qm(p) is d-compatible with saying that it may be that not-p, but is d-incompatible with saying that Qm(not-p); for example, "I am almost certain," "I believe," "It is likely." And a weak epistemic qualifier Qw may be defined as one such that saying that Qw(p) is d-compatible with saying that Qw(not-p); for example, "There is some evidence," "It is possible," ". . . may . . .".

Two epistemic qualifiers Q1 and Q2 can be incompatible in the sense that for no proposition p is any discourse coherent that contains both the remark that Q1(p) and the remark that Q2(p), neither having been withdrawn. (I refer to serious literal remarks here.) For example, the epistemic qualifiers ' know' and ' guess' are, I believe, incompatible in this sense.

I have reference throughout to the system expressible in English, which is I believe similar to that expressed in other familiar languages. This overall system of epistemic qualification seems to be largely constituted by relations of the various sorts that have so far been mentioned, together with further ones, some of which are mentioned later.

My epistemic qualification line on Moore's paradox is that these paradoxical utterances arise from epistemically qualifying propositional factors in an utterance in ways that are not consistent with the system of epistemic qualification.

Belief and Epistemic Qualification

"I think" and "I believe" do not allow all the same completions and constructions, but in the part of their use that I am concerned with here they seem to be synonymous, and I refer to them interchangeably. In particular, either can be used to formulate the original examples of Moore's paradox. Because they can also be so used for many of the additional examples dealt with here, it is important to clarify how the epistemic qualifier ' believe' functions in the system of epistemic qualification.

But the epistemic qualifying expression "I believe" is problematical in that it appears to correlate both with a less and also with a more comprehensive epistemic state. (By "more comprehensive" I mean that a given subject will almost always be in the more comprehensive state with respect to more propositions than he or she will be with respect to the other.) For example, it seems to have a broad sense in a question such as

(5) Did Aristotle believe that the earth was round?

where the fact that Aristotle regarded the matter as quite certain does not mean that sentence (5) should be answered in the negative; on the contrary, it would mean that it should be answered in the affirmative. Yet with respect to the question

(6) Did Aristotle know that the earth was round or just believe that it was?

that Aristotle knew or was certain that the earth was round would mean that the second alternative was a factually incorrect answer.

Thus, in the epistemic qualifier-epistemic state correlation, the narrow use of "I believe" seems not to correlate with the epistemic state referred to in historical belief reports but rather (if it correlates with any at all) with a narrower epistemic state. Yet we do not, I think, feel that "believe" has changed its meaning from the one case to the other or that we must ask for disambiguation of a sentence such as

(7) Aristotle believed that the earth was round.

I do not say that the string (7) is not, in context, associated with two different meanings (or propositional types), I think it is: one related to question (5), expressing an affirmative answer to it, and the other to a question such as question (6), giving the second alternative in answer. But I do not think these derive from two different lexical meanings of "believe."

We seem, then, to confront a puzzle about "I believe": How can it exhibit linguistic behavior of kinds associated with an ambiguous expression but apparently without being ambiguous?

Since Grice (1975), good method dictates that in analyzing conversational phenomena we try to explain as many effects as we can in terms of his maxims of conversation, rather than postulating a multiplicity of meanings of the expressions involved; this is vulgarly called "gricing" the use of the expression in question.

Let me try to do that here. Suppose that the actual lexical meaning of the verb "to believe" is that associated with the broader use. Then its curious behavior in Austinian position may be griced using the maxim of quality (or quantity), as follows. This epistemic qualifying expression functions to express an epistemic qualifier in situations in which it would be appropriate (according to this maxim) to say that Q'(p), with a stronger epistemic qualifier Q', if one candidly could (that is, if one was in a position to do so); because one does not, the inference is that one can not. So the force of "I believe that p" is (unless explicitly canceled) that of "I merely/just believe that p." The "merely/just" fits the case because the unused alternative is "I know" or "I'm certain," which involves stronger epistemic qualifiers. Because this sort of situation is often faced, what Grice calls a generalized conversational implicature would develop, so that the phrase would be regularly understood in the way inferred here, unless explicitly canceled. For the implicature to be a conversational (or generalized conversational) one, it does need to be cancelable; and it apparently can be canceled, as in "Aristotle not only believed that the earth was round; he was absolutely certain that it was."

(It is hard to know which of Grice's maxims is involved because he does not relate evidence and information to the epistemic qualification of the remarks involved. If epistemic qualification is included—and how can it not be?—then the maxim of quantity is presumably the maxim involved: "I believe that p" would convey less (purported) information than "I know that p," for example. If, somehow, the epistemic qualification is not included, then the maxim of quality would presumably be the one involved: The quality of "I believe that p," in the sense of the evidential basis for conveying that p (epistemic qualified that way) would not be as high as that of "I know that p." If the maxims were revised so as to deal explicitly with epistemic qualification, this problem would, I imagine, solve itself.)

If "I believe" can be griced in this way (or some other), it remains an open question whether or not knowledge entails belief. For simplicity, let us suppose that it does. Then personal epistemic qualifiers, at least, can be divided into those that entail belief (the doxastic epistemic qualifiers, as they may be called) and those that do not (although they are, perhaps, not incompatible with belief, for example, thinking it possible that p). There would be a broad epistemic state denoted by "believe," which, when its correlated epistemic qualifying expression occurred in Austinian position, would by the Gricean implicature (unless canceled) narrow down to the epistemic state correlated with just/merely believing, that is, to believing without (putatively) knowing. "I believe" would be correlated with the epistemic state of being in some doxastic epistemic state or other, which would be compatible (we are supposing) with knowing and also with not knowing but would be incompatible with not believing, whether in the sense of (A1) or in that of (A2).

III

John Searle's account of Moore's paradox is that, in a Moore-paradoxical utterance, such as

(1) It is raining, but I don't believe that it is

in the assertive utterance of the first conjunct, one does not say but rather expresses the putative fact that a certain "sincerity condition" holds, which the second conjunct explicitly denies. In the case of utterance (1), the sincerity condition is that the speaker believes that it is raining, and the denial of this condition is that the speaker does not believe it is raining. Thus Searle's account of Moore's paradox is explicitly tied to a speech act, the illocutionary act that would be performed in a normal assertive utterance of "It is raining" (and to the saying-and-disbelieving form of Moore's paradox).

Searle's account as he gives it cannot be correct, because his statement of sincerity conditions for assertive utterances is not correct. That it is not is not made especially salient by examples such as utterance (1), because "I don't believe that it is raining" expresses either disbelief (in the sense of believing that it is not the case that it is raining) or a denial of belief (its not being the case that one believes that it is raining), the utterance of either of which is d-incompatible with "it is raining." What one expresses by saying that it is raining is, however, not, as Searle says, that one believes that it is raining, if "believes" is taken in the narrow way, but rather the flat-statement epistemic qualifier, something similar to one's (not just thinking but) knowing that it is raining.

Others have said that what is implied or expressed by a speaker who says that p is either that the speaker believes or that the speaker knows that p, but I think this is pretty clearly wrong, because a Moore paradox of the given form (1) is forthcoming in any case. That something with this epistemic force is what is expressed is clear from the fact that, if what the speaker would express as

(8) I believe that it is raining

were what was expressed, that is, was the epistemically strongest thing expressed (as opposed to said, in a strict sense), then one could follow the utterance of "It is raining" with that of

(9) It may not be raining

without withdrawing or qualifying remark (8), just as one can follow an utterance of sentence (8) with an utterance of sentence (9) (linked with "but," say). But one cannot, just as one cannot follow an utterance of

(10) 1 know that it is raining

with an utterance of sentence (9), without withdrawing or qualifying remark (10).

If, on the other hand, it is the broader use of "believe" that Searle has in mind in connection with his sincerity condition, one that would cover claims to know that p, unadorned flat statements that p, saying that one believes that p, etc., then there is a problem about the sincerity condition of epistemically qualified statements. In the case of a statement such as (10), the sincerity condition would be (what the speaker could express as)

(11) 1 believe that I know that it is raining.

Similarly, for a statement such as

(12) I am certain that it is raining

the sincerity condition would be

(13) I believe that 1 am certain that it is raining.

That is, the epistemic qualifying expression of whatever qualifier was explicit in the original statement would turn up within the scope of "I believe" in the statement of the sincerity condition of the original statement. Now, although statements such as (11) and (13) have interpretations, they are not the right things to be sincerity conditions; rather, they are, as far as I can see, either metalinguistic statements (in the context equivalent to "I believe the right word to describe the situation is that I know that it is raining") or statements expressing uncertainty about the state one is in ("Yes, on further consideration it is certainty that I feel"), neither of which is necessarily present in sincere utterances of sentences (10) and (12) and certainly not as their sincerity conditions.

The correct sincerity condition in the case of utterance (12), surely, is just that the speaker is certain that it is raining. In the case of (10), where truth of the proposition said to be known is (presumably) logically entailed, the sincerity condition (which of course can be satisfied without entailing it) may be described as being that the speaker takes it that (or perhaps that the speaker assumes that) he or she knows that it is raining. In neither the case of statement (10) nor statement (12) is belief involved in the way Searle describes. Perhaps knowing and being certain that p are doxastic epistemic qualifiers. If so, then denying that one believes that p would be as Moore paradoxical as if believing that p were the sincerity condition of saying that it is raining. But "believe" taken broadly cannot, apparently, be used to state accurately an overall pattern of sincerity conditions of the kind Searle gives (1983, p. 9) for statemental utterances. The uniformity he finds in their sincerity conditions does not exist. These conditions are more specific and are relative to the epistemic qualifiers involved.

IV

According to the account of Moore's paradox that I wish to present here, Moore paradoxes are one kind of epistemic qualifier conflict, in the sense (already described) of d-incompatibility of epistemically qualified utterances. That is, Moore paradoxes are, on this account, cases in which epistemic qualifiers are so applied to propositional contents that d-incompatible (but not contradictory) illocutionary acts result. I now try to spell out what the conflict is in the case of Moore's original paradigms of the paradox. Their particular case is complicated by their involving the problematical concept of belief.

The system of epistemic qualification is constituted by various d-relations among epistemically qualified utterances; it is these relations that define the several roles of the epistemic qualifiers in the system. In some cases these relationships can be encapsulated in general principles or rules. I state several of these and use them in the explanations of Moore paradoxicality.

It appears to be the case that one of the general principles of the sort just mentioned is, briefly, that an epistemic qualifier d-incompatible with a given epistemic qualifier is d-incompatible with any stronger epistemic qualifier. More precisely stated:

(I) If saying that Q1(p) is d-incompatible with saying that Q2(q) and if Q3 is stronger than Q1, then saying that Q3(p) is also d-incompatible with saying that Q2(q).

Suppose that this is, in fact, a principle constitutive of or derivative from the system of epistemic qualifiers. Then one particular instance of principle (I) is that involved in a Moore paradox of form (Al), that is, the kind of case in which q = not-p, Q1 = Q2 = 'I believe', and Q3 = Qfs. 'I believe' is a moderately strong epistemic qualifier, so that (by definition of that strength) saying that Q1(p) is d-incompatible with saying that Q1(not-p), that is (in this case), with saying that Q2(q). Thus the first condition in the antecedent of (I) is satisfied for these choices of epistemic qualifiers and contents. Because Qfs is plainly stronger than 'I believe', the second condition is also satisfied, so that by (I) it follows that saying that Qfs(p) is d-incompatible with saying that one believes that not-p, which in form (Al) is what "I don't believe that p" means. Thus the two conjuncts in an instance of form (Al) are d-incompatible, and a fluent speaker is able to see right off that the utterance as a whole is something that in the relevant sense one cannot say. Furthermore, an instance of form (Al) satisfies the defining condition on Moore paradoxes that the utterance is not self-contradictory. As usual, we verify that such an instance is not self-contradictory by putting it into a non-Austinian position in its conjugation and finding that what it then expresses might well have been true; for example, it might well have been true that it was raining but I did not believe that it was.

Principle (I) may also be invoked to explain the paradoxicality of Moore paradoxes of the form (B). Because d-incompatibility is (presumably) a symmetric relation, the same choice of epistemic qualifiers, rewriting p as not-p and q as p, suffices.

Moore paradoxes of form (A2) require a more complicated treatment, because the scope of the negation in the second conjunct is here to be understood as wider than that of the epistemic qualifier 'I believe'. However, a minimal assumption from epistemic qualification-enriched speech-act theory seems to suffice, for there seems to be some such principle as

(II) Saying that Qfs(p) is d-incompatible with denying that Qm(p).

To take the case at hand, saying that it is raining (with flat-statement epistemic qualification) seems clearly d-incompatible with denying that one believes it is raining; thus the following conversation becomes odd at the point where such a denial would occur:

(C1) X: It is raining.

Y: You think it is raining?

X: No, it is not true that I think it is raining.

I mean, of course, ordinary or regular denying rather than metalinguistic denying, as in "It is not true that I think it is raining; I know it is." The almost overwhelming temptation to take X's second remark in dialogue (C1) to involve withdrawal of X's first remark would be explained by (Cl)'s otherwise being counter to principle (I). And a Moore paradox of form (A2) is an instance of principle (II), its two conjuncts being the incompatible utterances in question.

There is a problem inherent in trying to make plausible the postulation of principles such as (I) and (II), viz. that they must be (i) simple, familiar, and obvious, so as to make it plausible to impute knowledge of and obedience to them to speakers generally, and yet that (ii) they must fit the examples in question fairly simply, so that it may plausibly be supposed that their quick application is automatically made by a fluent speaker. Conditions (i) and (ii) operating together tend to make the principles look ad hoc. A methodological antidote is to show that there are many cases that can be explained by the principles, not just the ones one is working on. I do believe that their postulation will be borne out by such further consideration, although I cannot undertake it here.

As we saw and as Searle points out, his account of Moore's paradox would generalize to any illocutionary act with a sincerity condition, that is (if Searle is right (1969, p. 65, under 2)), to almost any illocutionary act whatever. But, unfortunately, his account is defective in at least the way indicated. However, some Moore-paradoxical-looking utterances do seem to be explicable in terms of his account but not mine, for example, his "I promise that p but I do not intend that p" (1979, p. 5). It may be that the propositional contents of these nonstatemental utterances are not epistemically qualified. If so, a different sort of explanation of their oddity than the present one is required. But even this sort of utterance can be odd in the way the present account claims that statemental Moore paradoxes are, for the present account of Moore's paradox generalizes not just to other sorts of illocutionary act (and not to them just because of their sincerity conditions, where, however, epistemic qualification may figure), but to utterances involving epistemic qualifier conflicts generally, which can manifest themselves in a variety of ways that are independent of illocutionary force, as I try to illustrate in a moment.

The question is, What exactly is necessary in order to have a paradoxical utterance resembling a Moore paradox (besides its being noncontradictory as a whole)? Because weak epistemic strength is not defined in terms of a d-incompatibility (but rather a d-compatibility), it might be thought that Moore-paradoxical utterances (in this perhaps extended sense) have to involve a pair of epistemically qualified propositions, at least one of which is at least moderately epistemically qualified. However, the following seems to be a counterinstance:

(14) It may be that his sister has gone, but it may be that he does not have a sister

where the first conjunct may (perfectly ordinarily and indeed as usual) be taken in a sense that accepts the pre-supposition (the denial of which is the proposition epistemically qualified in the second conjunct) that he has a sister, in which case the two conjuncts are d-incompatible.

I am not, in fact, at all sure what, in general, is necessary in order to have a Moore paradox. As we have seen and in the list to be given, a variety of features involved in or suggested by the original paradigms or by Searle's (or others') account of Moore's paradox are not, apparently, required in order to have what seems similar paradoxicality.

But let me say first what it is that on the present line is characteristic of these utterances: a conflict between two epistemically qualified propositions somehow or other involved in them. (The problem is to see how they must be involved in them.) Three things seem to figure in this, viz. (1) the unqualified content of the propositions, (2) their epistemic qualification (that is, how they are epistemically qualified, what epistemic force they are propounded with), and (3) the logical form, in a broad sense, of the utterance overall.

The following general remarks may be made concerning these factors: As to point 1, the contents may be contrary to one another in the familiar content-logical sense, that is, the truth-value-possibilities sense in which propositions not epistemically qualified may stand. This contrariety might, when these contents are epistemically qualified, make the overall utterance self-contradictory (in an utterance-logical sense, that is, in a sense appropriate to utterances involving epistemic qualification, rather than simply content-logical relationships), in which case it would (by definition) not be a Moore paradox. But the particular epistemic qualifiers involved might be personal epistemic qualifiers, in which case Moore's paradox could result, as in the original paradigms of forms (A) and (B). The statement "It is raining, but it is not raining" is just self-contradictory (content- and utterance-wise); but "It is raining, but I don't believe that it is" is a Moore paradox. "It is raining, but it may not be raining," with the impersonal epistemic qualifier '. . . may ... ' (= 'it may be') is perhaps problematic (because of its peculiar indirect-discourse properties), although maybe either a contradiction or a Moore paradox.

However, as to point 2, their epistemic qualification, the contents of two epistemically qualified propositions, may be consistent in the content-logical sense and yet produce a paradoxical utterance because of the relationship between the epistemic qualifiers (this being one constituent part of the system of epistemic qualifiers). The example in item (iii) on the following list, viz.

(15) It is raining, and/but it may be raining

involves one and the same self-consistent proposition as the contents involved, for example. The problem is with the epistemic qualifiers, the strong flat-statement qualifier and a weak epistemic qualifier, there being (to encapsulate this little part of the system) some principle, constitutive of these-respective epistemic qualifiers, to the effect that

(III) One cannot say that Qfs(p) and Qw(p).

So another sort of conflict between epistemically qualified propositions derives not from a conflict between their contents but rather from the ways those contents are there epistemically qualified.

But, to bring in point 3, the epistemically qualified propositions involved in the conflict need to occur in certain ways in the overall utterance; the kind of occurrence I call a matter of the "logical form" of the utterance, including semantic presupposition (if any) as one facet thereof. To illustrate, look at the logical form (in this sense) of an utterance of sentence (14); as there glossed, this form needs to be distinguished from that of an utterance involving the same sentence, but, as we may assume has been made clear by a different sort of context, not presupposing that the person referred to has a sister. More generally, the logical form of an utterance that is a conjunction or disjunction obviously needs to be distinguished from one that is a conditional or a generalization. As another illustration of how logical form can affect paradoxicality or the lack of it in an utterance, consider disjunctive utterances. The epistemic qualification of the disjuncts in such an utterance will, I believe, often not be epistemically qualified (although the disjunctive content as a whole will) and so a fortiori will not give rise to Moore's paradox in the ways discussed concerning points 1 and 2, whereas of course it was the epistemic qualification of the conjuncts of a conjuctive utterance that gave rise to the original paradigms of Moore paradoxicality.

Thus various restrictions on the form and/or the content of Moore-paradoxical utterances that, from the original paradigms and early exemplars, might have been expected to hold apparently do not do so universally. The following list contains at least some of these restrictions, with some illustrative examples. In such an utterance:

(i) there is no restriction on the propositional content epistemically qualified;

(ii) it is not required that one conjunct be a flat statement and the other be explicitly epistemically qualified; for example, "It is certain that it is raining, but it may not be";

(iii) it is not required that one conjunct be negative and the other affirmative; for example, "It is raining, and/but it may be raining";

(iv) it is not required that one propositional content be logically incompatible with the other—take the last example;

(v) it is not required that both conjuncts be statements, for example, "1 realize that it is raining, but may it not be?";

(vi) it is not required that the speaker, in one conjunct, say something conflicting with what saying the other only implies or expresses; compare the example in (iii);

and, in fact:

(vii) neither conjunct need be a statement; for example (assuming a bequest is not a statement), "To my only son I bequeath my gold watch and to my third son, if it turns out that there is such a person, the rest of my estate";

(viii) the utterance need not be conjunctive in form; for example, "If, as is likely, her second child is a boy, she must regret having had only one child";

and of course:

(ix) sincerity conditions of illocutionary acts need not be involved.

V

What if the component parts of an odd utterance arc separated, as they might be in a speaker's contributions to a conversation, say, or in a one-speaker discourse? It seems clear from the sorts of example we have been considering that the two components would still be related in the way they were in the paradox. But of course the conflict is not perceived as a single, impossible, paradoxical, odd utterance but rather as a kind of inconsistency in one's remarks, as a situation in which, typically, not both of a certain pair of things can be said, because incoherence in the total discourse would result. Or perhaps the person is not speaking honestly or candidly; or perhaps one has fogotten what one said earlier or has changed one's mind about what one wants to say. There may be many interpretations of the situation. Whatever the cause of the incoherence, the speaker cannot rest with saying both of the things which, placed together, would be Moore paradoxical; the speaker must withdraw at least one of them. The speaker must, or his or her discourse, in this part, will be rendered incoherent; we will not be able to understand it or perhaps the speaker.

But now, despite the apparent fact that the requisite relationships between the components still hold, the present dimension is, for some reason, not one along which Moore paradoxicality is apparently preserved. For some reason utterances of this type are not perceived as odd or paradoxical but rather as rendering the discourse incoherent or inconsistent. Unless it is simply the fact that two bursts of speech are involved, I do not know why this is, but I suspect that it has to do with the fact that there has (typically) been time in the discourse for the speaker to forget or revise what s/he said earlier (as of course we often do without much if any notice).

Also, up to now, it has been a question of epistemic qualifier conflict in just a single speaker's discourse, whether compressed into a Moore paradox or similar odd utterance or (as just discussed) spread out across an inconsistent or incoherent discourse. Allowing more than one speaker to be involved—one making one of the component remarks and one making the other, results in more than just a pair of speakers taking turns giving their opinions to each other. Nor do we get incoherent discourse (necessarily). What we get is disagreement. I believe that the situation is, in fact, that any Moore-paradoxical utterance, including the odd utterances of various sorts considered, will, if one speaker illocutes one component and another speaker illocutes the other, constitute their disagreeing with each other, either of the direct confrontation variety or (for example) of the kind where one disagrees with something the other is assuming, or perhaps some other. Note, for example, that

(16) I think it is, but/and I think it is not

(not taken in the sense in which it is used to introduce a distinction or a hedge) is a Moore-paradoxical utterance and that in the two-person conversation

(C2) X: 1 think it is.

Y: 1 think it is not.

the speakers disagree despite the apparent fact that each is referring to himself, to what he himself thinks. Yet in the single utterance

(17) It may be and it may not be

there is no oddity; nor is there disagreement (but only a reminder) in the conversation

(C3) X: It may be.

Y: And it may not be.

Thus there are different sorts of discourse relations, among which are discourse incompatibilities. A particular case of the latter are epistemic qualifier conflicts. When epistemic qualifier conflicts occur at their minimal limits along any of several lines, one reaches hypothetical utterances that resemble the original paradigms of Moore's paradox. Thus what may be (or may be regarded as) linguistic oddity of the same sort they exhibit spreads out over language as a whole, appearing in different guises as it goes: sometimes as Moore paradoxicality, sometimes as self-contradiction, and finally as incoherence or inconsistency in one's discourse and as interpersonal disagreement.

NOTE

¹ Single quotes will be used to form names of what will be called epistemic qualifiers (for example, 'I know'); double quotes, to refer to linguistic expression (for example, "I know").

REFERENCES

Alston, W. P. 1964. Philosophy of Language. Englewood Cliffs, N.J.: Prentice-Hall.

Austin, J. L. 1962. How to Do Things with Words. Cambridge, Mass.: Harvard University Press.

Caton, C. E. 1966. "On the general structure of the epistemic qualification of things said in English." Foundations of Language 2: 37-66.

Caton, C. E. 1981. "Stalnaker on pragmatic presupposition," in Radical Pragmatics, Peter Cole, ed. New York: Academic Press, 83-100.

Grice, H. P. 1975. "Logic and conversation," in The Logic of Grammar, G. Harman and D. Davidson, eds. Encino, Calif.: Dickenson, 64-75. Also in Syntax and Semantics 3: Speech Acts, J, L. Morgan and P. Cole, eds. (New York: Academic Press, 1975), 41-58.

Searle, John R. 1969. Speech Acts. Cambridge, England: Cambridge University Press.

Searle, John R. 1979. Expression and Meaning. Cambridge England: Cambridge University Press.

Searle, John R. 1983. Intentionality. Cambridge, England: Cambridge University Press.

Toulmin, S. E. 1958. The Uses of Argument. Cambridge, England: Cambridge University Press,