Gilbert Harman (born 1938) is an American philosopher, teaching at Princeton University since 1963, who has published widely in philosophy of language, cognitive science, philosophy of mind, ethics, moral psychology, epistemology, statistical learning theory, and metaphysics.
Three Levels of Meaning.
Journal of philosophy, 65, 1968.
Philosophers approach the theory of meaning in at least three different ways.
First, Carnap, Ayer, Lewis, Firth, Hempel, Sellars, Quine, etc. take meaning to be connected with evidence and inference, a function of the place an expression has in one's 'conceptual scheme' or of its role in some inferential 'language game'. Second, Morris, Stevenson, Grice, Katz, etc. take meaning to be a matter of the idea, thought, feeling, or emotion that an expression can be used to communicate. Third, Wittgenstein (?), Austin, Hare, Nowell-Smith, Searle, Alston, etc. take meaning to have something to do with the speech acts the expression can be used to perform. (A fourth approach, which emphasizes truth conditions, is discussed in several of the following essays, particularly Essay 11.) 8.1 Familiar Objections to Each Type of Theory.
8.1.1 Meaning as Conceptual Role
Theories of the first sort, which take meaning to be specified by inferential and observational evidential considerations, are sometimes accused of ignoring the social aspect of language. Such theories, it is said, admit the possibility of a private language in which one might express thoughts without being able to communicate them to another; and this possibility is held to be absurd. More generally, it has been argued that even if meaning depends on considerations of evidential connection, the relevant notion of evidence involves intersubjective objectivity, which requires the possibility of communication among several people. Therefore it is argued that one could not account for meaning via the notion of evidence without also discussion of meaning in communication. Furthermore, there are many uses of language to which the notion of evidence has no application. If one asks a question or gives an order, it is not appropriate to look for the evidence for what has been said. But if there end p.155 can be no evidence for a question, in the way that there can be evidence for a conclusion, differences in meaning of different questions cannot be explicated by means of differences in what evidence can be relevant to such questions. So theories of the first sort seem vulnerable in several respects.
8.1.2 Meaning as Communicated Thought
On the other hand, theories of the second sort are threatened by circularity from at least two directions. According to Katz (1966b), one understands the words someone else says by decoding them into the corresponding thought or idea. But a person ordinarily thinks in words, often the same words he communicates with and the same words others use when they communicate with him. Surely the words mean the same thing when used in these different ways? But to apply Katz's account of meaning to the words one thinks with would seem at best to take us in a circle. Similarly, consider Grice's theory of meaning. According to Grice (1957, 1969), one means that P by one's words (in communication) if and only if one uses them with the intention of getting one's listener to think one thinks that P. But what is it to think that P? On one plausible view it is to think certain words (or some other representations) by which one means that P. If so, Grice's analysis would seem to be circular: one means that P by one's words if and only if one uses them with the intention of getting one's listener to think one has done something by which one means that P. Circularity and worse also threaten from another side, if the second type of approach is intended to explain what it is to promise to do something or if it is supposed to be adequate to exhibit the difference between asking someone to do something and telling him to do it, etc. The fact that saying something in a particular context constitutes one or another speech act cannot be represented simply as the speaker's communicating certain thoughts. For example, promising to do something is not simply communicating that you intend to do it, nor is asking (or telling) someone to do something simply a matter of communicating your desire that he do it. At the very least, to perform one or another speech act, one must communicate that one is intending to be performing that act; so, at the very least, to treat all speech acts as cases of communication would involve the same sort of circularity already mentioned. Furthermore, communication of one's intention to be performing a given speech act is not in general sufficient for success. The speaker may not be in a position to promise or to tell someone to do something, no matter what his intentions and desires.
8.1.3 Meaning as Speech-Act Potential.
Finally, theories of the third sort, which treat meaning as speech-act potential, are also subject to familiar objections. For example, Chomsky (1964, 1966), following Humboldt, argues that this third approach (and probably the second as well) ignores the 'creative aspect of language use'. Language exists primarily for the free expression of thought. Communication and other social uses of language are, according to Chomsky, of only secondary importance. One of the most important characteristics of human language is its unbounded character. Almost anything that one says has never been said by anyone before. Surely this unboundedness reflects the unbounded creative character of thought and is not simply a reflection of the more or less practical uses to which language can be put in a social context. Furthermore, approaches of the third sort seem to be at least as afflicted with circularity as are approaches of the second sort. For example, Alston (1964) suggests defining sameness of meaning as sameness of illocutionary-act potential, where illocutionary acts are the relevant subclass of speech acts. He claims that two expressions have the same meaning if and only if they can be used to perform the same illocutionary acts. Now, suppose we ask whether the expressions 'water' and 'H 2 O' have the same meaning. They do only if, for example, in saying 'Please pass the water' one performs the same illocutionary act as one does in saying 'Please pass the H 2 O'. But it can be argued that we are able to decide whether these acts are the same only by first deciding whether the expressions 'water' and 'H 2 0' have the same meaning. If so, Alston's proposal is circular.
8.2 Three Levels of Meaning.
Each of the preceding objections is based on the assumption that the three approaches to the theory of meaning are approaches to the same thing. I suggest that this assumption is false. Theories of meaning may attempt to do any of three different things. One theory might attempt to explain what it is for a thought to be a thought of a certain sort with a given content. Another might attempt to explain what it takes to communicate certain information. A third might offer an account of speech acts. As theories of language, the first would offer an account of the use of language (or other representations) in thinking; the second, an account of the use of language in communication; the third, an account of the use of language in certain institutions, rituals, or practices of a group of speakers. I shall refer to theories of meaning of level 1, of level 2, and of level 3, respectively. I believe that there is a sense in which later levels presuppose earlier ones. Thus a theory of level 2, i.e. a theory of communication (of thoughts), presupposes a theory of level 1 concerning the nature of thoughts. Similarly, a theory of level 3 (e.g. an account of promising) must almost always presuppose a theory of level 2 (since in promising one must communicate what it is one has promised to do). The objections I have just discussed show only that a theory of one level does not provide a good theory of another level. A theory of the nature and content of thoughts does not provide a good account of communication. A theory of meaning in communication does not provide a good account of speech acts. And so forth. On the other hand, I do not want to deny that proponents of the various theories have occasionally been confused about their objectives. In the third section of this essay I shall argue that such confusion has led to mistakes in all three types of theory. But first, from the point of view of the suggested distinctions between such levels of meaning, I shall briefly review the three approaches to the theory of meaning sketched at the beginning of this paper.
8.2.1 Level 1
A theory of level 1 attempts to explain what it is to think that P, what it is to believe that P, to desire that P, etc. Let us suppose we are concerned only with thinking done in language. Such a supposition will not affect the argument so long as thinking makes use of some system of representation, whether or not the system is properly part of any natural language. Even if we do not know what the various expressions of a subject's language mean, we can still describe him as having thoughts, beliefs, desires, and other psychological states. It seems reasonable to assume that the subject has the thought that P if and only if he thinks certain words or other representations that have on this occasion the meaning or content that P; he believes that P if and only if his belief involves a sentence or other representation that has the content that P; and similarly for desires and other psychological attitudes. The problem of saying what it is to think, believe, desire, etc. that P can be reduced to the problem of saying what it is for certain words or other representations used in thinking to have the content that P. Another way to put the same point is this. A theory of the nature of thought, belief, desire, and other psychological attitudes can appear in the guise of the theory of meaning or content. That is the best way to interpret . the first sort of theory discussed at the beginning of this paper. Extreme positivists claim that the content of a thought, i.e. what thought it is, is determined by its conditions of verification and refutation. Its meaning or content is determined by the observational conditions under which the subject would acquire the corresponding belief plus those conditions under which he would acquire the corresponding disbelief. Other empiricists argue that what a thought is or means is determined by its position in a whole structure of thoughts and other psychological attitudes, i.e. its place in a subject's conceptual scheme, including not only relations to experience but also relations to other things in that same scheme. Several philosophers have argued a similar thesis that makes no explicit reference to meaning or content. Fodor (1965), Putnam (1960), and Scriven (1966) have each taken psychological states to be 'functional states' of the human organism. What is important about such states is not how they are realized; for my psychological states may well be realized in a different neurophysiological way from yours. What is important is that there is a certain relationship among the various states a person can be in, between such states and observational 'input', and between such states and action 'output'. In this regard persons are sometimes compared with nondeterministic automata (Miller, Galanter, and Pribram, 1960). Just as a particular program or flow chart may be instantiated by various automata made from quite different materials, so too the 'same' person (a person with the same psychological characteristics and dispositions) might be instantiated by different neurophysiological set-ups and perhaps even by some robot made of semiconductors, printed circuits, etc. For a person to be in a particular psychological state is like the automaton's being at a certain point in its program or flow chart rather than like something's happening at one or another transistor. If we conceive the automaton's operation to consist largely in the formation, transfer, and 'storage' of certain representations, the analogy is even better. To say that such an automaton is at a certain point in a particular program is to say, first, that the automaton has various possible states related to one another and to input and output in such a way that it instantiates a particular program and, second, that it is in a particular one of the states or collections so indicated. For the automaton in question, the same point can be made by first specifying the role of various representations it uses in its internal operation, its reaction to input, and its influence on output. Second, one may describe the present state of the automaton by indicating what representations are where (Harman, 1967a, 1968).
It is obvious how such an account may offer a functional account of psychological states via a person's use of language. Thus, according to Sellars (1963), the meanings of one's words are determined by the role of the words in an evidenceinference- action game, which includes the influence of observation on thought, the influence of thought on thought in inference, and the influence of thought on action via decision and intention. Sellars offers a functional account of psychological states in the guise of a theory of meaning. (Sellars, 1974, explicitly agrees.) It is important that the analogy be with nondeterministic rather than deterministic automata. According to Sellars, the meaning of an expression is given by its role in the evidence-inference-action game, where this role is not causal but rather defined in terms of possible (i.e., more or less legitimate) moves that can be made. A similar point would have to be accepted by anyone who would identify psychological states with functional states (Harman, 1973: 51-3). Quine's thesis of indeterminacy, discussed above in Essay 5, says that, functionally defined, the content of a thought is not uniquely determined. The thesis ought to be expressible directly as the following claim about instantiations of nondeterministic automata: when a set of possible states of some device can be interpreted in a particular way as instantiations of a given nondeterministic automaton, that interpretation will not in general be the only way to interpret those physical states as instantiations of the given automaton. I hope I have said enough to show how theories of the first sort may be treated as theories about the nature of the contents of thoughts and other psychological ('intentional') states.
8.2.2 Level 2.
A theory of level 2 attempts to say what communication is and what is involved in a message's having a particular meaning. Communication is communication of thoughts and ideas; and Katz's description of it (1966b) is perfectly acceptable provided that his talk about 'decoding' is not taken too literally. It is true that Katz's description of communication would have us explain meaning in terms of meaning; but the two sorts of meaning are different. Katz would have us explain the meaning of a message in terms of the meaning or content of a thought, which is to explain meaning of level 2 in terms of meaning of level 1. And there is nothing wrong with that. Similarly, Grice's theory of meaning avoids the charge of circularity by explaining the meaning of a message (what the speaker means) in terms of the content of the thought communicated (which the speaker intends the hearer to think the speaker has). Communication need not involve use of language. When it does, the language used need not be one either speaker or hearer is able to think in. And even whenthe language used is one both participants think in, it may (for the purposes of certain communications) be used arbitrarily as a code. But ordinary communication makes use of a language which both participants think in and which is not being used arbitrarily as a code. In such a case the hearer typically assigns, as his interpretation of what the speaker says, either (a) a thought that the hearer expresses using the same words the speaker has used (with possible minor modification, e.g. for first- and second-person in pronouns) or (b) a thought that is some simple function of a thought in those words, where the function is determined by context (e.g. irony). Similarly, the speaker standardly uses in communication (almost) the same words he uses in expressing to himself the thought he intends to communicate. This is no accident, and one will fail to understand the nature of linguistic communication unless one grasps this point. It is obscured when linguistic communication is described as if it involved processes of coding and decoding. We would not be able to use language in communication as we do if communication really involved coding and decoding. Similarly, it would be a mistake to treat learning one's first language as simply a matter of learning how to communicate one's thoughts to others and how to understand others when they attempt to communicate. When a child is exposed to language he acquires two things. First he acquires a new system of representation for use in thinking and in the formation of various psychological attitudes. This is the primary thing he acquires. Second he acquires the ability, alluded to above, to communicate with and understand other speakers of the language. This ability relies heavily on the fact that the language has been acquired as an instrument of thought. No very complicated principles of interpretation need to be learned to support this ability. All the child needs to do, at first, is to assume that other speakers express by their words thoughts the child would think using those same words. More complicated principles of interpretation are learned later to allow for lying, irony, metaphor, etc. But it would surely be a mistake to think of the child as having an ability to perform a certain sort of complicated decoding. Aside from that point, I hope it is now clear how, for example, Grice's theory may be treated as a promising attempt at a level 2 theory of meaning; and I hope it is clear why it should not be criticized for failing to do what can be done only by a theory of meaning of another level.
8.2.3 Level 3
A theory of level 3 would be a theory of social institutions, games, practices, etc. The theory would explain how the existence of such things can make certain acts possible, for example, how the existence of a game of football can make possible scoring a touchdown or how the existence of an institution of banking can make possible writing a cheque. In a sense such a theory is a theory of meaning. The game or institution confers meaning on an act like carrying a ball to a certain place or writing one's name on a piece of paper. Some institutions, games, practices, etc. involve the use of language and can therefore confer meaning (significance) on such uses of language. But this is a different sort of meaning from that involved in levels 1 and 2. And, typically, use of certain words within an institution, practice, or game presupposes that the words have meaning as a message (which standardly presupposes that they have meaning when used to express one's thoughts). Despite the priority of levels 1 and 2, meaning on those levels can sometimes presuppose meaning on level 3, but only because one can think and communicate about practices, games, and institutions.