Huw Price Speech act theory is one of the more lasting products of the linguistic movement in philosophy of the mid-Twentieth century. Within philosophy itself the movement’s products did not in general prove so durable. Particularly striking in this respect is the perceived fate of what was one of the most characteristic applications of the linguistic turn in philosophy, namely the view that many traditional philosophical problems are such as to yield to an understanding of the distinctive function of a particular part of language. Most typically, the crucial insight was held to be that despite appearances, the function of the part of language in question is not assertoric, or descriptive, and that the traditional problems arose at least in part from a failure to appreciate this point. Thus problems in moral philosophy were thought to yield to an appreciation that moral discourse is expressive rather than descriptive, problems in the philosophy of mind to an understanding of distinctive r™le of psychological ascriptions, and so on. The philosophical journals of the 1950s are rich with views like these. (No general term for this approach seems to have become widely accepted at the time. I shall call it “non-factualism”, for what it denies, most characteristically, is the fact-stating r™le of language of a certain kind.)
At the time, many of these non-factualist endeavours drew on the new terminology of speech act theory, taking their lead at least in part from J. L. Austin. It is therefore somewhat ironic that when non-factualism came to seen as discredited, one of the works responsible was Searle’s Speech Acts.2 Non-factualism was thus disowned by the movement from which, at least in part, it drew its inspiration. So it is that while speech act theory prospered outside philosophy, its early pretensions to application within philosophy were reviled or forgotten. Non-factualism was widely thought to have fallen victim to objections urged in the 1960s by Searle, and independently by Peter Geach (who took his inspiration from an argument of Frege’s).
Philosophical demise is rarely complete or permanent, however, and non-factualism has been receiving renewed attention more recently, particularly in a relatively new application to the problem of linguistic and psychological content.3 It would now be easy for a newcomer to fail to notice that for almost a generation the approach was commonly taken to be discredited. It therefore seems worth re-examining the supposedly fatal objection. After all, perhaps non-factualism really is dead, or as dead as a philosophical view can be, and its new devotees simply haven’t noticed. If not, then it would be nice to know how it managed to recover from what many took to be a mortal blow.
The paper thus begins with a brief reassessment of what I shall call the Frege argument (though I shall draw on the versions of the argument advanced by Geach and Searle). One possible outcome of this investigation would be a reaffirmation of the conclusions drawn by Geach and Searle, and thus a return to the status quo circa 1965—perhaps an unexciting result, but a useful one, if the Frege objection succeeds, given non-factualism’s current reluctance to lie down. The actual outcome is rather more interesting, however. For one thing the Frege argument turns out to be considerably less powerful than it has been taken to be, so that non-factualism remains a live option. Given the perceived importance of the Frege argument to the “overthrow” of linguistic philosophy, this conclusion suggests that contemporary philosophy might do well to reconsider. There are many contemporary metaphysical debates which would have looked sterile and misconceived to the linguistic philosophers of the 1950s. Without the Frege argument to fall back on, it would be a brave—or perhaps foolhardy—philosopher who would dismiss out of hand the linguistic point of view.4
In the present paper, however, I want to emphasise a different benefit of re-examining the Frege argument. As we shall see, the issues thereby thrown open are ones of fundamental concern in the philosophy of language and the foundations of speech act theory. In hindsight I think it is clear that when speech act theory detached itself from philosophy in the 1960s, a cluster of central issues concerning the nature of assertion, judgement, description, and the like, were left largely unresolved. I hope to show that to re-examine the Frege argument is to re-open these issues in a particularly fruitful way.
The paper is in three main parts. In the first (sections 1 to 4) I argue that the Frege argument is far from conclusive. It imposes certain constraints on the non-factualist, but fails to show that these constraints cannot be satisfied. I shall mention work by some prominent non-factualists that went some way towards showing how their view might meet these constraints. The upshot seems to be that the worst that the non-factualist can be convicted of is a degree of complexity in linguistic theory that factualist views seem to avoid—and for all its unpleasantness, complexity is rarely a fatal complaint.
All the same, the desire to free non-factualism of this complexity motivates the second part of the paper (sections 5 to 7). This part draws on recent interest in what I here call minimal semantics, extending the terminology employed in discussions of so-called minimal theories of truth.5 Briefly, I suggest that non-factualists might (i) concede that moral claims (or whatever) are statements in some minimal sense, and use this concession to meet the requirements identified by the Frege argument in the same direct and simple way that is available to a factualist; but (ii) reformulate their point about the character of moral claims in such a way that it does not conflict with the proposition that such claims are statements in the minimal sense. The move to a minimal semantics thus enables the non-factualist to sidestep the Frege argument.
I want to suggest that in the process we achieve a fresh and illuminating view of the relationship between truth-conditional semantics and the sort of pragmatic considerations about language often thought to be the proper concern of a theory of force, or speech act theory more generally. As reformulated non-factualism directs our attention to the function of particular parts of discourse. (This functional side of non-factualism is not new, of course; what is new is that it should be clearly divorced from a claim about the semantic status of the utterances in question.) The recognition that non-factualism need not be a semantic doctrine then enables us to regard functional pragmatics not as an addition tacked on to deal with the problems of force and tone, but as a complement to the theory of sense whose task is to explain how there come to be uses of language with senses of particular sort—how there come to be utterances with the sense of moral judgements, for example.
True, it is not clear that the reformulated doctrine should really be called non-factualism. As we shall see, it no longer involves the denial that the utterances of some disputed class are factual, or assertoric. Instead it treats these as relatively superficial and uninteresting linguistic categories, overlying diversity of a different kind. It is this separation of semantic and functional categories which seems to me of most interest to speech act theory. It suggests for example that assertion is a very much less fundamental linguistic category than has usually been assumed. At best it is a kind of higher-order category, grouping together some very diverse linguistic activities.
All the same, the question arises as to what these diverse activities have in common, in virtue of which they all come to be part of this single higher-order category. In the third part of the paper (section 8) I conclude by drawing attention to this central issue, an issue which has tended to be overlooked in earlier work. I note that there is a sense in which the issue embodies some of the insights of the Frege argument, and hence that things are not quite so easy for my reconstituted non-factualist as they earlier appeared; but I also note that the issue is not one that the non-factualist’s opponents can afford to shirk, so that the dialectical burden of the new issues is evenly spread.
1. The Frege-Geach-Searle arguments, and Searle’s unused loophole.
The Frege argument begins by observing that non-factualist accounts characteristically propose an interpretation of just those (canonical) sentences or utterances in which constructions of the relevant type—“It is probable that ...”, “It is good that ...”, “It is true that ...”, or whatever—are not part of any clause other than a complete sentence. It is noted that there are many other (subsidiary) occurrences of such constructions, and argued that the proposed accounts are unable to deal with at least some of these new cases, though obliged to do so. As Geach says,
Theories of non-descriptive performances regularly take into account only the use of a term ‘P’ to call something ‘P’; the corroboration theory of truth, for example, considers only the use of ‘true’ to call a statement true, and the condemnation theory of ‘bad’ considers only the way it is used to call something bad; predications of ‘true’ and ‘bad’ in if or then clauses, or in the clauses of a disjunction, are just ignored.
One could not write off such uses of the terms as calling for a different explanation from their use to call things true or bad; for that would mean that arguments of the pattern ‘if x is true (if w is bad), then p; but x is true (w is bad); ergo p’ contained a fallacy of equivocation, whereas in fact they are clearly valid.6
Searle’s version of the argument is somewhat different, in that he admits a possibility which Geach’s appeal to the validity of modus ponens would appear to exclude. Searle is objecting to what he calls “the speech act analysis” of words such as “good”, “true”, “know” and “probably”, the general form of which he takes to be: “The word W is used to perform the speech act A.” Searle says that
any analysis of the meaning of a word (or morpheme) must be consistent with the fact that the same word (or morpheme) can mean the same thing in all the grammatically different kinds of sentences in which it can occur.
the word ‘true’ means or can mean the same thing in interrogatives, indicatives, conditionals, negations, disjunctions, optatives, etc. (Searle 1969, p. 137)
However, Searle recognises that in order to meet this “condition of adequacy”, speech act analysts are
not committed to the view that every literal utterance of W is a performance of A, but rather [may claim] that utterances which are not performances of the act have to be explained in terms of utterances which are. (Searle 1969, p. 138)
Searle thus appears to acknowledge that it need not be said that the contribution the clause makes to the meaning of a conditional in which it occurs as antecedent is identical to the meaning it has when used canonically; but only that the former contribution depends in a rule-governed way (the rule being associated with the conditional form) on the meaning the clause has in the latter case. If Geach’s appeal to validity were successful, this view would seem untenable. The validity of modus ponens would depend on the meaning of such a clause being invariant between the two contexts.
Having admitted this possibility, however, Searle fails to take advantage of it. He rightly points out that
the speech act analysts ... need to show ... only ... that literal utterances which are not performances of the act A stand in a relation to performances of A in a way which is purely a function of the way the sentences uttered stand in relation to the standard indicative sentences, in the utterance of which the act is performed.
But he takes this to mean that if such sentences “are in the past tense, then the act is reported in the past; if they are hypothetical then the act is hypothesized, etc.” He then notes the obvious, namely that “the speech act analysis of the ... words: ‘good’, ‘true’, ‘probable’, etc. does not satisfy this condition. ... ‘If this is good, then we ought to buy it’ is not equivalent to ‘If I commend this, then we ought to buy it’; ‘This used to be good’ is not equivalent to ‘I used to commend this’ ”; and so on. (Searle 1969, pp. 138-9)
Although Searle himself does not canvas other ways in which the meaning of clauses such as “It is good that P” in various contexts may be systematically related to their meaning when they stand alone, it is clear that if the general objection is to be answered the solution will lie in this direction. However, the argument from modus ponens claims to bar the way. Let us test its strength.
2.The appeal to modus ponens.
As Geach notes, this argument is due originally to Frege (Frege 1960, at pp. 129-30), who uses it in arguing that a sentential negation operator cannot be construed as a sign of force; as an indication that a sentence, when uttered, has the force of a denial. Frege’s argument is in two parts:
Fr1 He notes that a negated sentence may occur as the antecedent of a conditional, where it does not amount to a denial, and concludes that in such a case the negation contributes to the sense of (or thought expressed by) the antecedent.
Fr2 He infers from this that if we want to allow that a case of modus ponens involving such a conditional is valid, we shall have to allow that the negation does not mark a denial, even when the negated sentence concerned stands alone.
The general principle invoked in Fr1 is something like this:
Embedded force exclusion (EFE)
Force modifiers cannot occur in embedded contexts.
We shall come back to this, but let us first consider Fr2. Here the argument might seem to be that the validity of modus ponens depends on the meaning of the antecedent clause in the conditional premiss being exactly the same as it is when the clause occurs alone (as in the categorical premiss). It would follow that because (according to Fr1) the negative clause is not a denial in the former context, it is not a denial in the latter. But as Hare points out (Hare 1971, p. 87) the same argument would show that when the clause stands alone it does not have the force of an assertion; for it lacks this force when used as an antecedent.
A more charitable interpretation is therefore that the argument for Fr2 depends on the following claim:
is valid only if the second premiss has the same sense (or expresses the
same thought) as the antecedent of the conditional premiss.
If we grant the conclusion of Fr1—i.e., that the negation operator has a sense-modifying r™le in determining the meaning of the conditional premiss in (1)—then SI implies that its r™le in the second premiss must also be to modify sense. Thus as Fr2 claims, the negation operator does not modify force, even in canonical cases.
The function of the appeal to modus ponens is therefore to extend the conclusion of Fr1 to canonical uses of the negation operator (and similarly for such things as modal and ethical operators, in Geach’s case). But how is SI to be justified? Not, on the face of it, by Geach’s remark that otherwise the inference would contain a fallacy of equivocation. Of course, there are fallacious arguments of the syntactic form “if P then Q; P; therefore Q” in which the fallacy turns on the fact that P is used with different senses in each premiss. However, to claim bluntly that any argument of this kind is fallacious is just to beg the question (given that both sides agree that (1) is valid). For both sides agree that this claim is incompatible with the view that the two occurrences of “not‑P” in (1) have different senses; but the disagreement is precisely as to which of these incompatible propositions must be given up.
In any case, the use Frege and Geach make of SI depends on the principle EFE. It is EFE which underpins the claim that in the antecedent of a conditional a negation operator modifies sense. But what are the grounds for accepting EFE? Apparently just the observation that in such a context no denial is being made. But this involves the very mistake we noted in the previous section, the loophole for avoiding which is recognised (if not adequately exploited) by Searle. In effect Searle recognises that in order to make sense of an occurrence of a denial operator in an embedded context, it is not necessary to say that such a subsidiary use has exactly the meaning it has when it stands alone. It is enough that its contribution to the meaning of the containing context should depend on the fact that it does signal a denial, when used canonically. For then there is a clear reason for including a force-indicator for denial in the subsidiary positions concerned: in order to show that the clause would have this force, if uttered alone.
We saw that Searle himself does not take advantage of this loophole. But so far we have found nothing in the argument from modus ponens that provides an obstacle to others doing so. On the contrary, the appeal to modus ponens has to this point depended on the assumption that no such loophole exists.
3. The attractions of uniformity.
Frege and Geach do have another argument for SI, however, also appealing to modus ponens. Unlike the above argument, this one does not rely on the sub-argument Fr1. Indeed it offers an independent argument for the conclusion of Fr1 (i.e., that in the antecedent of a conditional the negation operator modifies sense). This argument begins by noting that we evidently do have identity of sense in
(2) If P then Q; P; therefore Q (where P is not negated)
and moreover that this identity of sense is clearly crucial to the validity of the argument form. It then claims that if (1) is to exemplify the same form of inference—as in some sense it surely does—then identity of sense must play the same r™le. Uniformity seems to require that there be a common account of the conditional form, in the light of which identity of sense plays a constant r™le in guaranteeing validity. Thus this is an appeal not to a necessary condition for validity as such, but to the need for a uniform explanation of the validity of a class of inferences which evidently have a structural property in common.
Such theoretical uniformity is undoubtedly desirable, but is the only way to achieve it to treat (1) as a special case of (2)? Why not instead treat (1) and (2) as distinct sub-types of a single more general form of inference? It is not obvious that in that case the general criterion for validity would include the required identity of sense. There might rather be some more general condition, which reduced to identity of sense in the special case of (2). In the next section I outline an account of this kind.
4.Conditionals for non-factualists.
In summary then, the task of a non-factualist who wishes to evade the Frege argument seems to be twofold: first, to find a legitimate account of the significance of a force-modifying construction in a subsidiary clause; and second, to produce a general account of the linguistic function of the “if ... then ...” construction, such as to enable valid arguments to contain such force-modifiers in (at least) the antecedent position. The latter project is best tackled first, for the significance of a subsidiary force-modifier will inevitably depend on the nature of the subsidiary context in question. We shouldn’t expect a single account, applicable to any and every subsidiary context. The individual accounts will of course have something in common, but this may be nothing more than a common reference to the meaning that the force-modifier in question has in a canonical context.
Now in arguing that the utterances of some disputed class are not genuine assertions, non-factualists commonly rely on a distinction between beliefs and others sorts of propositional attitude. With this psychological distinction assumed in place, the non-factualist argues first, that we may characterise assertion as the linguistic expression of belief; and second, that the disputed utterances express some other sort of propositional attitude. Thus Frege’s opponent might tell us that negated sentences express disbeliefs rather than beliefs, the emotivist tells us that moral judgements express evaluative attitudes, the probabilistic subjectivist tells us that utterances of the form “It is probable that P” express the speaker’s high degree of confidence that P, and so on.
What concerns us here is not whether this is an adequate route to non-factualism in general, but the fact that by characterising force in terms of an associated type of propositional attitude, it provides the means to escape the Frege objection. The strategy requires that indicative conditionals themselves be treated non-assertorically. A sincere utterance of “If P then Q” will be said to indicate that a speaker possesses what may be called an “inferential disposition”—a mental state such that if the speaker were to adopt the mental attitude associated with the utterance “P”, she would be led to adopt the mental attitude associated with “Q”. For example the utterance “If it is not snowing, then Boris has gone swimming” will be said expresses a disposition to move from a state of disbelief that it is snowing, to a belief that Boris has gone swimming.
This suggestion provides a clear sense in which the force-modifying expression makes the same contribution to a canonical utterance, as to a conditional utterance in which it occurs in the antecedent or consequent. In each case it marks the association of the meaning of the whole utterance with a certain kind of propositional attitude: a disbelief, a degree of confidence, an evaluative attitude, or whatever. Other features of the particular occurrence of the expression in question determine firstly which particular propositional attitude of the given type is involved—its content, in other words—and secondly, how this propositional attitude stands in relation to the mental state associated with the utterance as a whole. For example in the canonical case for negation (an utterance of the form “Not P”) the fact that negation is the outermost operator indicates that the mental state associated with the utterance as a whole is just disbelief itself. While in the conditional case, the occurrence of the expression in (say) the antecedent position indicates that possession of the state of disbelief in question is the antecedent condition of the inferential disposition associated with the conditional. (This process of determination may be iterated, if the conditional itself occurs as a component of some larger utterance.)
It is important to distinguish this suggestion from the claim that a conditional reports a speaker’s possession of such an inferential disposition. If that were so, a conditional utterance would be an assertion about its speaker’s state of mind, and would be true or false according to whether the speaker concerned actually had such an inferential disposition. However, the proposal is intended to explain the meaning of the conditional in terms not of its truth conditions but its subjective assertibility conditions—i.e., in terms of the state of the speaker that normally licences its correct use. (The term subjective assertibility condition is being used in the sense involved in saying that the normal condition for the correct use of a statement P is that one believe that P. To say this is not to say that in asserting P one asserts that one believes that P.7)
The above proposal is similar to, though perhaps a little more psychologically explicit than, one made by Hare in answer to the Frege objection. Hare puts the common central insight rather nicely, saying that we know the meaning of the conditional “if we know how to do modus ponens.” In other words, the crucial thing is that we are in a position to affirm “If P then Q” “if we know that if we are in a position to affirm [P], we can go on to affirm [Q].” (Hare 1971, p. 87) Thus to say “If not-P then Q” is to indicate (though not to say) that one’s state of mind is such that if one were to deny that P, one would affirm (or be prepared to affirm) that Q. The correctness of the inference (1) thus amounts to the fact that (1) is the very inference a readiness to make which is signalled by the conditional premiss; and of course the same may be said about (2). In both cases the correctness of the inference is thus analytic: the standard use of the conditional is just such as to licence modus ponens. Moreover, the r™le of the force-modifying negation operator in the antecedent of the conditional is now clear. It helps to specify the nature of the circumstances in which the speaker indicates that she would be prepared to affirm the consequent—namely those circumstances in which she would be prepared to deny that P.
In Hare’s form or mine, this account is of course only a beginning. Much work would need to be done to show that the notion of an inferential disposition leads to a satisfactory account of ordinary language indicative conditionals, and of simple logical inferences in which they occur. And even if the account works for conditionals, it needs to be extended to the many other subsidiary contexts in which (what the non-factualist regards as) force-modifying operators may occur. For each such context we need a principle which links the general linguistic function of the context itself to the working hypothesis about operators in question, namely that their independent use is to signal a non-assertoric force of some kind. Even as it stands, however, the suggested account of conditionals does serve to establish a crucial general point. To paraphrase Hare (who is concerned with the moral case, of course):
The fact that sentences containing negation cannot be described without qualification as assertions, but have to be explained in terms of the more complex speech act of denial, is no bar to the appearance of negation in contexts where denial is not taking place, provided that the relation of these contexts to those in which it is taking place can be explained. (Hare 1971, p. 93)
Hare’s is not the only attempt in the literature to offer an account of conditionals with non-assertoric antecedents and consequents. I have already mentioned that of Michael Dummett (see foonote 7). Simon Blackburn also addresses the problem, again with the intention of defending a form of ethical non-factualism against the Frege argument. His suggestion is that
(3) If it is good that P then it is good that Q is itself an evaluative remark: roughly, it expresses a speaker’s approval of the disposition (or as Blackburn calls it the “moral sensibility”) to approve of Q, given that one approves of P.8 Like Hare’s theory and mine, this account has the crucial feature that it makes the significance of an embedded force-modifier dependent on but not identical to the significance it has in a canonical context. That said, however, it seems to me that Blackburn’s account is less plausible than the approach sketched above. It has the disadvantage that it does not give us a single unified account of conditional utterances, from which the required account of conditionals with embedded moral clauses falls out as a special case—conditionals in general are not expressions of moral sensibility. I suspect that Blackburn has confused two notions of endorsement, the first the semantic endorsement we give to any proposition when we assent to it, and the second the peculiarly moral endorsement we give to an act or state of affairs of which we approve. It is arguable that assent always involves an evaluative or normative element. To assent to a proposition is to take it to be right, correct, true. But this simply means that assent to an ethical proposition involves two sorts of evaluative attitude. To agree that war is evil is to take the proposition “War is evil” to be correct, to endorse it in that sense; and it is also to express one’s disapproval of war. With these notions of endorsement kept distinct, however, there seems no reason to say that accepting a moral conditional necessarily involves anything more than semantic endorsement. It need not itself express a moral attitude, even though it may indicate a certain structure of dependencies between the speaker’s moral and non-moral attitudes. “If all war is evil then the Gulf War was evil” is merely a logical truth.9