Forthcoming in Philosophia, vol. 32 (2004). Available online with the Editor's permission.

Is Solitary Rule-Following Possible?

Jussi Haukioja Department of Philosophy University of Turku FIN-20014 Turku Finland jussi.haukioja@iki.fi ABSTRACT
The aim of this paper is to discover whether or not a solitary individual, a human being isolated from birth, could become a rule-follower. The argumentation against this possibility rests on the claim that such an isolate could not become aware of a normative standard, with which her actions could agree or disagree. As a consequence, theorists impressed by this argumentation adopt a view on which the normativity of rules arises from corrective practices in which agents engage in a community. However, it has been suggested that an isolated individual could engage in such a practice by herself. Three prospective examples of such cases are considered, and the possibility of solitary rule-following is vindicated. Furthermore, the nature of the goals at which rule-following practices generally aim is clarified.

1. Introduction

The aim of this paper is, obviously, to answer the question asked in the title. It will hardly come as a surprise that the issue is in some way related to what is known as the 'Private Language Argument'. Unfortunately, it is not always clear what is meant by a 'private language', which is why I want to avoid that terminology. I shall begin with an account of what I take 'solitary rule-following' to be, and also, what kind of possibility it is that is under consideration in this paper.

A 'private language' is often taken to be a language devised to refer to one's own sensations, or one that can only be understood by a single individual.1 The possibility of such a language is not what I will be concerned with here. Rather, I will ask whether it would be possible for 'a born Crusoe', a human being isolated from birth, to engage in an activity which deserves to be called rule-following. Moreover, I will not ask whether Crusoe could speak or create a language. Instead, I will ask the more general question of whether he could follow a rule. It is generally agreed that in order to speak a language, one must be able to follow rules. However, there may well be additional considerations having to do with linguistic meaning which make speaking a language a more demanding task than rule-following in general. I will not discuss the possible existence and nature of such complications in this paper. The arguments against solitary language are usually based on general considerations on rulefollowing, without any reliance on features specific to linguistic meaning,2 and my intention is to examine whether those arguments succeed in establishing the impossibility of solitary rulefollowing. Rules, as I understand the term, are normative constraints on actions, judgements, and so on. That is, they set apart a subset of possible actions, judgements, and so on, as correct or as otherwise more approprate. By rule-following, in turn, I mean courses of action which aim at satisfying such constraints. Thus, an action, or a series of actions, may be in accordance with a rule without being an instance of rule-following, because the agent may not be aiming at conformity. Examples of rules include such familiar cases as traffic rules, rules of chess, and the grammar of English. Still, having restricted the problem in this way, there are at least two different questions that might be posed by asking whether solitary rule-following is possible. One might be interested in whether a solitary individual could actually come to follow rules, that is, whether it is psychologically plausible to suppose that this might happen. This question is not my

concern in this paper. Rather, I will inquire whether there is any conceptual incoherence in the supposition that a solitary individual could follow a rule. If the answer to this question turns out to be negative, that will in itself say nothing about the psychological question (except that it can be meaningfully raised). Of course, should it turn out that the notion of a solitary rulefollower is conceptually incoherent, the psychological question would lose its interest.3 Unlike a large proportion of the literature on the 'Private Language Argument', this paper will not be concerned with the interpretation of Wittgenstein's writings. The exegetical literature is full of controversy. Not only is there disagreement as to what form the argument should take and whether various formulations of it are sound or not, there is also disagreement as to what the argument is supposed to prove, where it is to be found in Wittgenstein's text, and whether the target is a solitary rule-follower or a private linguist (as described above). My interest lies in whether rule-following conceptually requires the presence of a community of agents, and although it is true that the centrality of this question in current philosophy of mind and language is largely due to Wittgenstein's later work, I believe that our understanding of the matter is better improved by focusing on the question itself rather than on exegetical issues.

2. Normativity

Why would anyone think that a solitary individual cannot be a rule-follower? At first sight, it may seem quite obvious that a born Crusoe could invent a language to describe his surroundings. In fact, this supposition has seemed so obvious to some philosophers that they have simply assumed that Wittgenstein must have been denying the possibility of a private language, the terms of which refer to an individual's own sensations. (See, for example, Ayer 1971: 55-56.)

The reason why many philosophers claim solitary rule-following to be impossible has to do with the normativity of rules. Rules are, by definition, not descriptive but normative. They lay down criteria for which choices, courses of action, and so on, are correct instances of following the rule. But they do not tell us what actually happens: a course of action and a rule that governs it can come apart. That is, rules can be broken; an agent's actions may go against a rule when she is trying to remain faithful to the rule, and even when she sincerely believes that she is successfully doing so. In order to make sense of errors, a standard of correctness must exist against which the particular applications of a rule can be compared. Moreover, the rule-follower must be (at least potentially) aware of the existence of a standard, and of the fact that her actions may or may not be in agreement with it. After all, we often speak of standards of correctness existing for the action and performance of non-human animals, even machines - think, for example, of faulty computer software. What makes rule-followers different is precisely their (potential) awareness of the standard, the fact that they aim to 'get things right', rather than just perform the way they have been conditioned or designed to do. When we engage in rule-following, we leave open the possibility that our judgements (about the applicability of a rule) may be incorrect, and that we may have to reverse them later. Crucially, a rule-follower has to be capable of making (and becoming aware of) a specific kind of error. She can be wrong, not just about whether a given case falls under a given rule, but about what the requirement of the rule is. Dorit Bar-On (1992: 33-34) makes a useful distinction between three kinds of error that may occur in using linguistic terms: recognitional errors - having one's perceptual data distorted in one way or another, and using the wrong term as a consequence; performance errors - slips of the tongue; and meaning errors - thinking that a word correctly applies to certain entities, when it in fact does not. Although Bar-On's distinction is between different kinds of linguistic error, it should be

obvious that all three kinds are possible with rule-following in general. I will adopt the term rule-requirement error to refer to the generalisation of meaning errors.4 If one cannot commit a rule-requirement error, this can only mean that all possible actions are equally correct - but this is to say that no standard of correctness exists, and no rule is being followed. One cannot commit recognitional or performance errors if there is no such thing as a correct action, hence these kinds of errors presuppose the possibility of rule-requirement errors. The notion of rulefollowing presupposes that one can be wrong about what the rule requires, in addition to being wrong about the applicability of a rule because one is mistaken about the specifics of a particular case.

3. The Argument against Solitary Rule-Following

The argument against solitary rule-following maintains that Crusoe could not commit a rulerequirement error. Were we to suppose that Crusoe is actually following a rule, there would seem to be no problem in explaining recognitional or performance errors, but that would already be to suppose that he can commit rule-requirement errors. But how could Crusoe possibly become aware of his failure to conform to a standard if all he has to go on are his own inclinations? No imaginable course of events, the argument holds, could lead to this there is no way Crusoe could be wrong about which actions accord with rules and which do not. And if that is the case, the notion of a rule seems to cease to apply. It is often assumed that this is what Wittgenstein meant with his often quoted remark (1968: 258): 'One would like to say: whatever is going to seem right to me is right. And that only means that here we can't talk about "right".' The point can be illustrated by thinking of rule-following as the act of continuing a series. In applying a rule in new instances, we are judging whether the new case is similar to

the known instances in the relevant respect (perceptual or otherwise). The trouble is, (nearly) anything is similar to anything else in some respect - how do we determine what the relevant respect is? A finite series does not by itself determine a single correct way of continuing it accordingly, a finite number of past instances do not by themselves determine the correct course of action to take in a new instance. How could there then, for a solitary agent, be a standard of correctness potentially different from the actual choice that she makes? If the past instances do not determine which series it is that should be continued, how can there be a wrong way of continuing a series? It is important to note that the problem has nothing to do with the unreliability of human memory. The argument against solitary rule-following does not rest on the claim that, because an agent can remember her past sensations and judgements incorrectly, she can (re)interpret, in the absence of surveillance from other agents, her past actions so that they come out correct, that is, as corresponding to the standard of correctness laid down by the rule she was following. The problem is more fundamental: the resources available to a solitary agent do not, it is claimed, succeed in singling out a standard of correctness in the first place, so there never was a rule that the agent was following. The problem is constitutive rather than epistemological, although it is easy to forget this because the constutive question is typically pursued by idealising our epistemic access to our past sensations and judgements. As a consequence of normativity considerations, many theorists conclude that the normativity of rules can only be made sense of in a communal context. We are almost constantly engaged in practices in which we compare our judgements with those made by others in our community. We are usually not consciously doing this, and for the most part we agree in our judgements. Our engagement in the practices becomes obvious when we do not agree. When we find that some of us do not, say, apply a linguistic term to the same set of entities as the rest of us, we instinctively take this to be a sign that some users of the term are

mistaken, and that their actions stand in need of correction. At times an agent may also find out that her own usage of a term (or application of some other kind of rule) is incorrect, and proceed to take steps to adjust her usage to fit the rest of the community. This idea is at the heart of the so-called 'community view' of rule-following, according to which agreement with (the majority of) one's peers is all there is to correctness (see, for example Wright 1980, Wright 1981). In other words, correctness is identified with consensus. Unsurprisingly, a community view will deny the possibility of solitary rule-following - the judgements made by an isolated individual ('a community of one') would be trivially correct, and hence correctness and actual performance could not diverge. It is not at all clear that the problems surrounding rules are solved simply by bringing in the community. A pure community view seems to be unable to account for our intuitive notion of objectivity, regarding our judgements (and other instances of rule-following). Intuitively, it does not seem to be the case that correctness always coincides with the majority opinion, as far as rules are concerned. There should, intuitively, be room for the possibility of an entire community going astray, at least for a while. While it is clear that the community view can account for an individual's error, a temporary community-wide error seems to be ruled out.5 Hence, mere consensus cannot be all there is to correctness. However, a more plausible version of the community view might hold that, as far as rule-applications are concerned, consensus and correctness may come apart, but that they cannot come apart for rule-requirements. That is, one might hold that the community, as a whole, could not commit a rule-requirement error, while still maintaining that the whole community might, due to some community-wide disturbance, become temporarily prone to systematic recognitional or performance error.6 For the special case of word meaning, this would be to say that the community, as a whole, cannot be wrong about what their words

mean, but they might be collectively wrong, at least temporarily, about whether the word in fact correctly applies to a particular entity. This may well be the right thing to say. Indeed, it would be strange to claim that, for example, one actually can defend against check by castling in a game of chess, and that the rules actually used in chess are wrong. However, I feel that more needs to be said if we want to achieve a full understanding of what rule-following consists in. In particular, this revised community view still leaves it a mystery how humans predominantly agree in their judgements about the applicability of a rule in new instances. Our judgements about rulerequirements must also be directed and constrained by factors other than our aiming at consensus (which is not to say that they are not constrained by that aim, only that it cannot be the whole story). It seems that individual, but largely shared, tendencies to see certain entities, situations, and so on, as primitively similar to each other will also need to be assumed in the explanation of rule-following. This is not, however, the place to go into details of such a theory.7 What matters for the question at issue here is that if we accept a theory which sees corrective practices as essential to rule-following, the possibility of solitary rule-following is threatened.

4. 'Time-Slice' Practices?

Not everyone would agree with the general points about rule-following made in the previous section. However, those points are among the premises on which the argument against solitary rule-following is generally based - proponents of other accounts of rule-following usually have no quarrel with the possibility of solitary rule-followers. For my purposes in this paper, we may grant that, although work remains to be done, practices of correction and approval do have a vital part to play in the explanation of normativity. It is only through such

corrective practices that we can become aware of a standard of correctness with which our judgements can either agree or disagree. When we fail to follow a rule correctly, our failure (or, at least, someone's failure) generally becomes evident when we find that we disagree with others about the applicability of the rule. And, conversely, agreement signals success (at least prima facie). To get an answer to the question asked in the title of this paper, we must then ask: is it necessary for corrective practices to be communal, or could Crusoe, in principle, engage in such a practice all by himself? Is the presence of a community of individuals, on purely conceptual grounds, required for corrective practices to take place? Or, to put it in other terms, could not the 'time-slices' of a single individual form a 'community', the practices of which give rise to normativity? I will now take a look at three different examples of 'timeslice' practices given in the literature.

Example 1: Crusoe and Interior Decoration Baker and Hacker (1984: 38-42) use the following example in arguing that solitary rulefollowing is possible: ... to take a simple example, [Crusoe] might use the pattern - - - as a rule or pattern to follow in decorating the walls of his house; when he notices four dots in a sequence he manifests annoyance with himself. He goes back and rubs one out, and perhaps checks carefully adjacent marks, comparing them with his 'masterpattern'. And so on. Of course, he is not merely following his 'inclinations', but rather following the rule. And it is his behaviour, including his corrective behaviour, which shows that he is following the rule, and what he counts as following the rule. (Baker & Hacker 1984: 39) The same example, which is a slightly modified version of an example used by Wittgenstein (Wittgenstein 1978: 344) is used by Meredith Williams to argue that solitary rule-following is not possible (Williams 1991: 110-112).

We are to suppose that the master pattern lays down the rule which Crusoe is following. But can Crusoe commit a rule-requirement error - that is, can he be wrong about what the correct application of the master pattern is in a given instance? Certainly we would call it an error if he did four dots in a sequence. But it seems that if Crusoe, for whatever reasons, happens to prefer painting a sequence of four dots at a certain point, nothing will force him to reconsider his decision (even if he sometimes does so in the manner Baker & Hacker describe.) There is nothing in the master pattern that will, all by itself, force one continuation of the series of dots and lines rather than another. In taking the master pattern to have a single interpretation Baker & Hacker are already supposing Crusoe to be a rule-follower. Without this supposition, the example simply does not work. If we do not take the master pattern to have an interpretation, Crusoe will be, at any given point, free to take it to suggest anything he happens to prefer. 'Whatever seems right to him is right', and the Crusoe of this example is not a rule-follower. On the basis of this example, Williams concludes that Crusoe could not follow rules, since anything playing a role analogous to the master pattern would itself need an interpretation in order to be a normative constraint on Crusoe's actions. For reasons that will become clear later, I find this generalisation unwarranted. It will be helpful to contrast this case with other examples, so let us move on.

Example 2: Crusoe and the Rubik's Cube Simon Blackburn argues for the possibility of solitary rule-following with the following story: Consider the example (due to Michael Dummett) of a born Crusoe who finds a Rubik's cube washed onto his island, and learns to solve it. He certainly doesn't solve it randomly, for he can do it on demand. It is natural to say that he follows principles (when there is a last corner to do...). Perhaps he has some rudimentary diagrams or

10

other mnemonics which he consults. With these he can do it, and without them he cannot. (Blackburn 1984a: 297) Suppose that one morning our Crusoe interprets his mnemonics in a different way from his previous usage. Is it really the case that 'whatever seems right to him is right'? Blackburn claims that it is not, because should this happen, Crusoe would no longer be able to solve the cube. Hence, Blackburn concludes, we have to say he is a rule-follower (see also Blackburn 1984b: 84). Presented as above, I am not convinced that we would be fully justified in saying that the Crusoe of this example would have to be a rule-follower. There is no doubt that we, judging Crusoe's actions from the outside, may describe his actions as correct or incorrect with respect to the goal of solving the cube ('no no, that symbol means he should twist that part of the cube in that direction, now he won't solve it.'). But the distinction between correctness and incorrectness need not be available to him. Perhaps the mnemonics he uses have come to exist due to a highly improbable incident, and Crusoe is just killing time by twisting the cube in ways somehow suggested by the symbols - he need not be aware of a correct way of applying those symbols in his cube-twisting activity. Nevertheless, a slight modification of the example may suffice to remedy this. Instead of imagining that Crusoe can, in fact, solve the cube and that he has a full system of principles available to help him solve it, imagine that he is formulating such a set of principles for himself.8 Furthermore, in order not to presuppose an ability to have internal states with a determinate propositional content (that is, in order not to suppose that Crusoe is already a rule-follower), we have to suppose that he cannot use precise descriptions in the formulation. Rather, he will have to rely on rough perceptual similarities between different configurations of the cube ('when the cube looks like that, do this manouvre', etc.). These similarities will have to be seen as primitive, in the sense that they are not similarities with respect to a

11

scheme of classification consciously applied by Crusoe, but rather similarities with respect to his instinctive dispositions to see entities, configurations, and so on, as similar to each other otherwise we would already be presupposing him to be a rule-follower. In such a situation, Crusoe will have to 'continue the series', that is, judge new configurations of the cube as either falling under one configuration-type or another (and similarly for the manoeuvres that he performs on the cube). A crucial point is that he must proceed by trial and error. Sometimes his classifications will result in his not being able to solve the cube on that attempt (or, perhaps more realistically, not achieving a subgoal such as solving one face of the cube). When that happens, he will have to revise his earlier classifications ('a configuration like that does not require a manouvre like this, after all'). As far as I can see, in such a situation we must take Crusoe to be a genuine rule-follower. (Of course, it is doubtful whether an isolated human could in fact manage to do this, but there is nothing conceptually incoherent in the supposition). His 'later selves' will sometimes find reason to think that his earlier categorisations have to be revised, thus the distinction between correctness and incorrectness will be available to him. In order to achieve a full system of instructions, his different 'time-slices' must seek consensus about which configurations count as relevantly similar to each other - relevantly with respect to the goal of solving the cube. What about the points made against the previous example? It may be objected, in parallel fashion, that a Rubik's cube does not, in itself, contain instructions about what should be done with it - outside a practice in which there is agreement on what counts as a solved cube, it is merely a multi-coloured object with moving parts. Are we not, then, illegitimately, assuming that Crusoe already has a representation with a conceptual content (namely, a representation of a solved cube)? Without such an assumption, will it not be the case that whatever configuration of the cube Crusoe ends up with, could be counted as a goal?

12

This objection misses an important difference between the two examples. Although we are assuming that Crusoe aims at (what we would call) solving the cube, we are not assuming that he is following a rule we would put as 'a cube should be manouvred so that each of the six faces has a uniform colour'. The rules we are assuming him to follow concern the particular manouvres that should be performed on the cube in its various possible configurations. Let me clarify the point by making a distinction between primitive and nonprimitive goals. Many of our goals are concept-involving or non-primitive in the sense that, in order for our success or failure at reaching the goal to have an effect on our subsequent behaviour, we need to use concepts - we need to classify (the relevant features of) the situation we find ourselves in. In the rule-following context this means, roughly, that the need we may feel to engage in a corrective practice derives from our judgement that the goal has not been reached. Such a judgement can, of course be incorrect - this is as it should be, since the goal itself is concept-involving, and thus requires rule-following. Primitive goals, in contrast, are goals which we have and aim at just by virtue of being the kind of organisms we are. Like other animals, we have primitive goals such as avoidance of physical pain, avoidance of hunger, and so on. Furthermore, like other animals, our failure to reach these goals can have an effect on our subsequent behaviour, even if no judgement is made that the goal has not been reached. In this example, we need to assume that the ultimate goal of Crusoe's practice, namely, solving the cube, is primitive. In other words, we need to assume that he simply finds a scrambled cube so annoying that he is inclined to take the trouble of developing a solving technique. Then, given that this is the sort of creature Crusoe is, he will need to develop a technique for solving the cube, and in this technique there will be a distinction between correct and incorrect manouvres. A similar move is not available in our first example, because

13

Crusoe's activity did not serve any further purpose which could be viewed as a primitive goal. We could, in that example, take Crusoe's preference for the repetition of three lines and three dots as primitive, but then the master pattern would no longer have a role to play. The assumption that Crusoe has the sort of primitive goal required in this example is, to say the least, psychologically implausible, but there is nothing conceptually incoherent about it, as far as I can tell. In the third and final example, the presumed ultimate goal of Crusoe's practice is much more naturally seen as primitive.

Example 3: Crusoe and Berries Bar-On (1992: 35-39) considers the example of a born Crusoe (or Super-Crusoe, as she calls him) who lives on an island with various kinds of berries available to him, some edible and some not. However, Crusoe has trouble telling them apart prior to eating them. (Perhaps edibles and inedibles look, smell and taste the same, or perhaps there are simply too many kinds of berries for his memory to handle.) To make life easier for himself, he begins marking bushes with coloured strings - one colour (say, blue) for those which produce edible berries and another (say, red) for those which do not. Bar-On wants to ask whether solitary language is possible. She concludes that this example is not a case of solitary language because, in the course of this practice, Crusoe would have no use for the Gricean intentions sometimes required for (non-natural) meaning. That is, Crusoe need not form the intention of inducing with a coloured string, in his future self, a belief via his later recognition of his (original) intention to induce that belief. As noted above, the denial of a solitary language is usually based on argumentation against solitary rule-following. Taken as an argument against the possibility of solitary rule-following, Bar-On's reasoning is obviously not successful: Gricean intentions already presuppose that the agent be capable of having propositional attitudes (namely, the intentions themselves). In other words, they presuppose that the agent be a rule-

14

follower. Of course, one might still maintain that Bar-On's argument presents a further problem for the possibility of a solitary language. However, one can hardly claim that Grice's account of linguistic meaning in terms of speaker's meaning (and, ultimately, the speaker's intentions) is universally accepted, and thus the argument will not trouble proponents of other accounts. As noted in section 2, in order for Crusoe to be a rule-follower it should be possible for him to commit a rule-requirement error. In the present context, it should be possible for him to be wrong about which groups of bushes red and blue strings are correctly attached to. It is obvious that Crusoe can commit what we would perceive as errors - he might, for example, attach a blue string on a bush with bad berries. But we perceive this as an error because we are already assuming that red strings, and not blue strings, are correctly attached to bad bushes. The question is, could Crusoe be mistaken about what red and blue strings 'mean', in his usage? It might seem that he could not. After all, he could extend his usage of coloured strings in any way he pleased, and no one would be there to correct him. We can imagine Crusoe to attach a blue string to a bush with bad berries, later eating berries from that bush, and becoming sick - something we would definitely call a recognitional or performance error, but only because we naturally assume that blue strings 'mean' edible berries. We can even imagine Crusoe going on to remove the blue string and replacing it with a red one. We would naturally call this behaviour corrective. Yet it is not clear that this is enough for Crusoe to count as a rule-follower, because it does not seem that he has revised his understanding of what the coloured strings mean - he has only revised his opinion about this particular bush. Consider, however, the following story. Imagine that Crusoe has been applying the coloured strings for a while, and it has become a routine matter. The technique is working

15

well for him, and he forgets how awful he feels when he eats more than a couple of the bad berries (suppose a couple is enough for Crusoe to recognise the symptoms). Suppose, furthermore, that he is a bit absent-minded and has an eye for colour combinations. Some of the bushes (both edible and inedible) have such colour that red strings look absolutely hideous when attached to them. The combination disturbs Crusoe so much that he cannot help himself - he attaches blue strings on these bushes, regardless of the berries they produce. When going around eating the berries, however, he picks berries from all bushes marked with blue strings, and as a result he becomes sick. The displeasure of becoming sick outweighs the displeasure of having to look at awful colour combinations, so Crusoe returns to marking all bushes with inedible berries with red strings, regardless of their appearance. This story would require some ingenuity from Crusoe, but I do not believe it requires him already to be a rule-follower. Let us take a closer look at what is happening here. At t0, before Crusoe starts to pay too much attention to colour combinations, he is aiming to attach red strings to all and only the inedibles (and, possibly apart from the occasional mistake, he is succeeding in this). At t1, Crusoe is aiming to attach red strings to bushes which produce inedibles and are not, in his opinion, unfavourably coloured.9 At t2, Crusoe eats berries from bushes marked with blue, and gets sick. Finally, at t3, he is again aiming to attach red strings to all and only the inedibles. As far as I can tell, the only way to make sense of this story is to say that Crusoe has committed, and corrected, a rule-requirement error. Between t0 and t1, his understanding of which bushes are correctly marked with red has changed, and between t2 and t3 he recognises this change as an error, and corrects himself. It is important to note that we are actually assuming Crusoe to be following two rules, which we (unlike Crusoe!) might verbally characterise roughly as follows: (R1) bad berries are to be marked with a red string

16

(R2)

berries marked with a red string are not to be eaten

This reflects the fact that Crusoe is, so to say, both the speaker and the hearer, or the producer and the consumer of symbols. At t0, t1 and t3, he is the producer (following R1), at t2 he is the consumer (following R2). Of course, his different time-slices cannot directly compare their understandings of what a red string 'means'. However, his different time-slices have a common goal in using the coloured strings, and the goal is such that failure to communicate properly across his different time-slices can result in a failure to achieve the goal. Getting sick, Crusoe fails to achieve the goal of his string-attaching activity. In our story, the root of the failure is not in a misperception of what the berries are like, or anything of the sort. Rather, he fails because his different time-slices have had different understandings about the correctness conditions for applying the coloured strings.10 The situation is analogous to cases of erroneous understanding of words in a community. Our differences become apparent only when we fail to achieve non-linguistic individual or collective goals because of failures of communication - because of different understandings between speakers and hearers. As long as Crusoe's activities accord with one of the rules, R1 or R2, he can commit (and become aware of committing) a rule-requirement error about the other. In our example, he is only eating berries from bushes not marked with red strings - that is, his actions accord with R2, and his failure to accord with R1 becomes evident when he becomes sick. The converse might also happen: were he in accord with R1, his failure to follow R2 would result in his becoming sick.11 Clearly, Crusoe can commit rule-requirement errors with respect to R1 and R2, and we can safely say that he is a rule-follower. It is misleading to take the crucial question to be that of whether or not Crusoe can be wrong about the fact that a red string 'means' inedible berries. In a sense, Crusoe could extend

17

his practice any way he pleased. He could start marking some inedibles with blue strings, and still reach his goal of not becoming sick by refraining from eating berries from these bushes by some other means. In this situation, a blue string would no longer 'mean' edibles. The situation is analogous to meaning change in language - both Crusoe-the-producer and Crusoethe-consumer have changed their understanding of which bushes the strings are correctly applied to. Hence, Crusoe would no longer be following the rules we would put as R1 and R2, but rather a different pair, R1* and R2*. But note that there would still be a possibility of Crusoe's committing rule-requirement errors with respect to R1* and R2*. The question is not whether the correctness conditions for Crusoe's practice are the same as those for rules we would find most natural, but whether such correctness conditions exist for Crusoe's stringattaching activity, in a way which he can be aware of. The crucial question is, can he have an erroneous understanding of what the coloured strings are supposed to be attached to. And in our story, Crusoe clearly can.12 Since Crusoe could, in the sense described, extend his practice as a whole in any way he pleases, we will have to acknowledge that there are no correctness conditions for his practice as a whole. True, it will either succeed or fail as a strategy of avoiding bad berries, but this is conceptually a wholly different matter. However, given that he is using one strategy, correctness conditions will come to exist for the rules within that practice, in the manner described above. Note that this is paralleled on the communal level - there is no sense in which, say, the grammar of English language, or the set of rules for chess, as a whole, are correct or incorrect, yet correctness conditions exist for particular rules within the grammar or the rules of chess. Furthermore, one can meaningfully raise the question of how efficient the English language is for some purpose, or whether the game of chess fulfils its purpose. One might still worry that we are implicitly presupposing Crusoe to be a rule-follower in setting up the example.13 That is, one might claim that in taking Crusoe to start marking

18

berries with coloured strings, we would already be taking him to be capable of seeing himself as a classifying agent - we would already assume him to have some notion of what it is to have symbols. A related worry could be raised about the previous example. We should be careful not to demand so much of Crusoe that a parallel argument would succeed in showing that it would be impossible for any agent, solitary or not, to become a rule-follower. Think about the communal case. Children learn symbols initially by having them presented to them in conjunction with their referents. After this stage their own usage is controlled by adults. Prior to this process, children are not able to follow rules, yet at some point they do become rule-followers. (When this happens, precisely, I do not know - I am not even sure there is a precise answer, since the process is likely to be gradual.) I claim that the process could be mirrored in the solitary case, at least with suitable modifications of the example. In the berry example, we might imagine that some good berries already are marked with blue strings, and some bad berries with red strings, for whatever reason. If the berries are otherwise indistinguishable (except for their effects on Crusoe after he has eaten some), surely Crusoe could start taking red strings as a warning sign - not in the sense of understanding a red string to be a symbol, but in the sense of coming to be indisposed to eat berries from bushes with red strings. This, I imagine, is something many non-human animals could learn to do. Crusoe would form a pre-conceptual connection between red strings and inedible berries. This would not yet be for him to use strings as symbols. However, we can imagine various possible (though improbable) events which might lead Crusoe towards actively using strings to mark berries. For example, he might see a string become untied in the wind. I do not see why we could not imagine Crusoe taking the string and retying it to the bush, without presupposing him to be a rule-follower. Similar contingencies could gradually lead Crusoe to apply strings to previously 'unclassified' bushes, after which the road would be open for him to engage in a corrective practice, as explained

19

above. In other words, we can enrich the examples so that Crusoe's natural surroundings play a role analogous to that of a 'teacher' or 'trainer', in whatever senses needed. The events we need to imagine might be highly improbable, but I see no reason to deny their possibility. In any case, the burden of proof is on those who claim that this cannot be done. If there is an argument to be found against solitary rule-following, based on a claim that natural surroundings could not play this role, this would be a new argument, not directly related to the usual grounds for denying Crusoe to be following rules, discussed here.

5. Conclusions

On the basis of the three examples considered above, we can draw some general conclusions. In order for a course of action to count as rule-following, it must serve some further end, where a failure to reach this end will be dispose the agent to reconsider her earlier actions. In the first example, the success or failure of Crusoe's painting dots and lines was directly connected to his own feelings of satisfaction or dissatisfaction (as noted, the existence of a master pattern does not make a difference, since we cannot presuppose it to have a definite interpretation). Thus, the appropriateness of Crusoe's actions (in the sense of satisfying his own aesthetic preferences) is evident straight away, and there will be no need to reconsider his earlier actions. Things are different in the other two examples. Crusoe must classify situations and entities (that is, configurations of the cube, and particular bushes), but his success or failure will only become evident later on. For this reason, Crusoe must leave open the possibility that he will have to reverse his initial classification. In other words, he must be aware of the possibility of a rule-requirement error, which is precisely what is required for him to be a rule-follower.

20

In the above explanation of solitary rule-following, I have assumed that the solitary agent has (reasonably) stable dispositions to see certain groups of entities (etc.) as primitively similar to each other. Without this assumption, we could not make sense of Crusoe treating, say, a certain configuration of the Rubik's cube, or a particular red string, as similar to others he has encountered earlier. Obviously, a similar assumption has to be made in the case of communal rule-following as well. Moreover, these dispositions have to be assumed to be stable, not only across times, but across persons as well. On this basis, Stephen Davies (1988) suggests that we can think of isolated individuals as having membership in the 'human community'. He reasons that a born Crusoe could be a rule-follower, provided that he is seen as belonging in 'the community of mature human beings' (Davies 1988: 61-64). While I certainly agree with Davies' general point about rule-following, I feel that this stretches the limits of the word 'community'. Surely it would be highly misleading to call this a 'community view' of rule-following. Furthermore, what is important is not the existence of others with the same nature (same patterns of response) but the (successful) pursuit of intertemporal agreement - this is the case in solitary as well as communal rule-following. If our Crusoe is a true isolate, never having had contact with other humans, how could the existence of others of the same nature have any bearing on whether he can become aware of a normative standard for his behaviour? We may not recognise Crusoe to be a rule-follower unless we (largely) share his nature - because we may not recognise his goals - but this is beside the point! Kripke famously concedes that we can think of a born Crusoe as following rules, but he claims that in doing so, we are 'taking him into our community and applying our criteria for rule following to him' (Kripke 1982: 110). Moreover, 'the falsity of the private model need not mean that a physically isolated individual cannot be said to follow rules; rather that an individual, considered in isolation [...] cannot be said to do so' (ibid.). These passages have puzzled many commentators. In my opinion, the crucial point is whether or not Crusoe

21

himself can aim at conforming to a normative standard - it makes no difference whether or not we would be inclined to ascribe correctness or incorrectness to his actions. However, perhaps Kripke's comment could be read in the following way. When we consider an individual in physical and temporal isolation, there is no sense in calling the agent a rule-follower. A single time-slice of an individual may have all the dispositional properties which enable the agent to engage in a corrective practice (socially or individually), but to be justified in calling that agent a rule-follower we must know whether or not she in fact does take part in such a practice. In the examples, we have tried to show how an agent could become a rule-follower, that is, we have assumed that Crusoe has not been a rule-follower prior to the events described in the examples. For this reason, we needed to assume that the goals pursued by Crusoe are primitive. That is, they are not goals which Crusoe pursues because he has formed a conception of how things should be. Rather, he pursues them because he is the kind of creature he is. There is no doubt that the majority of goals that we pursue, in a community, are non-primitive - we want something to be the case and we take actions that we believe to be necessary for our desire to be fulfilled. Nonetheless, I believe there is a more general conclusion to be made here. Our having these non-primitive goals will, in turn, require rulefollowing on our part, and hence they will have to be seen as being aimed at some further goal. Presumably this hierarchy of goals has a number of levels, but ultimately we will have to assume ourselves to have primitive goals in the communal case as well. These are likely to include such goals as happiness, avoidance of suffering, and so on. It may well be the case that some of our primitive goals are, in fact, social in nature - many of our non-primitive goals certainly are. The relevant rule-following practices will then, of course, be inherently social. But this is a fact about these practices, not about rule-following in general. Crusoe certainly

22

could not follow many of the rules we in fact follow, but he could, nonetheless, come to follow rules of his own.14

References
Ayer, A.J. (1971), Could Language Be Invented By a Robinson Crusoe?, in Jones (ed.) The Private Language Argument. London: Macmillan. Baker, G.P., and Hacker, P.M.S. (1984), Scepticism, Rules and Language. Oxford: Blackwell. Bar-On, D. (1992), On the Possibility of a Solitary Language, Nos 26: 27-46. Blackburn, S. (1984a), The Individual Strikes Back, Synthese 58: 281-301. Blackburn, S. (1984b), Spreading the Word. Oxford: Clarendon Press. Davies, S. (1988), Kripke, Crusoe and Wittgenstein, Australasian Journal of Philosophy 66: 52-66. Esfeld, M. (2001): Holism in the Philosophy of Mind and Philosophy of Physics. Dordrecht: Kluwer. Haukioja, J. (forthcoming): Hindriks on Rule-Following, Philosophical Studies. Hindriks, F. (forthcoming): A Modest Solution to the Problem of Rule-Following, Philosophical Studies. Kripke, S. (1982), Wittgenstein on Rules and Private Language. Oxford: Basil Blackwell. McDowell, J. (1984), Wittgenstein on Following a Rule, Synthese 58: 325-63. Pettit, P. (1990), The Reality of Rule-Following, Mind 99: 1-21. Pettit, P. (1996), The Common Mind: An Essay on Psychology, Society, and Politics (2nd edition). New York: Oxford University Press. Williams, M. (1991), Blind Obedience: Rules, Community and the Individual, in Puhl (ed.) Meaning Scepticism. Berlin: de Gruyter. Wittgenstein, L. (1968), Philosophical Investigations (3rd edition). Oxford: Basil Blackwell. Wittgenstein, L. (1978), Remarks on the Foundations of Mathematics (Revised edition). Cambridge: MIT Press. Wright, C. (1980), Wittgenstein on the Foundations of Mathematics . Cambridge: Harvard University Press. Wright, C. (1981), Rule-Following, Objectivity and the Theory of Meaning, in Holtzman and Leich (eds.) Wittgenstein: To Follow a Rule. London: Routledge & Kegan Paul. Wright, C. (1986), Does Philosophical Investigations I. 258-60 Suggest a Cogent Argument Against Private Language?, in Pettit and McDowell (eds.) Subject, Thought, and Context. Oxford: Clarendon Press.

23

1 See Wright (1986) for a detailed account of the distinction between private language and solitary language. 2 A notable exception is Bar-On (1992). I will discuss her argumentation below. 3 Bar-On (1992: 28-29) distinguishes three different senses of possibility in this context: psychological, conceptual and logical. She is interested in the question of whether solitary language is conceptually possible, that is, whether we can 'make sense of [Crusoe's] coming to employ signs with particular meanings.' She takes this to be distinct from the question of logical possibility, that is, the question of whether there is a contradiction in the idea of a solitary individual becoming a language speaker. About the latter question, Bar-On simply says that she assumes that there is no contradiction. As for the former, she assumes that an affirmative answer would have to 'come in the form of a plausible [...] story outlining a route [Crusoe] might take to language'. I am not convinced that there really are two distinct questions here. In any case, the usual way of discussing solitary language and rule-following lumps the two together: the denial of solitary language is invariably based on an alleged contradiction in the notion of a solitary rulefollower, while the claim that solitary rule-following is possible is generally accompanied by a fictitious story. If the former is successful, Bar-On's question of conceptual possibility is also answered negatively. If the latter is successful, we have established what Bar-On calls the logical possibility. Hence, if either side of the traditional dispute succeeds, we need not worry about Bar-On's distinction. 4 Thus, a rule-follower can, in general, commit recognitional, performance and rule-requirement errors, where the last category includes meaning errors, but also errors pertaining to rules where it would be quite inappropriate to speak of meanings. In a game of chess one might, for instance, mistake one's pawn for a bishop and move it along a diagonal (recognitional error), intend to make a legal move but accidentally place one's piece on the wrong square (performance error), or, say, think one can defend against check by castling (rule-requirement error). 5 John McDowell (1984: 359n10) brings out this point with his 'semi-paraphrase' of Wittgenstein's 258, quoted earlier: 'One would like to say: whatever is going to seem right to us is right. And that only means that here we can't talk about "right".' 6 In other words, this would be to claim that on the community level, unlike on the individual level, the possibility of rule-requirement errors is not necessary for the possibility of recognitional or performance errors. This may seem problematic, but in reality it is not. A community-wide recognitional error would, in fact, be the sum of individual recognitional errors, and these are made possible by the fact that each individual could, on her own, commit a rulerequirement error. To claim that a community-wide rule-requirement error is impossible would preclude the possibility of the majority of individuals committing a simultaneous rule-requirement error, but there is nothing objectionable in this, as far as I can see. 7 See Pettit (1990, 1996) for a general explanation of rule-following which combines individual dispositions with communal and/or solitary corrective practices. Hindriks (forthcoming) presents a series of objections to Pettit's view; see Haukioja (forthcoming) for an extension of Pettit's theory, intended to avoid these. 8 In Blackburn 1984b, the example is of 'a born Crusoe who over the years evolves a technique for solving a Rubik's cube washed onto his island' (p. 84). This slight difference in presentation makes a crucial difference. 9 Of course, these are our descriptions, not Crusoe's - we are not assuming him to be operating with an explicit definition. 10 Thus, his failure is not just a failure to follow what Bar-On (1992: 32-33) calls a teleological norm - he has not merely failed to follow a (natural or acquired) habit in order to achieve his goals. He has failed in a particular way - he has failed to follow the rule governing the correct use of coloured strings. The correctness conditions themselves are somewhat arbitrary (just as word meanings are), and not dictated by the goal of not getting sick. But the success of his whole string-using activity depends on his time-slices having a shared understanding of these correctness conditions. And it is here that he has failed. 11 It may seem as if we are already assuming Crusoe to be a rule-follower here, but that is not the case, for two separate reasons. Firstly, it is quite enough that Crusoe's actions conform to the other rule - we do not have to suppose that Crusoe is aiming at conformity with that rule. Secondly, it does not seem implausible at all to say that Crusoe starts following both rules simultaneously when he begins marking bushes with coloured strings. 12 Esfeld (2001, Ch. 3.2.2) insists that an asymmetry exists between the communal and solitary cases, because: 'the present disposition of the person always has a privileged position; for the past self cannot reply and give feedback.' This is, of course, true. But this is equally true of the 'past selves' of other members of one's community! If I, in a communal

24

setting, form an honest judgement about a particular case, surely I believe that I am correct in my judgement. Nonetheless, I leave open the possibility that later events (which may or may not involve others in my community) may lead me to reverse my judgement. This very same possibility can be left open in the solitary case. This act of 'leaving open' indicates that the agent is involved in a corrective practice, and hence counts as a rule-follower. Note that this applies both to judgements about rule-application and to judgements about rule-requirements. 13 I am grateful to Philip Pettit for stressing this point. 14 Most of this paper was written while I was a Visiting Fellow at the Research School of Social Sciences (RSSS), Australian National University, in 2001. I am grateful to the RSSS for having me as a visitor, and to the Alfred Kordelin Foundation, the Academy of Finland (projects 51319 and 52379), and the Turku University Foundation for making the visit financially possible. I also want to thank the following people for helpful comments and discussion: Michael Esfeld, Alex Miller, Philip Pettit, Michael Ridge, Juha Rikk, and two anonymous referees for this journal.

25