Linguistic Behaviour
Jonathan Bennett
The following is the ending section of Chapter I, "Programme for a Synthesis," from Bennett's book, Linguistic Behaviour (Cambridge University Press, 1976): 24-35. Appearing here by permission of Jonathan Bennett. (AC)§8. Thought and language
Any analysis leaves unanalysed materials on the right-hand side. Similarly, for my programme of analysis-by-synthesis I must help myself to an initial stock of concepts, without explaining them through anything more primitive. It is always a problem to know where to start. There is a line of conceptual dependence from 'language' to 'meaning' to 'intention' and 'belief', to 'organism', to 'physical object', and philosophical problems abound all the way down. But surely we can explain something without explaining its entire conceptual ancestry!
However, I shall not start with 'intention' and 'belief', because I want to lay foundations for those two concepts, showing how intentions and beliefs can bc attributed to creatures which lack a language. If they could not, I should be in trouble. My meaning-nominalist programme requires me to identify cases of meaning which are untouched by convention, and are thus not linguistic, so that I can move upwards from meaning to language through a stage where conventional meaning appears as the clean overlap of two species, convention and meaning. But if intentions and beliefs were prerogatives of creatures with language, then any case of meaning would involve an agent who possessed a language even if he was not using it just then; and that would spoil the bold picture which I hope to draw -- though my programme does not absolutely require it -- of a situation where we are fully entitled to say that U means something by x although we do not yet know whether U has a language.
So the plan is broadly speaking to start with thought and move from that through meaning to language. Here is a recent argument for the conclusion that the plan is a bad one:
If thought is a representational system analogous to public language, then it cannot be appealed to to explain how representational systems succeed in representing a world. Whatever the merits of agent-semantics as a component in an account of public linguistic performances, the analysis of representation must be conducted at a level undercutting the distinction between the overt and the covert, between public language and thought.{30}This reflects the author's assumption that there is a unitary notion of representation: he wants to discover how it is possible 'for language to represent the world'.{31} It seems also to assume that the difference between thought and language is just that between covert and overt representation. Since I reject that second assumption, I am free to inquire into what else is involved in the difference between thought and language, undeterred by threats about 'the ultimate sterility of agent-semantics and its intentionalist under-pinnings'.{32}
The above quotation occurs in a book which acknowledges the influence of Wilfrid Sellars. Sellars's own work is also marked by a refusal simply to start with thought and move on from it to language. A typical remark is that 'an uttering of p which is a primary expression of a belief that-p is not merely an expression of a thinking that-p but is itself a thinking, i.e., a thinking-out-loud that-p.{33} Sellars also warns of dangers which he sees inherent in 'approach[ing] language in terms of the paradigm of action',{34} that is in thinking of language as essentially a kind of (outer) behaviour which is in a certain way expressive of (inner) thought. I am unsure of my grasp of this, but Sellars's warnings seem to rest on the dubious assumption that his opponents will agree with him that 'thought is analogous to linguistic activity' in some strong though unexplained way.
The crucial difference between Sellars's approach and my Gricean one, however, seems to stem from the fact that his philosophy of language gives an absolutely fundamental place to the concept of a rule. I do not dispute that language is rule-governed (though I would rather say that it is conventional); but I think that Grice has shown us how to connect language with thought through a mediating concept -- meaning -- which does not essentially involve any such concept as that of a rule or convention; and so I am not even slightly tempted to think that the rules governing language must be intimately related to rules governing thought. This leaves me genuinely unable to bring Sellars's warnings to bear upon anything I am doing, though I realize that I have done scant justice to what is evidently a rich and subtle body of philosophical doctrine.
The position I am adopting, in opposition to the Sellarsian one, has been well expressed by Armstrong:
Once thought has been linguistically expressed the expressions can be perceived and react back upon the mind, eventually creating more complex and sophisticated thought which in turn can be given linguistic expression. And so speech gives birth to thought. It seems also to be at least a fact of nature, and may even be a logical necessity, that certain thoughts can be manifested in linguistic behavior only. But both these propositions are compatible with the 'distinct existence' in every case of mental states and the linguistic behavior in which they may be manifested.{35}In conformity with this, and following Grice, I shall start with thought -- specifically with intentions and beliefs -- and shall move from there to language through the concept of meaning. In my next three chapters I shall try to show positively and in detail how non-linguistic behaviour can support the concepts of intention and belief. First, though, I want to rebut arguments for saying that this is impossible.
§9. Thought without language?
Various writers have suggested analyses of 'x believes that P' in terms of 'x has such and such an attitude to S' where S is a sentence. That seems to tie belief to language but really it does not. Quine at any rate says that the proposed analysis 'is not, of course, intended to suggest that the subject of the [belief] speaks the language of the [sentence], or any language'.{36} Other proponents of that sort of approach, such as Carnap, appear not to have wondered what sorts of creatures can have beliefs. Rather they have considered the normal human situation where there are both beliefs and languages, and have offered theories about the relations between the two. I shall not discuss these theories.
Quine is indeed on record as thinking that there is something inescapably vague and impressionistic about any attribution of beliefs and intentions to lower animals -- he says that such attributions use an 'essentially dramatic idiom', which has no firm criterial foundations and depends upon our imaginatively putting ourselves in the animal's position and then using our (actual) language to express the beliefs we imagine we would have if thus placed.{37} In so far as this merely reflects pessimism about the chances of finding solid non-linguistic behavioural foundations for the concept of belief, the right response is just to build the foundations - which is what I shall attempt in the ensuing chapters. The other source of Quine's position involves skepticism about the legitimacy of 'He believes that . . .' even where 'he' does have a language. The reasons for that will be discussed in §78 below.
Bernard Williams has argued for a close link, of a kind, between belief and language. He presents a creature -- call it C -- some of whose belief-like epistemic states are manifested in assertion-like performances, with this peculiarity that C cannot 'assert' that P unless it 'believes' that P. That is, C cannot 'lie'. On the strength of this description of C, Williams contends (i) that C's performances are not genuine assertions, and (ii) that the states they manifest are not genuine beliefs.{38} I deny that he is entitled to claim (ii). He infers both conclusions from a single premiss, which he describes both as stating 'an extremely important fact about assertion' and as presenting 'a feature of belief'; and he words it both as 'Assertion can be insincere' and as 'The assertion of p is neither a necessary nor . . . a sufficient condition of having the belief that p'.{39} This, when construed in a certain way, does entail that it is not the case that C's performances are assertions which manifest C's beliefs; whence it follows that either (i) or (ii) is right. It is evidently implausible to deny (i) while maintaining (ii), that is, to say that C's performances are genuine assertions but that what they manifest are not genuine beliefs. We therefore seem to be driven by Williams's premiss to agree with him in accepting (i) that C's performances are not genuine assertions. But that leaves us with no basis whatsoever for also contending (ii) that the states manifested by those performances are nor genuine beliefs The premiss relating belief to assertion is simply irrelevant to the beliefs of a creature which, like C, does not assert.
Of any such creature, Williams says without qualification that 'Its states would not be beliefs', which suggests a very strong view about belief in relation to assertion and thus to language. I gather, though, that he wanted to deny the status of belief only to those states of C which are manifested in assertion-like performances, leaving open the possibility that C may have beliefs which it manifests in other ways. In fact, Williams attributes to languageless creatures only a concept of 'something rather like belief', or of 'belief' in an 'impoverished' sense; but his reasons for these cautions have to do not with the creature C but rather with other difficulties which I shall try to grapple with in chapter 4 below.{40}
The current situation regarding intentions and language is peculiar. Many philosophers would agree with this: 'Intention appears to be something that we can express but which brutes . . . can have, though lacking any distinct expression of intention'.{41} In conversation, however, one often meets the rival view that intentions need language; yet I can find almost no direct defenses of this in the literature.
One line of argument, which I have heard but never read, exploits a fact about intention which can easily seem to require intenders to have a language. Suppose that someone acts with a certain intention, and achieves what he intended to. Suppose further that what he achieves is getting food, emptying a rabbit-burrow, and contributing to the starvation of a bystanding buzzard. Although he achieved what he intended to. and achieved those three things, he may not have intended to achieve those three things -- e.g. he may have intended to get food without intending to empty the burrow or even knowing about the buzzard. This is true of intentions generally, it seems. One must always be prepared to say 'He intended to produce G but nor to produce H', even when producing G inevitably would involve producing H. For instance, he intended to drink that water but not to poison himself; and yet to drink that water would inevitably be to poison himself. Now if we express this point in the form 'One always intends to achieve something under a certain description', we may be tempted to infer that an intender must have a language.
But that is all wrong. The intender need not select descriptions: we do that when we describe his intention. Admittedly, we have to select because the intender has, as it were, selected aspects or features of the state of affairs resulting from his action; and I must explain what this feature-selection is. Of various prima facie workable accounts of the matter, I shall choose one using the concept of belief. We can be entitled, in speaking of a languageless creature, to distinguish 'He believes P' from 'He believes Q' even in cases where P and Q stand or fall together. So we can distinguish 'He intended to achieve G' from 'He intended to achieve H', even if achieving G must involve achieving H, thus: the agent did A intending to achieve G because his doing A is partly explained by his believing that this would lead to G; and he did not intend to achieve H because either he did not believe that doing A would lead to H or else his having this belief does nor help to explain his doing A. That, of course, is a mere promissory note: the details will be given in the following chapters. For the present, what matters is just that we are not compelled to restrict intentions to creatures with language just because intentions involve feature-selection.
The only extended written defence I know for the thesis that intentions require language is by Hampshire, who writes:
It is the possibility of our declaring . . . our intentions . . . that gives sense to the notion of intention itself. Without this possibility, the notion of intention becomes empty . . . it is characteristic of an intention that it may be formed long in advance of the action intended, and also that an intention may have existed without ever having been translated into action. It is senseless at speak of what a dog intended to do before it was interrupted or prevented or changed its mind, unless 'He intended to do so-and-so next' is certainly taken to mean the same as 'He would have done so-and-so next, if he had nor been prevented'. To say of a person 'He intended to do so and-so next' is certainly not equivalent to the statement that he would have done so-and-so if he had not been prevented . . . One might believe that someone had seriously and sincerely intended to do something, and at the same time be very doubtful whether he would in fact have done it, or even have tried to do it, if and when the occasion for action occurred.{42}Granted, the content of 'He intended to do A, but did not do it' is not equivalent to 'He would have done A if he had not been prevented'; but that simplistic equation fails for dogs as well as people, for reasons that have nothing to do with language. Hampshire simply asserts that it is 'senseless' to attribute an unfulfilled intention to a dog unless the attribution merely means a simple conditional of the form 'It would have done so-and-so if it had not been prevented'. Why should a dog's lacking a language debar it from being the subject of complex truths about its behavioural dispositions? And why should not these give the behavioural content of truths about the dog's unfulfilled intentions?
Hampshire's point cannot be that some attributions of (unfulfilled) intentions have no behavioural implications at all. Such an attribution, since it implies nothing about behaviour, could not be supported by any facts about behaviour; so we, in our present state of knowledge, could have no evidence for it; so it would be idle. Of course, a statement attributing to someone an intention at a certain time might be based on his earlier behaviour yet imply nothing about the future; but that goes for dogs as well as for people.
I think that Hampshire's point must be that someone who had an unfulfilled intention can say that he had it, whereas a dog cannot. Indeed, that means for revealing intentions is open only to creatures with languages, but what of it? Forget the fact that we do not ordinarily accept someone's reports on his unfulfilled intentions unless they have some backing in his non-linguistic behaviour; take the line between the linguistic and the non-linguistic revelation of intention to be as sharp, deep and absolute as you like; still, where does it lead? A little later Hampshire says: 'Intentions are something that may be concealed and disguised; but they can be concealed and disguised only because they naturally express themselves immediately either in words or in actions.' That is ambiguous, but it serves Hampshire's turn only if it implies that the concealment of intentions requires a language. Hampshire offers no defence of that, and it is certainly not self-evident. On the face of it, a language seems to offer an extra way of revealing intentions rather than indispensable means of concealing and disguising them.
Hampshire argues that intentions require language because it is of the essence of the concept of intention that 'an intention may have existed without ever having been translated into action', or that 'intentions . . . may be concealed and disguised'. The argument is not valid, I have contended; and now I question whether the premiss is even true.
The concept of intention has 'action uses', when it helps to explain behaviour in which the agent is said to act on an intention; and it has 'non-action uses', in which one attributes intentions which are concealed, disguised, thwarted or dropped. It is the non-action cases which Hampshire says require language; to which he must add the claim that the concept of intention should be applied only to agents who admit of both uses of the concept. There are two possible bases for this last claim.
(1) The action uses of the concept of intention somehow require or involve the non-action uses.
(2) The action uses can function without the others but the word 'intention' is improper except in describing a creature whose behaviour sometimes brings the non-action uses into play.
Hampshire offers no hint of an argument for (1), and I genuinely think that his basis is the relatively superficial and verbal claim (2). One pointer to this is the following remark, about a certain interpretation of the behaviour of a languageless creature: 'Here it might seem that the intention behind the activity is being stated, because at least the point and purpose of the activity are stated. But the more intellectual word "intention", since it is associated with the possibility of a declaration of intention, is out of place in the context of animal behaviour.' If Hampshire's basis really is just (2), a view about the meaning of 'intention', then I could harmlessly concede that his premiss is true and that his conclusion follows from it. At worst, that would imply that in my programme -- which needs only the action uses -- I ought not to use the word 'intention'. Instead of 'intention' then, I shall use a different word spelled in the same way. If Hampshire's premiss were about conceptual structures, this feint would be unavailing. But if his point is merely about the propriety of a word, my evasion is legitimate: verbal troubles admit of verbal remedies.
In Hampshire's book there is a hint of a different argument which he has developed for me in personal correspondence. He argues that intentions must sometimes be directed towards particular times in the non-immediate future; so an intender must be able to structure his future, e.g. to distinguish tomorrow from the day after; and Hampshire contends that this requires a language. Although his case for that last bit is not watertight, the whole line of thought is an interesting one which deserves to be developed further. But I need not let it block my present programme, which has no need for those uses of the concept of intention which involve ambitions, long-term contingency-plans and the like. Hampshire agrees, in his letter, that there is a sense of 'intention' in which 'the word can be made to march with "belief", and in step', that is a sense in which 'intentions' can be attributed to languageless creatures just so long as they have beliefs; and that somewhat attenuated sense is all I need. Hampshire would rather express it by 'purpose' than by 'intention', but I find it convenient -- and, in the context of my present work, harmless -- to retain my habit of using 'intention' for both sorts of case.
§10. Putnam' deceivers
I shall seek sufficient conditions in non-linguistic behaviour for the attribution of intention (or purpose) and belief. Those conditions are to be weak enough to cover all the intentions and beliefs that I shall need to support meaning, convention and language; but they need not be weak enough to be necessary for the application of 'intention' and 'belief'. My endeavour will thus be different from a reductive analysis of the concepts of intention and belief.
It differs in another way as well, because I am not looking for conditions which are logically sufficient for the attribution of intentions and beliefs, merely ones which are sufficient by normal, reasonable, everyday standards. The plan is to describe kinds of non-linguistic behaviour which would be best explained by supposing the agent to have certain intentions and beliefs. But that leaves one always vulnerable to the overlooked possibility, the equally good or even better explanation which does not involve intention and belief; and that is one reason why I do not claim to be adducing logically sufficient grounds. All I need to show is how those two concepts can have an active, coherent, working life in application to languageless creatures.
My programme therefore need not be embarrassed by the claim, which I indeed accept, that no statements about physical behaviour can logically entail any statements about minds. But Hilary Putnam has supported this claim with a particular argument which threatens not just that strong form of logical behaviourism but also my thesis that there is a weaker sufficiency relation -- one of the form 'P is sufficient by normal standards to entitle one to say that Q' -- between premisses about behaviour and conclusions about minds.{43} So although I am not defending logical behaviourism, I need to examine Putnam's attack on it.
Putnam describes creatures whose behaviour blocks the inference from behavioural premisses to mental conclusions because, he says, they have a 'pathological preference function'. That is, they assign a 'relatively infinite weight' to some kind of situation, which means that there is something they will do everything possible to achieve (or to avoid) in all circumstances. I would not label as 'pathological' the preference function of someone who would do anything to avoid being tortured, say; but that is a quibble. More important is the fact that someone's having a pathological preference function need not prevent his behaviour from manifesting facts about his mind. His behaviour could show us what he gave infinite weight to, and then we could work out his other preferences by observing his behaviour in circumstances where the 'pathological' preference was irrelevant. Someone who would do anything to avoid being tortured, or to become famous, may still be discovered to like vodka, swimming and Shakespeare, and to dislike cabbage, poker and Brahms.
In fact, Putnam's examples work only because his creatures assign relatively infinite weight not just to something but specifically to deceiving observers about their states of mind. He concedes that logical behaviourism 'constitutes a kind of "near miss" ',{44} but the miss is much nearer than he implies. Where he concedes that behaviourism may hold for creatures which do not have a pathological preference function, his argument lets it hold for any creatures which lack the extremely special pathological preference function which I have described.
Indeed, the miss is nearer still. We could even learn about the mind of someone who attached relatively infinite weight to deceiving us, for we might be able to observe his behaviour without his suspecting that we were doing so. If Putnam is to prevent this in the case of his supposed creatures, he must somehow guarantee them against ever being sure, wrongly, that they are unobserved. So in addition to their overwhelming distaste for being understood, he must endow them with either (a) perfect counter-intelligence skills or (b) total suspiciousness.
One might mink that (b) is a manageable supposal -- a mere matter of endowing psychopaths with a further pathological feature. But really it is not clear that such creatures are even logically possible. What does it mean to say that someone wants warmth, say, if there are no circumstances under which he would seek warmth with the sort of persistence that would show that he wants it? 'His wanting warmth consists in the fact that he would pursue it if he ever became sure that he was unobserved.' But that presupposes that he does always think that he may be under observation; and it is unclear what that alleged belief amounts to, given that it does not relate in any normal way to his sensory intake. Consider an agent who loves rum but never drinks it because whenever he sees rum he mistakenly thinks it is bouillon. What does it mean to credit him with that preference and that belief, rather than another pair (e.g. that he always recognizes rum, but hates to drink it) which would yield the same behaviour? If there is to be real content in the claim that his behaviour is explained by one particular preference-belief pair, the claim must support predictions about the conditions under which the behaviour would change -- such as that he would drink rum if he were given a chance to distinguish it from bouillon by its smell. But Putnam's creatures, on the version of them I am now examining, are said to have a belief which is not answerable to evidence. It is a blanket, programmatic, essentially psychotic or philosophical scepticism about chances of not being observed. There are no environmental conditions under which this 'belief' would predictably change; so there is no content to the claim that they have that belief, and thus no content, after all, to any claim about their preferences.
That argument shows my behaviourist bias, no doubt, but it also offers a legitimate challenge. It is fair to suggest what when readers think they understand Putnam's example, it is because they think of his creatures' wants as showing up in behaviour-patterns which are suspended only on rare occasions when the creatures think they are being observed. That shifts the story from (b) pathological suspicion back to (a) all-conquering ingenuity in finding out when they are being observed.
Now, Putnam's position is that from premisses about the behaviour of some creature we can never move by sheer logic to conclusions about its mind; because the creature may be of a peculiar kind for which no such move would be valid. But the kind has turned out to be more peculiar than at first appeared: for what we must allow for is just that our creature may have an all-conquering dislike of being understood and an infallible talent for knowing when it is being observed, even if we have no evidence for either supposition about it.
Logical behaviourism's 'near miss', if that is its measure, is as good as most hits. Certainly, there is nothing here to discomfit the claim that statements about behaviour can be sufficient, by normal evidential standards, to establish statements about minds. Nor does Putnam say that there is; but what he does offer is further from refuting the claim than his own treatment of it suggests.
In a helpful personal response to an earlier version of this section, Putnam argues like this: (a) The relationship between behaviour and psychological state depends upon facts about 'the actual functional organization that the creature has' -- its scale of preferences, system of beliefs, and so on. (b) These facts are a posteriori, and cannot be established by logic alone; and so (c) Even if we have the premiss that a given creature is not of the peculiar kind discussed above, nothing logically follows of the form, 'The creature will behave thus and so if and only if it is in psychological state P'. I fully concede (a) and (b), but deny that (c) follows. For (c) might be false, not because the facts about 'actual functional organization' are a priori but rather because they too could be established on the basis of behaviour. That, if it could be done, would let us use a creatures behaviour to establish its 'actual functional organization', and then to use that, and further facts about the behaviour, to establish all the creature's detailed psychological states.
Putnam points out that any such programme must grapple with the fact that any given bit of behaviour can be explained in terms of various different preference-belief pairs. I shall take this up in §15 below, where I shall try to show how one can select, from the various preference-belief pairs which fit a given single item of behaviour, the one pair which best fits the creature's behaviour-patterns as a whole. Perhaps this still won't license a logical move from behavioural premisses to mentalistic conclusions; but then logical moves are not what I am after.
[Return to Main Page]
Notes {30} Jay Rosenberg, Linguistic Representation (Dordrecht, 1974), p. 28. [Back]
{31} Ibid., p. 95. [Back]
{32} Ibid., p. 28. [Back]
{33} Wilfrid Sellars, 'Language as Thought and as Communication', reprinted in Essays in Philosophy and its History (Dordrecht, 1974). p. 107. [Back]
{34} Ibid., p. 98. For more on Sellars's basic approach to these matters see his 'Empiricism and the Philosophy of Mind' and the ensuing correspondence with Chisholm, in A. Marras (ed.), Intentionality, Mind, and Language (Urbana, Ill., 1972). [Back]
{35} D. M. Armstrong, "Meaning and Communication," Philosophical Review 80 (1971), pp. 427-8. The 'mentalism' of my present work is compatible with that defended in Jerrold J. Katz, 'Mentalism in Linguistics', Language, vol. 40 (1964), although our overall theories differ in method, aim and content. [Back]
{36} W. V. Quine, 'Quantifiers and Propositional Attitudes', in The Ways of Paradox (New York, 1966), p. 192. [Back]
{37} Quine, Word and Object (New York and London, 1960), p. 219. [Back]
{38} Bernard Williams, "Deciding to Believe," in Problems of the Self (Cambridge, 1973), p. 145. As well as the argument I am now criticizing, this paper contains much that is true, interesting, and relevant to the present work. [Back]
{39} Williams, p. 140. [Back]
{40} Ibid., pp. 139-40. On the whole matter of beliefs of languageless creatures I agree with the position and the argument of D. M. Armstrong, Belief, Truth and Knowledge (Cambridge, 1973), ch. 3. [Back]
{41} G. E. M. Anscombe, Intention (Oxford, 1957), p. 5; see also p. 85. [Back]
{42} Stuart Hampshire, Thought and Action (London, 1959), pp. 97-8. The remaining quotations in this section are from Ibid., pp. 97 and 99. [Back]
{43} Hilary Putnam, 'The Mental Life of Machines' and 'Brains and Behaviour', both in Mind, Language and Reality (Cambridge, 1975). [Back]
{44} Ibid., p. 422. [Back]
[Return to Main Page]