Karl R. Popper

First published in the Proceedings of the 11th International Congress of Philosophy, Vol. VII (1953): 101-107. Reprinted in Karl R. Popper, Conjectures and Refutations: The Growth of Scientific Knowledge (New York: Basic Books, 1962).
Transcribed into hypertext by Andrew Chrucky, Sept. 12, 1997.


This is a paper on the impossibility of a physicalistic causal theory of the human language.{1}

1.1 It is not a paper on linguistic analysis (the analysis of word-usages). For I completely reject the claim of certain language analysts that the source of philosophical difficulties is to be found in the misuse of language. No doubt some people talk nonsense, but I claim (a) that there does not exist a logical or language-analytical method of detecting philosophical nonsense (which, by the way, does not stop short of the ranks of logicians, language analysts and semanticists); (b) that the belief that such a method exists -- the belief more especially that philosophical nonsense can be unmasked as due to what Russell might have called 'type-mistakes' and what nowadays are sometimes called 'category-mistakes' -- is the aftermath of a philosophy of language which has since turned out to be baseless.

1.2 It is the result of Russell's early belief that a formula like 'x is an element of x' is (essentially or intrinsically) meaningless. We now know that this is not so. Although we can, indeed, construct a formalism F1 ('theory of types') in which the formula in question is 'not well-formed' or 'meaningless', we can construct another formalism (a type-free formalism) F2, in which the formula is 'well-formed' or 'meaningful'. The fact that a doubtful expression cannot be translated into a meaningful expression of a given F1 does not therefore establish that there exists no F2 such that the doubtful formula in question can be translated into a meaningful statement of F2. In other words, we are never able to say, in doubtful cases, that a certain formula, as used by some speaker, is 'meaningless' in any precise sense of this term; for somebody may invent a formalism such that the formula in question can be rendered by a well-formed formula of that formalism, to the satisfaction of the original speaker. The most one can say is, 'I do not see how such a formalism can be constructed'.

1.3 As for the body-mind problem, I wish to reject the following two different theses of the language analyst. (1) The problem can be solved by pointing out that there are two languages, a physical and a psychological language, but not two kinds of entities, bodies and minds. (2) The problem is due to a faulty way of talking about minds, i.e. it is due to talking as if mental states exist in addition to behaviour, while all that exists is behaviour of varying character, for example, intelligent and unintelligent behaviour.

1.31 I assert that (1), the two-language solution, is no longer tenable. It arose out of 'neutral monism', the view that physics and psychology are two ways of constructing theories, or languages, out of some neutral 'given' material, and that the statements of physics and of psychology are (abbreviated) statements about that given material, and therefore translatable into one another; that they are two ways of talking about the same facts. But the idea of a mutual translatability had to be given up long ago. With it, the two-language solution disappears. For if the two languages are not inter-translatable, then they deal with different kinds of facts. The relation between these kinds of facts constitutes our problem, which can therefore only be formulated by constructing one language in which we can speak about both kinds of facts.

1.32 Since (2) is so vague, we must ask: is there, or is there not, the station-master's belief that the train is leaving, in addition to his belief-like behaviour? Is there his intention to communicate a fact about the train to the signalman, in addition to his making the appropriate movements? Is there the signalman's understanding of the message in addition to his nderstanding-like behaviour? Is it possible that the signalman understood the message perfectly well but behaved (for some reason or other) as if he had misunderstood it?

1.321 If (as I think) the answer to these questions is 'yes', then the body-mind problem arises in approximately Cartesian form. If the answer 'no' is given, we are faced with a philosophical theory which may be called 'physicalism' or 'behaviourism'. If the questions are not answered but dismissed as 'meaningless'; if, more especially, we are told that to ask whether Peter has a toothache in addition to his toothache-like behaviour is meaningless because all that can be known about his toothache is known through observing his behaviour, then we are faced with the positivist's mistaken belief that a fact is (or is reducible to) the sum total of the evidence in its favour -- i.e. with the verifiability dogma of meaning. (Cf. 4.3, below, and my Logic of Scientific Discovery, 1959.)

1.4 An important assumption of what follows here is that the deterministic interpretation of physics, even of classical physics, is a misinterpretation, and that there are no 'scientific' reasons in favour of determinism. (Cf. my paper 'Indeterminism in Quantum Physics and in Classical Physics', Brit. Journ. Philos. of Science, 7, 1950.)


2. Karl Bühler appears to have been the first to propose, in 1918,{2} the doctrine of the three functions of language: (1) the expressive or symptomatic function; (2) the stimulative or signal function; (3) the descriptive function. To these I have added (4) the argumentative function, which can be distinguished{3} from function (3). It is not asserted that there are no other functions (such as prescriptive, advisory, etc.) but it is asserted that these four functions mentioned constitute a hierarchy, in the sense that each of the higher ones cannot be present without all those which are lower, while the lower ones may be present without the higher ones.

2.1 An argument, for example, serves as an expression in so far as it is an outward symptom of some internal state (whether physical or psychological is here irrelevant) of the organism. It is also a signal, since it may provoke a reply, or agreement. In so far as it is about something, and supports a view of some situation or state of affairs, it is descriptive. And lastly, there is its argumentative function, its giving reasons for holding this view, e.g. by pointing out difficulties or even inconsistencies in an alternative view.


3.1 The primary interest of science and philosophy lies in their descriptive and argumentative functions; the interest of behaviourism or physicalism, for example, can only lie in the cogency of their critical arguments.

3.2 Whether a person does in fact describe or argue, or whether he merely expresses or signals, depends on whether he speaks intentionally about something, or intentionally supports (or attacks) some view.

3.3 The linguistic behaviour of two persons (or of the same person at two different dates) may be indistinguishable; yet the one may, in fact, describe or argue, while the other may only express (and stimulate).

3.4 Any causal physicalistic theory of linguistic behaviour can only be a theory of the two lower functions of language.

3.5 Any such theory is therefore bound either to ignore the difference between the higher and lower functions, or to assert that the two higher functions are 'nothing but' special cases of the two lower functions.

3.6 This holds, more especially, for such philosophies as behaviourism, and the philosophies which try to rescue the causal completeness or self-sufficiency of the physical-world, such as epiphenomenalism, psycho-physical parallelism, the two-language solutions, physicalism, and materialism. (All these are self-defeating in so far as their arguments establish -- unintentionally, of course -- the non-existence of arguments.)


4.1 A wall-thermometer may be said not only to express its internal state, but also to signal, and even to describe. (A self-registering one does so even in writing.) Yet we do not attribute the responsibility for the description to it; we attribute it to its maker. Once we understand this situation, we see that it does not describe, any more than my pen does: like my pen it is only an instrument for describing. But it expresses its own state; and it signals.

4.2 The situation outlined in 4.1 is fundamentally the same for all physical machines, however complicated.

4.21 It may be objected that example 4.1 is too simple, and that by complicating the machine and the situation we may obtain true descriptive behaviour. Let us therefore consider more complex machines. By way of concession to my opponents, I shall even assume that machines can be constructed to any behaviouristic specification.

4.22 Consider a machine (invested with a lens, an analyser, and a speaking apparatus) which pronounces, whenever a physical body of medium size appears before its lens, the name of this body ('cat'; 'dog', etc.) or says, in some cases, 'I don't know'. Its behaviour can be made even more human-like (1) by making it do this not always, but only in response to a stimulus question, 'Can you tell me what this thing is?', etc.; (2) by making it in a percentage of cases reply, 'I am getting tired, let me alone for a while', etc. Other responses can be introduced, and varied -- perhaps according to inbuilt probabilities.

4.23 If the behaviour of such a machine becomes very much like that of a man, then we may mistakenly believe that the machine describes and argues; just as a man who does not know the working of a phonograph or radio may mistakenly think that it describes and argues. Yet an analysis of its mechanism teaches us that nothing of this kind happens. The radio does not argue, although it expresses and signals.

4.24 There is, in principle, no difference between a wall-thermometer and the 'observing' and 'describing' machine discussed. Even a man who is conditioned to react to appropriate stimuli with the sounds 'cat' and 'dog', without intention to describe or to name, does not describe, although he expresses and signals.

4.25 But let us assume that we find a physical machine whose mechanism we do not understand and whose behaviour is very human. We may then wonder whether it does not, perhaps, act intentionally, rather than mechanically (causally, or probabilistically), i.e. whether it does not have a mind after all; whether we should not be very careful to avoid causing it pain, etc. But once we realize completely how it is constructed, how it can be copied, who is responsible for its design, etc., no degree of complexity will make it different in kind from an automatic pilot, or a watch, or a wall-thermometer.

4.3 Objections to this view, and to the view 3.3, are usually based on the positivistic doctrine of the identity of empirically indistinguishable objects. Two clocks, the argument goes, may look alike, although the one works mechanically and the other electrically, but their difference can be discovered by observation. If no difference can be so discovered, then there simply is none. Reply: if we find two pound notes which are physically indistinguishable (even as to the number) we may have good reason to believe that one of them at least is forged; and a forged note does not become genuine because the forgery is perfect or because all historical traces of the act of forgery have disappeared.

4.4 Once we understand the causal behaviour of the machine, we realize that its behaviour is purely expressive or symptomatic. For amusement we may continue to ask the machine questions, but we shall not seriously argue with it -- unless we believe that it transmits the arguments, both from a person and back to a person

4.5 This, I think, solves the so-called problem of 'other minds'. If we talk to other people, and especially if we argue with them, then we assume (sometimes mistakenly) that they also argue: that they speak intentionally about things, seriously wishing to solve a problem, and not merely behaving as if they were doing so it has often been seen that language is a social affair and that solipsism, and doubts about the existence of other minds, become self-contradictory if formulated in a language. We can put this now more clearly. In arguing with other people (a thing which we have learnt from other people), for example about other minds, we cannot but attribute to them intentions, and this means, mental states. We do not argue with a thermometer.


5.1 But there are stronger reasons. Consider a machine which, every time it sees a ginger cat, says 'Mike'. It represents, we may be tempted to say, a causal model of naming, or of the name-relation.

5.2 But this causal model is deficient. We shall express this by saying that it is not (and cannot be) a causal realization of the name-relation. Our thesis is that a causal realization of the name-relation cannot exist.

5.21 We admit that the machine may be described as realizing what we may loosely call a 'causal chain'{4} of events joining Mike (the cat) with 'Mike' (its name). But there are reasons why we cannot accept this causal chain as a representation or realization of the relation between a thing and its name.

5.3 it is naive to look at this chain of events as beginning with the appearance of Mike and ending with the enunciation 'Mike'.

It 'begins' (if at all) with a state of the machine prior to the appearance of Mike, a state in which the machine is, as it were, ready to respond to the appearance of Mike. It 'ends' (if at all) not with the enunciation of a word, since there is a state following this. (All this is true of the corresponding human response, if causally considered.) It is our interpretation which makes Mike and 'Mike' the extremes (or terms) of the causal chain, and not the 'objective' physical situation. (Moreover, we might consider the whole process of reaction as name, or only the last letters of 'Mike', say, 'Ike'.) Thus, although those who know or understand the name-relation may choose to interpret a causal chain as a model of it, it is clear that the name-relation is not a causal relation, and cannot be realized by any causal model. (The same holds for all 'abstract', e.g. logical relations, even for the simplest one-one relation.)

5.4 The name-relation is therefore clearly not to be realized by, say, an association model, or a conditioned reflex model, of whatever complexity. It involves some kind of knowledge that 'Mike' is (by some convention) the name of the cat Mike, and some kind of intention to use it as a name.

5.5 Naming is by far the simplest case of a descriptive use of words. Since no causal realization of the name-relation is possible, no causal physical theory of the descriptive and argumentative functions of language is possible.


6.1 it is true that the presence of Mike in my environment may be one of the physical 'causes' of my saying, 'Here is Mike'. But if I say, 'Should this be your argument, then it is contradictory', because I have grasped or realized that it is so, then there was no physical 'cause' analogous to Mike; I do not need to hear or see your words in order to realize that a certain theory (it does not matter whose) is contradictory. The analogy is not to Mike, but rather to my realization that Mike is here. (This realization of mine may be causally, but not purely physically, connected with the physical presence of Mike.)

6.2 Logical relationships, such as consistency, do not belong to the physical world. They are abstractions (perhaps 'products of the mind). But my realization of an inconsistency may lead me to act, in the physical world, precisely as may my realization of the presence of Mike. Our mind may be said to be as capable of being swayed by logical (or mathematical, or, say, musical) relationships as by a physical presence.

6.3 There is no reason (except a mistaken physical determinism) why mental states and physical states should not interact. (The old argument that things so different could not interact was based on a theory of causation which has long been superseded.)

6.4 If we act through being influenced by the grasp of an abstract relationship, we initiate physical causal chains which have no sufficient physical causal antecedents. We are then 'first movers', or creators of a physical 'causal chain'.


The fear of obscurantism (or of being judged an obscurantist) has prevented most anti-obscurantists from saying such things as these. But this fear has produced, in the end, only obscurantism of another kind.

Go to Sellars' commentary


{1} This issue was first discussed by Karl Buhler in his Sprachtheorie, 1934, pp. 25-28. [Back]

{2} Referred to in his Sprachtheorie, loc. cit.[Back]

{3} Cf . "Towards a Rational Theory of Tradition", The Rationalist Annual, 1949. [Back]

{4} It does not matter for our present purposes whether or not the expression 'causal chain' is adequate for a more thorough analysis of causal relations. [Back]

Go to Sellars' commentary