PRIVATE LANGUAGES AND PRIVATE THEORISTS BY DAVID BAIN Simon Blackburn objects that Wittgenstein's private language argument overlooks the possibility that a private linguist can equip himself with a criterion of correctness by confirming generalizations about the patterns in which his private sensations occur. Crispin Wright responds that appropriate generalizations would be too few to be interesting. But I show that Wrighťs calculations are upset by his failure to appreciate both the richness of the data and the range of theories that would be available to the private linguist. Wittgenstein famously poses a problem for the idea of a private language, i.e., a language no two people could have reason to believe they share. A language for describing sensations would be private if sensations were in principle inaccessible to anyone but their subjects. The problem the aspiring speaker of such a language faces, according to Wittgenstein, is that he could never reasonably convict himself of incorrect uses of its terms. He would, Wittgenstein says, `have no criterion of correctness', and hence he would not really be speaking a language at all (PI §258). Simon Blackburn and Crispin Wright agree that this is Wittgenstein's point.1 But Blackburn thinks Wittgenstein overlooks the possibility that a speaker might regulate his use of a private sensation language by exploiting well-confirmed generalizations about the patterns in which his sensations occur.2 Wright offers Wittgenstein an intriguing response: even if an aspiring speaker might do this, not just any generalization will do ­ indeed, it turns out that the ratio of useful to useless generalizations is so small that there is only a negligible probability of one's being able to equip oneself to understand a language in the proposed way. In what follows, I argue that Wrighťs assessment of the aspiring private linguisťs chances is flawed. Though I suspect Wittgenstein can successfully be defended against Blackburn, my business in this paper is simply to show why, in doing so, one must not concede as much to Blackburn as Wright does. 1 See S. Blackburn, `The Individual Strikes Back', Synthese, 58 (1984), pp. 281­301; C.J.G. Wright, `Does Philosophical Investigations I §258 Suggest a Cogent Argument against Private Language?', in J. McDowell and P. Pettit (eds), Subject, Thought, and Context (Oxford UP, 1986), pp. 209­66. 2 See also R. Harrison, On What There Must Be (Oxford: Clarendon Press, 1974), p. 161; R. Walker, Kant (London: Routledge & Kegan Paul, 1978), p. 115; P. Carruthers, Introducing Persons (London: Croome Helm, 1986), ch. 6. The Philosophical Quarterly, Vol. 54, No. 216 July 2004 ISSN 0031­8094 The Editors of The Philosophical Quarterly, 2004. Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK, and 350 Main Street, Malden, MA 02148, USA. I. BLACKBURN'S PROPOSAL Let p1 be a phenomenological category of sensations. Suppose a subject A undergoes sensations at times t1 and t2. A judges at t1 s1. I am undergoing a p1 sensation and is inclined to judge at t2 both not-s1 and H. The sensation I am undergoing is of the same phenomenological type as the sensation I was undergoing at t1. A's inclinations at t2 are insufficient to justify a verdict that his earlier judgement s1 was false. He might just as well deny either (H) or not-s1. So the example does not show that A has a criterion of correctness. But, Blackburn argues (pp. 299­300), A would have more to go on than mere classificatory inclinations if he became a theorist about his sensations, engaged in a `project ... of ordering the expectation of the occurrence of sensation, with an aim at prediction, explanation, systematization'. Instead of (H)'s being a mere impression of the phenomenological identity of two sensations, for example, A might have established a correlation between two or more sensation types. Theories are ultimately answerable to observation, of course, but the correlation might be sufficiently well confirmed to warrant, in a given case, protecting it against a putative counter-example by rejecting a particular sensation judgement instead. A would still have to choose which particular sensation judgement to revoke, of course. (After all, in the example above, even if there were reason to protect (H), there would still be a choice as to which of s1 and not-s1 to revoke.) But if A confirms more correlations and has more classificatory inclinations, the idea is that he could make a principled decision on this further matter too. So equipped, Blackburn thinks, A could exploit such theoretical ideals as simplicity to underpin his verdicts about the correctness and incorrectness of his sensation judgements. Wright illustrates Blackburn's proposal as follows (pp. 239­41; I have changed some minor aspects of Wrighťs presentation). Suppose A undergoes three types of sensation, p1, p2 and p3. Let `S1' abbreviate `I underwent a p1 sensation at some point in the preceding six minutes', `not-S1' abbreviate `I did not undergo a p1 sensation during the preceding six minutes'; and read `S2', `not-S2', `S3', and `not-S3' similarly, mutatis mutandis (I intend the capital `S' to distinguish these past-tense judgements from the present-tense s1 above). Suppose that during an extended period, A confirms that the following pattern is exhibited over any six minutes: `If I did not undergo a p1 sensation in the preceding six minutes, then I underwent a p2 sensation; if I underwent a p3 sensation, then I did not undergo a p2 sensation'. This can be represented using the material conditionals H1. S1 S2 H2. S3 S2. There are eight internally consistent sets of judgement A might make about any six minutes. Wright represents these `diary types', as I shall call them, thus (the righthand side of the table is my elaboration, explained below): 428 DAVID BAIN The Editors of The Philosophical Quarterly, 2004 Diary type S1 S2 S3 Is the diary consistent with {(H1), (H2)}? 1 T T T No 2 T T F Yes 3 T F T Yes 4 T F F Yes 5 F T T No ­ OC for S3 6 F T F Yes 7 F F T No ­ OC for not-S1 8 F F F No Here an `F' under S1 on the fifth row means that one of the three judgements in a type 5 diary is not-S1; `OC' is short for `optimally correctable' (see below). Applying Blackburn's idea to the judgement type S3, suppose A records a type 5 diary, judging not-S1, S2 and S3. The conjunction of S2 and S3 is inconsistent with (H2), so a correction is needed. Since, unlike (H) in the original example, (H2) is a well confirmed correlation, A can reasonably try to preserve it, narrowing the candidates for revision to two: S2 and S3. Of these, S2 is corroborated by A's judgement not-S1: given (H1), revoking S2 (i.e., substituting not-S2) would require revoking notS1 too. So it is simpler for A to revise S3 instead. Hence Blackburn seems vindicated. A appears to have what Wittgenstein denied he could have, a criterion of correctness for S3. Again, given {(H1), (H2)}, A's recording a type 5 diary appears to be a circumstance in which he can reasonably revise a judgement of S3, thereby deciding that the correct account of his inner life over those six minutes was a diary not of type 5, but of type 6. II. WRIGHŤS OBJECTION Relative to a theory, a diary type is what I call `optimally correctable' (OC) for a type of sensation judgement Si if and only if any diary of that type is such that (i) It includes a judgement of Si (ii) It is inconsistent with the theory (iii) Consistency can be restored in a way that involves revising Si within that diary (iv) All other ways of restoring consistency involve more corrections to that diary than ways that involve revising Si within that diary.3 Hence the preceding paragraph shows that {(H1), (H2)} renders diary type 5 OC for S3, as it does type 7 for not-S1. But Wrighťs claim that not just any generalization will serve the private linguisťs purposes is brought out by the fact that {(H1), (H2)} fails, by contrast, to generate OC diary types for S1, S2, not-S2 and not-S3. For these, any diary type satisfying the first three conditions for being OC fails the fourth. In the case of S1 and not-S3, for example, they fail because if any diary inconsistent with {(H1), (H2)} were recorded, there would be a way of restoring consistency that involved fewer corrections to that diary than ways that involve revising the judgement, within that diary, either of S1 or of not-S3. For example, type 1 diaries PRIVATE LANGUAGES AND PRIVATE THEORISTS 429 The Editors of The Philosophical Quarterly, 2004 3 The terminology and formulation are mine, but see Wright, pp. 241, 246­8, 259. are the only type inconsistent with {(H1), (H2)} that involve S1. And, admittedly, if one were recorded, consistency could be restored by revising S1 and S3 together. But it could also be restored by revising either S2 or S3 alone. So S1 lacks an OC diary relative to {(H1), (H2)}. In the case of S2 and not-S2, relevant diaries fail the fourth condition, because if any diary inconsistent with {(H1), (H2)} were recorded, there would be a way of restoring consistency that involved the same number of corrections to that diary as ways that involve revising the judgement, within that diary, either of S2 or of not-S2. For example, the only diaries that are both recalcitrant and contain S2 are type 1 and type 5. If a type 1 diary were recorded, admittedly, consistency could be restored by revising S2, but it could also be restored by revising S3. As for type 5 diaries, I have already shown that if one were recorded, the way of restoring consistency that would involve fewest corrections to that diary would be revising its judgement of S3, not S2. So S2 lacks an OC diary relative to {(H1), (H2)}. Wrighťs objection to Blackburn crucially, if implicitly, involves the following conditional: W. A generalization determines a criterion of correctness for a putative judgement type Si only if it determines an OC diary type for Si. (W) can be seen to be operative, for example, in Wrighťs slide (pp. 241­2) from the preceding account of why S2 lacks an OC diary type to the view that there is no situation in which it would be reasonable for A to revoke a judgement of S2. Assuming that the simplest way to restore consistency is the most reasonable, and given that the only recalcitrant diaries involving S2 are the first and fifth, Wright clearly has the following idea. On the one hand, if a type 1 diary were recorded, then no correction would be reasonable, since although revoking S2 and revoking S3 would both be more simple corrections than alternatives would be, neither would be more simple than the other, and hence there would be no basis for choosing which to make. On the other hand, if a type 5 diary were recorded, the simplest and hence most reasonable way of restoring consistency would involve revoking S3, not S2. Thus, using (W), Wright concludes that relative to {(H1), (H2)}, A lacks a criterion of correctness for S2. And he draws the same conclusion for S1, not-S2 and not-S3. Hence these fail to be types of genuine judgement. In three steps, Wright reaches a more ambitious conclusion. First, he suggests that a judgement can be genuine only if its truth-functional compounds are, and that merely putative judgements could hardly render a diary inconsistent with a theory. Hence he argues (pp. 242­3) that the lack of criteria of correctness for S1, S2, not-S2 and not-S3 has a `rotten apple effecť, undermining the prima facie claim of the remaining types to being genuine. Secondly (p. 247), he thinks this rotten apple effect makes plausible a further conditional: a theory will generate criteria of correctness for judgements about any sensation types it concerns only if it generates criteria of correctness for judgements about all of those types. Given (W), this means that a theory must generate OC diaries for all such judgements. Thirdly, Wright presents extensive formal work (due largely to Warren Goldfarb), aiming to show that the ratio of theories that meet this condition to theories that do not is very small, and is 430 DAVID BAIN The Editors of The Philosophical Quarterly, 2004 the smaller the more types of sensation the theories concern (pp. 258­66). Thus, he concludes, A has a very low chance of confirming a theory equipping him to speak a private language. If, for instance, the theory in question is to range over four sensation types, then on Wrighťs calculations there is a one in 8,192 chance of an aspiring private linguist confirming a correlation that fits the bill! Wright (p. 250) thinks this conclusion will worry friends of privacy, for two reasons: first, because he has shown that Blackburn's theorizing proposal makes the possibility of a subjecťs speaking a private language contingent on the precise patterns in which his sensations occur; and secondly, because Wright thinks he has shown that the probability of a subjecťs sensations exhibiting an appropriate pattern is very small. Even if these two points are right, however, it is unclear why friends of privacy need be anxious. For one thing, it is surely Wittgenstein, rather than friends of privacy, who would reject the dependence of private sensation languages on the patterns the sensations exhibit. As Wright concedes, Wittgenstein seems to think a private language is logically impossible; he would surely not be insouciant about its being merely improbable. For another thing, friends of privacy might be. They might point out a parallel: Wittgenstein's own rule-following considerations show the possibility of a public language to be highly contingent.4 That contingency is tolerable, they might say, if only because the actual world is patently one in which public language is possible; and they might suggest that Wrighťs probabilities in the private case are tolerable too, on parallel grounds. Be all that as it may, the objection I shall develop against Wright is different, namely, that even if the private language issue were one concerning the aspiring linguisťs odds, Wright has underestimated them. III. CONDITIONAL (W) One important reason why Wright underestimates the private linguisťs chances is that (W), the crux of Wrighťs calculations, is false. To take one counter-example, {(H1), (H2)} does determine circumstances in which A would have grounds for correcting a judgement of S2, notwithstanding the fact that {(H1), (H2)} does not provide an OC diary for S2. (Or, to exercise proper caution, {(H1), (H2)} determines criteria of correctness for that judgement type unless such criteria are undermined by the rotten apple effect, to which I return below.) After all, {(H1), (H2)} is a theory confirmed as holding over any six minutes. Wright is thinking of A's diaries as being recorded in serial succession, concerning consecutive periods of six minutes. This undermines the natural reply against him that if A records a recalcitrant diary that cannot be non-arbitrarily revised now, then A might for the time being continue to record his classificatory inclinations until he provides himself with sufficient data to enable a later principled revision of that earlier diary. This reply is undermined, because when we think of diaries recorded in serial succession, it is difficult to see how a collection of diaries which individually provide no reason to change S2 could fare any better collectively. However, if {(H1), (H2)} holds over any six-minute period, there is no reason why we should follow PRIVATE LANGUAGES AND PRIVATE THEORISTS 431 The Editors of The Philosophical Quarterly, 2004 4 See A. Moore, `On the Right Track', Mind, 112 (2003), pp. 307­22. Wright in thinking of it as being applied only to diaries recorded in serial succession. Surely A can start a new diary as soon after its predecessor commences as he likes. To flesh out this possibility, suppose that the past-tense judgements constituting a diary, such as `I underwent a p1 sensation at some point in the preceding six minutes' (that is, judgement S1), are based on present-tense judgements made during the six minutes in question, such as `I am undergoing a p1 sensation' (this judgement I abbreviate with the lower-case `s1'). Suppose, then, that every two minutes, starting at t1, A undergoes a sensation, about which he makes a present-tense judgement; and every two minutes, starting at t0, he begins a new six-minute diary. The crucial upshot is that, after six minutes, every token present-tense judgement A makes will contribute not to one diary, but to three. In the diagram, each horizontal row of three squares represents a diary, named with a letter to its left (strictly, a diary is a set of three past-tense, not present-tense, judgements, but my usage is unproblematic, provided past-tense judgements made at the end of the six-minute period reflect present-tense judgements made during it). The abbreviation above each column represents the presenttense judgement which A makes when undergoing each sensation. Moreover, suppose that when A judges that a sensation of one type occurs, he judges simultaneously that sensations of the other types do not. When he judges s1, for example, he also judges not-s2 and not-s3. Hence by t6, for example, A has recorded a type 3 diary C, since the record of his present-tense judgements between t0 and t6 determines the pre-theoretical past-tense conclusion that, while sensations of types p1 and p3 have occurred over that period, no sensation of type p2 has. A `Ť in a box indicates a post-theoretical confirmation of a pre-theoretical judgement; an `F' indicates a post-theoretical revision, the new judgement being written beneath the `F'; a question mark indicates that a judgement is one of a pair in that diary such that, though one should be revised, there is no basis at the time of completing the diary for a principled decision as to which one it should be. The significance of the diaries' overlapping is this: if A revokes the t13 judgement of s2 in diary I, for example, he thereby revokes it in diaries G and H, since the judgement s2 in all three diaries is one and the same token judgement. Crucially, then, overlapping diaries create the possibility that some alterations to a recalcitrant diary will solve up to three diaries (including itself), and some alterations to one diary will cause up to three diaries (including itself) to become recalcitrant. This provides more leverage for making principled revisions. 432 DAVID BAIN The Editors of The Philosophical Quarterly, 2004 TC T T D E F G H T T T I J K T T T ?T T T T T ?T T T T T T ?F T ?T T T s3 s1 s1 s3 t0 t4 t6 t8 t10 s1 t12 14 t s2 16 t s1 18 t s3 20 t 22 t s1 s3 s3 ?F ?F s2 s2 s2 type 3 type 3 type 3 type 3 type 3 type 3 type 1 type 1 type 1 t2 The example which counters (W) emerges from the diagram's details. After a series of four overlapping type 3 diaries (C­F), A judges s2 at t13. This is the last entry in diary G (completed at t14), which is a recalcitrant type 1 diary, needing revision. At t14 (that is, looking only at diaries C­G), there is no principled way of deciding which of s2 and s3 to revoke within G. The subsequent completion of diary H is of no help either, since these two candidate corrections to G, between which A could not choose at t14, are identical with the two candidate corrections to H (also type 1) between which there is still no choosing. The ratio of solved diaries to revised judgements would be 2:1 for each of s2 and s3. However, the completion at t18 of diary I (type 1 again) is helpful, since it is now the case that changing one token judgement of s2 (at t13) would solve three diaries (G, H, and I) whereas a revision to s3 could achieve this reward only at the greater cost of changing two token judgements (at t11 and t17). Moreover, a provisional correction of s2 would not be upset by the completion of diaries J and K, since these are not recalcitrant, and thus cannot be solved (since they do not need solving) by a change to the t17 judgement of s3. On the face of it, this is a situation in which, guided by the ideal of simplicity, A has precisely what Wright thinks he could never have, a reason to revise his judgement S2 (and the present-tense s2), generated by the correlation {(H1), (H2)}, despite the fact that S2 lacks an OC diary. A similar example can be given for not-S2. Therefore Wrighťs conditional (W), which makes an OC diary a necessary condition for a criterion of correctness, is false. IV. NEGLECTED THEORIES My conclusion might seem premature, given Wrighťs claim that a theory will generate criteria of correctness for judgements about any of the sensation types it concerns only if it generates them for judgements about all those types. Overlapping diaries mean that we could generate prima facie criteria of correctness for more judgement types than Wright could, but since I doubt that we can use overlapping diaries to generate even prima facie criteria for s1 and not-s3, the rotten apple threat remains. Here, then, it is important that Wright not only overlooks the use of overlapping diaries to enrich the putative linguisťs data: he overlooks the range of theories that might be available to the linguist. Why, for example, might the linguist not consider theories concerning the temporal order of A's sensations? Suppose, for instance, that instead of {(H1), (H2)} (a theory comprising material conditionals), A confirmed the following: H3. p1 p2 H4. p2 p3 H5. p3 p1. Read (H3) as `A sensation of type p1 will be succeeded by a sensation of type p2 before a sensation of another type', and (H4) and (H5) similarly, mutatis mutandis. Now imagine that A records the following series of pre-theoretical judgements: PRIVATE LANGUAGES AND PRIVATE THEORISTS 433 The Editors of The Philosophical Quarterly, 2004 t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 t11 t12 s1 s2 s3 s1 s2 s3 s1 s1 s3 s1 s2 s3 As before, suppose that every time A judges that he is undergoing a sensation of one type, he simultaneously judges that he is not undergoing either of the other types. Having recorded this set of pre-theoretical judgements between t1 and t12, surely A could decide that, since {(H3), (H4), (H5)} is well confirmed, he must have been wrong in two of the three judgements he made at t8, namely, both s1 and not-s2. And there are similar examples in which A makes principled revisions to tokens of the remaining four types of judgement. Another case: Wright argues (p. 248) that there are no theories about two sensation types which generate criteria of correctness for all of the judgements a subject might make. But this seems false, once we enlarge the range of theories on offer, as, for example, with H6. p1 p2. Having recorded the series of judgements t1 t2 t3 t4 t5 t6 t7 t8 s1 s2 s1 s2 s1 s1 s1 s2 A could reasonably conclude that both of his judgements at t6 were incorrect, viz s1 and not-s2. And having recorded the series of judgements t1 t2 t3 t4 t5 t6 t7 t8 s1 s2 s1 s2 s2 s2 s1 s2 he could reasonably conclude that both of his judgements at t5 were incorrect, viz s2 and not-s1. Wright has not explained why this theory and its more complex cousins fall short of generating criteria of correctness for all the judgements whose subjectmatter they concern. -------- So Wright underestimates Blackburn's objection to Wittgenstein. If we concede that the aspiring private linguist might at least attempt to establish a criterion of correctness by theorizing about his sensations, we cannot then defuse this concession's implications for the anti-privacy view by invoking Wrighťs meagre assessment of the linguisťs odds of succeeding. For Wrighťs calculations are mistaken: he underestimates both the richness of the data and the range of theories that would be available to the linguist. Hence those who doubt the possibility of a private language, as I do, must not allow the issue to come down to such a calculation of odds.5 University of Nottingham 434 DAVID BAIN The Editors of The Philosophical Quarterly, 2004 5 For discussion and comments, I am extremely grateful to Simon Blackburn, Bill Child, James Ladyman, Doug Long, Jay Rosenberg, Andrew Woodfield and Crispin Wright.