Strawson on 'if' and ⊃

Strawson on ‘if’ and É1 Gunnar Björnsson Department of Philosophy University of Gothenburg Box 200 SE 405 30 Göteborg Sweden Email: gunnar.bjornsson@filosofi.gu.se Abstract This paper is concerned with Sir Peter Strawson’s critical discussion of Paul Grice’s defence of the material implication analysis of conditionals. It argues that although Strawson’s own ‘consequentialist’ suggestion concerning the meaning of conditionals cannot be correct, a related and radically contextualist account is able to both account for the phenomena that motivated Strawson’s consequentialism, and to undermine the material implication analysis by providing a simpler account of the processes that we go through when interpreting conditionals. 1. Introduction In his (1986) paper, ‘If’ and É’, Sir Peter Strawson made a number of interesting moves and remarks with relevance to both the Gricean programme of semantic Occamism and the theory of conditionals. Interestingly, Strawson argues that if the material implication analysis – or ‘materialism’, as I will call it – can be defended for indicative conditionals, it should be a viable theory of counterfactuals as well. His paper also sketches a negative argument against the Gricean defence of materialism, and a positive argument for what we might call ‘consequentialism’, the view that conditionals conventionally imply that some ground-consequence relation holds between the proposition expressed by the if-clause and the proposition expressed by the main clause. In this paper, I will do four things, all of which relate to Strawson’s argument. I will bolster Strawson’s case for a consequentialist account of conditionals; I will argue that this account, as it stands, faces exactly the same problems as materialism; I will explain how a radically contextualist account of conditionals avoids these problems while preserving what seems reasonable about consequentialism; finally, I will argue that such a radical contextualism surpasses materialism in what has been seen as its forte: explanatory simplicity. 2. Materialism within the Gricean programme According to materialism, the meaning of an indicative conditional, ‘if P, Q’ is given by the truth-conditions of the material implication: the conditional is false if P is true and Q is false; otherwise it is true. The standard reason for accepting this view is that 1 This paper was made possible by grants from The Bank of Sweden Tercentenary Foundation. S. Afr. J. Philos. 2008, 27(3) 25 it is the only truth-functional account of the indicative conditional that preserves the validity of modus ponens and modus tollens and explains how we can know that if P, Q solely on the basis of knowing that it is not the case that both P and not-Q. There are well-known problems with materialism, of course. For example, conditionals (1) through (4) below are true or highly likely according to materialism, but they are false, objectionable, or unintelligible according to common sense: (1) If Paris is the capital of Spain, then Cambridge has at least as many inhabitants as Tokyo. (2) If Paris is the capital of Spain, then Tokyo has at least as many inhabitants as Cambridge. (3) If Cambridge has at least as many inhabitants as Tokyo, then Paris is not the capital of Spain. (4) If Tokyo has at least as many inhabitants as Cambridge, then Paris is not the capital of Spain. The antecedents of (1) and (2) are known to be false, and the consequents of (3) and (4) are known to be true, so materialism implies that all these conditionals are known to be true. But apparently, normal speakers of English fail to see this. And they fail to see this because it is hard to see how the consequent would follow from the antecedent. By contrast, ordinary non-puzzling conditionals of the same form provide this consequentialist feature: (5) If Paris is the capital of Spain, then I am badly confused. (6) If you have two feet, then you have two legs. (7) If Sarah has measles, then she has a fever. (8) If the barometer is falling, then we will have rain. The Gricean notion of conversational implicature promises an elegant way of explaining our reactions to (1) through (4), without giving up materialism or postulating semantic ambiguities. Paul Grice (1989) suggested that interpretation labours under the presumption that the speaker is cooperative, assuming among other things that her assertions are as informative as needed, and not more (adhering to the maxim of quantity), and that she makes them on good evidence (adhering to the maxim of quality). This often leads us to understand utterances and sentences as meaning to communicate something other, or more, than what our words conventionally express. Such extra content is what is typically called ‘conversational implicature’, since it is implied by the act of uttering a certain sentence rather than by the sentence itself. To illustrate: even though its two conjuncts might both be true, the sentence ‘There is beer in the fridge, but I don’t believe it’ sounds contradictory, because on pain of violating the maxim of quality, someone who asserts ‘There is beer in the fridge’ must think that there is good enough reason to believe the statement, and this implicature is undermined by the second conjunct. Apply the assumption of cooperation to materialism. To satisfy the maxim of quality, the speaker needs to have good evidence that not both the antecedent and the negation of the consequent are true. Such evidence could take three possible forms: (A) Evidence that the antecedent is false. (B) Evidence that the consequent is true. 26 S. Afr. J. Philos. 2008, 27(3) (C) Evidence that is independent of the first two kinds of evidence: evidence of some fact that would allow one to infer the consequent from the antecedent, or the negation of the antecedent from the negation of the consequent. If the evidence is of the first two kinds, the maxim of quantity dictates that the speaker should assert that the antecedent is false, or that the consequent is true, rather than make the conditional statement. From the assumption that the speaker is cooperative, hearers can then conclude that the consequent can be inferred from the antecedent. This is especially clear when it is common knowledge that the antecedent is false, or that the consequent is true. No wonder, then, that (1) through (4) prompt puzzlement or rejection: they conversationally implicate that there is evidence of kind (C) that would let us infer the consequents from the antecedents, but we fail to see how there could be such evidence. If this is correct, some cases in which speakers are known to not be fully cooperative should reveal the ‘bare’ material truth-conditions. And it appears that they do. I hold out my hands to a child with candy hidden in the right hand, saying: ‘If it isn’t in the right hand, then it is in the left hand, and if it isn’t in the left hand, it is in the right hand’. Here it is clear to the child that I know where the candy is, but that I am withholding that information. And here, it seems, what I am saying is merely that there is candy in one of my hands: it is not both the case that the candy isn’t in the right hand and that it isn’t in the left hand. Should we accept this Gricean explanation of our reactions to conditionals (1) through (4), there would be no need to assume that ‘if P, then Q’ conditionals conventionally indicate that the consequent is inferable from the antecedent: materialism added to completely general principles of communication would be quite sufficient. Materialism would be saved by the Gricean credo of semantic Occamism, which tells us to refrain from postulating conventional meaning to expressions beyond what is necessary for explanatory purposes. This story is a paradigm of theoretical beauty. Moreover, as Strawson argues, it has an equally elegant but rarely noticed sequel: it straightforwardly extends to cover subjunctive or counterfactual conditionals. The problem in applying materialism to subjunctive conditionals has always been that subjunctives tend to be uttered in contexts where the material implication is obviously true, because the antecedent is taken to be false. Furthermore, they often seem to suggest counterfactuality by their very form: ‘If he had been at home now, then we could have called him’ strongly suggests that he is not home. Since we do not take counterfactuals to be trivially true, materialism seems inappropriate. But supposing that Gricean mechanisms can explain our reactions to (1) through (4), there is excellent reason to believe that we would take counterfactuals to convey some kind of consequence relation between antecedent and consequent: it is very clear that uttering the conditional would otherwise violate the maxim of quantity (Strawson 1986: 238-9). 3. From Materialism to Conventionalised Consequentialism Although seeing its beauty, Strawson finds the Gricean story unrealistic and ultimately untenable, in that it divorces meaning from actual linguistic practice. If our use of ‘if P, then Q’ conditionals in cooperative contexts always aims to communicate an inferential relation between P and Q, it would not be surprising if this became a conventionalised part of the meaning of that expression. And since we can expect such a process of conventionalisation, ‘the meaning conventionally assigned to ‘p É q’ is inher- S. Afr. J. Philos. 2008, 27(3) 27 ently unstable, and could not preserve itself unmodified in the natural condition of language-use’ (Strawson 1986: 241). If this argument doesn’t compel, we have more direct reasons to think that conversational implicature fails to explain common reactions to (1) through (4). Compare these reactions with our reactions to the following negated conjunctions, which are equivalent to (1) through (4), given materialism: (9) It is not both the case that Paris is the capital of Spain and that Cambridge has fewer inhabitants than Tokyo. (10) It is not both the case that Paris is the capital of Spain and that Tokyo has fewer inhabitants than Cambridge. (11) It is not both the case that Cambridge has at least as many inhabitants as Tokyo and that Paris is the capital of Spain. (12) It is not both the case that Tokyo has at least as many inhabitants as Cambridge and that Paris is the capital of Spain. When asked specifically for a true-false verdict and given a few seconds to think things over, people commonly say that all four negated conjunctions are true, sometimes protesting that we would not say things in this form. By contrast, the common verdict on sentences (1) through (4) is that they are false or unintelligible on the very ground that there is no connection between antecedent and consequent. Whereas the material implication reading is readily available for (9) through (12), this was not the case for the corresponding conditionals. Strawson’s suggestion, instead, is that ‘if P, then Q’ means the same as ‘P, so Q’ with the exception that only the latter asserts P and Q: (13) [If] Sarah has measles, [then/so] she has a fever. (14) [If] the barometer is falling, [then/so] we will have rain. Both kinds of expression conventionally imply the existence of a ground-consequence relation between the two propositions ‘without conventionally implying the relation to be of one kind rather than another’ (Strawson 1986: 241). Although this makes straightforward sense of reactions to (1) through (4), it is easy to see why Strawson restricted his thesis to the ‘if P, then Q’ conditionals. Take away ‘then’, and we open the door to statements that in no way imply a consequence relation between antecedent and consequent: (15) If you are hungry, there is still food in the kitchen. (16) If you look to your right, the book is on the top shelf. (17) The show was quite a success, if I may say so myself. (18) I’ll still be polite if you insult me, but I won’t be polite if you insult my wife. (Lycan 2001: 63) Neither conditional (15), (16) or (17), nor the first conditional in (18) seems to suggest that the truth of the consequent would follow from the truth of the antecedent. Although prepending ‘then’ to the consequent will rule out some counterexamples to consequentialism, others remain: (19) If the rain pours down, then the game is still on. This particular exception seems well handled by consequentialism if recourse to Gricean mechanisms is allowed for in cases that are rare enough to avoid conventionalisation. Since it is unlikely that there would be a ground-consequent relation between heavy rain and the game’s being on, we look for some neighbouring relevant content. 28 S. Afr. J. Philos. 2008, 27(3) Although the consequent of (19) does not follow from its antecedent, ‘then’ might indicate that the consequent is a conclusion drawn given the premise that it rains, although it is drawn without relying on that premise. However, even if this is a legitimate explanation, there are other problems for consequentialism besides a few apparent exceptions. 4. From Conventionalised to Contextual Consequentialism One apparent problem with consequentialism, even as restricted to ‘if P, then Q’ conditionals, is that it is in conflict with the widely accepted thesis originally suggested by Ernest Adams (1975): Adams’ thesis: A person’s credence for an indicative conditional equals her subjective conditional probability for the consequent given the antecedent. The conflict is at its clearest where antecedent and consequent are probabilistically independent of each other. In such cases, we do not take the consequent to follow from the antecedent: according to consequentialism, then, these conditionals are false. According to Adams’ thesis, however, such conditionals have a high credence whenever the subjective probability for the consequent is high. We have already seen two such cases: sentences (3) and (4). Here, commonsense reactions seem to accord with consequentialism: these conditionals are typically judged to be false rather than highly credible. This does not in itself provide a problem for Adams’ thesis. Both (3) and (4) are ‘if P, then Q’ conditionals, and a defender of Adams’ thesis might explain these reactions with reference to the fact that one conventional linguistic function of ‘then’ is such that it indicates a conclusion drawn when appended to the consequent of a conditional. But dropping ‘then’ appears to make little difference in these cases: (20) If Cambridge has at least as many inhabitants as Tokyo, Paris is not the capital of Spain. (21) If Tokyo has at least as many inhabitants as Cambridge, Paris is not the capital of Spain. According to everyday reactions, these conditionals are false or objectionable. Indeed, everyday reactions appear to take (20) and (21) to be largely synonymous with (3) and (4): all ‘then’ manages to do in (3) and (4) is to stress an inferential relation that is equally part of the message in (20) and (21). Conventionalism therefore appears to survive the encounter with Adams’ thesis. (It is worth noting that materialism is also in conflict with Adams’ thesis, as it takes the improbability of the antecedent to increase the credence of the conditional, ceteris paribus. But here the error lies in materialism rather than Adams’ thesis. For example: it is very unlikely that I will win the lottery, but the conditional ‘If I win the lottery, it will rain tomorrow’ seems no more likely than ‘If I don’t win the lottery, it will rain tomorrow’.) At the same time that (20) and (21) raise a problem for Adams, they also create a deep worry for consequentialism: it becomes unclear what linguistic form conventionally conveys the ground-consequence relation. As we have seen, the form ‘if P, Q’ does not in itself imply that there is a consequential relation between P and Q: now it appears that ‘then’ is not needed. One possibility is that ‘if P, Q’ is ambiguous, and that it is used, in one sense, to convey a relation of consequence. Before I try to settle this issue, however, there is a more pressing problem. As Strawson is careful to point out, the ground-consequence relation that we have in mind when making a conditional claim might be weak. If we have two lists of proposi- S. Afr. J. Philos. 2008, 27(3) 29 tions and believe that everything on one list is true while everything on the other is false, then for any two propositions, P and Q on the same list, we can make the claim that if P then Q (1986: 231-2). Or more generally: whenever ¬P and Q exhaust the epistemic options, we can conclude that if P, then Q, and if ¬Q, then ¬P. And this, of course, is what has motivated materialism. With this in mind, let us return to the non-cooperative conditional ‘if the candy isn’t in the right hand, then it is in the left hand’. For Griceans, this provides a case where the lack of full communicative cooperation allows us to interpret what the speaker wants to convey as the mere material implication: here, there is no pressure to infer the intention to communicate a consequence relation. For consequentialists, the same case is one in which the ground-consequence relationship is the minimal, materialist one that Strawson in effect acknowledges. Unfortunately, Strawson’s consequentialism is robbed of all content beyond materialism when it is allowed that a conditional can be true in virtue of a ground-consequence relation no stronger than the material implication. For the condition that there is a ground-consequence relation between the two propositions is trivially fulfilled for conditionals that satisfy the truth-condition of the material implication. If consequentialism were correct, then, we should find conditionals (1) through (4) as acceptable as the corresponding negated conjunctions (9) through (12). Since we do not, consequentialism should not be formulated as a mere existential claim. Nor can it be formulated in terms of some stronger consequence relation because of minimal cases in which our ground for accepting the conditional is that we accept the material implication. The only viable alternative appears to be that some particular kind of ground-consequence relation becomes part of the content on a case-by-case basis, in the context of utterance. In the case of conditionals (1) through (4), the material implication does not provide the right sort of relation between antecedent and consequent; in the candy case, it does. The conventional meaning of ‘if P, Q’ conditionals would not be that some ground-consequence relation holds between P and Q, but that the contextually relevant kind of ground-consequence relation holds. Such a view would mimic Roderick Chisholm’s (1955) analysis of subjunctive conditionals. On Chisholm’s account, the meaning of a given conditional is determined by whatever condition, C, the speaker has in mind, such that the consequent follows by statements of law from the antecedent and C. Consider a case in which a speaker believes all the following: (22) All gold is malleable. (23) Nothing is both malleable and not malleable. (24) That is not gold. (25) That is not malleable. A speaker who has (22) in mind might say: (26) If that were gold, it would be malleable. To understand (26) uttered by that speaker, we would have to understand it as conveying that (22) holds, and that the consequent follows from the antecedent given that statement. A speaker who has (23) and (25) in mind might say: (27) If that were gold, not all gold would be malleable. To understand (27) uttered by that speaker, we would have to understand it as conveying that (23) and (25) hold, and that the consequent follows from the antecedent given these two statements. Now, it is clear that our everyday and scientific use of subjunc- 30 S. Afr. J. Philos. 2008, 27(3) tives allow for this kind of contextualism or relativism: both (26) and (27) would be meaningful and useful parts of explicit abductive reasoning. On its own, (27) might seem odd, but it is perfectly acceptable in the right context: ‘If it were gold, not all gold would be malleable, but I am quite certain that all gold is malleable. So it cannot be gold.’ 5. Some Problems for Contextualist Consequentialism Although a speaker relativist or contextualist view of counterfactuals just like Chisholm’s has been defended more recently by van Fraassen (1980: Ch. 5), it has remained a fringe suggestion. There could be a number of reasons for this. Since these reasons would offer evidence against the form of speaker relativism that covers both indicatives and counterfactuals, we should therefore consider the main complaints, if nothing else to clarify the issues involved. Some complaints are based on misrepresentations of the speaker-relativist view. For example, Paul Horwich (1987: 162-3) raises the problem that, given van Fraassen’s form of speaker relativism, ‘we would be hard pressed to explain disagreement over counterfactual theses’, and that our practice of settling such disagreements with reference to empirical evidence ‘would be inappropriate if counterfactuals were compressed assertions of logical entailment.’ As far as I can tell, both complaints misunderstand the relativist view. The counterfactual conditional does not merely assert a logical entailment from P to Q given some conjunction of propositions, C, but also implies that C is true. Given this understanding of speaker relativism, disagreement over counterfactuals is easily explained as disagreements over the truth of C, and insofar as C involves empirical propositions, it is no wonder that we try to settle such disagreements with reference to empirical evidence. Other complaints are more serious. Jonathan Bennett (2003: 306-7) argues that Chisholm’s view would give speakers freedom that they patently do not have in determining the meaning of a conditional. For example, we cannot make the following true merely by having (22) and (25) in mind when uttering it: (28) If that were gold, something would be both malleable and not malleable. Bennett is of course right that (28) seems wrong. But this could be because we fail to intuitively grasp that it rests on (22) and (25). We can test this by providing a context that makes its ground clear: ‘This isn’t malleable. But we know that all gold is malleable, so if it were gold, something would be both malleable and not malleable. Since nothing is both malleable and not malleable, this isn’t gold.’ In this context, (28) makes more sense, just as contextualist consequentialism predicted. However, whereas (27) seemed perfectly alright when embedded in a reductio, (28) remains quite awkward – a joke, almost. This is something that contextualist consequentialism needs to explain, but such an explanation is fairly easy to come by. It is hard to take (28) seriously, because the facts on which the conditional rest are metaphysically incompatible with the antecedent, and because such conditionals can never allow for sound arguments from antecedent to consequent. (Something similar is true for perfectly acceptable conditionals such as ‘if he confessed before he died, we will never know’. Although both the conditional and its antecedent might be true, we cannot know that they are. However, we can employ this kind of conditional to rationally infer the consequent, for we can know that either the antecedent or its negation is true, and that the consequent would follow from each.) S. Afr. J. Philos. 2008, 27(3) 31 In general, speaker relativism or contextualism is compatible with the existence of quite strong constraints on the kind of content that can be communicated using conditionals in normal contexts. We might have a very strong bias for having certain kinds of facts or generalisations in mind when making counterfactual judgments, being prone to think first and foremost of generalisations that are central to our web of belief and better entrenched in our inferential practices and discourse. Such constraints will not only affect what speakers will tend to have in mind, but also what they can hope to get across, since hearers are likely to take the conditional to be concerned with whatever ground-consequence relation first springs to mind. Another concern about speaker relativism, again voiced by Bennett, is that we can accept conditionals without having any idea of what the relevant premises are that allow us to infer the consequent from the antecedent. Chisholm’s theory cannot be correct, because it presupposes that speakers have such ideas and that hearers pick up these ideas in successful communication involving counterfactuals: I am quite sure that if I had pressed the button again, the red light would have gone on again; yet I know none of the facts about the wiring that make the conditional true (Bennett 2003: 306). Fortunately for speaker relativism, there is no need to assume that speakers can provide a detailed account of the mechanisms involved. It is enough that speakers can identify a kind of condition under which the antecedent kind of event is always accompanied by the consequent kind of event. Apply this to Bennett’s case: I have pressed the button once, and the red light went on; I assume that the mechanism is the same now, that it is reliable, and that there is no abnormal external interference. From this and the assumption that I press the button, I can infer that the light goes on again, thus grounding Bennett’s conditional. Moreover, this ground is something that speakers can expect hearers to grasp in the normal context of communication. There is a more general methodological reason that suggestions such as Chisholm’s and van Fraassen’s are shunned. To many people, counterfactual conditionals have seemed suitable for analysing laws, causation, and dispositional properties. If Chisholm were right, however, they would seem incapable of providing a stable explanatory basis, and whatever explanations they offered would be more precisely brought out by making explicit the categorical basis of the conditional. As I see it, this is an advantage rather than a problem for Chisholm’s view. First, it is clear that counterfactuals display the kind of context-relativity exemplified by (26) and (27), and this is something that any view of counterfactuals will have to deal with. What David Lewis (1979), Michael McDermott (1999), Bennett (2003) and others have done, is to say that there is one kind of conditional that is suitable for this purpose, and it is perhaps the ‘standard’, or ‘non-degenerate’ kind. It is however still up to context to determine the truth-conditions for a particular conditional, thus diminishing the contrast to Chisholm’s brand of speaker relativism. Second, proposed counterfactual analyses of laws and causation are deeply controversial. Counterfactual analyses of causation have great difficulty making sense of various cases of overdetermination and intransitivity for example (Björnsson 2007a), and laws of nature might be best understood as generalisations figuring in the best systematic description of the world, along the lines suggested by Ramsey and Lewis (1973: 73-5), or as relations between universals, as suggested by David Armstrong (1983). It is not clear that speaker relativism about counterfactuals provides a real cost if it forces us to forego these counterfactual analyses. 32 S. Afr. J. Philos. 2008, 27(3) Third, when we are tempted by counterfactual analyses of some philosophical concept, this might well be because they bring out generalisations or relations between universals that are fundamental to our understanding of that concept. It is not obvious that they offer more than that, even where they seem to be extensionally adequate. There might be other reasons to reject speaker relativism or contextualism than those mentioned here. A proper assessment would have to develop the details of the view and see how it handles the variety of puzzle cases that have provided difficulties for theories of conditionals. Indeterministic cases in particular might prove a difficult problem for consequentialism, and other issues might arise once we look more closely at differences between indicatives and counterfactuals. Unfortunately, tackling these issues would go far beyond what I can hope to accomplish here. But there is one issue that I have already promised to address. 6. Consequentialism about What? Earlier, I said that for many ‘if P, then Q’ conditionals, removing ‘then’ makes little difference to our feeling that the conditional communicates a ground-consequence relation. We also saw that in some conditionals, ‘then’ does not indicate a consequence relation at all, or not one between antecedent and consequent. So while it is true that ‘then’ indicates a conclusion in many cases, ‘if P, Q’ often communicates relations of consequence between P and Q, and does so in a way that provides no solace for materialism: (1) through (4) became no more acceptable when ‘then’ was removed. On the other hand, we saw that there are many cases where ‘if P, Q’ does not communicate a ground-consequence relation. It seems, then, that if consequentialism is true, it is true about a subclass of sentences of the ‘if P, Q’ form. But why is it that only some conditionals are read in this way? One explanation would be that ‘if P, Q’ is ambiguous; it has different meanings, and a hearer needs to determine which the speaker has in mind. A problem with this view is that it would have to allow for quite a few meanings, since the relations indicated by sentences (15) through (18) are quite different. In saying that there is food in the kitchen if you are hungry, I assert that there is food in the kitchen and indicate a possibility that makes this assertion relevant. In saying that the book is on the top shelf if you look to your right, I am again asserting the main clause – the book is on the top shelf – but indicate a possibility that helps you identify the relevant top shelf. In saying that I will be polite if Joan insults me, by contrast, I am neither asserting that I will, in fact, be polite, nor that my being polite would follow from her insulting me. Joan might do other things that prompt impoliteness, such as insulting my wife, and I have not ruled out those by my statement. Apparently, ‘if P, Q’ would have to be multiply ambiguous. But it might be more plausible to say that in all these cases, what is conventionally indicated by the conditional form is merely that the if-clause introduces a proposition the truth of which is not presupposed or asserted by the sentence as a whole, and that the main clause is to be understood in relation to that introduction. Call this a ‘functionalist’ account of conditional statements, as it specifies their general conventional communicative function rather than, say, their conventional truth-conditions. Depending on what P and Q are and depending on the context, we might take the conditional to communicate that there is a certain ground-consequence relation (as we seem to do in the case of (1) through (8)), that a certain possibility is relevant to what is asserted by the main clause (as in (15)), or that the negation of Q does not follow from P in some salient way (as S. Afr. J. Philos. 2008, 27(3) 33 in the first conditional in (18)), to mention some possibilities. On this functionalist account, consequentialism would no longer be a thesis about what statements of the form ‘if P, Q’ conventionally mean, but a description of a kind of message that is frequently communicated with such statements. 7. Occam’s Razor in New Hands: Cutting out the Materialist Middle-Man I have argued that the combination of materialism and the Gricean account of conversational implicature fails to do justice to our reactions to a variety of conditionals. But suppose that I were wrong on that score, and suppose also that materialism could do a good job explaining the meaning of the various non-consequential conditionals exemplified by (15) through (18). In that case, would there be any reason to prefer materialism to functionalist contextualism? The Gricean reason for preferring materialism over consequentialism is simplicity. Starting from a weak and simple account of the meaning of conditionals, and helping oneself only to principles for cooperative communication that were needed to explain a host of other phenomena, one can explain what consequentialists want to explain without building the particular message into the meaning. And if there is no explanatory need to postulate a stronger meaning, we shouldn’t. If this brand of semantic Occamism is correct, and if Gricean explanations of our everyday intuitions concerning conditionals are successful, Strawson’s brand of consequentialism cannot be correct, since it postulates a stronger meaning than necessary. But Strawson believed that this result was implausible: Clearly, there could be a community about which some form of consequentialism were correct, so something about the argument must be amiss. In contrast to Bennett (2003: 26-7), I think that there might be some truth to Strawson’s complaint, but it should be noted that the argument from simplicity would have no bearing against the functionalist and contextualist view sketched here. Just like a materialism expanded to cover both indicative and counterfactual conditionals, functionalist contextualism promises to give a uniform account of all conditionals, although in terms of the communicative function of conditionals rather than their truth-conditions. And just like a wide materialism, functionalism would explain differences in what people intuitively take the various ‘if P, Q’ sentences to communicate in terms of contextual factors that steer hearers to one interpretation rather than another. Moreover, neither account postulates content that is logically weaker than that postulated by the other: since the functionalist account is not an account in terms of truth-conditions, the conventional content that it postulates cannot be logically stronger or weaker than that postulated by materialism. (This does not mean that the two accounts are compatible. They are both meant to be exhaustive claims about the conventional meaning of ‘if P, Q’ sentences.) There is, however, one strong reason to prefer the functionalist view to materialism: as I will now argue, the process of interpretation required given the functionalist contextualism suggested here is less roundabout than that required given Gricean materialism. And it is reasonable to assume that our interpretive mechanisms will converge on the least roundabout interpretive process, other things being equal. Under both views, the interpretation process starts with the assumption that the ‘if P, Q’ sentence conveys something with communicative relevance. Under the functionalist view, the sentence indicates that the if-clause introduces a proposition that is neither asserted nor presupposed, and that the main clause should be understood in rela- 34 S. Afr. J. Philos. 2008, 27(3) tion to that introduction. The interpretation mechanism looks for the most reasonable interpretation of the relation between the two clauses, and outputs the resulting speech act. Under the materialist view, the sentence provides a truth-condition, and the default interpretation is that the speaker asserts this truth-condition. In most cases, the default interpretation violates the maxim of quantity, so the interpretation mechanisms will search for a message that would make the utterance cooperative. Often, that message will be that there is a certain ground for taking the material conditional to be true, a ground independent of the truth or falsehood of antecedent and consequent. At other times, the message is that the consequent is true, and relevant in relation to the possibility introduced by the antecedent, and so forth. Comparing these two stories, it is evident that the functionalist story is more economic. If we start out processing conditionals along the lines of the materialist view, we could introduce two shortcuts without losing communicative precision, ending up with the functionalist view. First, we could go straight for the non-assertoric introduction of the if-clause. On materialism, grasping the default interpretation of conditionals already involves understanding that by uttering the if-clause the speaker has introduced a proposition without asserting it: this is part of asserting the truth-functional content. Secondly, we could go straight to a search for the most relevant message without first ruling out a default, bare material conditional content. In cases where the speaker just wants to communicate the material conditional, this would perhaps be a less efficient procedure. But since such cases are extremely rare, being restricted to the not-fully-cooperative cases and certain technical or formal contexts, and since the material conditional content will be salient in those cases, the extra interpretive costs that a functionalist interpretation would involve in those cases would be negligible compared to what is gained in normal cases. If this is correct, it seems that spontaneous linguistic reform driven by the economics of communication would move us from materialism to functionalism: the meaning postulated by materialism would indeed be ‘unstable … in the natural condition of language-use’, as Strawson claimed. Notice that this argument was made on the assumption that Gricean explanations of apparent counterexamples to materialism are successful. Earlier, I questioned this assumption, but we have now seen that even if we ignore these arguments, there is at least one account of conditionals that is simpler than materialism. Contrary to what the arguments for Gricean materialism are meant to show, it appears that wielding Occam’s Razor for explanatory simplicity cuts out the materialist middle-man, leaving a theory that in many ways resembles that which Strawson proposed. On the functionalist and contextualist view, conditionals very often do convey a ground-consequence relation, and we identify this content without first understanding the conditionals as asserting some weaker, non-consequential content.2 2 The functionalist contextualism proposed here shares some traits with suppositional or illocutionary theories of conditionals according to which declarative conditionals make no categorical statements, but assert their consequents conditional on, or under the supposition of, their antecedents (Barker 1995; Barnett 2006). But functionalist contextualism makes better sense of the intuitive idea that many declarative conditionals literally make claims that are categorically true or false depending on whether a consequence relation obtains between antecedent and consequent, as in the case of (5) through (8), or just depending on whether the consequent is true of false, as in the case of (15) and (17). Since it does not take the interpretation of declarative conditionals to proceed by first understanding them to be conditional S. Afr. J. Philos. 2008, 27(3) 35 References Adams, Ernest W. 1975. The Logic of Conditionals, Dordrecht: Reidel. Armstrong, David M. 1983. What Is a Law of Nature?, Cambridge: Cambridge U. P. Barker, Stephen J. 1995. ‘Towards a Pragmatic Theory of ‘If’’. Philosophical Studies 79, pp. 185-211. Barnett, David 2006. ‘Zif is If’. Mind 115:459, pp. 519-65. Bennett, Jonathan 2003. A Philosophical Guide to Conditionals, Oxford U. P. Björnsson, Gunnar 2007a. ‘How Effects Depend on Their Causes, Why Causal Transitivity Fails, and Why We Care about Causation’. Philosophical Studies 133:3, pp. 349-90. Björnsson, Gunnar 2007b. “Commentary on Lycan's ‘Conditional-Assertion Theories of Conditionals”. Philosophical Communications 43, pp. 1-8. Chisholm, Roderick M. 1955. ‘Law Statements and Counterfactual Inference’. Analysis 15.5, pp. 97-105. Grice, Paul H. 1989. Studies in the Way of Words, Harvard U. P. Horwich, Paul 1987. Asymmetries in Time, Cambridge MA: The MIT Press. Lewis, David 1973. Counterfactuals, Cambridge MA: Harvard U. P. Lewis, David 1979. ‘Counterfactual Dependence and Time’s Arrow’. Nous 13, pp. 455-76. Lycan, William G. 2001. Real Conditionals, Oxford U. P. Lycan, William G. 2006. ‘Conditional-Assertion Theories of Conditionals’. In Content and Modality, vol. Thomson, J.J. (ed) Oxford U. P. McDermott, Michael 1999. ‘Counterfactuals and Access Points’. Mind 108:430, pp. 291-334. Strawson, Peter F. 1986. “If’ and ‘É”. In Philosophical Grounds of Rationality: Intentions, Categories, Ends, vol. Grandy, R. and Warner, R. (eds) Clarendon Press, pp. 229-42. van Fraassen, Bas C. 1980. The Scientific Image, Oxford: Clarendon Press. assertions and then pragmatically inferring that they are categorical assertion, it offers a sim- pler view of the interpretation process. For criticism of conditional assertion theories of conditionals, see (Lycan 2006) and (Björnsson 2007b)