*EXTERNAL AND INTERNAL PRIVATE LANGUAGE ARGUMENTS* Leslie *Stevenson* University of St. Andrews, Schottland ABSTRACT The external version of the PLA (private language argument) argues that there can be no such thing as x using 'K' meaningfully unless it is logically possible for any other person y to acquire reason to believe that x is doing so. The internal version claims that it must be logically possible for x himself to acquire reason for belief in the meaningfulness of his own usage. Both versions of Wittgenstein's argument are here set out explicitly, and defended against various objections that have been made over the last forty years. I Much philosophical ink has been spilled both for and against Wittgenstein's private language argument ("PLA" hereafter). Now that discussion has got to the stage where one philosopher's account of the PLA ends in a blaze of mathematical pyrotechnics,*1* another devotes to it three lengthy chapters of a subtle two volume study,*2* and a third writes a dense 400 page book after a conversion experience in John McDowell's sitting room,*3* it may well be wondered whether there is anything left to say. But even if there is no single substantially new point in what follows, it may be useful to offer an account of the PLA which is surveyable in a finite time by philosophers who have not devoted most of their research effort to the topic. Surprisingly few interpreters set out the argument quite explicitly,*4* as I will try to do here. Following a suggestion by Travis,*5* I will distinguish 3rd person and 1st person versions, and will defend both. But first, why is the PLA so important? It promises to "deconstruct" the Cartesian conception of knowledge and mind which so many people find natural as soon as they start to think philosophically. It can seem uneniable that everyone has an immediate, incorrigible awareness of their own present states of consciousness, in contrast with our fallible judgments about material objects and other people. Even if we do not proceed to raise Cartesain sceptical questions, it can seem to be an ineluctable truth that each of us has, within the public world of objects, speech and action (about which no sane person has any serious doubt), an inner mental domain of necessary privacy - containing sensations, feelings, and thoughts which have aspects ("qualia") which it is not merely difficult, but somehow logically impossible to express in a way that anybody else can understand. If we take that thought seriously, further reflection can make us slide further, into the traditional epistemological problems about other minds (POM) and the external world (PEW). However, if the PLA can show an incoherence in the solipsistic starting-point that the Cartesian approach requires, then it promises to dissolve both PEW and POM. These problems can both be put as questions about whether we can have knowledge (orjustification ) for the relevant claims, or - more radically - about whether we can even understand them. As usually expressed, the PEW presupposes a form of foundationalism called "radical empiricism", according to which one can have basic, non-inferentially justified beliefs only about one's own present mental states (and logical truths). The "justification" question (PJEW) is whether on this basis one can have good reason for any beliefs about material objects and events. An "epistemic priority" thesis is thereby assumed, that one can know the contents of one's own experiences even if one lacks all knowledge of any external world. One could have no knowledge of other minds or of their language-use, so one would be classifying one's own mental states and thereby following rules, but without any possibility of any check for agreement with other people's judgements. The "meaning" version (PMEW) arises from the radical phenomenalist claim (first mooted by Berkeley and Hume) that since all that we are presented with are our own "ideas" or "perceptions", our beliefs can only be about them, and so we cannot even understand claims about anything existing unperceived - unless they can be analyzed into complex hypothetical propositions about perceptions. This assumes that one can follow rules in classifying one's own mental states, without even any concept of other rule-followers, let alone any knowledge of them. The PLA threatens to exclude both PJEW and PMEW as based on incoherent assumptions. The "justification" form of the other minds problem (PJOM) arises for radical empiricism, at the second level: suppose that beliefs about material objects (including human bodies) could be justified in terms of the supposedly basic perceptual beliefs, can beliefs about the mental states of others be justified by a further inferential step? PJOM also arises within a less radical form of foundationalism (more favoured in recent years), according to which we can have basic, non-inferential beliefs about material bodies: for the problem remains how to justify beliefs about other minds on this basis. For either form, an "argument from error" alleges that because we sometimes make mistaken judgments about other's mental states, we can never be adequately justified in any such belief. (So far, this presupposes that there are other minds, and that we can understand propositions about them.) The "argument from analogy" attempts to solve PJOM by claiming that anyone can note correlations between her own mental states and her bodily states and behaviour, and that ordinary canons of scientific reasoning can then justify inferences that when other human bodies behave similarly, then (probably) they have similar mental states. But the PLA implies that nobody could be in the presupposed initial position - (a) having justified beliefs about other bodies, but as yet no justified beliefs about other minds, while (b) understanding what it is for another person to have mental states, simply by entertaining hypotheses that they have states similar to those one experiences in oneself. If the PLA is sound, no concept can be formed purely on the basis of introspective "ostensive" definitions. The "meaning" version of the other minds problem (PMOM) assumes that one might understand propositions about one's own experience without yet understanding any claims about other minds. That is, one could be a solipsist in the radical sense that one would have no concept of "oneself" as a particular subject of experience distinct from other conscious beings. The PLA rules out such solipsism: it claims that a single individual cannot understand a symbol, follow a rule, or have thoughts with propositional content, unless it is possible for him to interact with others who can understand that term, follow that rule, or entertain thoughts with that content. Can there really be a single valid argument with such sweeping philosophical powers, able to knock out all versions of long- standing philsophical conceptions and problems at a single blow? Some caution about such overweening ambition may well be wise counsel. For one thing, there may be forms of the argument from error - specifically from dreaming - which do not presuppose exactly the kind of foundationalism or epistemic priority thesis which lies within the PLA's target area. But even if the Wittgensteinian PLA fails single-handedly to eliminate all versions of PEW and POM, or all defences of private "qualia", it clearly has fundamental implications for central issues in epistemology, theory of meaning, and philosophy of mind, and is as important an argument as any in philosophy. II No doubt because of its philosophical importance, opinion about the soundness of the PLA remains sharply divided, as it has done since Wittgenstein's own day. There have been two main lines of objection to it. One is that it involves an unargued and implausible assumption of verificationism: versions of this have been voiced by Strawson, Thomson, and Craig.*6* The other objection admits need for some measure of verifiability for attributions of meaning, but claims that the requirements can be met in a private language, in much the same way as they are in public language. Variations on this theme have been developed by Ayer, Harrison, Blackburn, and Carruthers.*7* Part of the reason for the continuing controversy is surely the brevity and aphoristic style of Wittgenstein's presentation of the argument in Philosophical Investigations 243-271. To get the full idea, one needs to study the whole book, and other texts of his later philosophy, especially his lecture notes on the topic.*8* Obviously the PLA needs to be taken in the context of the preceding treatment of rule-following in Investigations 180-242, even if it is not merely a corollary of that discussion (as Kripke has claimed*9*). At 243, Wittgenstein identifies the target of his PLA as follows: The individual words of this language are to refer to what can only be known to the person speaking; to his immediate private sensations. So another person cannot understand this language. This conception had been vividly expressed by Russell, when he argued (with playful enjoyment of the paradox involved) that two people can never mean the same thing by a word, since "the meaning you attach to your words must depend on the nature of the objects you are acquainted with", and "different people are acquainted with different objects".*10* Russell assumed a radical empiricist epistemology according to which the only things anyone can be directly "acquainted" with are his own mental states. We can surely widen the referents of the words in a putative private language from "immediate private sensations " to emotions and thoughts and all states of consciousness with which, supposedly, one can be "directly acquainted". What then was Wittgenstein's argument? Here is his section 258, which contains what is usually seen as the nerve of the argument, but here put into explicit dialogue form between the author and his interlocutor, labelled "PL" - the would-be user of a private language. [Other relevant changes are marked by square brackets.] PL: Let us imagine the following case. I want to keep a diary about the recurrence of a certain sensation. To this end I associate it with the sign "S" and write this sign in a calendar for every day on which I have the sensation. W: I will remark first of all that a definition of the sign cannot be formulated. PL: But still I can give myself a kind of ostensive definition. W: How? Can [you] point to the sensation? PL: Not in the ordinary sense. But I speak, or write the sign down, and at the same time I concentrate my attention on the sensation - and so, as it were, point to it inwardly. W: But what is this ceremony for? for that is all it seems to be! A definition surely serves to establish the meaning of a sign. PL: Well, that is done precisely by the concentration of my attention; for in this way I impress on myself the connexion between the sign and the sensation. W: But "I impress it on myself" can only mean: this process brings it about that [you] remember the connexion right in the future. But in the present case [you] have no criterion of correctness. One would like to say: whatever is going to seem right to [you] is right. And that only means that here we can't talk about 'right'. In Wittgenstein's text the first-person pronoun is used throughout, and the dialogue has to be read as internal to a single thinker, between the voice which expresses the idea of a private language, and the voice which argues that there is no such coherent conception. This is appropriate, for to serve its dialectical purpose the argument mujst be one that anyone can administer to himself, to show that despite first appearances the tempting conception involves an incoherence. But it is also possible to interpret the PLA as showing that one person (W) could never know that another (PL) is using words with a meaning knowable only to PL. A distinction can thus be made between "internal" and "external", or 1st person and 3rd person, versions of the PLA. III Can we set out the premisses of the PLA explicitly? Some time ago, J.J.Thomson offered a formulation, and criticized it for assuming verificationism. This is still a useful starting-point for reformulating the PLA, and reassessing the charge. She set out three steps:*11* (1) If a sign 'K' which a man uses is to be a name of a kind of thing in a language, his use of it must be governed by a rule of the form: X's and only X's are to be called 'K's'. (2) If his usage is governed by such a rule, it must be possible that he should call a thing a 'K' thinking it is an X when it is not an X. (3) There is no such thing as a man's thinking a thing is of the kind to be called 'K' and it not being so, unless it is logically possible that it be found out that it is not so. This obviously implies that for any general term, it must be logically possible to find out whether or not any given thing is of the relevant kind. Such a sweeping assumption of verificationism is highly disputable, but much will depend on exactly how "finding out" is interpreted - is it to yield conclusive reason, or just good but defeasible reason for belief? - is it to be possible in practice, or just in principle? - and possible for us now, or just for someone at some time? On the weakest interpretation (that a sign cannot be a kind-name unless it is logically possible for someone at some time to have some reason to believe that a thing is of that kind), the principle is perhaps not so implausible. Before going any further, we should change something else in Thomson's formulation: for it is false that for every term 'K' which one understands, one can give an explicit rule for its application, using another term 'X' (this would lead to a vicious infinite regress). There are very many words we use and understand, e.g. 'green', 'curved', 'hot', 'sour', 'painful', 'ticklish', etc., for which we cannot give any such rule. There can still be an objective difference between correct and incorrect applications of a word for which we have only a recognitional capacity, not a verbal definition. Let us therefore reformulate as follows: (1') If a sign 'K' is to be a general term in someone's language, his use of it must be rule-governed (in the sense that there is a difference between correct and incorrect applications of it). (2') If so, it must be possible that someone can call something a K when it is not correct to do so. (3') There is no such thing as calling something a K and it not being one, unless it is logically possible to find that out. Another problem now arises. (2') may be accepted as applying to all words for kinds of thing (or property, or states of affairs) in the physical world (noperceptual judgment is infallible); but does it apply to words used to classify one's own present mental states? Is it possible to think that one feels cold, thirsty, or resentful (or that one is calculating, daydreaming, or afterimaging) when one is not ? Of course, one can say one is in one of these states when one is not, either by a slip of the tongue, or by being insincere - but surely one cannot have false beliefs on these matters. Since the PLA claims to show that one cannot with one's own unaided resources give meaning to words classifying one's own mental states, we must formulate the PLA in such a way as to apply to such words, and to their use in "avowals". We need therefore to disambiguate the notion of 'calling' in (2'): is it to imply belief, or just the use - the mere utterance or inscription - of words? Let us reformulate in terms of the weaker notion: (1'') same as (1') above (2'') If there is such a difference, it must be possible for someone to use the term 'K' in a situation where it is not correct to do so. (3'') There is no such thing as using a term incorrectly unless it is logically possible to find out that that use is incorrect. There are various possible explanations of an incorrect use of a term. There can be a slip of the tongue, or the telling of a lie (the intentional expression of a belief the speaker does not have), and with words for public states of affairs there can also be sincere error (false belief). These possibilities all presuppose that the speaker knows the meaning of the term; but there can also be cases where he does not, either because he attaches awrong meaning to it, or because (like a parrot or a babbling infant) he attaches no meaning to it. We distinguish between using a term with some meaning (however idiosyncratic), and uttering it without meaning anything thereby. False belief would seem to be impossible for the use of mental words in avowals. But it is possible for a parrot, or a young child, or someone in an advanced stage of Alzheimer's disease, to use such words (even perhaps with a subjective feeling of making assertions thereby), butwithout understanding them. The difference we want to highlight now is that between uses governed by some rule and those in which there is no rule-following at all. Let us reformulate to make the nerve of the argument depend on this: (1''') If 'K' is a general term in x's usage, it has a meaning for him, i.e. it is used by him in a rule-governed way. (2''') For any use by x of a sign 'K', it must be possible to think x is using 'K' meaningfully when he is not doing so. (3''') There can be no such thing as x using 'K' meaningfully unless it is logically possible to find this out. The verificationism implied by (3''') is more restricted than that noted above: it is not about any claims at all, that something is or is not a K, but about whether someone is using a term 'K' meaningfully . So in this last formulation the PLA does not appeal to a general Verification Principle like the logical positivism of old, but only to a much more restricted (and arguably, more plausible) principle to the effect that meaning may be ascribed to an expression (or understanding of it to a person) only if that meaning or understanding is somehow manifestable in usage of the expression. Even if theological, cosmological or psychological speculations are allowed to outrun verification, propositions about semantics must, on this view, be justifiable in terms of perceptible facts about usage of the relevant expressions. But as noted above, there are still some important ambiguities left. Firstly, must the "finding out" be conclusive, or may it be defeasible? In view of the absence of logically conclusive proof anywhere outside logic and mathematics (and indeed of the fallibility of our recognition of proofs even there), we had better require only the possibility of getting some good (but defeasible) reason for belief about meaningfulness. If x uses a certain sign meaningfully, this is a contingent fact: it must be made true by something about x, and there must surely be some way in which the relevant facts can be known, or reasonably believed. Secondly, must the acquiring of such reasons be possible for anyone, or would it suffice that only one person - the putative private linguist himself - could do so? According to which option we take, we will have a premise for an external or internal version of the PLA: (Ext) There can be no such thing as x using 'K' meaningfully unless it is logically possible for any other person y to acquire reason to believe that x is doing so. (Int) There can be no such thing as x using 'K' meaningfully unless it is logically possible for x himself to acquire reason to believe that he is doing so. After all this work, we still do not have a complete argument explicitly ruling out private language, but it is obvious what is needed, namely to add appropriate premises saying that the requirements just expressed cannot be met in private "language", and to draw the conclusion. IV For the external or 3rd person version of the PLA, the formulation we at last arrive at is: (IE) If x is using a sign meaningfully, it must be logically possible for any other person y to have good reason to believe that x is doing so. (IIE) If someone x were using a sign privately, it would be logically impossible for any other person y to have any good reason for belief about the meaningfulness of x's usage of that sign. Ergo, the signs in a putative private language could not be used meaningfully, and a private "language" could not be a language. This external PLA wears its (limited) verificationism on its sleeve. (IE) says that semantic facts about someone's usage cannot float free of all conceivable knowledge or justified belief, and it claims, more particularly, that the fact of the meaningfulness of x's usage must be publicly accessible. This is not the stronger claim that the specific meanings of x's term must be publicly accessible. (IE) implies that if x is using a private language, the fact that he is doing so should be knowable by others. For if the putative language is not publically understandable, that fact is itself presumably a public fact, but if the "language" were nevertheless meaningful, (IE) says this should be generally knowable. As Travis puts it,*12* the technical term 'private language' had better be a term of public English, if we philosophers who discuss the matter are to know what we are talking about! So if somone is using a private language, there should be some publically accessible facts about their usage which make it true to say that that usage reports some further, unspecified, private facts. The second premise (IIE) is not merely definitional of the term 'private' as so far understood, so it needs a little further defence. Wittgenstein called a language "private" if nobody but its user could understand it, because it contained words which refer to what could be known only to him. However there is an important ambiguity in the notion of 'understanding' here. In one sense, A understands a term used by B if A attaches to it the same meaning as B does. But this is not enough for mutual understanding, since A and B might happen to use a word in the same way, but each without any knowledge of the other's usage (for example, A and B might quite independently have selected similar coding systems). In such a case, A can hardly be said to understand B, or to understand the term as used by B , unless he knows that B is using it with the same meaning as himself. Without such knowledge one could not gain new information from someone else's utterances, nor could we use language to co-ordinate our actions. On the Cartesian conception, one person's states of mind are unknowable to anyone else, but this does not rule out people somehow coming to use synonymous terms to describe their own similar mental states - it is just that they could never know they were doing so. Accordingly, we need to redefine "private language", as use of signswith meanings which it is logically impossible for anyone other than the user to have reason to believe that the user attaches to those signs .*13* Does (IIE) follow from this concept of privacy? If x were using terms with meanings which y could never know that x attached to them, how could y know that x was using those terms with any meanings at all? What possible public evidence about x could justify a belief on our part that he attaches private meanings to some of his terms, but without giving us reason for any more specific beliefs about what those meanings were? At best, the evidence would have to be indirect. We might think that if x displays plenty of competence in classifying external, perceptible states of affairs (we find ourselves in agreement with him about the cases on which we check up), then when he says that he has certain recurring feelings which he recognizes but cannot describe in any words we understand, we should give him the benefit of the doubt, and allow that he is indeed following private rules of classification. After all, we are often prepared - and surely quite reasonably - to credit people with recognitional capacities and with understanding words for public states of affairs in cases where for some contingent reason we are unable to observe the facts for ourselves. (Were we not justified in believing the perceptual reports of the first men on the moon?) So could it not be the best explanation of someone's linguistic behaviour that they are using terms with private meanings? But if such an explanation is to be reasonable it must at least be a possible explanation, internally coherent. The PLA denies this, so the defender of private language cannot simply apply the notion of best explanation here without begging the question. On the other hand, of course, it is not enough just to assert that such an explanation is incoherent - this must argued for. Travis has argued the case at some length.*14* We should not take x's word for it, just because x claims to be attaching private meanings to some of his words, for there is no reason to treat anyone as particularly authoritative on such subtle philosophical questions about the semantics of their language. For private meanings to be the best explanation of someone's usage, there would have to be public evidence about regularities in that usage, but in so far as there is such evidence, this would surely begin to justify attributing public meanings to the relevant terms; whereas if there is no such evidence, we have no reason to attribute any meanings, not even private ones. There seems in the end to be no conceivable state of public information which could give reason to believe in the existence of meanings without ipso facto giving reason to attribute specific public meanings. If so, (IIE) stands. This external PLA may be thought to beg the whole question rather swiftly, but it is not without some persuasive power. For anyone who denies (IE) and (IIE) must realize that they are committed to holding that whether or not someone is using a private language is a fact beyond all possibility of even the weakest sort of confirmation or disconfirmation by anyone other than the putative user. Not just the meanings, but even the meaningfulness and hence the very existence of a putative private language, would be publicly unknowable. However, if anyone is hardy enough to defend that conception, he will still have to face the internal version of the PLA, which argues that the private linguist himself could have no good reason to believe in the meaningfulness of his usage. V The internal PLA argues for the same conclusion as the external version, but from apparently weaker premisses: (II) If x is using a sign meaningfully, it must be logically possible for x to have good reason to believe that he is doing so. (III) If x were using a sign privately, it would be logically impossible for him to have reason for belief about the meaningfulness of his own usage of that sign. Ergo, the signs in a putative private language could not be used meaningfully, and a private "language" could not be a language. This internal PLA is plainly valid: the conclusion follows from the premises.*15* But neither of these is obviously true, so both need further defence. (II) involves a first-personal kind of verificationism about attributions of meaningfulness. (III) may still be found controversial, for it makes the crucial claim about what is lacking in private language but which must be present in ordinary public language if we are to avoid the impossible conclusion that meaning is not possible anywhere. I will now try to justify both premisses. Recall first that we have redefined "private language" as use of signswith meanings which it is logically impossible for anyone other than the user to have reason to believe that he attaches to those signs. In such a language interpersonal communication obviously cannot be achieved, but if it is to count as language at all, it must be usable in some way by a single subject for his own purposes - presumably to record information, to construct theories, or to make plans to guide his own actions. But this will surely be possible only if he can know (or at least, justifiably believe) that his use of signs is regular and meaningful, rather than random and meaningless. If he is to rely on his signs as expressing beliefs or plans which he formed in the past, the meaning he attaches to them at the later time had better be the same as before. If he entertains any suspicion that this might not be the case (e.g. if he is worried that he might be affected by a drug, mental illness, senility, a Cartesian evil demon - or any other fundamental mental incapacity) then he is in danger of losing all reason to believe that he is using signs with a definite, stable meaning. This is our support for (II). We must now discuss at more length why (III) has been disputed and how it might be defended. If we expand it to incorporate our amended definition of privacy, it amounts to the following: If x were using a sign with a meaning which nobody else could have reason to believe he attaches to it, it would be logically impossible for x to have any good reason for belief in the meaningfulness of his own usage. But what argument can be given for this? Is there really no way the putative private linguist could ever justify belief in the meaningfulness of his own usage? For the PLA to be sound, the answer must be no, whereas the answer to the parallel question about public language must be yes. Yet what, if anything, gives us reason to believe we are using terms with stable meanings in our ordinary use of language? Obviously, this not something we normally worry about: we take it for granted that we have learnt the meanings of words as standardly used, and that we continue to use them with those meanings. If doubt is cast on one's use of any particular term, one can appeal to a dictionary, or more directly to one's continuing agreement with other speakers in a wide variety of applications. This resource is necessarily lacking in a private language, so it is tempting to say that we have decisive reason for (III). But there remains a danger that the support offered for this answer about private language may extend to a similar scepticism about the meaningfulness of all public language. The arguments that Kripke has expressed so vividly (and almost made to seem Wittgenstein's own) - that nothing present to consciousness, nor anything remembered about past usage, nor any combination of past or present mental or physical facts, can ever justify the attribution of rule-governed usage - seem to imply the self- defeating conclusion of ruling out all meaning in any language.*16* Kripke concluded that the only way to avoid this is to accept a "sceptical solution", according to which talk of meaning is not strictly factual (it lacks truth-conditions), but is nevertheless a useful way of dignifying each other as competent rule-followers, given a background of widespread agreement in a community of language-users. So the dispute about the soundness of the internal PLA cannot be resolved quite so quickly. It has not been proved that there is no other way (other than appeal to community agreement, that is) that a private linguist could rationally support confidence in the meaningfulness of his own usage. Imagine someone completely isolated from all human society (e.g. on a desert island), and let him be in this lone condition from birth (unlike the Robinson Crusoe of Defoe's story). If he somehow develops a practice (what Wittgenstein also calls a "technique" or "custom") of using signs for his own guidance, could he not find reason to believe in the meaningfulness or regularity of his usage in the practical success which his usage achieved, giving him predictions which were useful because they were largely true ? But this would not be "private" language in the sense defined above, for the unshared nature of the meanings is only contingent: the story as imagined leaves it logically possible for others to arrive on the island, learn the usage of our linguistic Crusoe, and come to agree with him about new cases. Such fables have however been used to suggest by analogy that there would be nothing disabling in principle about the situation of the true private linguist. Many philosophers have vigorously argued*17* that there is an indirect way in which he could justify belief that his use of any one sign is meaningful, in terms of the contribution of that sign-use to his wider system of classificatory capacities. If he finds that his judgements fit together into a coherent network of relatively stable beliefs, then surely this can provides the justification. Let us scrutinize this line of thought. The evidence is supposed to come from the "stability" or "coherence" of judgments - but what does this mean? What people seem to have had in mind is that the subject could first notice correlations between recurring features of his private experiences, and assemble inductive support for generalizations which may thus acquire law- like status for him. It is suggested that he could then justify correcting any one singular judgement. If he has found that a mental state of type F has always (or nearly always) been quickly followed by one of type G, then he could rationally use this fact to reject a judgement that what he has just felt was an F, if he now judges that he does not experience a G. (He would need a reason for retaining the latter judgement and preferring to withdraw the former - but this might perhaps be allowed to consist in a more confident "feel" of the latter.) The private linguist is here cast in the role of a scientific theorist about patterns in his own subjective experience - at least in the minimal sense of employing inductive inference, and evaluating particular judgements in the light of the generalizations they exemplify. This obviously involves a fallibilist epistemology: each particular judgement is to be defeasible in the light of holistic considerations about its relation to others. Descartes' view that claims about the contents of one's own present states of consciousness are incorrigible has been abandoned. But the conception remains Cartesian in the more general sense that it is still radically solipsist: the subject is to find reason for correction of any one judgment purely in terms of what is immediately available to him, without invoking anything to do with objects or minds distinct from himself. But is there not a kind of circularity involved in this proposed way of justifying the belief that one is using terms meaningfully? Certainly, it is standard fallibilist epistemology to say that one can have reason to withdraw any particular judgement and thus revise one's opinion about its truth-value, but it is quite a different matter to suggest that one could find ground for doubt about whether one meant anything by an expression. If inductive assembly of evidence could do the latter, the evidence would surely be thereby subverting its own meaningulness, not just its truth. One would lose all reason to rely on the generalizations involving the relevant expressions (F and G), because one would be committed to doubting whether those expressions, and sentences containing them, still mean anything for one (or ever did so). If one is in doubt about whether one's use of a sign is rule-governed in the first place, how could one be in a position to assemble empirical evidence for generalizations which employ it, and to make inferences invoking them? One can only do this if one is using the contained expressions meaningfully - but that is just what the private linguist is seeking ground for believing. Could this alleged circularity be only apparent, arising from a too narrowly temporal understanding of the assembly of grounds for judgment? If we think of the proposed justification as conferring the dignity of meaning and rule-following on a temporally extended pattern of behavioural or mental events (none of which can deserve such titles when considered in isolation), and as conferring it because of complex patterns of interconnection which can only retrospectively be discerned, then it may be suggested that what we have is a certain measure of holism, but not circularity - not vicious circularity anyway. Could someone acquire reason to believe that they possess genuine classificatory capacities, not by getting independent evidence for each one separately, but from the overall fact that he uses a system of (apparently meaningful) signs which leads to a "coherent network of relatively stable beliefs"? But while such holistic justification is surely possible for specific beliefs, expressible in sentences, it is not at all so clear that it makes sense for concepts or classificatory capacities, which are general skills in the use of words or phrases. Any check on acquisition of the latter would seem to require feedback on success or failure at each stage, so the various skills involved must necessarily be acquired stepwise (as Pears has argued).*18* The vicious circularity returns into view when one considers how such holistic considerations could justify either doubt or reassurance about whether one meant anything by the expressions. If they could support doubt, they would thereby be subverting the very meaningfulness of the considerations adduced for doubt. If this is correct, the suggested holistic way of getting evidence for or against the meaningfulness of his use of an expression is not in the end available to the private linguist. Does there remain any other way for him to do so? Suppose he finds himself sometimes inclined, for no reason that he can express, to doubt some of his former judgements (perhaps he rejects some of them). His temporal pattern of judgements might become hesitant and unstable. Could this constitute reason for him to think that he does not mean anything by one or more of his signs? And could the mere absence of doubt or instability give him reason to believe that he is following rules? But if such hesitations have no rational support, there is no way to distinguish his condition from neurosis - which is how we would describe someone in real life who was subject to recurring doubts about his judgments, even when others agreed with him. There seems to remain no way in which the private linguist can avoid accepting (III), and hence the force of our internal PLA. VI The fundamental issue at stake in the PLA (in external or internal version) is what makes it the case that someone is following rules. The usual linguistic idiom of words and meaning (which we have used above) can be transposed throughout into talk of mental states with conceptual content - concepts, beliefs and judgments. (Thus the PLA could equally well be called the private meaning argument, or the private concept argument.) The question then becomes what makes it the case that any mental episodes are judgments that P rather than that Q, or indeed judgments at all, rather than mental states without conceptual content? (What makes them representations , in the language of Kant and of much recent cognitive science?*19*) If someone is following rules he must be following some particular rules rather than others: what then makes it the case that he is following those rules? For any particular speaker and noise or mark this is a contingent truth. Parrots, machines, and infants may utter a noise, which may happen to be recognizable phonetically as a word of some language, but they do not mean anything by it - not unless some further conditions are satisfied, anyway. In our actual use of language, we can have criteria for saying of another language-user that he means something by his use of an expression, as Wittgenstein notes at Investigations 269: Let us remember that there are certain criteria in a man's behaviour for the fact that he does not understand a word: that it means nothing to him, that he can do nothing with it. And criteria for his 'thinking he understands', attaching some meaning to the word, but not the right one. And, lastly, criteria for his understanding the word right. . . . Such criteria must surely involve the observable use made of language in various situations. But in a private language, we have argued, not even the user can have reason to believe that he is following a rule. In ordinary public language we use terms and find other people agreeing with our applications of them (at least in situations favourable for the relevant perceptions). We take this normal background (what Wittgenstein calls "peaceful agreement") so much for granted that its presence hardly registers with us as reason for thinking that we mean something by an expression. But for any given word, the continuation of such agreement in its application is a contingent matter, about which we have no absolute guarantee (as Wittgenstein's treatment of rule- following even in simple arithmetic makes dramatically clear). The onset of endemic disagreement is always logically possible. If such eventualities would give one reason to doubt whether one still meant something by an expression, presumably their absence ("peaceful agreement" about its use) can give one reason to think one does mean something by it. *20* We can imagine a Crusoe or wolf-child, socially isolated from birth but somehow developing a language to help himself to cope with the world. As noted above, such usage would be only contingently unshared, and thus not private in the Wittgensteinian sense. We can even envisage such a lone linguist assuring himself about the meaningfulness of his own usage by appealing to the success he had had in using it to predict events in his environment. To that extent, we can conceive of detaching meaningfulness from communal agreement and letting it rest on the stability of the single user's judgments about the world - though whether this imagined story is really logically possible is not obvious (the question is one which Wittgenstein himself seems to have steered away from).*21* Endemic, inexplicable disagreement with other speakers - or in their absence, recurring failure to apply any stable vocabulary and make any predictive sense of the material world - would seem in the end to be the only possible good reasons for doubting whether one is following a rule. Since in private usage these would not be even logical possibilities, we may infer that there could not be any positive reason to think that one is following a rule in any given case. "Epistemic" and "ontological" or "constitutive" formulations of the PLA's conclusion can be verbally distinguished.*22* We may assert the impossibility of justifying belief about private rules, or we may make the apparently stronger claim of their conceptual impossibility - that there could be no facts which would constitute someone's following a private rule. But if we apply the modicum of verificationism which we have concluded is necessary about attributions of rule-following, there will in the end be no substantial difference between the two versions. We conclude that "private rules" cannot be rules at all, so there can be no private concepts, meanings, or language (in the amended Wittgensteinian meaning of 'private'). NOTES *1* C.Wright, Does Philosophical Investigations I.258-60 suggest a Cogent Argument against Private Language?" in Subject, Thought, and Context, edited by P.Pettit and J.McDowell (Oxford: University Press 1986). *2* D.Pears, The False Prison: A Study of the Development of Wittgenstein's Philosophy (Oxford: University Press 1988), see Volume II, Ch.13-15. *3* C.Travis, The Uses of Sense: Wittgenstein's Philosophy of Language (Oxford: University Press 1989). *4* Pears acknowledges the need to set out the premises of a reductio argument explicitly, so that the finger can be pointed at the one to be blamed, but he does not seem to follow his own advice. *5* See Travis, op. cit., Ch.8, section 4. *6* P.F.Strawson, "Review of Wittgenstein's Philosophical Investigations ", Mind 63 (1954), and "Scruton and Wright on Anti-realism etc.", Proceedings of the Aristotelian Society 77 (1976-77); J.J.Thomson, "The Verification Principle and the Private Language Argument", American Philosophical Quarterly 1 (1964); E.J.Craig, "Privacy and Rule-following", in Language, Mind and Logic, edited by J.Butterfield (Cambridge, England: University Press 1986). *7* A.J.Ayer, "Can There Be a Private Language?" in Proceedings of the Aristotelian Society , Supplementary Volume 78 (1954); R.Harrison, On What There Must Be (Oxford: University Press 1989), Ch.3 & 6; S.Blackburn, "The Individual Strikes Back", Synthese 58 (1984); P. Carruthers, Understanding Persons (London: Croom Helm 1986), Ch.6. *8* L.Wittgenstein, "Notes for Lectures on "Private Experience", Philosophical Review 77 (1968); "The Language of Sense Data and Private Experience", Philosophical Investigations 7 (1984), pp. 1-45, 101-140. *9* S.Kripke, Wittgenstein on Rules and Private Language (Oxford: Blackwell 1982). *10* B.Russell, The Philosophy of Logical Atomism, Lecture II, p.195 in Logic and Knowledge, ed. R.C.Marsh (London: Allen and Unwin 1956). *11* Judith J. Thomson, op.cit., pp. 22, 23, 26-7. *12* Travis, Ch.8.4 *13* Cf. Wright, op. cit., section1. *14* Cf. Travis, Ch.8.3. Travis makes important play with "S-use sensitivity" - a subtle kind of context-dependence of meaning, which I have no space to do justice to here. *15* Our internal PLA is very close to that which Carruthers set out as follows (op. cit., p.171): (1) Meaningful use of a sign implies the capacity to use it in a regular (rule-governed) way. (2) A putative speaker of a private language cannot know whether or not they are capable of using their signs in a regular way. (C1) So a putative speaker of a private language cannot know whether or not their use use of their signs is meaningful. (3) Meaning requires knowledge: meaningful use of a sign requires (the capacity for) knowledge that one's use of that sign is meaningful. (C2) So a putative speaker of a private language cannot use the signs of their "language" meaningfully (i.e. private language is impossible). Carruthers offers what he takes to be sufficient support for (1) and (3), but finds that the considerations offered in support of (2) - that nothing present to consciousness or remembered about past usage could show that one's use of a sign is meaningful - far too strong, because they would seem to apply equally to public language. He suggests that (2) is false, since a private linguist can justify the meaningfulness of any one of his signs in terms of the contribution that it makes to a wider system of classificatory capacities. We will scrutinize this idea below. *16* S.Kripke, op. cit. *17* See note 7. *18* Pears, op.cit., p.395. *19* But note that Kant uses the term 'Vorstellungen' (translated by Kemp Smith as "representations") in a wide sense to cover any mental state, with or without conceptual content, and uses 'Erkentnisse' for those which do literally represent something. So not all "represenations" represent! See the Critique of Pure Reason A320/B376, and Patricia Kitcher, Kant's Transcendental Psychology (New York: Oxford University Press, 1990) p.66. *20* The PLA defended here is a version of the third of three interpretations which Wright canvasses but rejects (op. cit., pp.219-222). Though admitting the promise of such an interpretation, Wright presents two obstacles to it. One is that it presupposes a general verificationism - but as we have seen, this is not the case: it needs only a restricted verifiability for attributions of meaningfulness. Wright's own PLA appeals to the latter too - he takes it that two sentences can have different content only if they convey different information, i.e. if there could be a situation in which there was reason to believe one but not the other (pp.230, 235). He thus presupposes that for some sentences at least, there can be no difference in meaning without an epistemological difference. His first objection would seem to be a case of pot, kettle, and an imputation of blackness. Wright's second objection was that claims of the form "I mean something in my use of expressions" or "I am following rules" may be in the same epistemological class as "framework" statements, such as: "There exist other minds", "The world has existed for more than five minutes" and "There are material bodies". All these seem to be beyond evidential support in that no imaginable course of future experience could either confirm or disconfirm them, and yet (we want to think) it is reasonable for each one of us to accept them. So can't there be rational acceptability without empirical grounds? But if what we have argued is correct, there are possible courses of experience which would disconfirm the assumption that one means something by a particular expression. (Similarly, of course, there are courses of experience which disconfirm specific claims about particular material objects, other minds, or the age of the world.) So even if nothing could rationally motivate suspicion that one meant nothing by any of one's expressions, some possible experiences (of endemic disagreement, or of recurring predictive failure) might give one reason to doubt whether one meant something by a given expression. If so, it seems reasonable to require the putative private linguist to be able rationally to quell such specific doubts, as (II) requires. *21* Cf. Wittgenstein's "Notes ..." p.263 in Jones; and Rhees' notes p.118. See Pears, op.cit., Chapter 14, for a careful discussion of this. *22* Travis, op. cit., pp. 342-3.